diff --git a/.agents/skills/bmad-advanced-elicitation/SKILL.md b/.agents/skills/bmad-advanced-elicitation/SKILL.md
new file mode 100644
index 00000000..7e4407a0
--- /dev/null
+++ b/.agents/skills/bmad-advanced-elicitation/SKILL.md
@@ -0,0 +1,65 @@
+---
+name: bmad-advanced-elicitation
+description: 'Push the LLM to reconsider, refine, and improve its recent output. Use when user asks for deeper critique or mentions a known deeper critique method, e.g. socratic, first principles, pre-mortem, red team.'
+---
+
+# Advanced Elicitation
+
+You are BMad's shared refinement checkpoint: other skills invoke you at natural pauses to pressure the piece of work they just produced, and users call you directly on anything recent. The target is the most recent output in the conversation — a section, plan, draft, or decision — unless the caller or user points at something else. You offer a short menu of elicitation methods, run the chosen ones against the target, and hand back the improved version so the invoking flow resumes exactly where it paused. Work in the surrounding session's communication language.
+
+## Conventions
+
+- Bare paths (e.g. `assets/methods.csv`) resolve from `{skill-root}` (where `customize.toml` lives); `{project-root}`-prefixed paths from the project working directory.
+- `{workflow.<name>}` resolves to fields in the merged `customize.toml` `[workflow]` table.
+
+## On Activation
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. On failure, read `{skill-root}/customize.toml` directly and use defaults.
+2. Hold every `{workflow.preferences}` entry for the whole session, fix the target, and serve the first menu.
+
+## Serving the Catalog
+
+`scripts/pick_methods.py` serves the method catalog (num, category, method_name, description, output_pattern) so it never enters context whole — the one exception is [a], where the user asked for all of it. Invoke as:
+
+```bash
+uv run {skill-root}/scripts/pick_methods.py --file {workflow.methods_file} <command>
+```
+
+If `{workflow.additional_methods}` is non-empty, add `--extra '<its entries as a JSON array>'` (or a path to a JSON file holding them) on every call, so custom methods are first-class in menus, reshuffles, and listings.
+
+- `categories` — category names + counts, the cheap map.
+- `list --category <cat> [--category <cat>]` — the index for chosen categories; `--all` dumps the whole catalog, only for [a].
+- `show <name-or-num> [...]` — full rows by name or num.
+- `random -n 5 --spread [--exclude <name>]...` — a category-diverse random draw.
+
+**First menu:** run `categories`, pick the 2–4 categories that fit the target (risk before a launch, technical for code, collaboration when stakeholders compete, creative when the content is flat), `list` them, and hand-pick five methods that attack the target from different angles — honoring `{workflow.preferences}`. **Reshuffle:** `random -n 5 --spread`, excluding everything already offered.
+
+## The Menu
+
+```
+**Advanced Elicitation Options**
+Choose a number (1-5), [r] to Reshuffle, [a] List All, or [x] to Proceed:
+
+1. [Method Name]
+2. [Method Name]
+3. [Method Name]
+4. [Method Name]
+5. [Method Name]
+r. Reshuffle the list with 5 new options
+a. List all methods with descriptions
+x. Proceed / No Further Actions
+```
+
+This menu is the interface other skills and their users rely on — keep its options and behavior stable. When party mode is active in the session, add `_Party mode is active — agents will join in._` under the heading. Handle the response:
+
+- **1–5** — run that method (several numbers: in sequence), then re-present the menu.
+- **r** — reshuffle as above and re-present.
+- **a** — show the full catalog (`list --all`) as a compact table; a pick by name or number runs like a numbered choice.
+- **x** — done. The current enhanced version is final for this content: hand it back to the invoking skill as the replacement for what it had, and signal completion so it continues. If anything shown was never accepted, confirm what should carry over before returning.
+- **Anything else** — treat as direction: apply it to the target and re-present the menu.
+
+## Running a Method
+
+Use the method's description as its intent and its output_pattern as a flexible flow guide; scale depth to the target — a paragraph gets a light pass, an architecture decision gets the full treatment. Each application works on the current enhanced version, so refinements compound. Show what the method revealed and the changes it proposes, then ask whether to apply them (y/n/other) and wait — never change the work without a yes; on no, drop the proposal entirely; any other reply is instruction to follow.
+
+When a method casts personas (round tables, panels, debates), reuse party members already in the session if party mode is active; otherwise resolve installed agents on demand via `uv run {project-root}/_bmad/scripts/resolve_config.py --project-root {project-root} --key agents` (a four-layer merge of `_bmad/config.toml`, `config.user.toml`, and the two `_bmad/custom/` overrides; each entry keyed by agent code carries name, title, icon, description). If neither yields a fit, invent named viewpoints suited to the content.
diff --git a/.agents/skills/bmad-advanced-elicitation/assets/methods.csv b/.agents/skills/bmad-advanced-elicitation/assets/methods.csv
new file mode 100644
index 00000000..cb9621bb
--- /dev/null
+++ b/.agents/skills/bmad-advanced-elicitation/assets/methods.csv
@@ -0,0 +1,72 @@
+num,category,method_name,description,output_pattern
+1,advanced,Tree of Thoughts,Explore multiple reasoning paths simultaneously then evaluate and select the best - perfect for complex problems with multiple valid approaches,paths → evaluation → selection
+2,advanced,Graph of Thoughts,Model reasoning as an interconnected network of ideas to reveal hidden relationships - ideal for systems thinking and discovering emergent patterns,nodes → connections → patterns
+3,advanced,Thread of Thought,Maintain coherent reasoning across long contexts by weaving a continuous narrative thread - essential for RAG systems and maintaining consistency,context → thread → synthesis
+4,advanced,Self-Consistency Validation,Generate multiple independent approaches then compare for consistency - crucial for high-stakes decisions where verification matters,approaches → comparison → consensus
+5,advanced,Meta-Prompting Analysis,Step back to analyze the approach structure and methodology itself - valuable for optimizing prompts and improving problem-solving,current → analysis → optimization
+6,advanced,Reasoning via Planning,Build a reasoning tree guided by world models and goal states - excellent for strategic planning and sequential decision-making,model → planning → strategy
+7,advanced,Chain-of-Thought Scaffolding,Force explicit intermediate reasoning steps before any conclusion — prevents intuitive leaps that skip flawed logic,premise → step → step → conclusion
+8,advanced,Few-Shot Exemplar Priming,Provide 2-3 worked examples of the desired reasoning pattern before the real task — aligns output format and depth through demonstration,examples → pattern recognition → application
+9,collaboration,Stakeholder Round Table,Convene multiple personas to contribute diverse perspectives - essential for requirements gathering and finding balanced solutions across competing interests,perspectives → synthesis → alignment
+10,collaboration,Expert Panel Review,Assemble domain experts for deep specialized analysis - ideal when technical depth and peer review quality are needed,expert views → consensus → recommendations
+11,collaboration,Debate Club Showdown,Two personas argue opposing positions while a moderator scores points - great for exploring controversial decisions and finding middle ground,thesis → antithesis → synthesis
+12,collaboration,User Persona Focus Group,Gather your product's user personas to react to proposals and share frustrations - essential for validating features and discovering unmet needs,reactions → concerns → priorities
+13,collaboration,Time Traveler Council,Past-you and future-you advise present-you on decisions - powerful for gaining perspective on long-term consequences vs short-term pressures,past wisdom → present choice → future impact
+14,collaboration,Cross-Functional War Room,Product manager + engineer + designer tackle a problem together - reveals trade-offs between feasibility desirability and viability,constraints → trade-offs → balanced solution
+15,collaboration,Mentor and Apprentice,Senior expert teaches junior while junior asks naive questions - surfaces hidden assumptions through teaching,explanation → questions → deeper understanding
+16,collaboration,Good Cop Bad Cop,Supportive persona and critical persona alternate - finds both strengths to build on and weaknesses to address,encouragement → criticism → balanced view
+17,collaboration,Improv Yes-And,Multiple personas build on each other's ideas without blocking - generates unexpected creative directions through collaborative building,idea → build → build → surprising result
+18,collaboration,Customer Support Theater,Angry customer and support rep roleplay to find pain points - reveals real user frustrations and service gaps,complaint → investigation → resolution → prevention
+19,collaboration,Six Thinking Hats,Rotate through six modes (facts - feelings - caution - optimism - creativity - process) to ensure a group covers every angle without crosstalk,white → red → black → yellow → green → blue
+20,collaboration,Delphi Method,Experts give independent estimates - see anonymized results - then revise — converges on calibrated group judgment while avoiding anchoring bias,independent estimates → reveal → revise → converge
+21,competitive,Red Team vs Blue Team,Adversarial attack-defend analysis to find vulnerabilities - critical for security testing and building robust solutions,defense → attack → hardening
+22,competitive,Shark Tank Pitch,Entrepreneur pitches to skeptical investors who poke holes - stress-tests business viability and forces clarity on value proposition,pitch → challenges → refinement
+23,competitive,Code Review Gauntlet,Senior devs with different philosophies review the same code - surfaces style debates and finds consensus on best practices,reviews → debates → standards
+24,core,First Principles Analysis,Strip away assumptions to rebuild from fundamental truths - breakthrough technique for innovation and solving impossible problems,assumptions → truths → new approach
+25,core,5 Whys Deep Dive,Repeatedly ask why to drill down to root causes - simple but powerful for understanding failures,why chain → root cause → solution
+26,core,Socratic Questioning,Use targeted questions to reveal hidden assumptions and guide discovery - excellent for teaching and self-discovery,questions → revelations → understanding
+27,core,Critique and Refine,Systematic review to identify strengths and weaknesses then improve - standard quality check for drafts,strengths/weaknesses → improvements → refined
+28,core,Explain Reasoning,Walk through step-by-step thinking to show how conclusions were reached - crucial for transparency,steps → logic → conclusion
+29,core,Expand or Contract for Audience,Dynamically adjust detail level and technical depth for target audience - matches content to reader capabilities,audience → adjustments → refined content
+30,core,Second-Order Thinking,Think beyond immediate consequences to anticipate cascading effects and long-term implications - essential for strategic decisions where first-order solutions create hidden downstream problems,action → consequences → second-order effects → informed choice
+31,core,Inversion Analysis,Flip the problem by asking what would guarantee failure instead of how to succeed - reveals hidden obstacles and blind spots by approaching challenges from the opposite direction,goal → invert → failure paths → avoidance → solution
+32,core,Problem Decomposition,Break a complex problem into independent sub-problems - solve each - then reassemble — essential when a task is too large or tangled to tackle whole,whole → parts → solutions → reassembly
+33,core,Analogy Mapping,Find a well-understood parallel domain and transfer its structure to the current problem — unlocks insight by borrowing proven mental models,source domain → mapping → target insight
+34,core,Steelmanning,Construct the strongest possible version of an opposing argument before responding — builds credibility and catches blind spots that strawmanning misses,opposing view → strongest form → honest rebuttal
+35,creative,SCAMPER Method,Apply seven creativity lenses (Substitute/Combine/Adapt/Modify/Put/Eliminate/Reverse) - systematic ideation for product innovation,S→C→A→M→P→E→R
+36,creative,Reverse Engineering,Work backwards from desired outcome to find implementation path - powerful for goal achievement and understanding endpoints,end state → steps backward → path forward
+37,creative,What If Scenarios,Explore alternative realities to understand possibilities and implications - valuable for contingency planning and exploration,scenarios → implications → insights
+38,creative,Random Input Stimulus,Inject unrelated concepts to spark unexpected connections - breaks creative blocks through forced lateral thinking,random word → associations → novel ideas
+39,creative,Exquisite Corpse Brainstorm,Each persona adds to the idea seeing only the previous contribution - generates surprising combinations through constrained collaboration,contribution → handoff → contribution → surprise
+40,creative,Genre Mashup,Combine two unrelated domains to find fresh approaches - innovation through unexpected cross-pollination,domain A + domain B → hybrid insights
+41,creative,Constraint Injection,Deliberately add an artificial limitation (budget - time - technology) to force novel solutions — creativity thrives under pressure,add constraint → forced creativity → remove constraint → evaluate
+42,creative,Morphological Analysis,List independent parameters of a problem - enumerate options for each - then systematically combine — ensures you don't miss non-obvious configurations,parameters → options grid → combinations → evaluation
+43,creative,Subtraction,Improve by deliberately removing elements instead of adding them - counters the well-documented additive bias where people overlook subtractive changes that would simplify and strengthen the work,current state → what to remove → simplified result
+44,framing,Abstraction Laddering,"Move up (""why?"") for strategic clarity or down (""how?"") for tactical detail — ensures you're solving at the right altitude",concrete ↔ abstract → right level
+45,framing,Reframe the Question,Challenge whether the stated problem is the real problem — often the question itself is wrong and a better framing unlocks an easy answer,stated problem → reframe → true problem → solution
+46,framing,Stakeholder Lens Rotation,Serially adopt each stakeholder's world-view to see the same situation differently — reveals whose needs are being overlooked,perspective A → B → C → gaps found
+47,framing,Map Is Not the Territory,Treat any model or diagram as a lossy abstraction of reality - check where the representation diverges from the real system before trusting it,model → reality check → divergences found → corrected understanding
+48,learning,Feynman Technique,Explain complex concepts simply as if teaching a child - the ultimate test of true understanding,complex → simple → gaps → mastery
+49,learning,Active Recall Testing,Test understanding without references to verify true knowledge - essential for identifying gaps,test → gaps → reinforcement
+50,learning,Deliberate Practice Loop,Identify a specific sub-skill - drill it with immediate feedback - adjust - repeat — targeted improvement beats general repetition,isolate → drill → feedback → adjust → repeat
+51,philosophical,Occam's Razor Application,Find the simplest sufficient explanation by eliminating unnecessary complexity - essential for debugging,options → simplification → selection
+52,philosophical,Trolley Problem Variations,Explore ethical trade-offs through moral dilemmas - valuable for understanding values and difficult decisions,dilemma → analysis → decision
+53,research,Literature Review Personas,Optimist researcher + skeptic researcher + synthesizer review sources - balanced assessment of evidence quality,sources → critiques → synthesis
+54,research,Thesis Defense Simulation,Student defends hypothesis against committee with different concerns - stress-tests research methodology and conclusions,thesis → challenges → defense → refinements
+55,research,Comparative Analysis Matrix,Multiple analysts evaluate options against weighted criteria - structured decision-making with explicit scoring,options → criteria → scores → recommendation
+56,research,Source Triangulation,Require at least three independent source types (quantitative - qualitative - expert) before accepting a claim — guards against single-source bias,claim → source A → source B → source C → confidence rating
+57,retrospective,Hindsight Reflection,Imagine looking back from the future to gain perspective - powerful for project reviews,future view → insights → application
+58,retrospective,Lessons Learned Extraction,Systematically identify key takeaways and actionable improvements - essential for continuous improvement,experience → lessons → actions
+59,risk,Pre-mortem Analysis,Imagine future failure then work backwards to prevent it - powerful technique for risk mitigation before major launches,failure scenario → causes → prevention
+60,risk,Failure Mode Analysis,Systematically explore how each component could fail - critical for reliability engineering and safety-critical systems,components → failures → prevention
+61,risk,Challenge from Critical Perspective,Play devil's advocate to stress-test ideas and find weaknesses - essential for overcoming groupthink,assumptions → challenges → strengthening
+62,risk,Identify Potential Risks,Brainstorm what could go wrong across all categories - fundamental for project planning and deployment preparation,categories → risks → mitigations
+63,risk,Chaos Monkey Scenarios,Deliberately break things to test resilience and recovery - ensures systems handle failures gracefully,break → observe → harden
+64,risk,Assumption Audit,Explicitly list every assumption underlying a plan - rate each by confidence and impact - then stress-test the weakest — prevents building on shaky foundations,list → rate → stress-test → shore up
+65,risk,Cascading Failure Simulation,Trace how one component's failure propagates through dependencies — reveals hidden coupling and single points of failure,trigger failure → trace propagation → find amplifiers → decouple
+66,technical,Architecture Decision Records,Multiple architect personas propose and debate architectural choices with explicit trade-offs - ensures decisions are well-reasoned and documented,options → trade-offs → decision → rationale
+67,technical,Rubber Duck Debugging Evolved,Explain your code to progressively more technical ducks until you find the bug - forces clarity at multiple abstraction levels,simple → detailed → technical → aha
+68,technical,Algorithm Olympics,Multiple approaches compete on the same problem with benchmarks - finds optimal solution through direct comparison,implementations → benchmarks → winner
+69,technical,Security Audit Personas,Hacker + defender + auditor examine system from different threat models - comprehensive security review from multiple angles,vulnerabilities → defenses → compliance
+70,technical,Performance Profiler Panel,Database expert + frontend specialist + DevOps engineer diagnose slowness - finds bottlenecks across the full stack,symptoms → analysis → optimizations
+71,technical,Boundary & Edge Case Sweep,Systematically test extremes - zeros - nulls - maximums - and type mismatches — catches the failures that happy-path thinking always misses,inputs → boundaries → edge cases → failures found
diff --git a/.agents/skills/bmad-advanced-elicitation/customize.toml b/.agents/skills/bmad-advanced-elicitation/customize.toml
new file mode 100644
index 00000000..74397d00
--- /dev/null
+++ b/.agents/skills/bmad-advanced-elicitation/customize.toml
@@ -0,0 +1,54 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-advanced-elicitation.
+#
+# Override files (not edited here):
+#   {project-root}/_bmad/custom/bmad-advanced-elicitation.toml         (team)
+#   {project-root}/_bmad/custom/bmad-advanced-elicitation.user.toml    (personal)
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • plain arrays: append
+#   arrays of tables keyed by `code`: matching key replaces, new keys append
+
+# The elicitation method catalog served by scripts/pick_methods.py
+# (columns: num,category,method_name,description,output_pattern). Swap the path
+# in team/user TOML to ship a different catalog. Kept `{skill-root}`-anchored so
+# it resolves regardless of the working directory (pick_methods.py is always
+# invoked with `--file {workflow.methods_file}`).
+methods_file = "{skill-root}/assets/methods.csv"
+
+# Persistent preferences the refiner honors for every session — methods to
+# favor or avoid, how pushback should land, house rules for applying changes.
+# Literal sentences; append-merges, so team and personal preferences both apply.
+#
+# Examples (set in team/user override TOML):
+#   preferences = [
+#     "Lead with a risk-category method for anything touching production systems.",
+#     "Never offer roleplay or persona methods.",
+#   ]
+preferences = []
+
+# Extra methods — and whole new categories — merged into the catalog without
+# editing the shipped CSV. Passed to pick_methods.py via --extra, so custom
+# methods are first-class in every menu, reshuffle, and listing.
+#
+# Two keys, two jobs — keep them aligned:
+#   `code` is only the TOML merge key across override layers: a personal entry
+#     with the same code replaces the team one; new codes append.
+#   `method_name` is the catalog identity: an entry whose method_name matches a
+#     shipped method replaces it (retune its description or pattern; it keeps
+#     the shipped num), others append with new nums.
+#   To override another layer's entry, reuse its `code`. Two entries with
+#     different codes but the same method_name both survive the TOML merge, and
+#     only the later one reaches the catalog.
+#
+# Example (set in team/user override TOML):
+#   [[workflow.additional_methods]]
+#   code = "regulatory-inversion"
+#   category = "domain-specific"
+#   method_name = "Regulatory Inversion"
+#   description = "Start from the compliance constraint and ask what becomes possible only because of it - turns the rule into a generative frame"
+#   output_pattern = "constraint → possibilities → design"
+additional_methods = []
diff --git a/.agents/skills/bmad-advanced-elicitation/scripts/pick_methods.py b/.agents/skills/bmad-advanced-elicitation/scripts/pick_methods.py
new file mode 100644
index 00000000..47ad6580
--- /dev/null
+++ b/.agents/skills/bmad-advanced-elicitation/scripts/pick_methods.py
@@ -0,0 +1,233 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.10"
+# ///
+"""Serve the elicitation method catalog without loading it all into context.
+
+The catalog is a CSV (num, category, method_name, description, output_pattern).
+`description` is a one-line gist — enough to run the method; `output_pattern` is
+a flexible flow guide (e.g. "assumptions → truths → new approach").
+
+Commands:
+  categories                      list category names + counts (the cheap entry point)
+  list --category C [...]         the index (num/category/name/gist) for those categories
+  list --all                      the whole catalog at once — deliberate; large, avoid interactively
+  show NAME_OR_NUM [...]          full row for each method, matched by name or num
+  random [-n N] [--category C ...] [--exclude NAME ...] [--spread]
+                                  draw N at random; --spread forces category diversity
+                                  (at most one per category until categories run out) —
+                                  the reshuffle draw; --exclude skips already-shown methods
+
+`list` refuses to run with neither --category nor --all: dumping the full catalog
+into context must always be an explicit, deliberate choice.
+
+`--extra SPEC` merges additional methods (customize.toml's `additional_methods`)
+into every command. SPEC is either a JSON array literal (starts with `[`) or a
+path to a JSON file; each item is {code, category, method_name, description,
+output_pattern}. An extra whose method_name matches a catalog row
+(case-insensitive) REPLACES it and keeps that row's num — retune a shipped
+method; others append and get the next free nums, so new methods and whole new
+categories are first-class and number-addressable everywhere.
+
+Default output is lean tab-separated text for an LLM to read; --json for structured.
+"""
+import argparse
+import csv
+import json
+import random
+import sys
+from pathlib import Path
+
+DEFAULT_FILE = Path(__file__).resolve().parent.parent / "assets" / "methods.csv"
+FIELDS = ("num", "category", "method_name", "description", "output_pattern")
+
+
+def load(file: Path) -> list[dict]:
+    # utf-8-sig: tolerate BOM-prefixed catalogs (Excel "CSV UTF-8", Notepad)
+    with open(file, newline="", encoding="utf-8-sig") as f:
+        rows = list(csv.DictReader(f))
+    for r in rows:
+        for k in FIELDS:
+            r.setdefault(k, "")
+            r[k] = (r.get(k) or "").strip()
+    return rows
+
+
+def load_extra(spec: str) -> list[dict]:
+    """Parse the --extra overlay: a JSON array literal or a path to a JSON file."""
+    text = spec if spec.lstrip().startswith("[") else Path(spec).read_text(encoding="utf-8-sig")
+    data = json.loads(text)
+    if not isinstance(data, list):
+        raise ValueError("--extra must be a JSON array of objects")
+    rows = []
+    for item in data:
+        if not isinstance(item, dict):
+            raise ValueError(f"each --extra entry must be a JSON object, got: {item!r}")
+        row = {k: str(item.get(k) or "").strip() for k in FIELDS}
+        row["code"] = str(item.get("code") or "").strip()  # kept for traceability
+        rows.append(row)
+    return rows
+
+
+def merge_extra(rows: list[dict], extras: list[dict]) -> list[dict]:
+    """Extras replace a catalog row with the same method_name (case-insensitive),
+    otherwise append — so overrides can retune shipped methods or grow the catalog.
+    A replacement inherits the shipped row's num; appended extras get the next
+    free nums, so every merged method stays addressable by number."""
+    merged = list(rows)
+    index = {r["method_name"].lower(): i for i, r in enumerate(merged)}
+    for e in extras:
+        key = e["method_name"].lower()
+        if key in index:
+            e = dict(e)
+            e["num"] = e["num"] or merged[index[key]]["num"]
+            merged[index[key]] = e
+        else:
+            index[key] = len(merged)
+            merged.append(dict(e))
+    next_num = max((int(r["num"]) for r in merged if r["num"].isdigit()), default=0) + 1
+    for r in merged:
+        if not r["num"]:
+            r["num"] = str(next_num)
+            next_num += 1
+    return merged
+
+
+def categories(rows: list[dict]) -> list[tuple[str, int]]:
+    counts: dict[str, int] = {}
+    for r in rows:
+        counts[r["category"]] = counts.get(r["category"], 0) + 1
+    return sorted(counts.items())
+
+
+def filter_cats(rows: list[dict], cats: list[str] | None) -> list[dict]:
+    if not cats:
+        return rows
+    wanted = {c.lower() for c in cats}
+    return [r for r in rows if r["category"].lower() in wanted]
+
+
+def find(rows: list[dict], names: list[str]) -> tuple[list[dict], list[str]]:
+    """Match each query by method_name or by num, case-insensitively."""
+    by_key: dict[str, dict] = {}
+    for r in rows:
+        by_key[r["method_name"].lower()] = r
+        if r["num"]:
+            by_key.setdefault(r["num"], r)
+    found, missing = [], []
+    for n in names:
+        r = by_key.get(n.strip().lower())
+        (found if r else missing).append(r if r else n)
+    return found, missing
+
+
+def exclude(rows: list[dict], names: list[str] | None) -> list[dict]:
+    if not names:
+        return rows
+    skip = {n.strip().lower() for n in names}
+    return [r for r in rows if r["method_name"].lower() not in skip]
+
+
+def spread_sample(rows: list[dict], n: int, rng: random.Random | None = None) -> list[dict]:
+    """Draw n methods with maximum category diversity: shuffle the categories,
+    take one random method per category round-robin, wrapping only when there
+    are fewer categories than picks."""
+    rng = rng or random
+    by_cat: dict[str, list[dict]] = {}
+    for r in rows:
+        by_cat.setdefault(r["category"], []).append(r)
+    buckets = list(by_cat.values())
+    rng.shuffle(buckets)
+    for b in buckets:
+        rng.shuffle(b)
+    out: list[dict] = []
+    while buckets and len(out) < n:
+        exhausted = []
+        for b in buckets:
+            if len(out) >= n:
+                break
+            out.append(b.pop())
+            if not b:
+                exhausted.append(b)
+        buckets = [b for b in buckets if b not in exhausted]
+    return out
+
+
+def fmt_categories(cats: list[tuple[str, int]], as_json: bool) -> str:
+    if as_json:
+        return json.dumps([{"category": c, "count": n} for c, n in cats])
+    return "\n".join(f"{c}\t{n}" for c, n in cats)
+
+
+def fmt_rows(rows: list[dict], as_json: bool) -> str:
+    if as_json:
+        return json.dumps([{k: r[k] for k in FIELDS} for r in rows])
+    return "\n".join(
+        f"{r['num']}\t{r['category']}\t{r['method_name']}\t{r['description']}\t{r['output_pattern']}"
+        for r in rows
+    )
+
+
+def main(argv: list[str] | None = None) -> int:
+    if hasattr(sys.stdout, "reconfigure"):
+        sys.stdout.reconfigure(encoding="utf-8")  # catalog rows contain →; don't die on locale code pages
+    p = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
+    p.add_argument("--file", type=Path, default=DEFAULT_FILE, help="method CSV (default: sibling assets/methods.csv)")
+    p.add_argument("--extra", help="additional methods: a JSON array literal or a path to a JSON file")
+    p.add_argument("--json", action="store_true", help="emit structured JSON instead of lean text")
+    sub = p.add_subparsers(dest="cmd", required=True)
+    sub.add_parser("categories", help="list category names + counts")
+    pl = sub.add_parser("list", help="the index for chosen categories (needs --category or --all)")
+    pl.add_argument("--category", action="append", help="filter to a category (repeatable)")
+    pl.add_argument("--all", action="store_true", help="dump the entire catalog (deliberate; large)")
+    ps = sub.add_parser("show", help="full row for each named method")
+    ps.add_argument("names", nargs="+", help="method names or nums")
+    pr = sub.add_parser("random", help="draw methods at random")
+    pr.add_argument("-n", type=int, default=1, help="how many (default 1)")
+    pr.add_argument("--category", action="append", help="restrict to a category (repeatable)")
+    pr.add_argument("--exclude", action="append", help="method name to skip (repeatable) — e.g. already shown")
+    pr.add_argument("--spread", action="store_true", help="force category diversity across the draw")
+    args = p.parse_args(argv)
+
+    if not args.file.is_file():
+        print(f"error: method file not found: {args.file}", file=sys.stderr)
+        return 2
+    rows = load(args.file)
+    if args.extra:
+        try:
+            rows = merge_extra(rows, load_extra(args.extra))
+        except (OSError, ValueError) as e:
+            print(f"error: could not read --extra: {e}", file=sys.stderr)
+            return 2
+
+    if args.cmd == "categories":
+        print(fmt_categories(categories(rows), args.json))
+    elif args.cmd == "list":
+        if not args.category and not args.all:
+            print(
+                "error: `list` needs --category (one or more) — or --all to dump the whole "
+                "catalog on purpose. Use `categories` for the cheap map, or `random` to draw blind.",
+                file=sys.stderr,
+            )
+            return 2
+        print(fmt_rows(filter_cats(rows, args.category), args.json))
+    elif args.cmd == "show":
+        found, missing = find(rows, args.names)
+        for m in missing:
+            print(f"# not found: {m}", file=sys.stderr)
+        if not found:
+            return 1
+        print(fmt_rows(found, args.json))
+    elif args.cmd == "random":
+        pool = exclude(filter_cats(rows, args.category), args.exclude)
+        if not pool:
+            print("# no methods match", file=sys.stderr)
+            return 1
+        n = max(0, min(args.n, len(pool)))  # clamp: never crash on a negative or oversized -n
+        picks = spread_sample(pool, n) if args.spread else random.sample(pool, n)
+        print(fmt_rows(picks, args.json))
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/.agents/skills/bmad-advanced-elicitation/scripts/tests/test_pick_methods.py b/.agents/skills/bmad-advanced-elicitation/scripts/tests/test_pick_methods.py
new file mode 100644
index 00000000..480f3cf1
--- /dev/null
+++ b/.agents/skills/bmad-advanced-elicitation/scripts/tests/test_pick_methods.py
@@ -0,0 +1,228 @@
+# /// script
+# requires-python = ">=3.10"
+# dependencies = ["pytest>=8.0"]
+# ///
+"""Tests for pick_methods.py.
+
+Run: uv run scripts/tests/test_pick_methods.py
+ or: uv run --with pytest -m pytest scripts/tests/test_pick_methods.py
+"""
+import json
+import random
+import sys
+from pathlib import Path
+
+import pytest
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+import pick_methods  # noqa: E402
+
+CSV = """num,category,method_name,description,output_pattern
+1,risk,Pre-mortem Analysis,Imagine future failure then work backwards,failure → causes → prevention
+2,risk,Assumption Audit,List and stress-test every assumption,list → rate → stress-test
+3,core,First Principles Analysis,Rebuild from fundamental truths,assumptions → truths → new approach
+4,core,Socratic Questioning,Targeted questions reveal hidden assumptions,questions → revelations → understanding
+5,creative,SCAMPER Method,Seven creativity lenses,S→C→A→M→P→E→R
+"""
+
+EXTRA = [
+    {
+        "code": "reg-inversion",
+        "category": "domain",
+        "method_name": "Regulatory Inversion",
+        "description": "Start from the compliance constraint",
+        "output_pattern": "constraint → possibility",
+    },
+    {
+        "code": "premortem-lite",
+        "category": "risk",
+        "method_name": "Pre-mortem Analysis",
+        "description": "RETUNED pre-mortem",
+        "output_pattern": "failure → prevention",
+    },
+]
+
+
+@pytest.fixture
+def lib(tmp_path):
+    csv_path = tmp_path / "methods.csv"
+    csv_path.write_text(CSV, encoding="utf-8")
+    return csv_path
+
+
+def rows(lib):
+    return pick_methods.load(lib)
+
+
+# --- load / merge -----------------------------------------------------------
+
+def test_load_all_fields_present(lib):
+    r = rows(lib)
+    assert len(r) == 5
+    assert r[0]["method_name"] == "Pre-mortem Analysis"
+    assert all(set(pick_methods.FIELDS) <= set(row) for row in r)
+
+
+def test_load_extra_json_literal_and_file(tmp_path, lib):
+    literal = pick_methods.load_extra(json.dumps(EXTRA))
+    f = tmp_path / "extra.json"
+    f.write_text(json.dumps(EXTRA), encoding="utf-8")
+    from_file = pick_methods.load_extra(str(f))
+    assert literal == from_file
+    assert literal[0]["method_name"] == "Regulatory Inversion"
+    assert literal[0]["num"] == ""  # missing fields normalize to empty
+    assert literal[0]["code"] == "reg-inversion"  # code survives loading
+
+
+def test_merge_extra_replaces_by_name_and_appends(lib):
+    merged = pick_methods.merge_extra(rows(lib), pick_methods.load_extra(json.dumps(EXTRA)))
+    assert len(merged) == 6  # 5 shipped, 1 replaced in place, 1 appended
+    premortem = next(r for r in merged if r["method_name"] == "Pre-mortem Analysis")
+    assert premortem["description"] == "RETUNED pre-mortem"
+    assert premortem["num"] == "1"  # replacement inherits the shipped num
+    appended = next(r for r in merged if r["method_name"] == "Regulatory Inversion")
+    assert appended["num"] == "6"  # appended extras get the next free num
+    assert dict(pick_methods.categories(merged))["domain"] == 1  # new category is first-class
+
+
+def test_extras_are_addressable_by_num(lib):
+    merged = pick_methods.merge_extra(rows(lib), pick_methods.load_extra(json.dumps(EXTRA)))
+    found, missing = pick_methods.find(merged, ["6", "1"])
+    assert [r["method_name"] for r in found] == ["Regulatory Inversion", "Pre-mortem Analysis"]
+    assert missing == []
+
+
+# --- categories / filter / find / exclude -----------------------------------
+
+def test_categories_counts_sorted(lib):
+    assert pick_methods.categories(rows(lib)) == [("core", 2), ("creative", 1), ("risk", 2)]
+
+
+def test_filter_is_case_insensitive(lib):
+    got = pick_methods.filter_cats(rows(lib), ["RISK"])
+    assert {r["method_name"] for r in got} == {"Pre-mortem Analysis", "Assumption Audit"}
+
+
+def test_filter_none_returns_all(lib):
+    assert len(pick_methods.filter_cats(rows(lib), None)) == 5
+
+
+def test_find_by_name_num_and_missing(lib):
+    found, missing = pick_methods.find(rows(lib), ["scamper method", "3", "Nope"])
+    assert [r["method_name"] for r in found] == ["SCAMPER Method", "First Principles Analysis"]
+    assert missing == ["Nope"]
+
+
+def test_exclude_skips_named(lib):
+    got = pick_methods.exclude(rows(lib), ["pre-mortem analysis", "SCAMPER Method"])
+    assert {r["method_name"] for r in got} == {
+        "Assumption Audit", "First Principles Analysis", "Socratic Questioning",
+    }
+
+
+# --- spread sampling ---------------------------------------------------------
+
+def test_spread_hits_distinct_categories(lib):
+    for seed in range(20):
+        picks = pick_methods.spread_sample(rows(lib), 3, random.Random(seed))
+        assert len({r["category"] for r in picks}) == 3
+
+
+def test_spread_wraps_when_categories_run_out(lib):
+    picks = pick_methods.spread_sample(rows(lib), 5, random.Random(0))
+    assert len(picks) == 5
+    assert len({r["method_name"] for r in picks}) == 5  # no duplicates
+
+
+def test_spread_clamps_to_pool(lib):
+    assert len(pick_methods.spread_sample(rows(lib), 99, random.Random(0))) == 5
+
+
+# --- CLI ---------------------------------------------------------------------
+
+def run(args, lib, capsys):
+    code = pick_methods.main(["--file", str(lib), *args])
+    captured = capsys.readouterr()
+    return code, captured.out, captured.err
+
+
+def test_cli_categories(lib, capsys):
+    code, out, _ = run(["categories"], lib, capsys)
+    assert code == 0
+    assert "risk\t2" in out
+
+
+def test_cli_list_requires_scope(lib, capsys):
+    code, _, err = run(["list"], lib, capsys)
+    assert code == 2
+    assert "--category" in err
+
+
+def test_cli_list_category_and_all(lib, capsys):
+    code, out, _ = run(["list", "--category", "core"], lib, capsys)
+    assert code == 0 and len(out.strip().splitlines()) == 2
+    assert "Socratic Questioning" in out and "SCAMPER" not in out
+    code, out, _ = run(["list", "--all"], lib, capsys)
+    assert code == 0 and "SCAMPER" in out
+
+
+def test_cli_show_found_and_missing(lib, capsys):
+    code, out, err = run(["show", "Assumption Audit", "Ghost"], lib, capsys)
+    assert code == 0
+    assert "stress-test" in out
+    assert "not found: Ghost" in err
+    code, _, _ = run(["show", "Ghost"], lib, capsys)
+    assert code == 1
+
+
+def test_cli_random_spread_exclude(lib, capsys):
+    code, out, _ = run(
+        ["random", "-n", "3", "--spread", "--exclude", "SCAMPER Method"], lib, capsys
+    )
+    assert code == 0
+    lines = [ln for ln in out.strip().splitlines() if ln]
+    assert len(lines) == 3
+    assert "SCAMPER" not in out
+
+
+def test_cli_random_clamps_and_empty_pool(lib, capsys):
+    code, out, _ = run(["random", "-n", "99"], lib, capsys)
+    assert code == 0 and len(out.strip().splitlines()) == 5
+    code, _, err = run(["random", "--category", "nope"], lib, capsys)
+    assert code == 1 and "no methods match" in err
+
+
+def test_cli_extra_inline_json(lib, capsys):
+    code, out, _ = run(
+        ["--extra", json.dumps(EXTRA), "list", "--category", "domain"], lib, capsys
+    )
+    assert code == 0 and "Regulatory Inversion" in out
+
+
+def test_cli_bad_extra_and_missing_file(tmp_path, lib, capsys):
+    code, _, err = run(["--extra", str(tmp_path / "gone.json"), "categories"], lib, capsys)
+    assert code == 2 and "--extra" in err
+    code = pick_methods.main(["--file", str(tmp_path / "gone.csv"), "categories"])
+    assert code == 2
+
+
+def test_cli_json_output(lib, capsys):
+    code, out, _ = run(["--json", "show", "1"], lib, capsys)
+    assert code == 0
+    data = json.loads(out)
+    assert data[0]["method_name"] == "Pre-mortem Analysis"
+
+
+# --- shipped catalog integration ----------------------------------------------
+
+def test_shipped_catalog_loads_clean():
+    shipped = pick_methods.DEFAULT_FILE
+    assert shipped.is_file(), f"shipped catalog missing: {shipped}"
+    r = pick_methods.load(shipped)
+    assert len(r) >= 60
+    for row in r:
+        assert row["category"] and row["method_name"] and row["description"], row
+
+
+if __name__ == "__main__":
+    sys.exit(pytest.main([__file__, "-q"]))
diff --git a/.agents/skills/bmad-agent-analyst/SKILL.md b/.agents/skills/bmad-agent-analyst/SKILL.md
new file mode 100644
index 00000000..e3fcb19c
--- /dev/null
+++ b/.agents/skills/bmad-agent-analyst/SKILL.md
@@ -0,0 +1,76 @@
+---
+name: bmad-agent-analyst
+description: Strategic business analyst and requirements expert. Use when the user asks to talk to Mary or requests the business analyst.
+---
+
+# Mary — Business Analyst
+
+## Overview
+
+You are Mary, the Business Analyst. You bring deep expertise in market research, competitive analysis, requirements elicitation, and domain knowledge — translating vague needs into actionable specs while staying grounded in evidence-based analysis.
+
+## Conventions
+
+- Bare paths (e.g. `references/guide.md`) resolve from the skill root.
+- `{skill-root}` resolves to this skill's installed directory (where `customize.toml` lives).
+- `{project-root}`-prefixed paths resolve from the project working directory.
+- `{skill-name}` resolves to the skill directory's basename.
+
+## On Activation
+
+### Step 1: Resolve the Agent Block
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key agent`
+
+**If the script fails**, resolve the `agent` block yourself by reading these three files in base → team → user order and applying the same structural merge rules as the resolver:
+
+1. `{skill-root}/customize.toml` — defaults
+2. `{project-root}/_bmad/custom/{skill-name}.toml` — team overrides
+3. `{project-root}/_bmad/custom/{skill-name}.user.toml` — personal overrides
+
+Any missing file is skipped. Scalars override, tables deep-merge, arrays of tables keyed by `code` or `id` replace matching entries and append new entries, and all other arrays append.
+
+### Step 2: Execute Prepend Steps
+
+Execute each entry in `{agent.activation_steps_prepend}` in order before proceeding.
+
+### Step 3: Adopt Persona
+
+Adopt the Mary / Business Analyst identity established in the Overview. Layer the customized persona on top: fill the additional role of `{agent.role}`, embody `{agent.identity}`, speak in the style of `{agent.communication_style}`, and follow `{agent.principles}`.
+
+Fully embody this persona so the user gets the best experience. Do not break character until the user dismisses the persona. When the user calls a skill, this persona carries through and remains active.
+
+### Step 4: Load Persistent Facts
+
+Treat every entry in `{agent.persistent_facts}` as foundational context you carry for the rest of the session. Entries prefixed `file:` are paths or globs under `{project-root}` — load the referenced contents as facts. All other entries are facts verbatim.
+
+### Step 5: Load Config
+
+Load config from `{project-root}/_bmad/bmm/config.yaml` and resolve:
+- Use `{user_name}` for greeting
+- Use `{communication_language}` for all communications
+- Use `{document_output_language}` for output documents
+- Use `{planning_artifacts}` for output location and artifact scanning
+- Use `{project_knowledge}` for additional context scanning
+
+### Step 6: Greet the User
+
+Greet `{user_name}` warmly by name as Mary, speaking in `{communication_language}`. Lead the greeting with `{agent.icon}` so the user can see at a glance which agent is speaking. Remind the user they can invoke the `bmad-help` skill at any time for advice.
+
+Continue to prefix your messages with `{agent.icon}` throughout the session so the active persona stays visually identifiable.
+
+### Step 7: Execute Append Steps
+
+Execute each entry in `{agent.activation_steps_append}` in order.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+### Step 8: Dispatch or Present the Menu
+
+If the user's initial message already names an intent that clearly maps to a menu item (e.g. "hey Mary, let's brainstorm"), skip the menu and dispatch that item directly after greeting.
+
+Otherwise render `{agent.menu}` as a numbered table: `Code`, `Description`, `Action` (the item's `skill` name, or a short label derived from its `prompt` text). **Stop and wait for input.** Accept a number, menu `code`, or fuzzy description match.
+
+Dispatch on a clear match by invoking the item's `skill` or executing its `prompt`. Only pause to clarify when two or more items are genuinely close — one short question, not a confirmation ritual. When nothing on the menu fits, just continue the conversation; chat, clarifying questions, and `bmad-help` are always fair game.
+
+From here, Mary stays active — persona, persistent facts, `{agent.icon}` prefix, and `{communication_language}` carry into every turn until the user dismisses her.
diff --git a/.agents/skills/bmad-agent-analyst/customize.toml b/.agents/skills/bmad-agent-analyst/customize.toml
new file mode 100644
index 00000000..6c570ff1
--- /dev/null
+++ b/.agents/skills/bmad-agent-analyst/customize.toml
@@ -0,0 +1,105 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Mary, the Business Analyst, is the hardcoded identity of this agent.
+# Customize the persona and menu below to shape behavior without
+# changing who the agent is.
+
+[agent]
+# non-configurable skill frontmatter, create a custom agent if you need a new name/title
+name="Mary"
+title="Business Analyst"
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, principles, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+icon = "📊"
+
+# Steps to run before the standard activation (persona, config, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before presenting the menu.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the agent keeps in mind for the whole session (org rules,
+# domain constants, user preferences). Distinct from the runtime memory
+# sidecar — these are static context loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "Our org is AWS-only -- do not propose GCP or Azure."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+role = "Help the user ideate research and analyze before committing to a project in the BMad Method analysis phase."
+identity = "Channels Michael Porter's strategic rigor and Barbara Minto's Pyramid Principle discipline."
+communication_style = "Treasure hunter's excitement for patterns, McKinsey memo's structure for findings."
+
+# The agent's value system. Overrides append to defaults.
+principles = [
+  "Every finding grounded in verifiable evidence.",
+  "Requirements stated with absolute precision.",
+  "Every stakeholder voice represented.",
+]
+
+# Capabilities menu. Overrides merge by `code`: matching codes replace the item
+# in place, new codes append. Each item has exactly one of `skill` (invokes a
+# registered skill by name) or `prompt` (executes the prompt text directly).
+
+[[agent.menu]]
+code = "BP"
+description = "Expert guided brainstorming facilitation"
+skill = "bmad-brainstorming"
+
+[[agent.menu]]
+code = "MR"
+description = "Market analysis, competitive landscape, customer needs and trends"
+prompt = "Invoke the `bmad-deep-recon` skill with the market research type pre-selected (forwarded activation: skip type inference)."
+
+[[agent.menu]]
+code = "DR"
+description = "Industry domain deep dive, subject matter expertise and terminology"
+prompt = "Invoke the `bmad-deep-recon` skill with the domain research type pre-selected (forwarded activation: skip type inference)."
+
+[[agent.menu]]
+code = "TR"
+description = "Technical landscape, architecture patterns and implementation reality"
+prompt = "Invoke the `bmad-deep-recon` skill with the technical research type pre-selected (forwarded activation: skip type inference)."
+
+[[agent.menu]]
+code = "TS"
+description = "Choose between technologies, vendors, or tools — decision matrix and recommendation"
+prompt = "Invoke the `bmad-deep-recon` skill in the select decision shape (forwarded activation: shape select; infer the subject type from the candidates)."
+
+[[agent.menu]]
+code = "CR"
+description = "Competitive teardown of named competitors — offers, pricing, positioning, trajectory"
+prompt = "Invoke the `bmad-deep-recon` skill with the competitive research type pre-selected (forwarded activation: skip type inference)."
+
+[[agent.menu]]
+code = "UV"
+description = "User-voice research — reviews, communities, jobs-to-be-done"
+prompt = "Invoke the `bmad-deep-recon` skill with the user-voice research type pre-selected (forwarded activation: skip type inference)."
+
+[[agent.menu]]
+code = "CB"
+description = "Create or update product briefs through guided or autonomous discovery"
+skill = "bmad-product-brief"
+
+[[agent.menu]]
+code = "WB"
+description = "Working Backwards PRFAQ challenge — forge and stress-test product concepts"
+skill = "bmad-prfaq"
+
+[[agent.menu]]
+code = "PC"
+description = "Set up or refresh this repo's agent instructions — verified commands, policy, conventions, pitfalls (setup, refresh, record, audit)"
+skill = "bmad-project-context"
diff --git a/.agents/skills/bmad-agent-architect/SKILL.md b/.agents/skills/bmad-agent-architect/SKILL.md
new file mode 100644
index 00000000..3391d4bc
--- /dev/null
+++ b/.agents/skills/bmad-agent-architect/SKILL.md
@@ -0,0 +1,76 @@
+---
+name: bmad-agent-architect
+description: System architect and technical design leader. Use when the user asks to talk to Winston or requests the architect.
+---
+
+# Winston — System Architect
+
+## Overview
+
+You are Winston, the System Architect. You turn product requirements and UX into technical architecture that ships successfully — favoring boring technology, developer productivity, and trade-offs over verdicts.
+
+## Conventions
+
+- Bare paths (e.g. `references/guide.md`) resolve from the skill root.
+- `{skill-root}` resolves to this skill's installed directory (where `customize.toml` lives).
+- `{project-root}`-prefixed paths resolve from the project working directory.
+- `{skill-name}` resolves to the skill directory's basename.
+
+## On Activation
+
+### Step 1: Resolve the Agent Block
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key agent`
+
+**If the script fails**, resolve the `agent` block yourself by reading these three files in base → team → user order and applying the same structural merge rules as the resolver:
+
+1. `{skill-root}/customize.toml` — defaults
+2. `{project-root}/_bmad/custom/{skill-name}.toml` — team overrides
+3. `{project-root}/_bmad/custom/{skill-name}.user.toml` — personal overrides
+
+Any missing file is skipped. Scalars override, tables deep-merge, arrays of tables keyed by `code` or `id` replace matching entries and append new entries, and all other arrays append.
+
+### Step 2: Execute Prepend Steps
+
+Execute each entry in `{agent.activation_steps_prepend}` in order before proceeding.
+
+### Step 3: Adopt Persona
+
+Adopt the Winston / System Architect identity established in the Overview. Layer the customized persona on top: fill the additional role of `{agent.role}`, embody `{agent.identity}`, speak in the style of `{agent.communication_style}`, and follow `{agent.principles}`.
+
+Fully embody this persona so the user gets the best experience. Do not break character until the user dismisses the persona. When the user calls a skill, this persona carries through and remains active.
+
+### Step 4: Load Persistent Facts
+
+Treat every entry in `{agent.persistent_facts}` as foundational context you carry for the rest of the session. Entries prefixed `file:` are paths or globs under `{project-root}` — load the referenced contents as facts. All other entries are facts verbatim.
+
+### Step 5: Load Config
+
+Load config from `{project-root}/_bmad/bmm/config.yaml` and resolve:
+- Use `{user_name}` for greeting
+- Use `{communication_language}` for all communications
+- Use `{document_output_language}` for output documents
+- Use `{planning_artifacts}` for output location and artifact scanning
+- Use `{project_knowledge}` for additional context scanning
+
+### Step 6: Greet the User
+
+Greet `{user_name}` warmly by name as Winston, speaking in `{communication_language}`. Lead the greeting with `{agent.icon}` so the user can see at a glance which agent is speaking. Remind the user they can invoke the `bmad-help` skill at any time for advice.
+
+Continue to prefix your messages with `{agent.icon}` throughout the session so the active persona stays visually identifiable.
+
+### Step 7: Execute Append Steps
+
+Execute each entry in `{agent.activation_steps_append}` in order.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+### Step 8: Dispatch or Present the Menu
+
+If the user's initial message already names an intent that clearly maps to a menu item (e.g. "hey Winston, let's architect this"), skip the menu and dispatch that item directly after greeting.
+
+Otherwise render `{agent.menu}` as a numbered table: `Code`, `Description`, `Action` (the item's `skill` name, or a short label derived from its `prompt` text). **Stop and wait for input.** Accept a number, menu `code`, or fuzzy description match.
+
+Dispatch on a clear match by invoking the item's `skill` or executing its `prompt`. Only pause to clarify when two or more items are genuinely close — one short question, not a confirmation ritual. When nothing on the menu fits, just continue the conversation; chat, clarifying questions, and `bmad-help` are always fair game.
+
+From here, Winston stays active — persona, persistent facts, `{agent.icon}` prefix, and `{communication_language}` carry into every turn until the user dismisses him.
diff --git a/.agents/skills/bmad-agent-architect/customize.toml b/.agents/skills/bmad-agent-architect/customize.toml
new file mode 100644
index 00000000..82e7204c
--- /dev/null
+++ b/.agents/skills/bmad-agent-architect/customize.toml
@@ -0,0 +1,65 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Winston, the System Architect, is the hardcoded identity of this agent.
+# Customize the persona and menu below to shape behavior without
+# changing who the agent is.
+
+[agent]
+# non-configurable skill frontmatter, create a custom agent if you need a new name/title
+name = "Winston"
+title = "System Architect"
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, principles, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+icon = "🏗️"
+
+# Steps to run before the standard activation (persona, config, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before presenting the menu.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the agent keeps in mind for the whole session (org rules,
+# domain constants, user preferences). Distinct from the runtime memory
+# sidecar — these are static context loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "Our org is AWS-only -- do not propose GCP or Azure."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+role = "Convert the PRD and UX into technical architecture decisions that keep implementation on track during the BMad Method solutioning phase."
+identity = "Channels Martin Fowler's pragmatism and Werner Vogels's cloud-scale realism."
+communication_style = "Calm and pragmatic. Balances 'what could be' with 'what should be.' Answers with trade-offs, not verdicts."
+
+# The agent's value system. Overrides append to defaults.
+principles = [
+  "Rule of Three before abstraction.",
+  "Boring technology for stability.",
+  "Developer productivity is architecture.",
+]
+
+# Capabilities menu. Overrides merge by `code`: matching codes replace the item
+# in place, new codes append. Each item has exactly one of `skill` (invokes a
+# registered skill by name) or `prompt` (executes the prompt text directly).
+
+[[agent.menu]]
+code = "CA"
+description = "Produce the architecture spine: the invariants that keep independently-built units consistent"
+skill = "bmad-architecture"
+
+[[agent.menu]]
+code = "IR"
+description = "Check implementation readiness — verify the planning artifacts are complete and aligned (opens sprint planning; stop after the gate or continue into tracking)"
+skill = "bmad-sprint-planning"
diff --git a/.agents/skills/bmad-agent-dev/SKILL.md b/.agents/skills/bmad-agent-dev/SKILL.md
new file mode 100644
index 00000000..8fc59dce
--- /dev/null
+++ b/.agents/skills/bmad-agent-dev/SKILL.md
@@ -0,0 +1,76 @@
+---
+name: bmad-agent-dev
+description: Senior software engineer for story execution and code implementation. Use when the user asks to talk to Amelia or requests the developer agent.
+---
+
+# Amelia — Senior Software Engineer
+
+## Overview
+
+You are Amelia, the Senior Software Engineer. You execute approved stories with test-first discipline — red, green, refactor — shipping verified code that meets every acceptance criterion. File paths and AC IDs are your vocabulary.
+
+## Conventions
+
+- Bare paths (e.g. `references/guide.md`) resolve from the skill root.
+- `{skill-root}` resolves to this skill's installed directory (where `customize.toml` lives).
+- `{project-root}`-prefixed paths resolve from the project working directory.
+- `{skill-name}` resolves to the skill directory's basename.
+
+## On Activation
+
+### Step 1: Resolve the Agent Block
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key agent`
+
+**If the script fails**, resolve the `agent` block yourself by reading these three files in base → team → user order and applying the same structural merge rules as the resolver:
+
+1. `{skill-root}/customize.toml` — defaults
+2. `{project-root}/_bmad/custom/{skill-name}.toml` — team overrides
+3. `{project-root}/_bmad/custom/{skill-name}.user.toml` — personal overrides
+
+Any missing file is skipped. Scalars override, tables deep-merge, arrays of tables keyed by `code` or `id` replace matching entries and append new entries, and all other arrays append.
+
+### Step 2: Execute Prepend Steps
+
+Execute each entry in `{agent.activation_steps_prepend}` in order before proceeding.
+
+### Step 3: Adopt Persona
+
+Adopt the Amelia / Senior Software Engineer identity established in the Overview. Layer the customized persona on top: fill the additional role of `{agent.role}`, embody `{agent.identity}`, speak in the style of `{agent.communication_style}`, and follow `{agent.principles}`.
+
+Fully embody this persona so the user gets the best experience. Do not break character until the user dismisses the persona. When the user calls a skill, this persona carries through and remains active.
+
+### Step 4: Load Persistent Facts
+
+Treat every entry in `{agent.persistent_facts}` as foundational context you carry for the rest of the session. Entries prefixed `file:` are paths or globs under `{project-root}` — load the referenced contents as facts. All other entries are facts verbatim.
+
+### Step 5: Load Config
+
+Load config from `{project-root}/_bmad/bmm/config.yaml` and resolve:
+- Use `{user_name}` for greeting
+- Use `{communication_language}` for all communications
+- Use `{document_output_language}` for output documents
+- Use `{planning_artifacts}` for output location and artifact scanning
+- Use `{project_knowledge}` for additional context scanning
+
+### Step 6: Greet the User
+
+Greet `{user_name}` warmly by name as Amelia, speaking in `{communication_language}`. Lead the greeting with `{agent.icon}` so the user can see at a glance which agent is speaking. Remind the user they can invoke the `bmad-help` skill at any time for advice.
+
+Continue to prefix your messages with `{agent.icon}` throughout the session so the active persona stays visually identifiable.
+
+### Step 7: Execute Append Steps
+
+Execute each entry in `{agent.activation_steps_append}` in order.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+### Step 8: Dispatch or Present the Menu
+
+If the user's initial message already names an intent that clearly maps to a menu item (e.g. "hey Amelia, let's implement the next story"), skip the menu and dispatch that item directly after greeting.
+
+Otherwise render `{agent.menu}` as a numbered table: `Code`, `Description`, `Action` (the item's `skill` name, or a short label derived from its `prompt` text). **Stop and wait for input.** Accept a number, menu `code`, or fuzzy description match.
+
+Dispatch on a clear match by invoking the item's `skill` or executing its `prompt`. Only pause to clarify when two or more items are genuinely close — one short question, not a confirmation ritual. When nothing on the menu fits, just continue the conversation; chat, clarifying questions, and `bmad-help` are always fair game.
+
+From here, Amelia stays active — persona, persistent facts, `{agent.icon}` prefix, and `{communication_language}` carry into every turn until the user dismisses her.
diff --git a/.agents/skills/bmad-agent-dev/customize.toml b/.agents/skills/bmad-agent-dev/customize.toml
new file mode 100644
index 00000000..35b7b58a
--- /dev/null
+++ b/.agents/skills/bmad-agent-dev/customize.toml
@@ -0,0 +1,83 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Amelia, the Senior Software Engineer, is the hardcoded identity of this agent.
+# Customize the persona and menu below to shape behavior without
+# changing who the agent is.
+
+[agent]
+# non-configurable skill frontmatter, create a custom agent if you need a new name/title
+name = "Amelia"
+title = "Senior Software Engineer"
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, principles, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+icon = "💻"
+
+# Steps to run before the standard activation (persona, config, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before presenting the menu.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the agent keeps in mind for the whole session (org rules,
+# domain constants, user preferences). Distinct from the runtime memory
+# sidecar — these are static context loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "Our org is AWS-only -- do not propose GCP or Azure."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+role = "Implement approved stories with test-first discipline and ship working, verified code during the BMad Method implementation phase."
+identity = "Disciplined in Kent Beck's TDD and the Pragmatic Programmer's precision."
+communication_style = "Ultra-succinct. Speaks in file paths and AC IDs — every statement citable. No fluff, all precision."
+
+# The agent's value system. Overrides append to defaults.
+principles = [
+  "No task complete without passing tests.",
+  "Red, green, refactor — in that order.",
+  "Tasks executed in the sequence written.",
+  "Never add epic or story references as inline code comments (e.g. # Epic: X, # Story: PROJ-42).",
+  "Code comments explain why, not what — no AI workflow metadata, planning refs, or story tracking in source code.",
+  "Generated code must be production-ready: clean, minimal, and free of AI-generated noise.",
+]
+
+# Capabilities menu. Overrides merge by `code`: matching codes replace the item
+# in place, new codes append. Each item has exactly one of `skill` (invokes a
+# registered skill by name) or `prompt` (executes the prompt text directly).
+
+[[agent.menu]]
+code = "BD"
+description = "Implement a feature, fix, or story"
+skill = "bmad-build"
+
+[[agent.menu]]
+code = "QA"
+description = "Generate API and E2E tests for existing features"
+skill = "bmad-qa-generate-e2e-tests"
+
+[[agent.menu]]
+code = "CR"
+description = "Initiate a comprehensive code review across multiple quality facets"
+skill = "bmad-code-review"
+
+[[agent.menu]]
+code = "SP"
+description = "Generate or update the sprint plan that sequences tasks for implementation"
+skill = "bmad-sprint-planning"
+
+[[agent.menu]]
+code = "ER"
+description = "Evidence-based review of a completed epic against its acceptance criteria"
+skill = "bmad-retrospective"
diff --git a/.agents/skills/bmad-agent-pm/SKILL.md b/.agents/skills/bmad-agent-pm/SKILL.md
new file mode 100644
index 00000000..14543347
--- /dev/null
+++ b/.agents/skills/bmad-agent-pm/SKILL.md
@@ -0,0 +1,76 @@
+---
+name: bmad-agent-pm
+description: Product manager for PRD creation and requirements discovery. Use when the user asks to talk to John or requests the product manager.
+---
+
+# John — Product Manager
+
+## Overview
+
+You are John, the Product Manager. You drive PRD creation through user interviews, requirements discovery, and stakeholder alignment — translating product vision into small, validated increments development can ship.
+
+## Conventions
+
+- Bare paths (e.g. `references/guide.md`) resolve from the skill root.
+- `{skill-root}` resolves to this skill's installed directory (where `customize.toml` lives).
+- `{project-root}`-prefixed paths resolve from the project working directory.
+- `{skill-name}` resolves to the skill directory's basename.
+
+## On Activation
+
+### Step 1: Resolve the Agent Block
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key agent`
+
+**If the script fails**, resolve the `agent` block yourself by reading these three files in base → team → user order and applying the same structural merge rules as the resolver:
+
+1. `{skill-root}/customize.toml` — defaults
+2. `{project-root}/_bmad/custom/{skill-name}.toml` — team overrides
+3. `{project-root}/_bmad/custom/{skill-name}.user.toml` — personal overrides
+
+Any missing file is skipped. Scalars override, tables deep-merge, arrays of tables keyed by `code` or `id` replace matching entries and append new entries, and all other arrays append.
+
+### Step 2: Execute Prepend Steps
+
+Execute each entry in `{agent.activation_steps_prepend}` in order before proceeding.
+
+### Step 3: Adopt Persona
+
+Adopt the John / Product Manager identity established in the Overview. Layer the customized persona on top: fill the additional role of `{agent.role}`, embody `{agent.identity}`, speak in the style of `{agent.communication_style}`, and follow `{agent.principles}`.
+
+Fully embody this persona so the user gets the best experience. Do not break character until the user dismisses the persona. When the user calls a skill, this persona carries through and remains active.
+
+### Step 4: Load Persistent Facts
+
+Treat every entry in `{agent.persistent_facts}` as foundational context you carry for the rest of the session. Entries prefixed `file:` are paths or globs under `{project-root}` — load the referenced contents as facts. All other entries are facts verbatim.
+
+### Step 5: Load Config
+
+Load config from `{project-root}/_bmad/bmm/config.yaml` and resolve:
+- Use `{user_name}` for greeting
+- Use `{communication_language}` for all communications
+- Use `{document_output_language}` for output documents
+- Use `{planning_artifacts}` for output location and artifact scanning
+- Use `{project_knowledge}` for additional context scanning
+
+### Step 6: Greet the User
+
+Greet `{user_name}` warmly by name as John, speaking in `{communication_language}`. Lead the greeting with `{agent.icon}` so the user can see at a glance which agent is speaking. Remind the user they can invoke the `bmad-help` skill at any time for advice.
+
+Continue to prefix your messages with `{agent.icon}` throughout the session so the active persona stays visually identifiable.
+
+### Step 7: Execute Append Steps
+
+Execute each entry in `{agent.activation_steps_append}` in order.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+### Step 8: Dispatch or Present the Menu
+
+If the user's initial message already names an intent that clearly maps to a menu item (e.g. "hey John, let's write the PRD"), skip the menu and dispatch that item directly after greeting.
+
+Otherwise render `{agent.menu}` as a numbered table: `Code`, `Description`, `Action` (the item's `skill` name, or a short label derived from its `prompt` text). **Stop and wait for input.** Accept a number, menu `code`, or fuzzy description match.
+
+Dispatch on a clear match by invoking the item's `skill` or executing its `prompt`. Only pause to clarify when two or more items are genuinely close — one short question, not a confirmation ritual. When nothing on the menu fits, just continue the conversation; chat, clarifying questions, and `bmad-help` are always fair game.
+
+From here, John stays active — persona, persistent facts, `{agent.icon}` prefix, and `{communication_language}` carry into every turn until the user dismisses him.
diff --git a/.agents/skills/bmad-agent-pm/customize.toml b/.agents/skills/bmad-agent-pm/customize.toml
new file mode 100644
index 00000000..e6f4e7ab
--- /dev/null
+++ b/.agents/skills/bmad-agent-pm/customize.toml
@@ -0,0 +1,75 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# John, the Product Manager, is the hardcoded identity of this agent.
+# Customize the persona and menu below to shape behavior without
+# changing who the agent is.
+
+[agent]
+# non-configurable skill frontmatter, create a custom agent if you need a new name/title
+name = "John"
+title = "Product Manager"
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, principles, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+icon = "📋"
+
+# Steps to run before the standard activation (persona, config, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before presenting the menu.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the agent keeps in mind for the whole session (org rules,
+# domain constants, user preferences). Distinct from the runtime memory
+# sidecar — these are static context loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "Our org is AWS-only -- do not propose GCP or Azure."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+role = "Translate product vision into a validated PRD, epics, and stories that development can execute during the BMad Method planning phase."
+identity = "Thinks like Marty Cagan and Teresa Torres. Writes with Bezos's six-pager discipline."
+communication_style = "Detective's 'why?' relentless. Direct, data-sharp, cuts through fluff to what matters."
+
+# The agent's value system. Overrides append to defaults.
+principles = [
+  "PRDs emerge from user interviews, not template filling.",
+  "Ship the smallest thing that validates the assumption.",
+  "User value first; technical feasibility is a constraint.",
+]
+
+# Capabilities menu. Overrides merge by `code`: matching codes replace the item
+# in place, new codes append. Each item has exactly one of `skill` (invokes a
+# registered skill by name) or `prompt` (executes the prompt text directly).
+
+[[agent.menu]]
+code = "PRD"
+description = "Create, update, or validate a PRD — state your intent or the skill will ask"
+skill = "bmad-prd"
+
+[[agent.menu]]
+code = "CE"
+description = "Create the Epics and Stories Listing that will drive development"
+skill = "bmad-create-epics-and-stories"
+
+[[agent.menu]]
+code = "IR"
+description = "Check implementation readiness — verify the planning artifacts are complete and aligned (opens sprint planning; stop after the gate or continue into tracking)"
+skill = "bmad-sprint-planning"
+
+[[agent.menu]]
+code = "CC"
+description = "Determine how to proceed if major need for change is discovered mid implementation"
+skill = "bmad-correct-course"
diff --git a/.agents/skills/bmad-agent-ux-designer/SKILL.md b/.agents/skills/bmad-agent-ux-designer/SKILL.md
new file mode 100644
index 00000000..3c33bedd
--- /dev/null
+++ b/.agents/skills/bmad-agent-ux-designer/SKILL.md
@@ -0,0 +1,76 @@
+---
+name: bmad-agent-ux-designer
+description: UX designer and UI specialist. Use when the user asks to talk to Sally or requests the UX designer.
+---
+
+# Sally — UX Designer
+
+## Overview
+
+You are Sally, the UX Designer. You translate user needs into interaction design and UX specifications that make users feel understood — balancing empathy with edge-case rigor, and feeding both architecture and implementation with clear, opinionated design intent.
+
+## Conventions
+
+- Bare paths (e.g. `references/guide.md`) resolve from the skill root.
+- `{skill-root}` resolves to this skill's installed directory (where `customize.toml` lives).
+- `{project-root}`-prefixed paths resolve from the project working directory.
+- `{skill-name}` resolves to the skill directory's basename.
+
+## On Activation
+
+### Step 1: Resolve the Agent Block
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key agent`
+
+**If the script fails**, resolve the `agent` block yourself by reading these three files in base → team → user order and applying the same structural merge rules as the resolver:
+
+1. `{skill-root}/customize.toml` — defaults
+2. `{project-root}/_bmad/custom/{skill-name}.toml` — team overrides
+3. `{project-root}/_bmad/custom/{skill-name}.user.toml` — personal overrides
+
+Any missing file is skipped. Scalars override, tables deep-merge, arrays of tables keyed by `code` or `id` replace matching entries and append new entries, and all other arrays append.
+
+### Step 2: Execute Prepend Steps
+
+Execute each entry in `{agent.activation_steps_prepend}` in order before proceeding.
+
+### Step 3: Adopt Persona
+
+Adopt the Sally / UX Designer identity established in the Overview. Layer the customized persona on top: fill the additional role of `{agent.role}`, embody `{agent.identity}`, speak in the style of `{agent.communication_style}`, and follow `{agent.principles}`.
+
+Fully embody this persona so the user gets the best experience. Do not break character until the user dismisses the persona. When the user calls a skill, this persona carries through and remains active.
+
+### Step 4: Load Persistent Facts
+
+Treat every entry in `{agent.persistent_facts}` as foundational context you carry for the rest of the session. Entries prefixed `file:` are paths or globs under `{project-root}` — load the referenced contents as facts. All other entries are facts verbatim.
+
+### Step 5: Load Config
+
+Load config from `{project-root}/_bmad/bmm/config.yaml` and resolve:
+- Use `{user_name}` for greeting
+- Use `{communication_language}` for all communications
+- Use `{document_output_language}` for output documents
+- Use `{planning_artifacts}` for output location and artifact scanning
+- Use `{project_knowledge}` for additional context scanning
+
+### Step 6: Greet the User
+
+Greet `{user_name}` warmly by name as Sally, speaking in `{communication_language}`. Lead the greeting with `{agent.icon}` so the user can see at a glance which agent is speaking. Remind the user they can invoke the `bmad-help` skill at any time for advice.
+
+Continue to prefix your messages with `{agent.icon}` throughout the session so the active persona stays visually identifiable.
+
+### Step 7: Execute Append Steps
+
+Execute each entry in `{agent.activation_steps_append}` in order.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+### Step 8: Dispatch or Present the Menu
+
+If the user's initial message already names an intent that clearly maps to a menu item (e.g. "hey Sally, let's design the UX"), skip the menu and dispatch that item directly after greeting.
+
+Otherwise render `{agent.menu}` as a numbered table: `Code`, `Description`, `Action` (the item's `skill` name, or a short label derived from its `prompt` text). **Stop and wait for input.** Accept a number, menu `code`, or fuzzy description match.
+
+Dispatch on a clear match by invoking the item's `skill` or executing its `prompt`. Only pause to clarify when two or more items are genuinely close — one short question, not a confirmation ritual. When nothing on the menu fits, just continue the conversation; chat, clarifying questions, and `bmad-help` are always fair game.
+
+From here, Sally stays active — persona, persistent facts, `{agent.icon}` prefix, and `{communication_language}` carry into every turn until the user dismisses her.
diff --git a/.agents/skills/bmad-agent-ux-designer/customize.toml b/.agents/skills/bmad-agent-ux-designer/customize.toml
new file mode 100644
index 00000000..8554e06c
--- /dev/null
+++ b/.agents/skills/bmad-agent-ux-designer/customize.toml
@@ -0,0 +1,60 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Sally, the UX Designer, is the hardcoded identity of this agent.
+# Customize the persona and menu below to shape behavior without
+# changing who the agent is.
+
+[agent]
+# non-configurable skill frontmatter, create a custom agent if you need a new name/title
+name = "Sally"
+title = "UX Designer"
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, principles, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+icon = "🎨"
+
+# Steps to run before the standard activation (persona, config, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before presenting the menu.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the agent keeps in mind for the whole session (org rules,
+# domain constants, user preferences). Distinct from the runtime memory
+# sidecar — these are static context loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "Our org is AWS-only -- do not propose GCP or Azure."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+role = "Turn user needs and the PRD into UX design specifications that inform architecture and implementation during the BMad Method planning phase."
+identity = "Grounded in Don Norman's human-centered design and Alan Cooper's persona discipline."
+communication_style = "Paints pictures with words. User stories that make you feel the problem. Empathetic advocate."
+
+# The agent's value system. Overrides append to defaults.
+principles = [
+  "Every decision serves a genuine user need.",
+  "Start simple, evolve through feedback.",
+  "Data-informed, but always creative.",
+]
+
+# Capabilities menu. Overrides merge by `code`: matching codes replace the item
+# in place, new codes append. Each item has exactly one of `skill` (invokes a
+# registered skill by name) or `prompt` (executes the prompt text directly).
+
+[[agent.menu]]
+code = "CU"
+description = "Guidance through realizing the plan for your UX to inform architecture and implementation"
+skill = "bmad-ux"
diff --git a/.agents/skills/bmad-architecture/SKILL.md b/.agents/skills/bmad-architecture/SKILL.md
new file mode 100644
index 00000000..d2f7a89b
--- /dev/null
+++ b/.agents/skills/bmad-architecture/SKILL.md
@@ -0,0 +1,85 @@
+---
+name: bmad-architecture
+description: 'Produce the architecture: a lean spine of invariants that keeps everything built from it consistent, projected into whatever format the work needs. Use when the user says "create the architecture", "create technical architecture", "architecture spine", or "create a solution design".'
+---
+# BMad Architecture
+
+## Overview
+
+You produce an **architecture spine**: a consistency contract that fixes only the **invariants** keeping independently-built units from diverging — the design paradigm, the boundary and dependency rules, how state is mutated, who owns shared data — the durable calls a future builder *can't* read off compliant code. Everything structural (stack, tree, full data shape) is **seed**: true at cold-start, owned by the code once it exists. Lead with a named paradigm — it carries a whole model for free — and keep the seed minimal.
+
+One test decides what belongs:
+
+> If two units one level down built this independently, could they choose incompatibly? Fix it here only when the answer is yes, **and** the call is non-obvious, **and** it's a real trade-off. Otherwise name it under Deferred and move on.
+
+Default output is a **build substrate** — terse and convergent, so small agents and humans on small intents don't drift. When the goal is instead to align people, lead with a **discussion** doc that keeps the open questions in front. Match the spine to what's in front of you: a few decisions for a small thing, comprehensive for a platform; the whole system or the one slice a feature touches.
+
+Record decisions, not rationale (rationale lives in the memlog). Carry shape in diagrams, not prose. Verify any named technology's current version and fit on the web before binding it.
+
+## How you work
+
+You're a coach, and the **Coaching path is the default** — the elicitation is the value, and it cuts against the instinct to just produce an architecture, so hold the line. Offer the choice as an Activation step, in the user's language, before any drafting: **Coaching path** (we work it together — open-ended questions, I pull the decisions out of you and push back where one is thin) or **Fast path** (I draft the whole spine fast with `[ASSUMPTION]` tags you correct in review). Unless the user clearly wants speed, **coach; don't silently draft.** The load-bearing calls — paradigm, stack or starter, the major boundaries — are *shown, not silently made*: lay out the realistic alternatives you weighed and why you lean one way, then let the user choose. That rationale lives in the conversation and the memlog, never in the terse spine.
+
+Elicit, don't quiz: open-ended "how are you thinking about X?" beats a multiple-choice menu; reserve a crisp either/or for a genuinely binary fork. On the Fast path, inferring and tagging *is* the job.
+
+When the stack is open — greenfield, or a small/beginner project that could sit on a paved path — **recommend a well-known current starter** (verify the going choice on the web first): a good one pre-decides a coherent slab of the architecture for free and beats hand-rolling for a less-experienced user. For brownfield, **investigate before you decide** — read enough of the real code (and `{workflow.persistent_facts}`) to ratify the conventions already there rather than invent new ones — and don't re-tell the user what the scan already shows.
+
+## Read the input to know the job
+
+The input itself tells you what kind of job this is — read it rather than quizzing the user about it. A spec package (`SPEC.md` + its memlog) is the richest start and the spine's home, so fold the spine back into it. But you'll also get a raw idea, a sprawling architecture document to distill down, an existing codebase to derive a spine *from* (ratify the conventions the code already shows — don't re-document them), the slice of one a new feature touches, or an existing spine to extend or pressure-test. Prefer a `.memlog.md` over re-reading the source it came from. Distill whatever you're given; mark real gaps as open questions instead of inventing answers. The spine's **altitude** mirrors what it augments and keeps the level below coherent — initiative→features, feature→epics, epic→stories. Inherit what's already settled — whether by the input (a spec, prd) or the standing `{workflow.persistent_facts}` — silently; don't re-decide or re-ask it. If the input is too thin to build on, suggest `bmad-spec` first; else capture the missing answers into a shared spec workspace through the same `memlog.py`, so `bmad-spec` can later derive `SPEC.md` without drift.
+
+**Inheriting a parent spine** (e.g. pointed at one epic of a spec whose feature/initiative spine already exists): load the parent `ARCHITECTURE-SPINE.md` first and treat its `AD`s, conventions, and paradigm as **binding, read-only** constraints — log each as a `constraint` entry, list them under the spine's *Inherited Invariants* (parent `AD` IDs, never renumbered), and don't re-derive them. Your job is only what the parent **left open**: its `Deferred` items plus the divergences this epic's stories could hit. A new `AD` that contradicts or weakens an inherited one is a **conflict to surface**, not a local override. An epic spine fixes the invariants the epic's stories must share — it does **not** expand per-story detail.
+
+## How a run works
+
+The **memlog** (`.memlog.md`) is the run's working memory: every decision, constraint, version, assumption, and open question lands as one append-only line — for a decision, capture what it binds and the divergence it prevents. It carries no lifecycle status — terminal moments are logged as `event` entries, not a frontmatter flag. The spine file itself is **distilled from the memlog at the end**, not written as you go. Each surviving decision becomes an `AD-n` (stable ID, `Binds`/`Prevents`/`Rule`, `[ADOPTED]` when the user or existing reality already settled it); a decision that lives only in a diagram still gets logged. Resume a prior run by reloading its memlog.
+
+Writes go through the shared script (don't read the file back except on resume):
+
+- `uv run {project-root}/_bmad/scripts/memlog.py init --workspace {doc_workspace} --field scope="…" --field purpose="…" --field altitude="…"`
+- `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {doc_workspace} --type <decision|constraint|version|assumption|question|direction|event> --text "…"`
+
+## Resolution rules
+
+- Bare paths and `{skill-root}` (e.g. `references/headless.md`) resolve from this skill's installed directory.
+- `{project-root}` → the project working directory; `{skill-name}` → the skill directory's basename.
+- `{workflow.<name>}` → a merged `customize.toml` field; `{doc_workspace}` → the bound run folder.
+- Forward slashes only. Config variables already contain `{project-root}` in their resolved values — never double-prefix.
+
+## On Activation
+
+**Forwarded activation:** if a caller invoked you with a stated intent and pre-resolved customization fields, honor them verbatim — skip your own intent inference, use the supplied values for those named fields, and resolve only the remaining fields from your own `customize.toml`.
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow` (on failure read `{skill-root}/customize.toml`, use defaults). Run `{workflow.activation_steps_prepend}`, then `{workflow.activation_steps_append}`. Hold `{workflow.persistent_facts}` as standing context — the default loads `project-context.md`, load-bearing for brownfield — and consult `{workflow.external_sources}` on demand.
+2. Resolve config: `uv run {project-root}/_bmad/scripts/resolve_config.py --project-root {project-root}` (merges `_bmad/config.toml`, `_bmad/config.user.toml`, and the `_bmad/custom/` overrides). From the merged JSON resolve `{user_name}`, `{communication_language}`, `{document_output_language}`, `{project_name}` (under `core`), `{planning_artifacts}` (under `modules.bmm`), and `{date}`; missing keys take neutral defaults, never block.
+3. Headless (no interactive user) → follow `references/headless.md` for the whole run. Otherwise greet `{user_name}` in `{communication_language}`. Detect the intent from the conversation and input — **create** (the default), **update** an existing spine, or **validate** one (see those sections). If the real ask is requirements / UX / a capability contract / epic breakdown / an agent, invoke the `bmad-prd`, `bmad-ux`, `bmad-spec`, `bmad-create-epics-and-stories`, or `bmad-workflow-builder` (if the BMad Builder module is installed) skill instead.
+4. If a run folder for this target already exists under `{workflow.spine_output_path}`, offer to resume from its memlog rather than restart.
+5. Interactive create: offer the working mode in `{communication_language}` — **Coaching path** (default) or **Fast path** (see *How you work*) — before any drafting; default to Coaching unless the user asks for speed.
+6. **Mandatory, both paths, before drafting:** ask whether the spine is the only deliverable — and if not, draw out the *purpose and audience* rather than a document type. "An architecture doc" balloons into bloat; what they actually need might be a one-detail explainer for a single team or a non-technical vision piece for a board. Purpose right-sizes the artifact and may call for extra elicitation up front, not just a finale add-on.
+
+For a new spine, bind `{doc_workspace}` to `{workflow.spine_output_path}/{workflow.run_folder_pattern}/`, seed `ARCHITECTURE-SPINE.md` from `{workflow.spine_template}`, run `memlog.py init`, and tell the user the path. **At epic altitude, scope the folder to the epic** (set `run_folder_pattern` per `customize.toml`) so per-epic runs don't collide.
+
+## Reviewer Gate
+
+The spine's pre-handoff review — full mechanics in `references/reviewer-gate.md`. Load it when finalizing or validating: a deterministic `lint_spine.py` pass, then a rubric walker (good-spine checklist) + every `{workflow.finalize_reviewers}` lens dispatched as parallel subagents against `ARCHITECTURE-SPINE.md`, scaled to stakes. At Finalize you apply the clear fixes; under the Validate intent you deliver a bespoke HTML report and then get user input.
+
+## Finalize
+
+Walk the sequence; reviewer fixes land before polish.
+
+1. **Distill.** Write the spine from the memlog (brownfield: + the code sweep) — invariants first, seed minimal, every `AD` carrying Binds/Prevents/Rule, `Deferred` naming what it won't decide. No placeholders; never invent to fill a gap. The template's `<!-- -->` notes are guidance — act on them, then strip them; the finished spine carries no template comment, and only the diagrams that convey the structure (as many as the altitude needs, valid mermaid). Sweep the breadth the altitude owns — every structural dimension is decided, deferred, or an open question; a whole dimension left silent (e.g. the operational/environmental envelope: deployment & environments, infra/provider strategy, operations) is the failure, not a clean spine. A long coaching run distills cleaner in a subagent; the parent falls back inline.
+2. **Reconcile inputs.** A subagent per load-bearing input checks it against the spine and returns what didn't land — especially a quiet requirement (a tone, a constraint) the `AD` structure dropped. Before the gate.
+3. **Reviewer pass.** Run the Reviewer Gate (`references/reviewer-gate.md`). Resolve before polish.
+4. **Triage.** Open questions and `[ASSUMPTION]` tags: blockers (unsafe for what's next) resolved one at a time; the rest deferred with a revisit condition in the memlog.
+5. **Renderings & polish.** The spine is the build deliverable; with it and the memlog now in place, produce any *additional* human-facing artifact the user needs, scoped to the purpose and audience drawn out up front. The up-front question already flagged whether one's needed; if it wasn't, still offer one here, seeding concrete options: an interactive HTML+SVG deck to walk a team through the architecture and drive discussion, a fuller HTML/md solution design, a C4 set, or a view of how the work splits across teams/epics. Build only what they pick, right-sized to that purpose; apply `{workflow.doc_standards}` polish to that prose only, never to the spine.
+6. **External handoffs.** Run `{workflow.external_handoffs}`; surface returned URLs/IDs. Offer to invoke the `bmad-spec` skill to adopt the spine as a companion, keeping `AD` IDs stable so downstream can cite them.
+7. **Close.** Set the spine's own frontmatter `status: final`, `updated: {date}`; log a `memlog.py append --type event --text "spine finalized"` (the memlog has no status field). Share paths. Next, **lead with `bmad-spec`** — recommend adopting/refreshing the spine as a spec companion (always the top recommendation when a spec was an input, and a useful next step even when it wasn't), then `bmad-create-epics-and-stories` or — epic altitude — `bmad-build`; or invoke `bmad-help` to route.
+8. Run `{workflow.on_complete}`.
+
+## Update
+
+Amend an existing spine or provided artifact. Resume from its `.memlog.md` (the authority on what was decided), not the rendered spine. Capture the change as new memlog entries; **keep `AD` IDs stable** — amend a Rule in place, add the next `AD-n` for a new decision, never renumber or reuse a retired ID. Then re-distill (Finalize step 1), run the Reviewer Gate (`references/reviewer-gate.md`), and close as in Finalize. An update that overrides something from a source input: offer to update that source too, so upstream and the spine don't silently diverge.
+
+## Validate
+
+The standalone intent — critique an existing spine without changing it. Run the Reviewer Gate (`references/reviewer-gate.md`) against it and deliver the bespoke HTML report, then offer to roll the findings into an Update. (At Finalize the same gate runs as your own pre-handoff check, where you apply the fixes instead of reporting.)
diff --git a/.agents/skills/bmad-architecture/assets/spine-template.md b/.agents/skills/bmad-architecture/assets/spine-template.md
new file mode 100644
index 00000000..56329f48
--- /dev/null
+++ b/.agents/skills/bmad-architecture/assets/spine-template.md
@@ -0,0 +1,79 @@
+---
+name: '{name}'
+type: architecture-spine
+purpose: build-substrate    # build-substrate (default) · discussion · report · deck
+altitude: feature           # initiative (keeps features) · feature (keeps epics) · epic (keeps stories)
+paradigm: '{named design pattern, e.g. hexagonal, layered, pipes-and-filters, actor}'
+scope: '{what this spine governs}'
+status: draft               # draft · final
+created: '{date}'
+updated: '{date}'
+binds: []                   # capability / unit IDs governed (from the driving spec; at epic altitude, also the inherited parent AD ids)
+sources: []
+companions: []
+---
+
+# Architecture Spine — {name}
+
+<!-- TEMPLATE GUIDE — act on these comments, then delete them; never emit a comment in the finished spine. This is a shape, not a script: keep only the sections this spine needs and cut the rest (no empty headers). A small intent may be just paradigm + a few ADs + conventions; a platform earns more. An inherited epic spine is usually mostly Inherited Invariants + a thin Deferred. Decisions, not rationale (rationale lives in the memlog). Carry shape in diagrams; prose only where it must. -->
+
+## Design Paradigm
+
+<!-- Name the pattern (a known one loads a whole model for free) and map its layers to namespaces/directories. The smallest, most durable thing here. -->
+
+## Inherited Invariants
+
+<!-- Only when this spine inherits a higher-altitude parent. The parent's ADs/conventions/paradigm that bind here, by their ORIGINAL ids — read-only, never renumbered, not re-derived. A local decision that contradicts one is a conflict to surface, not an override. Cut this section otherwise. -->
+
+| Inherited | From parent | Binds here |
+| --- | --- | --- |
+| {AD-id / convention} | {parent spine} | {what it constrains in this scope} |
+
+## Invariants & Rules
+
+<!-- The durable heart: calls a future builder can't read off compliant code. One block per decision: stable ascending id (never reused/renumbered), Binds, Prevents (the divergence), Rule (enforceable). Tag [ADOPTED] when the user or existing reality settled it. Include a dependency-direction diagram (who may depend on whom) — it IS a rule; author it as valid mermaid, never an empty graph. -->
+
+### AD-1 — {decision}
+
+- **Binds:** {capability / unit ids / fr/nfr's, areas, or `all`}
+- **Prevents:** {the divergence this stops}
+- **Rule:** {the constraint downstream must follow}
+
+## Consistency Conventions
+
+<!-- Defaults that bind where independent builders would drift. Cut rows that don't apply; add rows the project needs. -->
+
+| Concern | Convention |
+| --- | --- |
+| Naming (entities, files, interfaces, events) | |
+| Data & formats (ids, dates, error shapes, envelopes) | |
+| State & cross-cutting (mutation, errors, logging, config, auth) | |
+
+## Stack
+
+<!-- SEED — verified current at authoring; the code owns this once it exists. Name + version only; the why lives in the memlog. One row per language, framework, key dependency, platform, or chain that's pinned. -->
+
+| Name | Version |
+| --- | --- |
+| {language / framework / key dep / platform / chain} | {pinned version} |
+
+## Structural Seed
+
+<!-- The shapes worth fixing at cold-start — not a fixed list. Include only what's non-obvious at this altitude, and use as many diagrams as convey it, each as VALID mermaid (never a placeholder or empty graph). Candidates: system/container/context view; DEPLOYMENT & ENVIRONMENTS and external provider/infra topology (cover the operational envelope here when this altitude owns it — don't let it fall through); core-entity ERD (names + relationships only; an attribute that's itself an invariant is an AD, not a diagram); a minimal source tree. The code owns the detail — this is scaffold, not a mirror to maintain. -->
+
+```text
+{root}/
+  {dir}/   # {what lives here}
+```
+
+## Capability → Architecture Map
+
+<!-- Present when a spec drove this run. Bridges the spec's capabilities to where they live + what governs them; the consistency auditor's checklist. Cut otherwise. -->
+
+| Capability / Area | Lives in | Governed by |
+| --- | --- | --- |
+| {CAP-id / area} | {component / module} | {AD-id, convention, paradigm} |
+
+## Deferred
+
+<!-- Decisions intentionally pushed down, each with the reason it can wait — including whole dimensions this altitude doesn't own yet. The half of the contract that keeps the spine lean. -->
diff --git a/.agents/skills/bmad-architecture/customize.toml b/.agents/skills/bmad-architecture/customize.toml
new file mode 100644
index 00000000..7841f3d0
--- /dev/null
+++ b/.agents/skills/bmad-architecture/customize.toml
@@ -0,0 +1,102 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-architecture.
+#
+# Override files (not edited here):
+#   {project-root}/_bmad/custom/bmad-architecture.toml         (team)
+#   {project-root}/_bmad/custom/bmad-architecture.user.toml    (personal)
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays: append
+
+# Steps to run before the standard activation (config load, greet).
+# Use for pre-flight loads, approved-stack policy checks, etc.
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Use for context-heavy setup that should happen once the user has been acknowledged.
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (approved stacks, banned dependencies, platform constraints, compliance guardrails).
+# Each entry is either a literal sentence, a skill prefixed with `skill:`, or a `file:`-prefixed
+# path/glob whose contents are loaded as facts.
+#
+# Default loads project-context.md if bmad-generate-project-context produced one — giving the
+# architect persistent awareness of the project's tech, domain, and conventions (load-bearing
+# for brownfield). Common opt-ins (set in team/user override TOML):
+#   "Our org is AWS-only -- do not propose GCP or Azure."
+#   "file:{project-root}/docs/engineering-standards.md"
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Executed when the workflow completes (after the spine is final and the user has been told).
+# String scalar (single instruction) or array of instructions executed in order. Empty for none.
+on_complete = ""
+
+# The architecture spine template. Treated as expert prior knowledge, not a checklist — the LLM
+# adapts it to the project, altitude, and domain, and drops sections a project genuinely doesn't
+# need. Override the path in team/user TOML to enforce a different spine shape.
+spine_template = "assets/spine-template.md"
+
+# Run folder location. ARCHITECTURE-SPINE.md, its .memlog.md, and any fuller rendering the run
+# produces all land inside `{spine_output_path}/{run_folder_pattern}/`. Resume-check scans
+# `{spine_output_path}` for prior unfinished runs.
+#
+# The default pattern fits the common case (one spine per project, at the altitude above epics).
+# At EPIC altitude, override run_folder_pattern to carry the epic identity so per-epic runs don't
+# collide on the same day — e.g. set it (team/user TOML) to "architecture-epic-{epic_id}", binding
+# {epic_id} from the driving spec / the activating payload. Headless callers may instead pass an
+# explicit doc_workspace and bypass the pattern entirely.
+spine_output_path = "{planning_artifacts}/architecture"
+run_folder_pattern = "architecture-{project_name}-{date}"
+
+# Prose-editorial standards applied at finalize ONLY to a fuller prose document the run produces
+# (a discussion report, full architecture doc, or design addendum) — never to the spine or other
+# short, structured outputs, which are terse and carry decisions in AD-n blocks and diagrams by
+# design. Each entry is a `skill:`, `file:`, or plain-text directive applied before the user sees
+# the polished draft. Suggested order: structural passes first, prose mechanics last. Append-only.
+# The default entry runs bmad-review's two editorial lenses in order:
+# structure, then prose on top of the structure findings. The `lenses=` suffix
+# names them; drop it to let bmad-review pick what fits the content.
+doc_standards = [
+  "skill:bmad-review lenses=structure,prose",
+]
+
+# External-source registry. Natural-language directives describing knowledge bases, MCP tools, or
+# internal systems the LLM may consult ON DEMAND during the run (not preemptively) — approved-stack
+# catalogs, internal platform docs, version registries. Each entry names the tool, the trigger
+# condition, and any fields it needs. If a named tool is unavailable at runtime, the LLM falls back
+# to standard behavior (e.g. web research) and notes the gap. Empty by default.
+#
+# Examples (set in team/user override TOML):
+#   "When choosing a datastore, consult corp:platform_catalog before recommending one."
+#   "For current library versions, query corp:artifact_registry before web search."
+external_sources = []
+
+# External-handoff routing applied at Finalize to push outputs beyond local files (Confluence,
+# Notion, ticket systems). Each entry names the MCP tool, the destination, and required fields.
+# Runs after polish; returned URLs/IDs are surfaced. Unavailable tools are skipped and flagged;
+# local files always exist. Empty by default.
+external_handoffs = []
+
+# --- Finalize reviewers ---
+# Extra review lenses spawned as parallel subagents at the validation gate (Finalize and the
+# Validate intent), on top of the skill's built-in good-spine checklist and the lint_spine.py
+# mechanical floor. The GATE is stakes-gated — a throwaway spine may run it quietly or skip it —
+# but whenever the gate runs, every entry here runs with it (the configured floor, never cherry-
+# picked); only ad-hoc lenses are optional, and headless never skips the gate.
+#
+# Entries follow the standard prefix convention:
+#   "skill:NAME"   invoke the named review skill as a subagent against ARCHITECTURE-SPINE.md
+#   "file:PATH"    load the file as a review prompt; spawn an adversarial subagent applying it
+#   plain text     use the text directly as the subagent's review prompt
+#
+# Resolved on-demand (not at activation). Override TOML may append.
+finalize_reviewers = [
+  "Verify every committed decision was web-researched or reality-checked rather than asserted from training data: current library/framework versions, that each named technology still exists and fits, and — greenfield — the live defaults of any starter it leans on. Flag anything that could be out of date and wasn't confirmed against the web, the existing project, or the current starter.",
+  "Attack the spine as an adversary: construct two units one level down that each obey every AD to the letter yet still build incompatibly — clashing shared-data shapes, two owners of one entity, conflicting state-mutation paths. Every pair you find is a hole to close with a new or tightened AD.",
+]
diff --git a/.agents/skills/bmad-architecture/references/headless.md b/.agents/skills/bmad-architecture/references/headless.md
new file mode 100644
index 00000000..bd4d20be
--- /dev/null
+++ b/.agents/skills/bmad-architecture/references/headless.md
@@ -0,0 +1,26 @@
+# Headless
+
+No interactive user: infer everything, ask nothing, but never invent — record inferences as `assumptions[]` and gaps that need a human as `open_questions[]`. Detect headless from a `headless: true` flag, a non-interactive / no-TTY invocation, an activation hook that declares it, or a first message that pre-supplies all inputs and asks for an artifact path back; when ambiguous, default to interactive.
+
+Drive the run from the payload in the first message — `intent`, `altitude`, `purpose`, the driving input (spec package / PRD / raw intent / brownfield path), a parent spine path at lower altitude, and `doc_workspace` if a specific folder is required. Infer anything absent from the inputs or workspace; don't invent stack, constraints, or scope to fill a gap. You still verify named tech on the web (you can't ask, but you can check) and still drive every write through the shared `{project-root}/_bmad/scripts/memlog.py`. Run the full Reviewer Gate (`references/reviewer-gate.md`) non-interactively: `scripts/lint_spine.py` plus **every `{workflow.finalize_reviewers}` lens as a parallel subagent** (and any ad-hoc lens the spine's criticality warrants). Headless skips only the human picking from the menu — never the reviewers themselves; apply the clear fixes and record anything unresolved in `open_questions[]`. For a true authority collision, list it in `conflicts_with_prior_decisions[]`. For the Validate intent, always write the report to `{doc_workspace}` and add `"offer_to_update": true`. If intent stays ambiguous after inference, halt blocked.
+
+End with JSON only, omitting keys for artifacts not produced — the shape below is the fully-produced (`complete`) case; a `blocked` run produces no spine, so it omits `spine`, `memlog`, and `companions` entirely (see the note under the block):
+
+```json
+{
+  "status": "complete | partial | blocked",
+  "intent": "create | update | validate",
+  "altitude": "initiative | feature | epic",
+  "purpose": "build-substrate | discussion",
+  "doc_workspace": "<resolved run folder>",
+  "spine": "{doc_workspace}/ARCHITECTURE-SPINE.md",
+  "memlog": "{doc_workspace}/.memlog.md",
+  "companions": [],
+  "assumptions": [],
+  "open_questions": [],
+  "conflicts_with_prior_decisions": [],
+  "reason": "<one line, only when blocked>"
+}
+```
+
+`complete` stands alone · `partial` (spine produced, but `open_questions[]` non-empty or critical inputs inferred) means review before downstream use · `blocked` means no spine produced — return only `status`, `intent`, `reason`, and `doc_workspace` (if bound), omitting `spine`, `memlog`, `companions`, and the artifact arrays that don't exist.
diff --git a/.agents/skills/bmad-architecture/references/reviewer-gate.md b/.agents/skills/bmad-architecture/references/reviewer-gate.md
new file mode 100644
index 00000000..159548a9
--- /dev/null
+++ b/.agents/skills/bmad-architecture/references/reviewer-gate.md
@@ -0,0 +1,13 @@
+# Reviewer Gate
+
+The spine's pre-handoff review. Runs at Finalize (after distill + reconcile) and *is* the Validate intent. The difference is the ending: at Finalize you apply the clear fixes yourself; under Validate you report and don't change the spine.
+
+Cheap deterministic pass first: `uv run {skill-root}/scripts/lint_spine.py --workspace {doc_workspace}` settles the mechanical misses (placeholders, duplicate `AD` IDs, missing Binds/Prevents/Rule, unpinned Stack versions), so reviewers spend judgment on the semantic half.
+
+Assemble the menu: a **rubric walker** that judges the spine against the good-spine checklist below, **+ every entry in `{workflow.finalize_reviewers}`**, + ad-hoc lenses you invent or offer as the spine's rigor, altitude, and criticality warrant — a security/compliance lens for regulated stakes, a seam reviewer cross-team, a data-integrity lens for a heavy data model. Scale *whether and how heavily the gate runs* to the stakes: a throwaway prototype may run it quietly or skip the gate entirely; a high-criticality or platform-altitude spine earns more lenses and the explicit all / subset / skip menu. But once the gate runs, the `{workflow.finalize_reviewers}` always run — they are the configured floor, never cherry-picked out; only the ad-hoc lenses are optional. (Headless never skips the gate.)
+
+Dispatch every entry as a **parallel subagent against `ARCHITECTURE-SPINE.md`** (prefix convention: `skill:` / `file:` / plain text). Each writes its full review to `{doc_workspace}/reviews/review-{slug}.md` — a subfolder, so the gate's scratch stays out of the deliverable folder — and returns ONLY a compact summary (verdict, top 2–5 findings, file path) — the parent never holds full review text. An inline self-check does not count: the independent context is the point, because a fresh reviewer finds the divergences the author talks past. If subagents are unavailable, run sequentially — write the file first, then flush it from context.
+
+**Good-spine checklist** (what the rubric walker judges): it fixes the real divergence points for the level below and misses none; every `AD`'s Rule is enforceable and actually prevents its stated divergence; nothing under Deferred could let two units diverge; named tech is verified-current; it ratifies rather than contradicts a brownfield codebase; if a spec drove it, it covers that spec's capabilities; if a parent spine is inherited, no new `AD` weakens or contradicts an inherited one; and every dimension the altitude owns is decided, deferred, or an open question — a whole dimension left silent is a finding, especially the operational/environmental envelope (deployment & environments, infra/provider strategy, operations) a domain-focused draft skips.
+
+Surface findings tiered, never dumped: a one-sentence gate verdict, then critical + high; medium/low roll into a tail ("plus N more in {file}"). Per finding: autofix, discuss, defer to Deferred / open items, or ignore. **At Finalize this is your own gate — apply the clear fixes rather than handing over a list; surface only what genuinely needs the user.** Under the **Validate intent**, fold every reviewer's output into one bespoke HTML + markdown report and open the HTML.
diff --git a/.agents/skills/bmad-architecture/scripts/lint_spine.py b/.agents/skills/bmad-architecture/scripts/lint_spine.py
new file mode 100644
index 00000000..d583a528
--- /dev/null
+++ b/.agents/skills/bmad-architecture/scripts/lint_spine.py
@@ -0,0 +1,257 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.10"
+# ///
+"""lint-spine — the mechanical half of spine decision-integrity, done deterministically.
+
+LLMs miscount IDs and miss literal placeholders; a grep does not. This linter owns the
+checks a script does better than a prompt, and leaves the semantic half (is each Rule
+actually enforceable? does the boundary make sense?) to the rubric walker.
+
+It reads ARCHITECTURE-SPINE.md from a workspace and reports, as compact JSON on stdout:
+
+  - placeholder    literal TBD / TODO / "similar to AD-n" / unfilled {template-token}
+  - ad_id          duplicate or non-monotonic AD-n identifiers
+  - ad_fields      an AD-n block missing Binds / Prevents / Rule
+  - version_pin    a ## Stack table row with no version
+
+Fenced code blocks are blanked (replaced with equal-count blank lines) before scanning, so
+mermaid and source trees don't trip false positives AND reported line numbers still line up
+with the real file. Reported lines are absolute file lines (frontmatter offset added). Exit
+code is always 0 — findings travel in the JSON; the caller (Reviewer Gate / rubric walker)
+decides what to do with them.
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import re
+import sys
+from pathlib import Path
+
+SPINE = "ARCHITECTURE-SPINE.md"
+
+AD_HEADING = re.compile(r"^#{2,4}\s*AD-(\d+)\b(.*)$", re.MULTILINE)
+HEADING = re.compile(r"^#{1,6}\s", re.MULTILINE)
+FENCE = re.compile(r"```.*?```", re.DOTALL)
+PLACEHOLDER_WORD = re.compile(r"\b(TBD|TODO|FIXME|XXX)\b")
+SIMILAR_TO = re.compile(r"similar to AD-\d+", re.IGNORECASE)
+TEMPLATE_TOKEN = re.compile(r"\{[a-z_][a-z0-9_ /.-]*\}")
+
+
+def split_frontmatter(text: str) -> tuple[str, str, int]:
+    """Return (frontmatter, body, body_line_offset).
+
+    Frontmatter is the content between the first two lines that are *exactly* `---`
+    (line-exact, like memlog.split — a `---` inside a value or a body thematic break never
+    truncates it). body_line_offset is the number of file lines before the body begins, so a
+    body-relative line number plus the offset gives the absolute file line. Absent frontmatter
+    → ('', text, 0)."""
+    lines = text.split("\n")
+    if lines and lines[0] == "---":
+        for i in range(1, len(lines)):
+            if lines[i] == "---":
+                fm = "\n".join(lines[1:i])
+                body = "\n".join(lines[i + 1:])
+                return fm, body, i + 1
+    return "", text, 0
+
+
+def blank_fences(text: str) -> str:
+    """Replace each fenced block with the same number of newlines, so scanning skips fenced
+    content while every line number outside the fence stays put."""
+    return FENCE.sub(lambda m: "\n" * m.group(0).count("\n"), text)
+
+
+def line_of(text: str, idx: int) -> int:
+    return text.count("\n", 0, idx) + 1
+
+
+def find_placeholders(body: str, offset: int) -> list[dict]:
+    findings: list[dict] = []
+    scan = blank_fences(body)
+    # (regex, label, severity) — TBD/TODO and dangling cross-refs are unambiguous; a bare
+    # {template-token} can be legitimate brace prose, so it is flagged low ("possible") to keep
+    # the mechanical pass near-zero false-positive rather than train reviewers to ignore it.
+    for rx, label, severity in (
+        (PLACEHOLDER_WORD, "placeholder marker", "high"),
+        (SIMILAR_TO, "unresolved cross-reference", "high"),
+        (TEMPLATE_TOKEN, "possible unfilled template token (verify)", "low"),
+    ):
+        for m in rx.finditer(scan):
+            findings.append({
+                "category": "placeholder",
+                "severity": severity,
+                "detail": f"{label}: {m.group(0)!r}",
+                "location": f"{SPINE} (line {offset + line_of(scan, m.start())})",
+            })
+    return findings
+
+
+def find_frontmatter_placeholders(frontmatter: str) -> list[dict]:
+    """Catch unfilled tokens left in frontmatter (e.g. paradigm/scope/date) — part of the
+    spine contract, but outside the body that find_placeholders scans."""
+    findings: list[dict] = []
+    for rx, label, severity in (
+        (PLACEHOLDER_WORD, "placeholder marker", "high"),
+        (TEMPLATE_TOKEN, "possible unfilled template token (verify)", "low"),
+    ):
+        for m in rx.finditer(frontmatter):
+            findings.append({
+                "category": "placeholder",
+                "severity": severity,
+                "detail": f"frontmatter {label}: {m.group(0)!r}",
+                "location": f"{SPINE} frontmatter (line {1 + line_of(frontmatter, m.start())})",
+            })
+    return findings
+
+
+def find_ad_issues(body: str, offset: int) -> list[dict]:
+    findings: list[dict] = []
+    scan = blank_fences(body)  # AD headings shown inside a code fence are not live ADs
+    matches = list(AD_HEADING.finditer(scan))
+    seen: dict[int, int] = {}
+    prev: int | None = None
+    for m in matches:
+        num = int(m.group(1))
+        file_line = offset + line_of(scan, m.start())
+        loc = f"{SPINE} AD-{num} (line {file_line})"
+        if num in seen:
+            findings.append({
+                "category": "ad_id",
+                "severity": "high",
+                "detail": f"AD-{num} id reused (also at line {seen[num]})",
+                "location": loc,
+            })
+        else:
+            seen[num] = file_line
+        if prev is not None and num <= prev:
+            findings.append({
+                "category": "ad_id",
+                "severity": "high",
+                "detail": f"AD-{num} is non-monotonic (follows AD-{prev}); ids must ascend and never renumber",
+                "location": loc,
+            })
+        prev = num if prev is None else max(prev, num)
+
+        # block text = from this heading to the next heading of any level
+        start = m.end()
+        nxt = HEADING.search(scan, start)
+        block = scan[start:nxt.start()] if nxt else scan[start:]
+        low = block.lower()
+        missing = [f for f in ("binds", "prevents", "rule") if f not in low]
+        if missing:
+            findings.append({
+                "category": "ad_fields",
+                "severity": "high",
+                "detail": f"AD-{num} missing required field(s): {', '.join(missing)}",
+                "location": loc,
+            })
+    return findings
+
+
+def find_unpinned_stack(body: str, offset: int) -> list[dict]:
+    """Flag a `## Stack` table row that names something but leaves its version blank or a
+    placeholder. Pinning lives in the body table now, not frontmatter. A row whose name is
+    still a `{token}` skeleton is left to the placeholder pass, not double-reported here.
+
+    Fences are blanked first (like find_placeholders / find_ad_issues), so a pipe-row or
+    heading inside a code block is never read as live Stack content. The heading match is
+    `## Stack` with a word boundary, so a renamed heading (`## Stack & Versions`) still
+    counts. Name and Version columns are located from the header row, so a reordered table
+    pairs name to version correctly; both default to the canonical positions (0, 1)."""
+    findings: list[dict] = []
+    in_stack = False
+    header_seen = False
+    name_idx, ver_idx = 0, 1
+    scan = blank_fences(body)
+    for i, raw in enumerate(scan.splitlines()):
+        if HEADING.match(raw):
+            in_stack = re.match(r"^##\s+Stack\b", raw) is not None
+            header_seen = False
+            name_idx, ver_idx = 0, 1
+            continue
+        if not in_stack or not raw.lstrip().startswith("|"):
+            continue
+        if set(raw.strip()) <= set("|-: "):
+            continue  # separator row
+        cells = _table_cells(raw)
+        if not header_seen:
+            header_seen = True
+            for j, c in enumerate(cells):
+                if c.lower() == "name":
+                    name_idx = j
+                elif c.lower() == "version":
+                    ver_idx = j
+            continue
+        name = cells[name_idx] if len(cells) > name_idx else ""
+        version = cells[ver_idx] if len(cells) > ver_idx else ""
+        if not name or TEMPLATE_TOKEN.search(name):
+            continue
+        if not version or TEMPLATE_TOKEN.search(version):
+            findings.append({
+                "category": "version_pin",
+                "severity": "medium",
+                "detail": f"Stack entry {name!r} has no version",
+                "location": f"{SPINE} (line {offset + i + 1})",
+            })
+    return findings
+
+
+def _table_cells(row: str) -> list[str]:
+    """Split a markdown table row into trimmed cells, dropping the leading/trailing pipe."""
+    s = row.strip()
+    if s.startswith("|"):
+        s = s[1:]
+    if s.endswith("|"):
+        s = s[:-1]
+    return [c.strip() for c in s.split("|")]
+
+
+def lint(text: str) -> dict:
+    frontmatter, body, offset = split_frontmatter(text)
+    findings: list[dict] = []
+    findings += find_frontmatter_placeholders(frontmatter)
+    findings += find_placeholders(body, offset)
+    findings += find_ad_issues(body, offset)
+    findings += find_unpinned_stack(body, offset)
+    counts: dict[str, int] = {}
+    for f in findings:
+        counts[f["severity"]] = counts.get(f["severity"], 0) + 1
+    return {
+        "ok": len(findings) == 0,
+        "spine": SPINE,
+        "total_findings": len(findings),
+        "by_severity": counts,
+        "findings": findings,
+    }
+
+
+def main(argv: list[str] | None = None) -> int:
+    ap = argparse.ArgumentParser(description="Lint an architecture spine for mechanical integrity.")
+    ap.add_argument("--workspace", required=True, help="run folder containing ARCHITECTURE-SPINE.md")
+    ap.add_argument("-o", "--output", help="write JSON here instead of stdout")
+    args = ap.parse_args(argv)
+
+    spine_path = Path(args.workspace) / SPINE
+    if not spine_path.exists():
+        result = {"ok": False, "error": f"{spine_path} not found", "findings": [], "total_findings": 0}
+    else:
+        try:
+            text = spine_path.read_text(encoding="utf-8")
+        except (OSError, UnicodeDecodeError) as e:
+            # honor the "exit code is always 0" contract: a read/decode failure travels in JSON
+            result = {"ok": False, "error": f"could not read {spine_path}: {e}", "findings": [], "total_findings": 0}
+        else:
+            result = lint(text)
+
+    out = json.dumps(result, indent=2)
+    if args.output:
+        Path(args.output).write_text(out + "\n", encoding="utf-8")
+    else:
+        print(out)
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/.agents/skills/bmad-architecture/scripts/tests/test_lint_spine.py b/.agents/skills/bmad-architecture/scripts/tests/test_lint_spine.py
new file mode 100644
index 00000000..55cf7482
--- /dev/null
+++ b/.agents/skills/bmad-architecture/scripts/tests/test_lint_spine.py
@@ -0,0 +1,270 @@
+# /// script
+# requires-python = ">=3.10"
+# dependencies = ["pytest>=8.0"]
+# ///
+"""Tests for lint_spine.py. Run: uv run --with pytest pytest scripts/tests/test_lint_spine.py
+
+The spine under test: a clean spine lints empty; the linter catches exactly the
+mechanical defects a prompt is unreliable at — literal placeholders, AD-n id breakage,
+AD-n blocks missing required fields, and unpinned Stack versions.
+"""
+import importlib.util
+import json
+import re
+import sys
+from pathlib import Path
+
+import pytest
+
+_SPEC = importlib.util.spec_from_file_location(
+    "lint_spine", Path(__file__).resolve().parent.parent / "lint_spine.py"
+)
+lint_spine = importlib.util.module_from_spec(_SPEC)
+sys.modules["lint_spine"] = lint_spine
+_SPEC.loader.exec_module(lint_spine)
+
+
+CLEAN = """---
+name: 'Demo'
+---
+
+## Invariants & Rules
+
+### AD-1 — single write path
+
+- **Binds:** all
+- **Prevents:** divergent mutation
+- **Rule:** state changes only through the command bus
+
+### AD-2 — layered deps `[ADOPTED]`
+
+- **Binds:** all
+- **Prevents:** import cycles
+- **Rule:** ui -> app -> domain, never backward
+
+```mermaid
+flowchart LR
+  A --> B{decision}
+```
+
+## Stack
+
+| Name | Version |
+| --- | --- |
+| fastapi | 0.115 |
+| pydantic | 2.9 |
+"""
+
+
+def cats(result):
+    return sorted(f["category"] for f in result["findings"])
+
+
+def test_clean_spine_passes():
+    result = lint_spine.lint(CLEAN)
+    assert result["ok"] is True
+    assert result["total_findings"] == 0
+
+
+def test_mermaid_braces_not_flagged():
+    # the {decision} node lives in a fenced block and must not read as a template token
+    result = lint_spine.lint(CLEAN)
+    assert "placeholder" not in cats(result)
+
+
+def test_placeholder_markers_caught():
+    text = CLEAN.replace("the command bus", "TBD")
+    result = lint_spine.lint(text)
+    assert "placeholder" in cats(result)
+
+
+def test_similar_to_caught():
+    text = CLEAN.replace("import cycles", "similar to AD-1")
+    result = lint_spine.lint(text)
+    assert any("cross-reference" in f["detail"] for f in result["findings"])
+
+
+def test_unfilled_template_token_caught():
+    text = CLEAN.replace("single write path", "{decision}")
+    result = lint_spine.lint(text)
+    assert any(f["category"] == "placeholder" for f in result["findings"])
+
+
+def test_duplicate_ad_id_caught():
+    text = CLEAN.replace("### AD-2 — layered deps `[ADOPTED]`", "### AD-1 — layered deps")
+    result = lint_spine.lint(text)
+    assert "ad_id" in cats(result)
+
+
+def test_non_monotonic_ad_id_caught():
+    text = CLEAN.replace("### AD-2 — layered deps `[ADOPTED]`", "### AD-5 — layered deps").replace(
+        "### AD-1 — single write path", "### AD-9 — single write path"
+    )
+    result = lint_spine.lint(text)
+    assert any("non-monotonic" in f["detail"] for f in result["findings"])
+
+
+def test_missing_field_caught():
+    text = CLEAN.replace("- **Rule:** state changes only through the command bus\n", "")
+    result = lint_spine.lint(text)
+    assert any(f["category"] == "ad_fields" and "rule" in f["detail"] for f in result["findings"])
+
+
+def test_unpinned_dep_caught():
+    text = CLEAN.replace("| fastapi | 0.115 |", "| fastapi |  |")
+    result = lint_spine.lint(text)
+    assert "version_pin" in cats(result)
+
+
+def test_placeholder_version_caught():
+    text = CLEAN.replace("| fastapi | 0.115 |", "| fastapi | {pin} |")
+    result = lint_spine.lint(text)
+    assert any(f["category"] == "version_pin" and "fastapi" in f["detail"] for f in result["findings"])
+
+
+def test_no_stack_section_ok():
+    text = CLEAN.split("## Stack")[0]
+    result = lint_spine.lint(text)
+    assert "version_pin" not in cats(result)
+
+
+def test_stack_skeleton_row_not_version_pinned():
+    # a leftover {token} name is the placeholder pass's job, not a double-reported version_pin
+    text = CLEAN.replace("| fastapi | 0.115 |", "| {language / framework} | {pinned version} |")
+    result = lint_spine.lint(text)
+    assert "version_pin" not in cats(result)
+
+
+def test_stack_html_comment_not_parsed_as_row():
+    text = CLEAN.replace("## Stack\n", "## Stack\n\n<!-- SEED — verified current 2026-06 -->\n")
+    result = lint_spine.lint(text)
+    assert "version_pin" not in cats(result)
+
+
+def test_template_token_is_low_severity():
+    # a bare {token} can be legitimate brace prose; it is flagged, but low (not high) so the
+    # mechanical pass stays near-zero false-positive
+    text = CLEAN.replace("single write path", "{decision}")
+    result = lint_spine.lint(text)
+    toks = [f for f in result["findings"] if f["category"] == "placeholder" and "template token" in f["detail"]]
+    assert toks and all(f["severity"] == "low" for f in toks)
+
+
+def test_no_frontmatter_body_still_scanned():
+    text = "## Invariants\n\n### AD-1 — x\n\n- **Binds:** all\n- **Prevents:** drift\n- **Rule:** TBD\n"
+    result = lint_spine.lint(text)
+    assert "placeholder" in cats(result)  # TBD caught even with no frontmatter
+
+
+def test_frontmatter_value_with_dashes_not_truncated():
+    # a value containing '---' must not be read as the closing fence (line-exact close)
+    text = ("---\nname: 'x'\nscope: 'phase 1 --- phase 2'\n---\n\n"
+            "## Stack\n\n| Name | Version |\n| --- | --- |\n| fastapi |  |\n")
+    result = lint_spine.lint(text)
+    assert any(f["category"] == "version_pin" for f in result["findings"])  # read past the inline ---
+
+
+def test_ad_heading_in_fence_not_counted():
+    text = (
+        "---\nname: 'x'\n---\n\n"
+        "### AD-1 — real\n\n- **Binds:** all\n- **Prevents:** drift\n- **Rule:** do x\n\n"
+        "## Docs\n\n```text\n### AD-2 — illustrative only, no fields\n```\n"
+    )
+    result = lint_spine.lint(text)
+    assert result["ok"] is True  # the fenced AD-2 is not a live AD → no ad_fields/ad_id finding
+
+
+def test_stack_table_flags_only_the_unpinned_row():
+    text = ("---\nname: 'x'\n---\n\n## Stack\n\n| Name | Version |\n| --- | --- |\n"
+            "| fastapi | 0.115 |\n| redis |  |\n")
+    result = lint_spine.lint(text)
+    pins = [f for f in result["findings"] if f["category"] == "version_pin"]
+    assert len(pins) == 1 and "redis" in pins[0]["detail"]
+
+
+def test_stack_table_all_pinned_ok():
+    text = ("---\nname: 'x'\n---\n\n## Stack\n\n| Name | Version |\n| --- | --- |\n"
+            "| fastapi | 0.115 |\n")
+    result = lint_spine.lint(text)
+    assert "version_pin" not in cats(result)
+
+
+def test_fenced_stack_rows_not_parsed():
+    # an illustrative fenced table under ## Stack must not be read as live rows (fences are
+    # blanked first, like every other pass) — a blank-version row inside a fence is not a finding
+    text = ("---\nname: 'x'\n---\n\n## Stack\n\n| Name | Version |\n| --- | --- |\n"
+            "| fastapi | 0.115 |\n\n```text\n| example |  |\n```\n")
+    result = lint_spine.lint(text)
+    assert "version_pin" not in cats(result)
+
+
+def test_fenced_stack_heading_not_live():
+    # a `## Stack` heading shown inside a code fence is not the live Stack section
+    text = ("---\nname: 'x'\n---\n\n## Docs\n\n```md\n## Stack\n\n| foo |  |\n```\n")
+    result = lint_spine.lint(text)
+    assert "version_pin" not in cats(result)
+
+
+def test_renamed_stack_heading_still_scanned():
+    # the heading match is word-boundary, so a varied `## Stack` heading still counts
+    text = ("---\nname: 'x'\n---\n\n## Stack & Versions\n\n| Name | Version |\n| --- | --- |\n"
+            "| redis |  |\n")
+    result = lint_spine.lint(text)
+    pins = [f for f in result["findings"] if f["category"] == "version_pin"]
+    assert len(pins) == 1 and "redis" in pins[0]["detail"]
+
+
+def test_reordered_columns_pair_name_to_version():
+    # Version-then-Name header: the unpinned row must still be flagged by its real name
+    text = ("---\nname: 'x'\n---\n\n## Stack\n\n| Version | Name |\n| --- | --- |\n"
+            "| 0.115 | fastapi |\n|  | redis |\n")
+    result = lint_spine.lint(text)
+    pins = [f for f in result["findings"] if f["category"] == "version_pin"]
+    assert len(pins) == 1 and "redis" in pins[0]["detail"]
+
+
+def test_placeholder_line_number_is_absolute():
+    # a TBD after a multi-line fence reports its real file line (fence blanked, not collapsed)
+    text = (
+        "---\nname: 'x'\n---\n\n"
+        "## A\n\n"
+        "```text\nf1\nf2\nf3\n```\n\n"
+        "TBD here\n"
+    )
+    result = lint_spine.lint(text)
+    ph = next(f for f in result["findings"] if "TBD" in f["detail"])
+    n = int(re.search(r"line (\d+)", ph["location"]).group(1))
+    assert n == 13
+
+
+def test_missing_spine_file_reports_error(tmp_path, capsys):
+    rc = lint_spine.main(["--workspace", str(tmp_path)])
+    out = json.loads(capsys.readouterr().out)
+    assert rc == 0 and out["ok"] is False and "not found" in out["error"]
+
+
+def test_frontmatter_unfilled_token_caught():
+    # an unfilled {scope}/{paradigm}/{date} in frontmatter is part of the contract and must lint
+    text = "---\nname: 'x'\nscope: '{what this spine governs}'\n---\n\n## Invariants\n"
+    result = lint_spine.lint(text)
+    fm = [f for f in result["findings"] if f["category"] == "placeholder" and "frontmatter" in f["detail"]]
+    assert fm and any("template token" in f["detail"] for f in fm)
+
+
+def test_frontmatter_tbd_caught():
+    text = "---\nname: 'x'\nstatus: TBD\n---\n\n## Invariants\n"
+    result = lint_spine.lint(text)
+    assert any(f["category"] == "placeholder" and "frontmatter" in f["detail"] and "TBD" in f["detail"]
+               for f in result["findings"])
+
+
+def test_unreadable_spine_returns_error_not_crash(tmp_path, capsys):
+    # a spine that exists but can't be UTF-8 decoded must yield error JSON + exit 0, not a traceback
+    (tmp_path / lint_spine.SPINE).write_bytes(b"\xff\xfe bad bytes not utf-8")
+    rc = lint_spine.main(["--workspace", str(tmp_path)])
+    out = json.loads(capsys.readouterr().out)
+    assert rc == 0 and out["ok"] is False and "could not read" in out["error"]
+
+
+if __name__ == "__main__":
+    sys.exit(pytest.main([__file__, "-q"]))
diff --git a/.agents/skills/bmad-brainstorming/SKILL.md b/.agents/skills/bmad-brainstorming/SKILL.md
new file mode 100644
index 00000000..a4bf8314
--- /dev/null
+++ b/.agents/skills/bmad-brainstorming/SKILL.md
@@ -0,0 +1,80 @@
+---
+name: bmad-brainstorming
+description: Facilitate a brainstorming session using diverse creative techniques. Use when the user says 'help me brainstorm' or 'help me ideate'.
+---
+
+# BMad Brainstorming
+
+## Overview
+
+You are a creative brainstorming coach. This skill runs a brainstorming session: someone brings a topic and wants to generate far more and far better ideas on it than they would alone — pushing past the obvious with sharper questions and harder constraints, with no rush to finish. The best sessions end with the user surprised by what came out.
+
+The session runs in one of three stances, chosen by the user — set explicitly at the start, or already implied by how they asked: **Facilitator** (you never supply ideas — a forcing function for theirs), **Creative Partner** (you facilitate *and* play along, trading ideas), or **Ideate for me** (you run the whole session yourself and show them the result). The chosen stance holds for the whole run.
+
+## Conventions
+
+- Bare paths (e.g. `references/headless.md`) resolve from `{skill-root}` (where `customize.toml` lives); `{project-root}`-prefixed paths from the project working directory.
+- `{workflow.<name>}` resolves to fields in the merged `customize.toml` `[workflow]` table.
+
+## On Activation
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. On failure, use a subagent to read `{skill-root}/customize.toml` directly with defaults.
+2. Run each `{workflow.activation_steps_prepend}` entry. Treat each `{workflow.persistent_facts}` entry as foundational context (`file:`-prefixed entries are paths/globs under `{project-root}` — load their contents; others are facts verbatim).
+3. Resolve central config: `uv run {project-root}/_bmad/scripts/resolve_config.py --project-root {project-root} --key core` (merges `_bmad/config.toml`, `_bmad/config.user.toml`, and the `_bmad/custom/` overrides); from the merged JSON resolve `{user_name}`, `{communication_language}`, `{document_output_language}`, `{output_folder}`, `{project_name}`; `{date}` is today. On failure or missing values → neutral defaults; never block.
+4. **If launched headless** (a machine signal, not a human asking for output — `references/headless.md` lists them): load `references/headless.md` and follow it for the whole run; never load it otherwise. Outside headless, you generate ideas yourself only in autonomous mode (`references/mode-autonomous.md`) — never in facilitator or partner mode.
+5. **Otherwise (interactive):** greet `{user_name}` in `{communication_language}` and stay in it. Note that `bmad-party-mode` and `bmad-advanced-elicitation` are available any time (mention only the ones installed; either may be absent). Glob `{workflow.output_dir}/*/.memlog.md`, read each frontmatter, and offer to resume any with `status` not `complete` (`## Resuming`) or start fresh (`## Run a Session`).
+
+Run each `{workflow.activation_steps_append}` entry; if either hook list was non-empty, confirm every entry ran before continuing.
+
+## Framing — hold this the whole run
+
+These fight your defaults, in every mode; hold them deliberately. The stance you pick adds one more frame (`references/mode-*.md`) on top.
+
+- **Aim past 100 ideas; resist concluding.** The urge to organize or wrap is the enemy of divergence — when in doubt, push for one more. Land only when the user is spent or the topic is mined out.
+- **Keep shifting the creative domain** — every 5–10 turns (or ~10 ideas when you're generating), usually by moving to the next technique.
+- **One prompt per message while in dialogue (Facilitator, Creative Partner); no multiple-choice menus.** Don't stack questions into a wall or hand a menu that invites lazy picking — both pull the user out of generating. The only exceptions are the two up-front *process* choices (stance, and the technique flow): *how* to run is theirs to pick; *what* to ideate never is.
+
+**The memlog** is the session's memory: the single source every output builds from, and the file a resume reloads. Whatever isn't in it is gone. Log every idea, decision, question, and bit of user direction — anything you'd regret losing if the window closed — one line each, the gist in the user's meaning, in time order; never edit or reorder. Skip your prompts and small talk. All writes to memlog are atomic and use the script `memlog.py` invoked as follows:
+
+- `uv run {project-root}/_bmad/scripts/memlog.py init --workspace {doc_workspace} --field topic="<topic>" --field goal="<goal>" --field mode="<facilitator|partner|autonomous>"` — create it once topic, goal, and stance are known.
+- `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {doc_workspace} --type <kind> --text "<one-line gist>"` — log one entry. `--type` ∈ `idea`/`insight`/`question`/`decision`/`direction`/`technique` (a switch: `--text "started <name>"`); omit for a plain note. Add `--by user`/`--by coach` to mark authorship — **required in Creative Partner mode** (renders `(idea by user)`); skip it otherwise.
+- `uv run {project-root}/_bmad/scripts/memlog.py set --workspace {doc_workspace} --key status --value complete` — flip status at wrap-up.
+
+## Run a Session
+
+Open with one compound question what are we brainstorming, and what's the goal or why behind it (along with asking if there are any inputs or special requests). The why shapes technique choice and synthesis (*kids' iPhone apps to build with your own kids* vs. *to win market share* point different ways). If the kickoff already made both clear, skip the question and confirm; read anything they point you to. Derive a kebab-case `{topic_slug}` and bind `{doc_workspace} = {workflow.output_dir}/{workflow.output_folder_name}/`.
+
+Now set the **stance** and the **technique batch** in one step — the composer page does both, so make it the default.
+
+**The composer page (primary).** The file is `{skill-root}/assets/brain-selector.html`. With a customized catalog (overridden `{workflow.brain_methods}` or any `{workflow.additional_techniques}`), regenerate it first: `uv run {skill-root}/scripts/brain.py --file {workflow.brain_methods} [--extra {doc_workspace}/extra-techniques.json] html --out {doc_workspace}/brain-selector.html` (pass `--extra`, a JSON list of `{category, technique_name, description}`, when there are additional techniques; the file is then `{doc_workspace}/brain-selector.html`). Try to open it (`open` / `xdg-open` / `start`), then say, in one message: *"It should open in your browser — compose your session, click **Copy prompt**, and paste the result back. If it didn't open, open `<path>` yourself, or say 'let's do it in chat'."* You can't see their browser, so never claim it opened.
+
+Read the pasted block: the **`Facilitation mode:`** line → the stance; the **listed techniques** (full category/name/description, some tagged `(random pick)`) → run them as given, no `list`/`show` needed; **`invent N`** / **`you choose N`** → see `## Choosing Techniques`.
+
+**Or in chat.** If they can't open the page or would rather not, pick the stance here and choose techniques per `## Choosing Techniques`.
+
+Either way, once the stance is known, create the memlog (the `init` above, with `--field mode=`) and load its frame for the rest of the run — Facilitator → `references/mode-facilitator.md`, Creative Partner → `references/mode-partner.md`, Ideate for me → `references/mode-autonomous.md`. Tell the user the memlog path: state is on disk now, so the session survives interruption.
+
+## Choosing Techniques
+
+For **Facilitator** and **Creative Partner**. (In **Ideate for me** you pick and run techniques yourself — see `references/mode-autonomous.md`.)
+
+Most sessions arrive with a batch already composed on the page — run it as given (each technique's full text is in the paste; no `list`/`show` needed). Two parts of a paste delegate back to you:
+
+- **`invent N`** (Inventive Flow) — invent N brand-new techniques on the fly. A line may scope an invention (`invent 1 new technique in the spirit of <category>`, from the page's per-category invent card) — when it does, honor that category's spirit. Announce the order, log each one's name + description, and offer to save a keeper to `{workflow.additional_techniques}` at wrap-up.
+- **`you choose N`** (Facilitator Chosen) — pick N techniques fitting the goal, `{workflow.favorite_techniques}` first; confirm exact names with a scoped `uv run {skill-root}/scripts/brain.py --file {workflow.brain_methods} list --category <cat>`. Never pull the library whole into context.
+
+If they didn't use the page, load `references/in-chat-techniques.md` and pick the batch in chat (**3–4 is the sweet spot**).
+
+Run each technique until it stops producing — log each idea, and the switch itself as a `technique` entry when you move on — then announce the new lens and let the change of technique do the domain-shifting. When the batch is spent, offer three paths: run another batch, **converge** to narrow and decide (`## Converging`), or wrap up (`## Wrap-Up`).
+
+## Converging
+
+The catalog is all *divergent* — built to generate. When the user is ready to narrow and decide (or asks to "pick"/"prioritize"/"make it real"), load `references/converge.md` and follow it; it ends by handing off to `## Wrap-Up`. Convergence is a distinct phase: never fold it into a generating batch, and don't push toward it while ideas are still flowing.
+
+## Resuming
+
+Picking up an existing session instead of starting fresh: load `references/resume.md` and follow it.
+
+## Wrap-Up
+
+Load `references/finalize.md` (after `## Converging`, or directly when the user is spent): synthesis, `status: complete`, artifacts.
diff --git a/.agents/skills/bmad-brainstorming/assets/brain-icons.json b/.agents/skills/bmad-brainstorming/assets/brain-icons.json
new file mode 100644
index 00000000..9a979d8d
--- /dev/null
+++ b/.agents/skills/bmad-brainstorming/assets/brain-icons.json
@@ -0,0 +1,166 @@
+{
+  "categories": {
+    "creative": {
+      "hue": "#6d5cf0",
+      "glyph": "<g stroke=\"currentColor\" stroke-width=\"2.4\" stroke-linecap=\"round\"><line x1=\"22\" y1=\"6.5\" x2=\"22\" y2=\"12.5\"/><line x1=\"22\" y1=\"31.5\" x2=\"22\" y2=\"37.5\"/><line x1=\"6.5\" y1=\"22\" x2=\"12.5\" y2=\"22\"/><line x1=\"31.5\" y1=\"22\" x2=\"37.5\" y2=\"22\"/><line x1=\"11.3\" y1=\"11.3\" x2=\"15.5\" y2=\"15.5\"/><line x1=\"28.5\" y1=\"28.5\" x2=\"32.7\" y2=\"32.7\"/><line x1=\"32.7\" y1=\"11.3\" x2=\"28.5\" y2=\"15.5\"/><line x1=\"15.5\" y1=\"28.5\" x2=\"11.3\" y2=\"32.7\"/></g><circle cx=\"22\" cy=\"22\" r=\"6.6\" fill=\"currentColor\" fill-opacity=\"0.25\"/><circle cx=\"22\" cy=\"22\" r=\"3.6\" fill=\"currentColor\"/>"
+    },
+    "deep": {
+      "hue": "#4658c9",
+      "glyph": "<g fill=\"none\" stroke=\"currentColor\"><circle cx=\"22\" cy=\"22\" r=\"13\" stroke-width=\"1.5\" stroke-opacity=\"0.4\"/><circle cx=\"22\" cy=\"22\" r=\"9\" stroke-width=\"1.7\" stroke-opacity=\"0.7\"/><circle cx=\"22\" cy=\"22\" r=\"5\" stroke-width=\"1.9\"/></g><circle cx=\"22\" cy=\"22\" r=\"2.4\" fill=\"currentColor\"/>"
+    },
+    "structured": {
+      "hue": "#3b6ea5",
+      "glyph": "<g fill=\"currentColor\"><rect x=\"11\" y=\"11\" width=\"9.5\" height=\"9.5\" rx=\"2\"/><rect x=\"23.5\" y=\"11\" width=\"9.5\" height=\"9.5\" rx=\"2\" fill-opacity=\"0.25\"/><rect x=\"11\" y=\"23.5\" width=\"9.5\" height=\"9.5\" rx=\"2\" fill-opacity=\"0.25\"/><rect x=\"23.5\" y=\"23.5\" width=\"9.5\" height=\"9.5\" rx=\"2\"/></g>"
+    },
+    "quantum": {
+      "hue": "#2b86d9",
+      "glyph": "<g stroke=\"currentColor\" stroke-width=\"1.8\" fill=\"none\"><ellipse cx=\"22\" cy=\"22\" rx=\"14.5\" ry=\"6\" transform=\"rotate(28 22 22)\"/><ellipse cx=\"22\" cy=\"22\" rx=\"14.5\" ry=\"6\" transform=\"rotate(-28 22 22)\"/></g><circle cx=\"22\" cy=\"22\" r=\"6.6\" fill=\"currentColor\" fill-opacity=\"0.18\"/><circle cx=\"22\" cy=\"22\" r=\"3.4\" fill=\"currentColor\"/><circle cx=\"33.2\" cy=\"17.4\" r=\"2\" fill=\"currentColor\"/>"
+    },
+    "speculative_future": {
+      "hue": "#0fb5c9",
+      "glyph": "<g stroke=\"currentColor\" stroke-width=\"2.2\" stroke-linecap=\"round\" stroke-linejoin=\"round\" fill=\"none\"><path d=\"M11 31 L 26.5 15.5\"/><path d=\"M20 14.5 H 28 V 22.5\"/></g><circle cx=\"31\" cy=\"12\" r=\"2.8\" fill=\"currentColor\"/><g stroke=\"currentColor\" stroke-width=\"1.4\" stroke-linecap=\"round\"><line x1=\"31\" y1=\"6.5\" x2=\"31\" y2=\"8.4\"/><line x1=\"31\" y1=\"15.6\" x2=\"31\" y2=\"17.5\"/><line x1=\"25.5\" y1=\"12\" x2=\"27.4\" y2=\"12\"/><line x1=\"34.6\" y1=\"12\" x2=\"36.5\" y2=\"12\"/></g>"
+    },
+    "collaborative": {
+      "hue": "#15a3a3",
+      "glyph": "<g stroke=\"currentColor\" stroke-width=\"1.8\"><line x1=\"14\" y1=\"16\" x2=\"30\" y2=\"16\"/><line x1=\"14\" y1=\"16\" x2=\"22\" y2=\"30\"/><line x1=\"30\" y1=\"16\" x2=\"22\" y2=\"30\"/></g><g fill=\"currentColor\" fill-opacity=\"0.22\"><circle cx=\"14\" cy=\"16\" r=\"4.6\"/><circle cx=\"30\" cy=\"16\" r=\"4.6\"/><circle cx=\"22\" cy=\"30\" r=\"4.6\"/></g><g fill=\"currentColor\"><circle cx=\"14\" cy=\"16\" r=\"2.4\"/><circle cx=\"30\" cy=\"16\" r=\"2.4\"/><circle cx=\"22\" cy=\"30\" r=\"2.4\"/></g>"
+    },
+    "biomimetic": {
+      "hue": "#1f9d6b",
+      "glyph": "<path d=\"M22 7.5 C 31.5 12.5, 31.5 29, 22 36.5 C 12.5 29, 12.5 12.5, 22 7.5 Z\" fill=\"currentColor\" fill-opacity=\"0.22\"/><path d=\"M22 9 V 35.5\" stroke=\"currentColor\" stroke-width=\"1.8\" stroke-linecap=\"round\" fill=\"none\"/><g stroke=\"currentColor\" stroke-width=\"1.5\" stroke-linecap=\"round\"><path d=\"M22 16 l5.6 -2.6\"/><path d=\"M22 16 l-5.6 -2.6\"/><path d=\"M22 22 l6.6 -2.6\"/><path d=\"M22 22 l-6.6 -2.6\"/><path d=\"M22 28 l5.6 -2.6\"/><path d=\"M22 28 l-5.6 -2.6\"/></g>"
+    },
+    "constraint": {
+      "hue": "#d9882b",
+      "glyph": "<g stroke=\"currentColor\" stroke-width=\"2.2\" stroke-linecap=\"round\" stroke-linejoin=\"round\" fill=\"none\"><path d=\"M17 11 H 11 V 17\"/><path d=\"M27 11 H 33 V 17\"/><path d=\"M17 33 H 11 V 27\"/><path d=\"M27 33 H 33 V 27\"/></g><circle cx=\"22\" cy=\"22\" r=\"5\" fill=\"currentColor\" fill-opacity=\"0.25\"/><circle cx=\"22\" cy=\"22\" r=\"2.6\" fill=\"currentColor\"/>"
+    },
+    "wild": {
+      "hue": "#e2562f",
+      "glyph": "<path d=\"M24.5 6.5 L 12.5 24 H 19.5 L 17.5 37.5 L 31.5 18.5 H 24 L 24.5 6.5 Z\" fill=\"currentColor\"/>"
+    },
+    "cultural": {
+      "hue": "#c75b39",
+      "glyph": "<circle cx=\"22\" cy=\"22\" r=\"13.5\" fill=\"currentColor\" fill-opacity=\"0.14\"/><g stroke=\"currentColor\" stroke-width=\"1.6\" fill=\"none\"><circle cx=\"22\" cy=\"22\" r=\"13.5\"/><ellipse cx=\"22\" cy=\"22\" rx=\"6\" ry=\"13.5\"/><line x1=\"8.5\" y1=\"22\" x2=\"35.5\" y2=\"22\"/><path d=\"M11 15 H 33\" stroke-opacity=\"0.55\"/><path d=\"M11 29 H 33\" stroke-opacity=\"0.55\"/></g>"
+    },
+    "theatrical": {
+      "hue": "#cf4d6f",
+      "glyph": "<path d=\"M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z\" fill=\"currentColor\" fill-opacity=\"0.18\"/><path d=\"M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z\" stroke=\"currentColor\" stroke-width=\"1.8\" fill=\"none\"/><g fill=\"currentColor\"><circle cx=\"18.5\" cy=\"21\" r=\"1.7\"/><circle cx=\"25.5\" cy=\"21\" r=\"1.7\"/></g><path d=\"M18 27 C 20 29.5, 24 29.5, 26 27\" stroke=\"currentColor\" stroke-width=\"1.8\" stroke-linecap=\"round\" fill=\"none\"/>"
+    },
+    "absurdist": {
+      "hue": "#e0529c",
+      "glyph": "<g transform=\"rotate(-12 22 22)\"><circle cx=\"22\" cy=\"22\" r=\"13\" fill=\"currentColor\" fill-opacity=\"0.14\"/><circle cx=\"22\" cy=\"22\" r=\"13\" stroke=\"currentColor\" stroke-width=\"1.6\" fill=\"none\"/><path d=\"M16 19 q 2 -2.4 4 0\" stroke=\"currentColor\" stroke-width=\"1.8\" stroke-linecap=\"round\" fill=\"none\"/><circle cx=\"26.5\" cy=\"18.8\" r=\"1.8\" fill=\"currentColor\"/><path d=\"M16.5 26 C 19 30, 25 30, 28 24.5\" stroke=\"currentColor\" stroke-width=\"1.8\" stroke-linecap=\"round\" fill=\"none\"/></g>"
+    },
+    "introspective_delight": {
+      "hue": "#b15ad6",
+      "glyph": "<circle cx=\"22\" cy=\"13.5\" r=\"4\" fill=\"currentColor\"/><path d=\"M10.5 31 C 12.5 23, 31.5 23, 33.5 31 Z\" fill=\"currentColor\" fill-opacity=\"0.22\"/><path d=\"M10.5 31 C 12.5 23, 31.5 23, 33.5 31\" stroke=\"currentColor\" stroke-width=\"1.7\" fill=\"none\"/><path d=\"M13.5 30 C 16 26.5, 20 25.5, 22 25.5 C 24 25.5, 28 26.5, 30.5 30\" stroke=\"currentColor\" stroke-width=\"1.5\" fill=\"none\" stroke-opacity=\"0.6\"/>"
+    }
+  },
+  "techniques": {
+    "Yes And Building": "<g fill=\"currentColor\"><rect x=\"8\" y=\"27\" width=\"12\" height=\"8\" rx=\"1.5\" fill-opacity=\".8\"/><rect x=\"14\" y=\"19\" width=\"12\" height=\"8\" rx=\"1.5\" fill-opacity=\".5\"/><rect x=\"20\" y=\"11\" width=\"12\" height=\"8\" rx=\"1.5\"/></g>",
+    "Brain Writing Round Robin": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M31 16 A10 10 0 1 0 32.5 22\"/><path d=\"M31 10 L31.5 16.3 L25 16.5\"/></g>",
+    "Random Stimulation": "<rect x=\"11\" y=\"11\" width=\"22\" height=\"22\" rx=\"4\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><g fill=\"currentColor\"><circle cx=\"17\" cy=\"17\" r=\"1.8\"/><circle cx=\"27\" cy=\"17\" r=\"1.8\"/><circle cx=\"22\" cy=\"22\" r=\"1.8\"/><circle cx=\"17\" cy=\"27\" r=\"1.8\"/><circle cx=\"27\" cy=\"27\" r=\"1.8\"/></g>",
+    "Role Playing": "<rect x=\"11\" y=\"9\" width=\"22\" height=\"26\" rx=\"3\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><circle cx=\"22\" cy=\"19\" r=\"4\" fill=\"currentColor\"/><path d=\"M15.5 30 c2 -4.5 11 -4.5 13 0\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\"/>",
+    "Ideation Relay Race": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><line x1=\"12\" y1=\"31\" x2=\"27\" y2=\"16\"/><line x1=\"8\" y1=\"22\" x2=\"14\" y2=\"22\" stroke-opacity=\".5\"/><line x1=\"8\" y1=\"27\" x2=\"13\" y2=\"27\" stroke-opacity=\".35\"/></g><circle cx=\"29\" cy=\"14\" r=\"3.4\" fill=\"currentColor\"/>",
+    "Idea Hot Potato": "<path d=\"M11 31 Q22 8 33 31\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-dasharray=\"2 3.5\" stroke-linecap=\"round\"/><circle cx=\"22\" cy=\"12.5\" r=\"4.2\" fill=\"currentColor\"/>",
+    "Steal And Upgrade": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M20 33 V14\"/><path d=\"M13 21 L20 14 L27 21\"/></g><path d=\"M30 27 l1 2.6 2.6 1 -2.6 1 -1 2.6 -1 -2.6 -2.6 -1 2.6 -1 z\" fill=\"currentColor\"/>",
+    "Fold The Paper": "<path d=\"M13 16 L21 12 V28 L13 32 Z\" fill=\"currentColor\" fill-opacity=\".22\"/><path d=\"M21 12 L29 16 V32 L21 28 Z\" fill=\"currentColor\" fill-opacity=\".45\"/><path d=\"M13 16 L21 12 L29 16 M21 12 V28\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.5\" stroke-linejoin=\"round\"/>",
+    "What If Scenarios": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M18 17 a4 4 0 1 1 4 4 v3\"/></g><circle cx=\"22\" cy=\"30\" r=\"1.6\" fill=\"currentColor\"/>",
+    "Analogical Thinking": "<circle cx=\"15\" cy=\"22\" r=\"6\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><rect x=\"25\" y=\"16\" width=\"12\" height=\"12\" rx=\"2\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><path d=\"M19 20 q2 2 0 4 M23 20 q-2 2 0 4\" stroke=\"currentColor\" stroke-width=\"1.6\" fill=\"none\"/>",
+    "First Principles Thinking": "<g fill=\"currentColor\"><rect x=\"10\" y=\"28\" width=\"8\" height=\"6\" rx=\"1\"/><rect x=\"18.5\" y=\"28\" width=\"8\" height=\"6\" rx=\"1\"/><rect x=\"27\" y=\"28\" width=\"7\" height=\"6\" rx=\"1\"/></g><path d=\"M22 25 L22 11 M16 17 L22 11 L28 17\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"/>",
+    "Forced Relationships": "<circle cx=\"12\" cy=\"22\" r=\"3.4\" fill=\"currentColor\"/><circle cx=\"32\" cy=\"22\" r=\"3.4\" fill=\"currentColor\"/><path d=\"M15 22 q7 -9 14 0\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-dasharray=\"1.5 3\"/>",
+    "Time Shifting": "<circle cx=\"22\" cy=\"22\" r=\"12\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><path d=\"M22 15 V22 L27 25\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\"/>",
+    "Metaphor Mapping": "<rect x=\"10\" y=\"14\" width=\"14\" height=\"14\" rx=\"2\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><circle cx=\"28\" cy=\"25\" r=\"7\" fill=\"currentColor\" fill-opacity=\".22\"/><circle cx=\"28\" cy=\"25\" r=\"7\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/>",
+    "Cross-Pollination": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M13 14 H27 a4 4 0 0 1 0 8 H17 a4 4 0 0 0 0 8 H31\"/><path d=\"M28 11 L31.5 14 L28 17 M16 27 L12.5 30 L16 33\"/></g>",
+    "Concept Blending": "<circle cx=\"18\" cy=\"22\" r=\"8\" fill=\"currentColor\" fill-opacity=\".25\"/><circle cx=\"26\" cy=\"22\" r=\"8\" fill=\"currentColor\" fill-opacity=\".25\"/><g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"><circle cx=\"18\" cy=\"22\" r=\"8\"/><circle cx=\"26\" cy=\"22\" r=\"8\"/></g>",
+    "Reverse Brainstorming": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M13 18 H28 a4 4 0 0 1 0 8 H16\"/><path d=\"M19 15 L13 18 L19 21 M22 23 L16 26 L22 29\"/></g>",
+    "Sensory Exploration": "<path d=\"M10 22 q12 -10 24 0 q-12 10 -24 0 z\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linejoin=\"round\"/><circle cx=\"22\" cy=\"22\" r=\"4\" fill=\"currentColor\"/>",
+    "Five Whys": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><circle cx=\"14\" cy=\"13\" r=\"2.4\"/><circle cx=\"22\" cy=\"22\" r=\"2.4\"/><circle cx=\"30\" cy=\"31\" r=\"2.4\"/><path d=\"M15.6 14.8 L20.4 20.2 M23.6 23.8 L28.4 29.2\"/></g>",
+    "Provocation Technique": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M24 9 L13 24 H21 L19 35 L31 19 H23 Z\"/></g>",
+    "Assumption Reversal": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M16 14 V30\"/><path d=\"M11.5 25 L16 30 L20.5 25\"/><path d=\"M28 30 V14\"/><path d=\"M23.5 19 L28 14 L32.5 19\"/></g>",
+    "Question Storming": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M14 16 a3.2 3.2 0 1 1 3.2 3.2 v2\"/><path d=\"M26 13 a3.6 3.6 0 1 1 3.6 3.6 v2.4\"/></g><circle cx=\"17.2\" cy=\"27\" r=\"1.5\" fill=\"currentColor\"/><circle cx=\"29.6\" cy=\"25.6\" r=\"1.6\" fill=\"currentColor\"/>",
+    "Constraint Mapping": "<path d=\"M11 14 L18 12 L26 14 L33 12 V30 L26 32 L18 30 L11 32 Z\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linejoin=\"round\"/><path d=\"M18 12 V30 M26 14 V32\" stroke=\"currentColor\" stroke-width=\"1.6\"/>",
+    "Failure Analysis": "<circle cx=\"20\" cy=\"20\" r=\"8\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><line x1=\"26\" y1=\"26\" x2=\"33\" y2=\"33\" stroke=\"currentColor\" stroke-width=\"2.4\" stroke-linecap=\"round\"/><path d=\"M20 16 V21 M20 24 V24\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\"/>",
+    "Emergent Thinking": "<g fill=\"currentColor\"><circle cx=\"11\" cy=\"31\" r=\"1.6\"/><circle cx=\"17\" cy=\"29\" r=\"1.6\"/><circle cx=\"16\" cy=\"23\" r=\"1.6\"/><circle cx=\"22\" cy=\"24\" r=\"1.8\"/><circle cx=\"23\" cy=\"17\" r=\"1.9\"/><circle cx=\"29\" cy=\"18\" r=\"1.7\"/><circle cx=\"28\" cy=\"12\" r=\"2.1\"/></g>",
+    "Causal Loop Mapping": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M16 16 a9 9 0 1 1 -2 12\"/><path d=\"M16 10.5 L16.5 16.5 L10.5 17\"/><path d=\"M30 28.5 L29.5 22.5 L35 22\"/></g>",
+    "Morphological Analysis": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.8\"><rect x=\"11\" y=\"11\" width=\"22\" height=\"22\" rx=\"2\"/><path d=\"M11 18.3 H33 M11 25.6 H33 M18.3 11 V33 M25.6 11 V33\"/></g><rect x=\"18.5\" y=\"18.5\" width=\"7\" height=\"7\" fill=\"currentColor\" fill-opacity=\".4\"/>",
+    "Laddering": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M16 9 V35 M28 9 V35 M16 15 H28 M16 22 H28 M16 29 H28\"/></g>",
+    "Inner Child Conference": "<circle cx=\"22\" cy=\"16\" r=\"6\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><path d=\"M22 22 V31\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\"/><path d=\"M19 34 q3 -3 6 0\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.8\" stroke-linecap=\"round\"/><g fill=\"currentColor\"><circle cx=\"20\" cy=\"15\" r=\"1\"/><circle cx=\"24\" cy=\"15\" r=\"1\"/></g>",
+    "Shadow Work Mining": "<circle cx=\"22\" cy=\"22\" r=\"12\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><path d=\"M22 10 a12 12 0 0 1 0 24 z\" fill=\"currentColor\" fill-opacity=\".85\"/>",
+    "Values Archaeology": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"><path d=\"M10 16 h24\" stroke-opacity=\".4\"/><path d=\"M10 22 h24\" stroke-opacity=\".6\"/></g><path d=\"M22 24 L16 30 L22 36 L28 30 Z\" fill=\"currentColor\"/>",
+    "Future Self Interview": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M14 10 H30 L24 22 L30 34 H14 L20 22 Z\"/></g><path d=\"M18 14 H26\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\"/>",
+    "Body Wisdom Dialogue": "<path d=\"M22 33 C12 26 9 19 13.5 15 C17 12 21 14 22 17 C23 14 27 12 30.5 15 C35 19 32 26 22 33 Z\" fill=\"currentColor\" fill-opacity=\".22\"/><path d=\"M22 33 C12 26 9 19 13.5 15 C17 12 21 14 22 17 C23 14 27 12 30.5 15 C35 19 32 26 22 33 Z\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/>",
+    "Permission Giving": "<rect x=\"10\" y=\"14\" width=\"24\" height=\"16\" rx=\"2.5\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><path d=\"M15 23 L19 27 L28 17\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2.2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"/>",
+    "Secret Wish Confession": "<rect x=\"13\" y=\"20\" width=\"18\" height=\"14\" rx=\"2.5\" fill=\"currentColor\" fill-opacity=\".22\"/><rect x=\"13\" y=\"20\" width=\"18\" height=\"14\" rx=\"2.5\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><path d=\"M16.5 20 v-3 a5.5 5.5 0 0 1 11 0 v3\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/>",
+    "Mood Weather Report": "<circle cx=\"17\" cy=\"17\" r=\"4.5\" fill=\"currentColor\" fill-opacity=\".5\"/><path d=\"M22 30 a5 5 0 0 1 0.5 -10 a6 6 0 0 1 11 2.5 a4 4 0 0 1 -1.5 7.5 z\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linejoin=\"round\"/>",
+    "SCAMPER Method": "<circle cx=\"22\" cy=\"22\" r=\"5.5\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><g stroke=\"currentColor\" stroke-width=\"2.4\" stroke-linecap=\"round\"><path d=\"M22 9 V13.5 M22 30.5 V35 M9 22 H13.5 M30.5 22 H35 M12.8 12.8 L16 16 M28 28 L31.2 31.2 M31.2 12.8 L28 16 M16 28 L12.8 31.2\"/></g>",
+    "Six Thinking Hats": "<path d=\"M14 26 q8 -5 16 0\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\"/><path d=\"M17 26 q-6 1 -8 3 q13 4 26 0 q-2 -2 -8 -3\" fill=\"currentColor\" fill-opacity=\".22\"/><path d=\"M17 26 c-1 -8 11 -8 10 0\" fill=\"currentColor\" fill-opacity=\".5\"/>",
+    "Decision Tree Mapping": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><circle cx=\"22\" cy=\"12\" r=\"2.6\"/><circle cx=\"14\" cy=\"32\" r=\"2.6\"/><circle cx=\"30\" cy=\"32\" r=\"2.6\"/><path d=\"M22 14.5 L22 20 M22 20 L14 29.4 M22 20 L30 29.4\"/></g>",
+    "Solution Matrix": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.8\"><rect x=\"11\" y=\"11\" width=\"22\" height=\"22\" rx=\"2\"/><path d=\"M11 22 H33 M22 11 V33\"/></g><path d=\"M24.5 14.5 L26.5 16.5 L30.5 12.5\" stroke=\"currentColor\" stroke-width=\"2\" fill=\"none\" stroke-linecap=\"round\" stroke-linejoin=\"round\"/>",
+    "Trait Transfer": "<path d=\"M12 16 l1.6 3.4 3.6 .4 -2.7 2.5 .7 3.6 -3.2 -1.8 -3.2 1.8 .7 -3.6 -2.7 -2.5 3.6 -.4 z\" fill=\"currentColor\"/><rect x=\"25\" y=\"23\" width=\"9\" height=\"9\" rx=\"2\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><path d=\"M17 22 L25 27\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-dasharray=\"1.5 2.5\"/>",
+    "Lotus Blossom": "<g fill=\"currentColor\"><circle cx=\"22\" cy=\"22\" r=\"3.4\"/></g><g fill=\"currentColor\" fill-opacity=\".4\"><circle cx=\"22\" cy=\"13\" r=\"2.8\"/><circle cx=\"22\" cy=\"31\" r=\"2.8\"/><circle cx=\"13\" cy=\"22\" r=\"2.8\"/><circle cx=\"31\" cy=\"22\" r=\"2.8\"/><circle cx=\"15.5\" cy=\"15.5\" r=\"2.5\"/><circle cx=\"28.5\" cy=\"15.5\" r=\"2.5\"/><circle cx=\"15.5\" cy=\"28.5\" r=\"2.5\"/><circle cx=\"28.5\" cy=\"28.5\" r=\"2.5\"/></g>",
+    "Worst Possible Idea": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M17 11 v10 h-5 l10 12 10 -12 h-5 v-10 z\"/></g>",
+    "Disney Method": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"><circle cx=\"14\" cy=\"22\" r=\"4.5\"/><circle cx=\"22\" cy=\"22\" r=\"4.5\"/><circle cx=\"30\" cy=\"22\" r=\"4.5\"/></g>",
+    "Starbursting": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M22 8 V15 M22 29 V36 M8 22 H15 M29 22 H36 M12 12 L17 17 M27 27 L32 32 M32 12 L27 17 M12 32 L17 27\"/></g><path d=\"M19.5 19 a3.2 3.2 0 1 1 3 4 v1.2\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.8\" stroke-linecap=\"round\"/>",
+    "Mind Mapping": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><circle cx=\"22\" cy=\"22\" r=\"4\"/><circle cx=\"11\" cy=\"13\" r=\"2.2\"/><circle cx=\"33\" cy=\"13\" r=\"2.2\"/><circle cx=\"10\" cy=\"28\" r=\"2.2\"/><circle cx=\"32\" cy=\"31\" r=\"2.2\"/><path d=\"M19 19.5 L12.5 14.5 M25 19.5 L31.5 14.5 M19 24.5 L11.5 27 M25.5 24 L30.5 29.5\"/></g>",
+    "Crazy 8s": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.7\"><rect x=\"9\" y=\"12\" width=\"8\" height=\"9\" rx=\"1.5\"/><rect x=\"18\" y=\"12\" width=\"8\" height=\"9\" rx=\"1.5\"/><rect x=\"27\" y=\"12\" width=\"8\" height=\"9\" rx=\"1.5\"/><rect x=\"9\" y=\"23\" width=\"8\" height=\"9\" rx=\"1.5\"/><rect x=\"18\" y=\"23\" width=\"8\" height=\"9\" rx=\"1.5\"/><rect x=\"27\" y=\"23\" width=\"8\" height=\"9\" rx=\"1.5\"/></g>",
+    "Time Travel Talk Show": "<rect x=\"18\" y=\"9\" width=\"8\" height=\"15\" rx=\"4\" fill=\"currentColor\" fill-opacity=\".25\"/><rect x=\"18\" y=\"9\" width=\"8\" height=\"15\" rx=\"4\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><path d=\"M14 21 a8 8 0 0 0 16 0 M22 29 V34 M17 34 H27\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\"/>",
+    "Alien Anthropologist": "<ellipse cx=\"22\" cy=\"30\" rx=\"13\" ry=\"4.5\" fill=\"currentColor\" fill-opacity=\".25\"/><path d=\"M22 11 c7 0 10 6 10 11 c0 5 -5 7 -10 7 c-5 0 -10 -2 -10 -7 c0 -5 3 -11 10 -11 z\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><g fill=\"currentColor\"><ellipse cx=\"18\" cy=\"22\" rx=\"1.6\" ry=\"2.4\"/><ellipse cx=\"26\" cy=\"22\" rx=\"1.6\" ry=\"2.4\"/></g>",
+    "Dream Fusion Laboratory": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M18 9 H26 M19.5 9 V18 L13 30 a2 2 0 0 0 2 3 H29 a2 2 0 0 0 2 -3 L24.5 18 V9\"/></g><path d=\"M16.5 26 H27.5\" stroke=\"currentColor\" stroke-width=\"2\"/><circle cx=\"20\" cy=\"29\" r=\"1.4\" fill=\"currentColor\"/><circle cx=\"25\" cy=\"28\" r=\"1.1\" fill=\"currentColor\"/>",
+    "Emotion Orchestra": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M17 30 V15 L31 12 V27\"/><circle cx=\"14\" cy=\"30\" r=\"3\"/><circle cx=\"28\" cy=\"27\" r=\"3\"/></g>",
+    "Parallel Universe Cafe": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"><circle cx=\"18\" cy=\"22\" r=\"9\"/><circle cx=\"26\" cy=\"22\" r=\"9\" stroke-dasharray=\"2.5 2.5\"/></g>",
+    "Persona Journey": "<path d=\"M14 33 q-2 -8 6 -9 q8 -1 6 -8\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-dasharray=\"0.1 4\"/><circle cx=\"14\" cy=\"33\" r=\"2.4\" fill=\"currentColor\"/><path d=\"M26 16 l3 -5 3 5 z\" fill=\"currentColor\"/>",
+    "Devil's Advocate Courtroom": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M22 10 V32 M14 32 H30\"/><path d=\"M11 16 H33 M11 16 L8 23 H14 Z M33 16 L30 23 H36 Z\"/></g>",
+    "Chaos Engineering": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M22 9 L25 18 L34 18 L27 24 L30 33 L22 27 L14 33 L17 24 L10 18 L19 18 Z\"/></g>",
+    "Guerrilla Gardening Ideas": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M22 33 V21\"/><path d=\"M22 22 c-7 0 -9 -6 -9 -9 c6 0 9 3 9 9 z\"/><path d=\"M22 24 c6 0 8 -4 8 -7 c-5 0 -8 2 -8 7 z\"/></g>",
+    "Pirate Code Brainstorm": "<path d=\"M22 10 c-7 0 -11 5 -11 11 c0 4 2 6 4 7 v4 h3 v-2 h2 v2 h4 v-2 h2 v2 h3 v-4 c2 -1 4 -3 4 -7 c0 -6 -4 -11 -11 -11 z\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linejoin=\"round\"/><g fill=\"currentColor\"><circle cx=\"17.5\" cy=\"21\" r=\"2.2\"/><circle cx=\"26.5\" cy=\"21\" r=\"2.2\"/></g>",
+    "Zombie Apocalypse Planning": "<circle cx=\"22\" cy=\"22\" r=\"4\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"><path d=\"M22 10 a12 12 0 0 1 6 3.5 M32 16 a12 12 0 0 1 0 12 M28 33.5 a12 12 0 0 1 -12 0 M12 28 a12 12 0 0 1 0 -12 M16 10.5 a12 12 0 0 1 6 -0.5\" stroke-dasharray=\"0.1 5.5\"/></g><g fill=\"currentColor\"><circle cx=\"22\" cy=\"11\" r=\"2\"/><circle cx=\"11\" cy=\"22\" r=\"2\"/><circle cx=\"33\" cy=\"22\" r=\"2\"/></g>",
+    "Drunk History Retelling": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M13 12 H31 L25 23 V31 H19 V23 Z\"/><path d=\"M19 31 H25\" /></g><circle cx=\"29\" cy=\"14\" r=\"1.4\" fill=\"currentColor\"/>",
+    "Anti-Solution": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M14 18 a8 8 0 1 1 -1 8\"/><path d=\"M14 12 L14 18.5 L20 18\"/></g>",
+    "Elemental Forces": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M22 8 L30 22 H14 Z\"/><path d=\"M14 30 L22 36 L30 30\"/><path d=\"M14 26 H30\"/></g>",
+    "Nature's Solutions": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M16 32 C12 24 12 20 16 12 M28 32 C32 24 32 20 28 12\"/><path d=\"M16 16 L28 14 M16 22 L28 20 M16 28 L28 26\"/></g>",
+    "Ecosystem Thinking": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><circle cx=\"22\" cy=\"13\" r=\"2.4\"/><circle cx=\"12\" cy=\"27\" r=\"2.4\"/><circle cx=\"32\" cy=\"27\" r=\"2.4\"/><circle cx=\"22\" cy=\"24\" r=\"2.4\"/><path d=\"M22 15.4 V21.6 M14 26 L20 24.5 M30 26 L24 24.5 M13.6 25.2 L30.4 25.2\"/></g>",
+    "Evolutionary Pressure": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M11 31 H17 M19 31 a6 6 0 0 1 6 -6 M25 25 a5 5 0 0 1 5 -5 M30 20 H33\"/><circle cx=\"11\" cy=\"31\" r=\"2\" fill=\"currentColor\"/><circle cx=\"33\" cy=\"20\" r=\"2.6\" fill=\"currentColor\"/></g>",
+    "Predator & Prey": "<path d=\"M22 9 L33 14 V23 C33 30 28 34 22 36 C16 34 11 30 11 23 V14 Z\" fill=\"currentColor\" fill-opacity=\".18\"/><path d=\"M22 9 L33 14 V23 C33 30 28 34 22 36 C16 34 11 30 11 23 V14 Z\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linejoin=\"round\"/>",
+    "Metamorphosis Stages": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M22 12 V32\"/><path d=\"M22 16 C14 12 10 18 14 22 C10 26 14 32 22 28 C30 32 34 26 30 22 C34 18 30 12 22 16\"/></g>",
+    "Swarm Logic": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M22 10 L29 14 V22 L22 26 L15 22 V14 Z\"/><path d=\"M15 24 L18 33 M29 24 L26 33 M22 28 V35\" stroke-opacity=\".6\"/></g>",
+    "Observer Effect": "<path d=\"M9 22 q13 -10 26 0 q-13 10 -26 0 z\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linejoin=\"round\"/><circle cx=\"22\" cy=\"22\" r=\"4.5\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><circle cx=\"22\" cy=\"22\" r=\"1.8\" fill=\"currentColor\"/>",
+    "Entanglement Thinking": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"><circle cx=\"15\" cy=\"22\" r=\"6\"/><circle cx=\"29\" cy=\"22\" r=\"6\"/></g><path d=\"M15 22 h14\" stroke=\"currentColor\" stroke-width=\"2\" stroke-dasharray=\"1.5 2.5\"/><g fill=\"currentColor\"><circle cx=\"15\" cy=\"22\" r=\"1.8\"/><circle cx=\"29\" cy=\"22\" r=\"1.8\"/></g>",
+    "Superposition Collapse": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M11 12 C20 16 24 16 33 12 M11 18 C20 22 24 22 33 18 M11 24 C20 28 24 28 33 24\"/><path d=\"M22 26 V34\"/></g><circle cx=\"22\" cy=\"34\" r=\"2\" fill=\"currentColor\"/>",
+    "Relativity Frame Shift": "<rect x=\"11\" y=\"11\" width=\"22\" height=\"22\" rx=\"2\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-opacity=\".4\"/><rect x=\"15\" y=\"15\" width=\"18\" height=\"18\" rx=\"2\" transform=\"rotate(-14 22 22)\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/>",
+    "Field Lines": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M12 13 V31 M16 13 C24 18 24 26 16 31 M22 13 C32 18 32 26 22 31\"/></g><circle cx=\"11\" cy=\"22\" r=\"2\" fill=\"currentColor\"/>",
+    "Quantum Tunneling": "<rect x=\"20\" y=\"9\" width=\"5\" height=\"26\" rx=\"1.5\" fill=\"currentColor\" fill-opacity=\".3\"/><path d=\"M10 22 H34\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2.2\" stroke-linecap=\"round\"/><path d=\"M28 17 L34 22 L28 27\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2.2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"/>",
+    "Indigenous Wisdom": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M26 11 C18 16 14 24 13 33 M26 11 C28 18 26 25 20 29\"/><path d=\"M26 11 C24 13 22 14 19 15 M24 16 C22 18 20 19 17 20 M22 21 C20 23 18 24 15 25\"/></g>",
+    "Fusion Cuisine": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M10 19 a12 7 0 0 0 24 0 Z\"/><path d=\"M22 19 V32 M16 32 H28\"/></g>",
+    "Ritual Innovation": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M12 33 V17 a10 10 0 0 1 20 0 V33\"/><path d=\"M12 33 H32 M22 33 V21\"/></g>",
+    "Mythic Frameworks": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M14 12 h13 a3 3 0 0 1 3 3 v17 l-3 -2 -3 2 -3 -2 -3 2 V15 a3 3 0 0 0 -3 -3 z\"/><path d=\"M14 12 a3 3 0 0 0 -3 3 h6\"/><path d=\"M20 18 H26 M20 23 H26\"/></g>",
+    "Proverb Mining": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M22 14 C17 10 11 11 11 11 V30 s6 -1 11 3 c5 -4 11 -3 11 -3 V11 s-6 -1 -11 3 z\"/><path d=\"M22 14 V31\"/></g>",
+    "Ancestor Council": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"><circle cx=\"22\" cy=\"14\" r=\"3.5\"/><circle cx=\"13\" cy=\"19\" r=\"3\"/><circle cx=\"31\" cy=\"19\" r=\"3\"/></g><g fill=\"currentColor\" fill-opacity=\".25\"><path d=\"M16 31 c0 -5 12 -5 12 0 z\"/><path d=\"M8 31 c0 -4 9 -4.5 9 0 z\"/><path d=\"M27 31 c0 -4.5 9 -4 9 0 z\"/></g>",
+    "Trickster's Gambit": "<rect x=\"11\" y=\"12\" width=\"13\" height=\"18\" rx=\"2\" transform=\"rotate(-10 17.5 21)\" fill=\"currentColor\" fill-opacity=\".2\" stroke=\"currentColor\" stroke-width=\"2\"/><rect x=\"20\" y=\"14\" width=\"13\" height=\"18\" rx=\"2\" transform=\"rotate(10 26.5 23)\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><path d=\"M26.5 19 l1.4 3 1.4 -3 -1.4 -1 z\" fill=\"currentColor\"/>",
+    "Villain's Monologue": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M12 20 C16 18 19 18 22 21 C25 18 28 18 32 20 C30 24 26 24 22 21 C18 24 14 24 12 20 Z\"/></g><circle cx=\"22\" cy=\"14\" r=\"2.4\" fill=\"currentColor\"/>",
+    "Explain It to a Golden Retriever": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M14 18 C12 12 17 13 18 17 M30 18 C32 12 27 13 26 17\"/><path d=\"M15 19 C13 28 18 33 22 33 C26 33 31 28 29 19 C26 16 18 16 15 19 Z\"/></g><g fill=\"currentColor\"><circle cx=\"19\" cy=\"24\" r=\"1.4\"/><circle cx=\"25\" cy=\"24\" r=\"1.4\"/><circle cx=\"22\" cy=\"28\" r=\"1.6\"/></g>",
+    "Infomercial at 3AM": "<rect x=\"9\" y=\"14\" width=\"26\" height=\"18\" rx=\"2.5\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><path d=\"M18 9 L22 14 L26 9\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"/><path d=\"M22 19 l1 2.6 2.8 .2 -2.1 1.9 .7 2.7 -2.4 -1.5 -2.4 1.5 .7 -2.7 -2.1 -1.9 2.8 -.2 z\" fill=\"currentColor\"/>",
+    "Drunk Uncle at Thanksgiving": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M11 16 L20 16 L27 11 V29 L20 24 L11 24 Z\"/><path d=\"M30 16 q3 4 0 8 M33 13 q5 7 0 14\"/></g>",
+    "Cursed Genie": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M10 30 h18 a2 2 0 0 0 2 -2 c0 -5 -6 -5 -8 -8 c5 -1 8 -3 8 -3 c-4 -2 -12 -2 -16 1 c-4 3 -5 9 -4 12 z\"/><path d=\"M30 17 L33 14 M31 21 L35 20\" stroke-opacity=\".6\"/></g>",
+    "Three Rounds of Stupid": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M13 30 V20 M9 24 L13 20 L17 24 M22 30 V15 M18 19 L22 15 L26 19 M31 30 V11 M27 15 L31 11 L35 15\"/></g>",
+    "Kill the Crown Jewel": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M11 28 L13 15 L19 22 L22 12 L25 22 L31 15 L33 28 Z\"/><path d=\"M11 28 H33\"/></g><path d=\"M14 12 L30 32\" stroke=\"currentColor\" stroke-width=\"2.4\" stroke-linecap=\"round\"/>",
+    "1000x Budget": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"><ellipse cx=\"22\" cy=\"14\" rx=\"9\" ry=\"3.5\"/><path d=\"M13 14 V22 a9 3.5 0 0 0 18 0 V14\"/><path d=\"M13 22 V30 a9 3.5 0 0 0 18 0 V22\"/></g>",
+    "Ship in 60 Minutes": "<circle cx=\"22\" cy=\"24\" r=\"11\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><path d=\"M22 24 V17 M22 24 L27 27\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\"/><path d=\"M18 8 H26 M22 8 V13\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\"/>",
+    "The $0 Mandate": "<circle cx=\"22\" cy=\"22\" r=\"11\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><path d=\"M22 14 V30 M18 18 a4 3 0 0 1 8 0 a4 3 0 0 1 -8 4 a4 3 0 0 0 8 0\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.8\" stroke-linecap=\"round\"/><path d=\"M14 30 L30 14\" stroke=\"currentColor\" stroke-width=\"2.2\" stroke-linecap=\"round\"/>",
+    "One Feature Only": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><circle cx=\"22\" cy=\"22\" r=\"4.5\"/><path d=\"M22 9 V13 M22 31 V35 M9 22 H13 M31 22 H35\" stroke-opacity=\".35\"/></g><circle cx=\"22\" cy=\"22\" r=\"2\" fill=\"currentColor\"/>",
+    "Crank the Dial to 11": "<path d=\"M11 28 A12 12 0 0 1 33 28\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\"/><path d=\"M22 28 L31 17\" stroke=\"currentColor\" stroke-width=\"2.2\" stroke-linecap=\"round\"/><circle cx=\"22\" cy=\"28\" r=\"2.6\" fill=\"currentColor\"/>",
+    "Constraint Roulette": "<circle cx=\"22\" cy=\"22\" r=\"12\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><circle cx=\"22\" cy=\"22\" r=\"12\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-dasharray=\"3 3.7\" stroke-opacity=\".5\"/><circle cx=\"22\" cy=\"22\" r=\"3\" fill=\"currentColor\"/><path d=\"M22 7 L25 12 H19 Z\" fill=\"currentColor\"/>",
+    "Time Horizon Ladder": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M9 30 H35\"/><path d=\"M14 30 V24 M22 30 V18 M30 30 V12\"/></g><g fill=\"currentColor\"><circle cx=\"14\" cy=\"24\" r=\"2\"/><circle cx=\"22\" cy=\"18\" r=\"2\"/><circle cx=\"30\" cy=\"12\" r=\"2\"/></g>",
+    "Post-Scarcity Test": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M15 22 a4.5 4.5 0 1 1 4.5 4.5 C16 26.5 14 18 11 18 a3.5 3.5 0 0 0 0 7 c4 0 5 -8 11 -8 a4.5 4.5 0 0 1 0 9 c-3 0 -4 -4.5 -7 -4.5\"/></g>",
+    "Utopia vs Dystopia Split-Screen": "<rect x=\"11\" y=\"11\" width=\"22\" height=\"22\" rx=\"3\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><path d=\"M22 11 V33\" stroke=\"currentColor\" stroke-width=\"2\"/><path d=\"M22 11 H33 a0 0 0 0 1 0 0 V33 H22 Z\" fill=\"currentColor\" fill-opacity=\".8\"/>",
+    "Sci-Fi Artifact From the Future": "<path d=\"M22 9 L33 15 V28 L22 35 L11 28 V15 Z\" fill=\"currentColor\" fill-opacity=\".15\"/><path d=\"M22 9 L33 15 V28 L22 35 L11 28 V15 Z M11 15 L22 21 L33 15 M22 21 V35\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linejoin=\"round\"/>",
+    "Emerging Tech Collision": "<rect x=\"15\" y=\"15\" width=\"14\" height=\"14\" rx=\"2\" fill=\"currentColor\" fill-opacity=\".22\"/><rect x=\"15\" y=\"15\" width=\"14\" height=\"14\" rx=\"2\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\"/><g stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\"><path d=\"M19 15 V10 M25 15 V10 M19 29 V34 M25 29 V34 M15 19 H10 M15 25 H10 M29 19 H34 M29 25 H34\"/></g>",
+    "What-If-The-World-Changed Card Flip": "<rect x=\"13\" y=\"10\" width=\"18\" height=\"24\" rx=\"2.5\" fill=\"currentColor\" fill-opacity=\".18\" stroke=\"currentColor\" stroke-width=\"2\"/><path d=\"M22 10 V34\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-dasharray=\"2 2.5\"/><path d=\"M27 16 a4 4 0 1 1 4 4\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.8\" stroke-linecap=\"round\"/>",
+    "Future Anthropologist Dig": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M22 27 a6 6 0 1 1 0.1 0 z\"/><path d=\"M22 23 a2.5 2.5 0 1 0 0.1 0 M19 30 a5 5 0 0 0 6 0\"/></g><path d=\"M12 16 L17 13 M32 16 L27 13\" stroke=\"currentColor\" stroke-width=\"1.6\" stroke-opacity=\".5\"/>",
+    "How Might We": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M11 13 H33 a2 2 0 0 1 2 2 V27 a2 2 0 0 1 -2 2 H24 L19 34 V29 H11 a2 2 0 0 1 -2 -2 V15 a2 2 0 0 1 2 -2 Z\"/></g><path d=\"M19 19 a3.2 3.2 0 1 1 3.4 3.4 v1.6\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.8\" stroke-linecap=\"round\"/><circle cx=\"22.4\" cy=\"27\" r=\"1.4\" fill=\"currentColor\"/>",
+    "Job to Be Done": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linejoin=\"round\"><rect x=\"9\" y=\"16\" width=\"26\" height=\"16\" rx=\"2.5\"/><path d=\"M17 16 v-3 a2 2 0 0 1 2 -2 h6 a2 2 0 0 1 2 2 v3\"/><path d=\"M9 23 H35\"/></g><circle cx=\"22\" cy=\"23\" r=\"1.8\" fill=\"currentColor\"/>",
+    "Empathy Map": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"1.8\"><rect x=\"11\" y=\"11\" width=\"22\" height=\"22\" rx=\"2\"/><path d=\"M11 22 H33 M22 11 V33\"/></g><path d=\"M22 27 c-2.6 -2.1 -4.2 -3.4 -4.2 -5.2 a2.1 2.1 0 0 1 4.2 -1 a2.1 2.1 0 0 1 4.2 1 c0 1.8 -1.6 3.1 -4.2 5.2 z\" fill=\"currentColor\"/>",
+    "Backcasting": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M30 11 V33\"/></g><path d=\"M30 13 L20 16.5 L30 20 Z\" fill=\"currentColor\"/><path d=\"M28 27 H14\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-dasharray=\"1.5 3\"/><path d=\"M18 23 L14 27 L18 31\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"/>",
+    "Scenario Cross": "<g stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\"><path d=\"M22 9 V35 M9 22 H35\"/></g><g fill=\"currentColor\"><circle cx=\"15\" cy=\"15\" r=\"2\"/><circle cx=\"29\" cy=\"15\" r=\"2\"/><circle cx=\"15\" cy=\"29\" r=\"2\"/><circle cx=\"29\" cy=\"29\" r=\"2\"/></g>",
+    "TRIZ Contradiction": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M9 15 H18 M14.5 11.5 L18 15 L14.5 18.5\"/><path d=\"M35 29 H26 M29.5 25.5 L26 29 L29.5 32.5\"/></g><path d=\"M22 17.5 l1.5 3.4 3.7 .3 -2.8 2.4 .9 3.6 -3.3 -1.9 -3.3 1.9 .9 -3.6 -2.8 -2.4 3.7 -.3 z\" fill=\"currentColor\"/>",
+    "Fishbone Diagram": "<g fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"><path d=\"M9 22 H33\"/><path d=\"M33 22 L36 18.5 M33 22 L36 25.5\"/><path d=\"M14 22 L18 15 M14 22 L18 29 M23 22 L27 15 M23 22 L27 29\"/></g><circle cx=\"9\" cy=\"22\" r=\"1.9\" fill=\"currentColor\"/>",
+    "Build on What Works": "<g fill=\"currentColor\"><rect x=\"11\" y=\"24\" width=\"6\" height=\"9\" rx=\"1\"/><rect x=\"19\" y=\"19\" width=\"6\" height=\"14\" rx=\"1\"/><rect x=\"27\" y=\"13\" width=\"6\" height=\"20\" rx=\"1\"/></g><path d=\"M11 19 L18 13 L24 16 L33 8\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"/><path d=\"M28 8 H33 V13\" fill=\"none\" stroke=\"currentColor\" stroke-width=\"2\" stroke-linecap=\"round\" stroke-linejoin=\"round\"/>"
+  }
+}
diff --git a/.agents/skills/bmad-brainstorming/assets/brain-methods.csv b/.agents/skills/bmad-brainstorming/assets/brain-methods.csv
new file mode 100644
index 00000000..32790e0f
--- /dev/null
+++ b/.agents/skills/bmad-brainstorming/assets/brain-methods.csv
@@ -0,0 +1,109 @@
+category,technique_name,description,detail,provenance,good_for,audience
+collaborative,Yes And Building,"Never negate; each person opens with ""Yes, and..."" and adds to the last idea, stacking a chain of accepted additions",,classic,novel|unstuck|planning,group
+collaborative,Brain Writing Round Robin,"Everyone writes ideas silently, then passes their sheet; you build on whatever lands in front of you, round after round",,classic,novel|feature,group
+collaborative,Random Stimulation,"Pull a random word or image and force a link to the problem: ""how does THIS spark a solution?""",,classic,unstuck|novel,either
+collaborative,Role Playing,"Each person speaks as a different stakeholder, voicing what that role wants, fears, and would demand of the idea",,classic,strategy|personal|feature,either
+collaborative,Ideation Relay Race,"30-second turns, no pausing: add one idea, slap it to the next person, keep the baton moving before anyone overthinks",,playful,unstuck,group
+collaborative,Idea Hot Potato,"One idea gets tossed around the circle; each catcher must mutate it in 10 seconds before passing, no repeats allowed",,playful,unstuck,group
+collaborative,Steal And Upgrade,"Pick a neighbor's idea you envy, claim it out loud, then make it visibly better before handing it back improved",,signature,novel|unstuck,group
+collaborative,Fold The Paper,"Each person adds one line to a hidden drawing or sentence, sees only the previous fragment, then unfold the surreal whole",,playful,unstuck|novel,group
+creative,What If Scenarios,"Detonate one constraint at a time — unlimited budget, opposite is true, problem vanished — and chase what rushes in",,signature,novel|strategy|unstuck,either
+creative,Analogical Thinking,Ask 'this is like what?' and steal the solution pattern from the domain that answers,,signature,feature|novel|diagnosis,either
+creative,First Principles Thinking,"Strip every assumption to bedrock facts, then rebuild the solution from scratch on truth alone",,classic,feature|novel|diagnosis|strategy,either
+creative,Forced Relationships,Grab two unrelated things at random and force a bridge between them until an idea falls out,,signature,novel|unstuck,either
+creative,Time Shifting,"Solve the problem as a 1900s artisan, then a 2150 colonist — harvest the era-bound constraints and tricks",,signature,novel|unstuck,either
+creative,Metaphor Mapping,"Declare the problem IS a chosen metaphor, extend the metaphor fully, map each part back to find insights",,signature,novel|diagnosis,either
+creative,Cross-Pollination,"Ask how a wildly different industry — casinos, ERs, beekeeping — would crack this, then adapt their move",,signature,novel|feature|strategy,either
+creative,Concept Blending,"Fuse two concepts into one new hybrid category and name what the merger becomes, not just combines",,signature,novel,either
+creative,Reverse Brainstorming,Generate problems instead of solutions — 'how could we make this fail?' — then mine each for its inverse,,classic,diagnosis|feature|unstuck,either
+creative,Sensory Exploration,"Interrogate the idea through each sense — its taste, smell, sound, texture — to surface non-analytical angles",,signature,novel|unstuck,either
+deep,Five Whys,"Ask ""why?"" five times in a chain, each answer feeding the next, until you hit the root cause beneath the symptom",,classic,diagnosis,either
+deep,Provocation Technique,"State something deliberately absurd, then mine it: ""how could this be useful?"" Extract the usable principle hiding inside",,classic,unstuck|novel,either
+deep,Assumption Reversal,"List every assumption baked into the problem, flip each to its opposite, then rebuild a solution on the inverted foundation",,classic,novel|diagnosis|strategy,either
+deep,Question Storming,"Generate only questions about the problem, zero answers allowed, until the real problem worth solving comes into focus",,classic,diagnosis|strategy|unstuck,either
+deep,Constraint Mapping,"Map every constraint, sort real from imagined, then attack each: dissolve it, route around it, or turn it into an asset",,signature,feature|strategy|diagnosis,either
+deep,Failure Analysis,"Dissect a relevant failure: what broke, why it broke, what lesson it leaves, and how to apply that wisdom here",,signature,diagnosis|strategy|feature,either
+deep,Emergent Thinking,Stop forcing a solution; watch what patterns the system keeps producing and name what's trying to emerge on its own,,signature,strategy|novel,either
+deep,Causal Loop Mapping,"Diagram the feedback loops linking causes and effects, find the reinforcing and balancing cycles, and target the leverage point",,classic,diagnosis|strategy,either
+deep,Morphological Analysis,"List the problem's independent parameters, generate options for each, then combine across them to surface untried configurations",,classic,feature|novel|planning,either
+deep,Laddering,"Ask 'and what would that give you?' up the chain until you reach the real underlying need, then ideate fresh at that level",,classic,personal|strategy|diagnosis,either
+introspective_delight,Inner Child Conference,"Answer as your 7-year-old self: ask naive 'why why why' questions, chase wonder, ban every boring adult thought",,signature,personal|unstuck,solo
+introspective_delight,Shadow Work Mining,"Name what you're avoiding, resisting, or scared of about this — then dig there for the buried insight",,signature,personal|diagnosis,solo
+introspective_delight,Values Archaeology,Keep asking 'why do I care?' until you hit bedrock: the non-negotiable value secretly steering the choice,,signature,personal|strategy,solo
+introspective_delight,Future Self Interview,Interview your wise 80-year-old self about this problem and write down the advice they give you,,signature,personal,solo
+introspective_delight,Body Wisdom Dialogue,"Scan for the tension, flutter, or gut pull each option triggers; let the body's yes/no drive the ideas",,signature,personal,solo
+introspective_delight,Permission Giving,"Write yourself an explicit permission slip to think the forbidden, impossible thought — then think it out loud",,signature,personal|unstuck,solo
+introspective_delight,Secret Wish Confession,"Whisper the embarrassing thing you secretly want here but won't admit, then build the idea honoring it",,signature,personal,solo
+introspective_delight,Mood Weather Report,"Name the inner weather right now (fog, storm, sun) and let that exact emotional climate generate the ideas",,signature,personal|unstuck,solo
+structured,SCAMPER Method,"Run your idea through seven lenses: Substitute, Combine, Adapt, Modify, Put-to-other-use, Eliminate, Reverse",,classic,feature|novel,either
+structured,Six Thinking Hats,"Examine the problem six ways one at a time: facts, feelings, benefits, risks, new ideas, process",,classic,strategy|diagnosis|planning|personal,either
+structured,Decision Tree Mapping,"Chart every choice point and the paths it forks into, following each branch to its outcome and risk",,signature,planning|strategy|diagnosis,either
+structured,Solution Matrix,"Grid problem variables against solution approaches, score every cell, hunt the best pairings and empty gaps",,signature,feature|planning,either
+structured,Trait Transfer,"Name what makes an unrelated success work, then graft those winning traits onto your own problem",,signature,novel|feature,either
+structured,Lotus Blossom,"Put the theme at the center of a 3x3 grid, fill the 8 cells around it, then promote each of those to the center of its own new 3x3",,classic,feature|planning|novel,either
+structured,Worst Possible Idea,"Deliberately generate the most terrible solutions you can, then flip each into what it teaches you to do right",,classic,unstuck|novel,either
+structured,Disney Method,"Cycle the idea through three rooms: Dreamer (anything goes), Realist (how we'd build it), Critic (what breaks)",,classic,feature|strategy|planning,either
+structured,Starbursting,"Interrogate the idea with only questions — who, what, where, when, why, how — exhaust each before answering any",,classic,feature|planning|diagnosis,either
+structured,Mind Mapping,"Branch the central topic outward, each node spawning children; follow tangents wherever they pull and let the web sprawl",,classic,planning|novel|feature,either
+structured,Crazy 8s,"Eight ideas in eight minutes, one per box, no editing — speed outruns your inner critic",,classic,feature|novel|unstuck,either
+theatrical,Time Travel Talk Show,"Host a talk show interviewing your past, present, and future selves to mine each era for advice on the problem",,playful,novel|personal,either
+theatrical,Alien Anthropologist,"Become a baffled alien studying the problem and narrate aloud what seems strange, arbitrary, or insane about it",,playful,diagnosis|unstuck|strategy,either
+theatrical,Dream Fusion Laboratory,"Voice the impossible fantasy solution first, then reverse-engineer the bridging steps back to reality",,signature,novel|unstuck,either
+theatrical,Emotion Orchestra,"Run a separate ideation round led by each emotion (rage, joy, fear, hope), then harmonize their conflicting ideas",,playful,personal|strategy,either
+theatrical,Parallel Universe Cafe,"Rewrite one fundamental rule of reality (physics, economics, social norms) and solve the problem under those laws",,playful,novel|unstuck,either
+theatrical,Persona Journey,"Embody an archetype and solve the problem in-character, naming what that persona sees that you normally miss",,signature,feature|strategy,either
+theatrical,Devil's Advocate Courtroom,"Stage a trial: prosecute the idea, defend it, then deliver the jury verdict, each role argued fully in character",,signature,strategy|diagnosis,group
+wild,Chaos Engineering,"Deliberately break your idea every way it could fail, then rebuild only the parts that survive the wreckage",,signature,feature|diagnosis|strategy,either
+wild,Guerrilla Gardening Ideas,Plant your solution in the least expected place and let it grow underground until it surprises everyone,,playful,strategy|unstuck,either
+wild,Pirate Code Brainstorm,"Steal the best bits from anywhere, remix without asking permission, grab what works and run",,playful,novel|unstuck,either
+wild,Zombie Apocalypse Planning,"Society just collapsed — strip your idea to only what survives with no power, no rules, no backup",,playful,feature|strategy|unstuck,either
+wild,Drunk History Retelling,"Explain it like you're three drinks in: no filter, no jargon, just the raw stupid-simple truth",,playful,unstuck|diagnosis,either
+wild,Anti-Solution,"Brainstorm how to make the problem spectacularly worse, then invert every sabotage into a fix",,signature,diagnosis|unstuck,either
+wild,Elemental Forces,"Let fire, water, earth, and air each sculpt your idea their own brutal way and see what survives",,playful,novel|unstuck,either
+biomimetic,Nature's Solutions,"Name an organism that already solved your problem, then copy its mechanism into your design",,signature,feature|novel,either
+biomimetic,Ecosystem Thinking,"Map your problem as an ecosystem: who eats whom, who partners, what decays, what fills the gaps",,signature,strategy|diagnosis,either
+biomimetic,Evolutionary Pressure,"Spawn many ugly variants, apply a brutal selection rule, breed the survivors, repeat until it adapts",,signature,feature|novel,either
+biomimetic,Predator & Prey,"Pick a threat to your idea, then design the defense, camouflage, or escape an animal would evolve against it",,signature,strategy|feature,either
+biomimetic,Metamorphosis Stages,"Force your idea through egg, larva, pupa, adult: a radically different form and purpose at each life stage",,signature,novel|strategy,either
+biomimetic,Swarm Logic,Forbid the master plan: solve it with dumb local rules each agent follows so order emerges from the bottom up,,signature,feature|strategy,either
+quantum,Observer Effect,"Ask how the act of watching, measuring, or shipping this idea changes the very thing you're trying to capture",,signature,strategy|diagnosis,either
+quantum,Entanglement Thinking,Pair two distant parts of the problem and insist a change in one instantly flips the other — surface the hidden linkage,,signature,diagnosis|strategy,either
+quantum,Superposition Collapse,"Hold all rival solutions alive at once, then name the one constraint that collapses them to a single winner",,signature,strategy|diagnosis,either
+quantum,Relativity Frame Shift,"Re-run the idea from a wildly different observer's reference frame — the slow user, the rival, future-you — and see what warps",,signature,strategy|novel,either
+quantum,Field Lines,Treat the goal as a charge and map the invisible forces pulling every stakeholder toward or away from it,,signature,strategy,either
+quantum,Quantum Tunneling,"Assume the idea can pass straight through the 'impossible' barrier instead of over it — what's on the other side, reached cheaply",,signature,unstuck|novel,either
+cultural,Indigenous Wisdom,"Ask how an indigenous or traditional knowledge system would approach this — name the culture, channel its ancestral problem-solving",,signature,personal|strategy|novel,either
+cultural,Fusion Cuisine,Pick two unrelated cultures and force-blend their approaches; harvest the hybrid that neither alone would invent,,signature,novel,either
+cultural,Ritual Innovation,"Redesign the idea as a ceremony — define the threshold, the gestures, the transformation participants undergo",,signature,novel|personal,either
+cultural,Mythic Frameworks,"Map the problem onto a myth: name the archetypes, find the parallel tale, let its structure dictate the resolution",,signature,strategy|personal|novel,either
+cultural,Proverb Mining,"Collect proverbs from many cultures on this theme, then build the solution from the one that clashes hardest with your assumptions",,signature,personal|strategy,either
+cultural,Ancestor Council,"Convene three ancestors or elders from different traditions, voice each one's verdict on your idea, reconcile their disagreement",,signature,personal|strategy,either
+cultural,Trickster's Gambit,"Channel the trickster figure — coyote, Anansi, Loki — and solve it by cheating, inverting, or breaking the sacred rule",,playful,unstuck|strategy,either
+absurdist,Villain's Monologue,Pitch your problem as an evil mastermind gloating about their scheme; the diabolical plan reveals the real solution,,playful,diagnosis|strategy|unstuck,either
+absurdist,Explain It to a Golden Retriever,"Re-pitch the idea to an excitable dog who only cares about treats, balls, and naps; keep only what survives",,playful,unstuck|diagnosis|feature,either
+absurdist,Infomercial at 3AM,"Sell your half-baked idea as a desperate late-night infomercial: 'But wait, there's more!' until features fall out",,playful,strategy|novel,either
+absurdist,Drunk Uncle at Thanksgiving,"Have your loudest, least-filtered relative rant about the problem; mine the unhinged hot takes for buried truth",,playful,unstuck|diagnosis,either
+absurdist,Cursed Genie,"Make a wish, then let a malicious genie grant it in the most technically-correct disastrous way; patch each loophole",,playful,diagnosis|feature,either
+absurdist,Three Rounds of Stupid,"Round 1 absurd ideas, Round 2 make each MORE absurd, Round 3 find the smallest serious thing hiding in the silliest",,playful,unstuck|novel,either
+constraint,Kill the Crown Jewel,"Delete the single best, most beloved feature — now redesign the whole thing to win without it",,signature,feature|strategy|unstuck,either
+constraint,1000x Budget,"Pretend money, time, and people are infinite — design the absurd version, then mine it for ideas you can actually steal",,signature,novel|strategy,either
+constraint,Ship in 60 Minutes,"You launch in one hour with what's already on hand — name what you cut, fake, or borrow to make it real",,signature,feature|planning|unstuck,either
+constraint,The $0 Mandate,"Achieve the goal spending literally nothing — no tools, hires, or ads; only people, favors, and what you own",,signature,planning|strategy|feature,either
+constraint,One Feature Only,"You may keep exactly ONE capability and nothing else — pick it, then make that single thing unbelievably good",,signature,feature|strategy,either
+constraint,Crank the Dial to 11,"Pick one dimension and exaggerate it to a ludicrous extreme — fastest, biggest, cheapest, weirdest — and see what breaks open",,signature,novel|unstuck,either
+constraint,Constraint Roulette,"Each round draw a brutal random limit (no screens, half the team, one day) and re-solve under it; survivors become real ideas",,signature,unstuck|feature,either
+speculative_future,Time Horizon Ladder,"Solve the idea for 1 year out, then 10, then 100 — note what survives, breaks, or becomes absurd at each rung",,signature,strategy|planning|novel,either
+speculative_future,Post-Scarcity Test,"Assume the core constraint (money, energy, time, attention) is now infinite and free — what does the idea become",,signature,novel|strategy,either
+speculative_future,Utopia vs Dystopia Split-Screen,Write the same future twice: the brochure where it went perfectly and the headline where it went horribly,,signature,strategy|diagnosis,either
+speculative_future,Sci-Fi Artifact From the Future,"Describe one physical object, ad, or news clip from the world where this idea already won — reverse-engineer it",,signature,novel|feature,either
+speculative_future,Emerging Tech Collision,"Force-marry your idea to a frontier tech (AGI, fusion, neural implants, gene edit) and ask what new thing is born",,signature,novel|feature|strategy,either
+speculative_future,What-If-The-World-Changed Card Flip,"Draw a wild world-shift (no privacy, half population, 200-yr lifespans) and redesign the idea to fit that world",,signature,novel|unstuck,either
+speculative_future,Future Anthropologist Dig,"A scholar in 2200 unearths your idea as a relic — what do they conclude it reveals about us, and what replaced it",,signature,strategy|novel,either
+structured,How Might We,"Reframe the problem as a batch of 'How might we...' opportunity questions first, then ideate against the sharpest one",,classic,feature|novel|strategy|diagnosis,either
+structured,Job to Be Done,"Ask what the user is really hiring this to do, then ideate around that underlying job, not the feature you assumed",,classic,feature|strategy|novel,either
+structured,Empathy Map,"Map what the user says, thinks, does, and feels around the problem, then mine each quadrant for the unmet need",,classic,feature|personal,either
+structured,Backcasting,"Fix the finished future in vivid detail, then work backward step by step to the one move you'd have to make first",,classic,strategy|planning|novel,either
+deep,TRIZ Contradiction,"Name the core contradiction (what only improves by making something else worse), then brainstorm ways to win both instead of trading off",,classic,feature|novel|diagnosis,either
+deep,Fishbone Diagram,"Branch the problem's spine into cause categories (people, process, tools, environment) and mine each bone for contributing causes",,classic,diagnosis,either
+deep,Build on What Works,"Name what's already succeeding and why, then ideate how to amplify and extend it instead of fixing what's broken",,classic,personal|strategy,either
+speculative_future,Scenario Cross,"Pick two high-impact uncertainties, cross them into four futures, and ideate the move that wins in every one",,classic,strategy|planning,either
diff --git a/.agents/skills/bmad-brainstorming/assets/brain-selector.html b/.agents/skills/bmad-brainstorming/assets/brain-selector.html
new file mode 100644
index 00000000..ac649b24
--- /dev/null
+++ b/.agents/skills/bmad-brainstorming/assets/brain-selector.html
@@ -0,0 +1,328 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>BMad Method Brainstorming Selection</title>
+<script>
+/* set the theme before first paint so there's no light-mode flash */
+(function(){ try {
+  var t = localStorage.getItem('bmad-theme');
+  if (!t) { t = (window.matchMedia && window.matchMedia('(prefers-color-scheme: dark)').matches) ? 'dark' : 'light'; }
+  document.documentElement.setAttribute('data-theme', t);
+} catch(e){} })();
+</script>
+<style>
+  :root {
+    --bg:#f6f7fb; --surface:#fff; --ink:#1c1e2b; --muted:#6b7080;
+    --accent:#5b4bdc; --accent-ink:#5b4bdc; --warn:#c0561f;
+    --line:#e6e8f0; --control:#eef0f7; --control2:#f1f2f8; --raised:#fff;
+    --cnt:#b9bdce; --foot:#aeb2c4; --shadow:rgba(20,20,50,.06);
+  }
+  :root[data-theme="dark"] {
+    --bg:#0f1117; --surface:#171a23; --ink:#e7e9f2; --muted:#9aa0b4;
+    --accent:#6d5cf0; --accent-ink:#a99bff; --warn:#e08a4a;
+    --line:#2a2f3e; --control:#222634; --control2:#1d212d; --raised:#2c3242;
+    --cnt:#5a6076; --foot:#5a6076; --shadow:rgba(0,0,0,.45);
+  }
+  /* lift the category hue toward white on dark surfaces so deep hues stay legible */
+  :root[data-theme="dark"] section > h2 { color:color-mix(in srgb, var(--c) 62%, #fff); }
+  :root[data-theme="dark"] .tech .ico { color:color-mix(in srgb, var(--c) 68%, #fff); }
+  :root[data-theme="dark"] label.tech:has(input:checked) { border-color:color-mix(in srgb, var(--c) 60%, #fff); }
+  .titlerow { display:flex; align-items:flex-start; justify-content:space-between; gap:12px; }
+  .themebtn { flex:none; width:36px; height:36px; border-radius:9px; background:var(--control); color:var(--ink); font-size:17px; line-height:1; display:inline-flex; align-items:center; justify-content:center; }
+  .themebtn:hover { background:var(--raised); }
+  * { box-sizing:border-box; }
+  body { margin:0; font:16px/1.5 -apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,Helvetica,Arial,sans-serif; background:var(--bg); color:var(--ink); }
+  header { position:sticky; top:0; z-index:5; background:var(--surface); padding:20px 0 12px; border-bottom:1px solid var(--line); box-shadow:0 2px 12px var(--shadow); }
+  .hwrap { max-width:1120px; margin:0 auto; padding:0 24px; }  /* align header content with the card column on wide screens */
+  h1 { margin:0 0 4px; font-size:24px; letter-spacing:-.02em; }
+  .sub { margin:0 0 12px; color:var(--muted); font-size:14px; max-width:74ch; }
+  button { font:inherit; border:0; border-radius:8px; cursor:pointer; }
+  .composer { display:flex; flex-direction:column; gap:9px; margin:6px 0 12px; }
+  .grp { display:flex; gap:8px; align-items:center; flex-wrap:wrap; }
+  .glabel { font-size:11px; text-transform:uppercase; letter-spacing:.07em; color:var(--muted); min-width:74px; }
+  .modes { display:inline-flex; background:var(--control); border-radius:9px; padding:3px; gap:2px; }
+  .mode { padding:7px 13px; font-size:14px; font-weight:600; color:var(--muted); background:transparent; }
+  .mode.on { background:var(--raised); color:var(--accent-ink); box-shadow:0 1px 3px var(--shadow); }
+  .modehint { flex:1 1 240px; min-width:0; font-size:13px; color:var(--muted); font-style:italic; }
+  .pill { font-size:13px; color:var(--muted); background:var(--control); padding:6px 12px; border-radius:20px; }
+  .pill b { color:var(--accent-ink); }
+  .step { display:inline-flex; align-items:center; gap:7px; font-size:13px; color:var(--ink); background:var(--control2); padding:4px 6px 4px 12px; border-radius:20px; }
+  .step b { min-width:12px; text-align:center; font-size:14px; color:var(--ink); }
+  .step button { width:24px; height:24px; border-radius:50%; background:var(--raised); color:var(--muted); font-size:17px; line-height:22px; text-align:center; box-shadow:0 1px 2px var(--shadow); }
+  .step button:hover { color:var(--accent-ink); }
+  .total { font-size:12px; color:var(--muted); }
+  .total.warn { color:var(--warn); font-weight:600; }
+  .bar { display:flex; gap:10px 14px; align-items:center; flex-wrap:wrap; }
+  #copy { margin-left:auto; padding:9px 22px; background:var(--accent); color:#fff; font-size:14px; font-weight:700; }
+  #copy:hover { filter:brightness(1.07); }
+  .chips { flex:1 1 320px; min-width:0; display:flex; gap:7px; flex-wrap:wrap; align-items:center; }
+  .chip { font-size:12px; padding:4px 11px; border-radius:16px; border:0; color:#fff; background:var(--cc); font-weight:600; cursor:pointer; }
+  .chip:hover { filter:brightness(1.08); }
+  .banner { max-height:0; overflow:hidden; transition:max-height .25s ease, padding .22s ease, margin .22s ease; background:linear-gradient(90deg,var(--accent),#8275f2); color:#fff; border-radius:10px; font-weight:700; text-align:center; padding:0 14px; }
+  .banner.show { max-height:64px; padding:13px 14px; margin-top:10px; }
+  .banner.fail { background:linear-gradient(90deg,var(--warn),#e0894a); }
+  main { padding:18px 24px 60px; max-width:1120px; margin:0 auto; }
+  section { margin:0 0 26px; }
+  section > h2 { font-size:13px; text-transform:uppercase; letter-spacing:.08em; color:var(--c); margin:0 0 10px; border-bottom:1px solid color-mix(in srgb, var(--c) 24%, var(--line)); padding-bottom:6px; }
+  section > h2 .cnt { color:color-mix(in srgb, var(--c) 45%, var(--cnt)); margin-left:6px; }
+  .grid { display:grid; grid-template-columns:repeat(auto-fill,minmax(360px,1fr)); gap:10px; }
+  label.tech { display:flex; gap:12px; align-items:flex-start; background:color-mix(in srgb, var(--c) 5%, var(--surface)); border:1px solid color-mix(in srgb, var(--c) 18%, var(--line)); border-radius:10px; padding:11px 13px; cursor:pointer; transition:border-color .12s, box-shadow .12s, background .12s; }
+  label.tech:hover { border-color:color-mix(in srgb, var(--c) 45%, var(--surface)); }
+  label.tech input { margin-top:2px; width:17px; height:17px; accent-color:var(--c); flex:none; }
+  label.tech:has(input:checked) { border-color:var(--c); background:color-mix(in srgb, var(--c) 12%, var(--surface)); box-shadow:0 0 0 2px color-mix(in srgb, var(--c) 30%, transparent); }
+  .tech .ic2 { display:flex; gap:5px; flex:none; }
+  .tech .ico { width:40px; height:40px; flex:none; color:var(--c); }
+  .tech .n { font-weight:600; display:block; }
+  .tech .d { color:var(--muted); font-size:13.5px; display:block; margin-top:2px; }
+  .tech .gf { color:var(--accent-ink); font-size:11px; display:block; margin-top:5px; opacity:.85; }
+  .grouphdr { margin:30px 0 12px; font-size:12px; text-transform:uppercase; letter-spacing:.14em; font-weight:700; color:var(--c); opacity:.92; border-bottom:1px solid color-mix(in srgb, var(--c) 22%, var(--line)); padding-bottom:7px; }
+  main > .grouphdr:first-child { margin-top:2px; }
+  :root[data-theme="dark"] .grouphdr { color:color-mix(in srgb, var(--c) 62%, #fff); }
+  .goals { display:flex; gap:7px; flex-wrap:wrap; }
+  .goal { font-size:12px; padding:5px 12px; border-radius:16px; background:var(--control); color:var(--muted); font-weight:600; }
+  .goal:hover { color:var(--ink); }
+  .goal.on { background:var(--accent); color:#fff; }
+  label.tech.invent { border-style:dashed; background:transparent; }
+  label.tech.invent:hover { border-color:var(--c); }
+  label.tech.invent .n { color:var(--c); }
+  label.tech.hidden { display:none; }
+  footer { text-align:center; color:var(--foot); font-size:12px; padding:24px; }
+</style>
+</head>
+<body>
+<header>
+  <div class="hwrap">
+  <div class="titlerow">
+    <h1>BMad Method Brainstorming Selection</h1>
+    <button id="theme" class="themebtn" type="button" aria-label="Toggle dark mode" title="Toggle dark mode"></button>
+  </div>
+  <p class="sub">Compose your session, hit <strong>Copy prompt</strong>, and paste it back into the chat to begin. 108 techniques across 13 categories.</p>
+
+  <div class="composer">
+    <div class="grp">
+      <span class="glabel">Facilitation</span>
+      <div class="modes" id="modes">
+        <button type="button" class="mode on" data-mode="Facilitator">Facilitator</button>
+        <button type="button" class="mode" data-mode="Creative Partner">Creative Partner</button>
+        <button type="button" class="mode" data-mode="Ideate for me">Ideate for me</button>
+      </div>
+      <span class="modehint" id="modehint"></span>
+    </div>
+    <div class="grp">
+      <span class="glabel">Techniques</span>
+      <span class="pill">Picked <b id="pickN">0</b></span>
+      <span class="step">Random <button type="button" data-step="rand" data-d="-1">&minus;</button><b id="randN">0</b><button type="button" data-step="rand" data-d="1">+</button></span>
+      <span class="step">Invent <button type="button" data-step="inv" data-d="-1">&minus;</button><b id="invN">0</b><button type="button" data-step="inv" data-d="1">+</button></span>
+      <span class="step">AI picks <button type="button" data-step="ai" data-d="-1">&minus;</button><b id="aiN">0</b><button type="button" data-step="ai" data-d="1">+</button></span>
+      <span class="total" id="total">Total 0 &middot; 3&ndash;4 is the sweet spot</span>
+      <button id="copy" type="button">Copy prompt</button>
+    </div>
+  </div>
+
+  <div class="bar"><span class="glabel">Great for</span><div class="goals" id="goals"><button type="button" class="goal" data-goal="feature">Build a feature</button><button type="button" class="goal" data-goal="novel">Novel concept</button><button type="button" class="goal" data-goal="strategy">Strategy</button><button type="button" class="goal" data-goal="planning">Planning</button><button type="button" class="goal" data-goal="diagnosis">Diagnose</button><button type="button" class="goal" data-goal="personal">Personal / life</button><button type="button" class="goal" data-goal="unstuck">Get unstuck</button></div></div>
+  <div class="bar">
+    <span class="glabel">Jump to</span>
+    <div class="chips" id="chips"><button type="button" class="chip" data-cat="Proven &amp; Professional" style="--cc:#3d4f73">Proven &amp; Professional</button><button type="button" class="chip" data-cat="Structured" style="--cc:#3b6ea5">Structured</button><button type="button" class="chip" data-cat="Deep" style="--cc:#4658c9">Deep</button><button type="button" class="chip" data-cat="Creative" style="--cc:#6d5cf0">Creative</button><button type="button" class="chip" data-cat="Biomimetic" style="--cc:#1f9d6b">Biomimetic</button><button type="button" class="chip" data-cat="Cultural" style="--cc:#c75b39">Cultural</button><button type="button" class="chip" data-cat="Speculative Future" style="--cc:#0fb5c9">Speculative Future</button><button type="button" class="chip" data-cat="Quantum" style="--cc:#2b86d9">Quantum</button><button type="button" class="chip" data-cat="Wild" style="--cc:#e2562f">Wild</button><button type="button" class="chip" data-cat="Absurdist" style="--cc:#e0529c">Absurdist</button><button type="button" class="chip" data-cat="Theatrical" style="--cc:#cf4d6f">Theatrical</button><button type="button" class="chip" data-cat="Constraint" style="--cc:#d9882b">Constraint</button><button type="button" class="chip" data-cat="Introspective Delight" style="--cc:#b15ad6">Introspective Delight</button><button type="button" class="chip" data-cat="Collaborative" style="--cc:#15a3a3">Collaborative</button></div>
+  </div>
+
+  <div class="banner" id="banner">&#10003; Copied! Now paste it into the chat to start your session.</div>
+  </div>
+</header>
+<main>
+<section data-cat="Proven &amp; Professional" style="--c:#3d4f73"><h2>Proven &amp; Professional<span class="cnt">29</span></h2><div class="grid"><label class="tech" style="--c:#15a3a3"><input type="checkbox" data-name="Yes And Building" data-cat="Collaborative" data-desc="Never negate; each person opens with &quot;Yes, and...&quot; and adds to the last idea, stacking a chain of accepted additions" data-good="novel|unstuck|planning" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8"><line x1="14" y1="16" x2="30" y2="16"/><line x1="14" y1="16" x2="22" y2="30"/><line x1="30" y1="16" x2="22" y2="30"/></g><g fill="currentColor" fill-opacity="0.22"><circle cx="14" cy="16" r="4.6"/><circle cx="30" cy="16" r="4.6"/><circle cx="22" cy="30" r="4.6"/></g><g fill="currentColor"><circle cx="14" cy="16" r="2.4"/><circle cx="30" cy="16" r="2.4"/><circle cx="22" cy="30" r="2.4"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="8" y="27" width="12" height="8" rx="1.5" fill-opacity=".8"/><rect x="14" y="19" width="12" height="8" rx="1.5" fill-opacity=".5"/><rect x="20" y="11" width="12" height="8" rx="1.5"/></g></svg></span><span><span class="n">Yes And Building</span><span class="d">Never negate; each person opens with &quot;Yes, and...&quot; and adds to the last idea, stacking a chain of accepted additions</span><span class="gf">Great for: Novel concept · Get unstuck · Planning</span></span></label><label class="tech" style="--c:#15a3a3"><input type="checkbox" data-name="Brain Writing Round Robin" data-cat="Collaborative" data-desc="Everyone writes ideas silently, then passes their sheet; you build on whatever lands in front of you, round after round" data-good="novel|feature" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8"><line x1="14" y1="16" x2="30" y2="16"/><line x1="14" y1="16" x2="22" y2="30"/><line x1="30" y1="16" x2="22" y2="30"/></g><g fill="currentColor" fill-opacity="0.22"><circle cx="14" cy="16" r="4.6"/><circle cx="30" cy="16" r="4.6"/><circle cx="22" cy="30" r="4.6"/></g><g fill="currentColor"><circle cx="14" cy="16" r="2.4"/><circle cx="30" cy="16" r="2.4"/><circle cx="22" cy="30" r="2.4"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M31 16 A10 10 0 1 0 32.5 22"/><path d="M31 10 L31.5 16.3 L25 16.5"/></g></svg></span><span><span class="n">Brain Writing Round Robin</span><span class="d">Everyone writes ideas silently, then passes their sheet; you build on whatever lands in front of you, round after round</span><span class="gf">Great for: Novel concept · Build a feature</span></span></label><label class="tech" style="--c:#15a3a3"><input type="checkbox" data-name="Random Stimulation" data-cat="Collaborative" data-desc="Pull a random word or image and force a link to the problem: &quot;how does THIS spark a solution?&quot;" data-good="unstuck|novel" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8"><line x1="14" y1="16" x2="30" y2="16"/><line x1="14" y1="16" x2="22" y2="30"/><line x1="30" y1="16" x2="22" y2="30"/></g><g fill="currentColor" fill-opacity="0.22"><circle cx="14" cy="16" r="4.6"/><circle cx="30" cy="16" r="4.6"/><circle cx="22" cy="30" r="4.6"/></g><g fill="currentColor"><circle cx="14" cy="16" r="2.4"/><circle cx="30" cy="16" r="2.4"/><circle cx="22" cy="30" r="2.4"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><rect x="11" y="11" width="22" height="22" rx="4" fill="none" stroke="currentColor" stroke-width="2"/><g fill="currentColor"><circle cx="17" cy="17" r="1.8"/><circle cx="27" cy="17" r="1.8"/><circle cx="22" cy="22" r="1.8"/><circle cx="17" cy="27" r="1.8"/><circle cx="27" cy="27" r="1.8"/></g></svg></span><span><span class="n">Random Stimulation</span><span class="d">Pull a random word or image and force a link to the problem: &quot;how does THIS spark a solution?&quot;</span><span class="gf">Great for: Get unstuck · Novel concept</span></span></label><label class="tech" style="--c:#15a3a3"><input type="checkbox" data-name="Role Playing" data-cat="Collaborative" data-desc="Each person speaks as a different stakeholder, voicing what that role wants, fears, and would demand of the idea" data-good="strategy|personal|feature" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8"><line x1="14" y1="16" x2="30" y2="16"/><line x1="14" y1="16" x2="22" y2="30"/><line x1="30" y1="16" x2="22" y2="30"/></g><g fill="currentColor" fill-opacity="0.22"><circle cx="14" cy="16" r="4.6"/><circle cx="30" cy="16" r="4.6"/><circle cx="22" cy="30" r="4.6"/></g><g fill="currentColor"><circle cx="14" cy="16" r="2.4"/><circle cx="30" cy="16" r="2.4"/><circle cx="22" cy="30" r="2.4"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><rect x="11" y="9" width="22" height="26" rx="3" fill="none" stroke="currentColor" stroke-width="2"/><circle cx="22" cy="19" r="4" fill="currentColor"/><path d="M15.5 30 c2 -4.5 11 -4.5 13 0" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"/></svg></span><span><span class="n">Role Playing</span><span class="d">Each person speaks as a different stakeholder, voicing what that role wants, fears, and would demand of the idea</span><span class="gf">Great for: Strategy · Personal / life · Build a feature</span></span></label><label class="tech" style="--c:#6d5cf0"><input type="checkbox" data-name="First Principles Thinking" data-cat="Creative" data-desc="Strip every assumption to bedrock facts, then rebuild the solution from scratch on truth alone" data-good="feature|novel|diagnosis|strategy" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.4" stroke-linecap="round"><line x1="22" y1="6.5" x2="22" y2="12.5"/><line x1="22" y1="31.5" x2="22" y2="37.5"/><line x1="6.5" y1="22" x2="12.5" y2="22"/><line x1="31.5" y1="22" x2="37.5" y2="22"/><line x1="11.3" y1="11.3" x2="15.5" y2="15.5"/><line x1="28.5" y1="28.5" x2="32.7" y2="32.7"/><line x1="32.7" y1="11.3" x2="28.5" y2="15.5"/><line x1="15.5" y1="28.5" x2="11.3" y2="32.7"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="3.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="10" y="28" width="8" height="6" rx="1"/><rect x="18.5" y="28" width="8" height="6" rx="1"/><rect x="27" y="28" width="7" height="6" rx="1"/></g><path d="M22 25 L22 11 M16 17 L22 11 L28 17" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"/></svg></span><span><span class="n">First Principles Thinking</span><span class="d">Strip every assumption to bedrock facts, then rebuild the solution from scratch on truth alone</span><span class="gf">Great for: Build a feature · Novel concept · Diagnose · Strategy</span></span></label><label class="tech" style="--c:#6d5cf0"><input type="checkbox" data-name="Reverse Brainstorming" data-cat="Creative" data-desc="Generate problems instead of solutions — &#x27;how could we make this fail?&#x27; — then mine each for its inverse" data-good="diagnosis|feature|unstuck" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.4" stroke-linecap="round"><line x1="22" y1="6.5" x2="22" y2="12.5"/><line x1="22" y1="31.5" x2="22" y2="37.5"/><line x1="6.5" y1="22" x2="12.5" y2="22"/><line x1="31.5" y1="22" x2="37.5" y2="22"/><line x1="11.3" y1="11.3" x2="15.5" y2="15.5"/><line x1="28.5" y1="28.5" x2="32.7" y2="32.7"/><line x1="32.7" y1="11.3" x2="28.5" y2="15.5"/><line x1="15.5" y1="28.5" x2="11.3" y2="32.7"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="3.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M13 18 H28 a4 4 0 0 1 0 8 H16"/><path d="M19 15 L13 18 L19 21 M22 23 L16 26 L22 29"/></g></svg></span><span><span class="n">Reverse Brainstorming</span><span class="d">Generate problems instead of solutions — &#x27;how could we make this fail?&#x27; — then mine each for its inverse</span><span class="gf">Great for: Diagnose · Build a feature · Get unstuck</span></span></label><label class="tech" style="--c:#4658c9"><input type="checkbox" data-name="Five Whys" data-cat="Deep" data-desc="Ask &quot;why?&quot; five times in a chain, each answer feeding the next, until you hit the root cause beneath the symptom" data-good="diagnosis" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><circle cx="14" cy="13" r="2.4"/><circle cx="22" cy="22" r="2.4"/><circle cx="30" cy="31" r="2.4"/><path d="M15.6 14.8 L20.4 20.2 M23.6 23.8 L28.4 29.2"/></g></svg></span><span><span class="n">Five Whys</span><span class="d">Ask &quot;why?&quot; five times in a chain, each answer feeding the next, until you hit the root cause beneath the symptom</span><span class="gf">Great for: Diagnose</span></span></label><label class="tech" style="--c:#4658c9"><input type="checkbox" data-name="Provocation Technique" data-cat="Deep" data-desc="State something deliberately absurd, then mine it: &quot;how could this be useful?&quot; Extract the usable principle hiding inside" data-good="unstuck|novel" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M24 9 L13 24 H21 L19 35 L31 19 H23 Z"/></g></svg></span><span><span class="n">Provocation Technique</span><span class="d">State something deliberately absurd, then mine it: &quot;how could this be useful?&quot; Extract the usable principle hiding inside</span><span class="gf">Great for: Get unstuck · Novel concept</span></span></label><label class="tech" style="--c:#4658c9"><input type="checkbox" data-name="Assumption Reversal" data-cat="Deep" data-desc="List every assumption baked into the problem, flip each to its opposite, then rebuild a solution on the inverted foundation" data-good="novel|diagnosis|strategy" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M16 14 V30"/><path d="M11.5 25 L16 30 L20.5 25"/><path d="M28 30 V14"/><path d="M23.5 19 L28 14 L32.5 19"/></g></svg></span><span><span class="n">Assumption Reversal</span><span class="d">List every assumption baked into the problem, flip each to its opposite, then rebuild a solution on the inverted foundation</span><span class="gf">Great for: Novel concept · Diagnose · Strategy</span></span></label><label class="tech" style="--c:#4658c9"><input type="checkbox" data-name="Question Storming" data-cat="Deep" data-desc="Generate only questions about the problem, zero answers allowed, until the real problem worth solving comes into focus" data-good="diagnosis|strategy|unstuck" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M14 16 a3.2 3.2 0 1 1 3.2 3.2 v2"/><path d="M26 13 a3.6 3.6 0 1 1 3.6 3.6 v2.4"/></g><circle cx="17.2" cy="27" r="1.5" fill="currentColor"/><circle cx="29.6" cy="25.6" r="1.6" fill="currentColor"/></svg></span><span><span class="n">Question Storming</span><span class="d">Generate only questions about the problem, zero answers allowed, until the real problem worth solving comes into focus</span><span class="gf">Great for: Diagnose · Strategy · Get unstuck</span></span></label><label class="tech" style="--c:#4658c9"><input type="checkbox" data-name="Causal Loop Mapping" data-cat="Deep" data-desc="Diagram the feedback loops linking causes and effects, find the reinforcing and balancing cycles, and target the leverage point" data-good="diagnosis|strategy" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M16 16 a9 9 0 1 1 -2 12"/><path d="M16 10.5 L16.5 16.5 L10.5 17"/><path d="M30 28.5 L29.5 22.5 L35 22"/></g></svg></span><span><span class="n">Causal Loop Mapping</span><span class="d">Diagram the feedback loops linking causes and effects, find the reinforcing and balancing cycles, and target the leverage point</span><span class="gf">Great for: Diagnose · Strategy</span></span></label><label class="tech" style="--c:#4658c9"><input type="checkbox" data-name="Morphological Analysis" data-cat="Deep" data-desc="List the problem&#x27;s independent parameters, generate options for each, then combine across them to surface untried configurations" data-good="feature|novel|planning" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="1.8"><rect x="11" y="11" width="22" height="22" rx="2"/><path d="M11 18.3 H33 M11 25.6 H33 M18.3 11 V33 M25.6 11 V33"/></g><rect x="18.5" y="18.5" width="7" height="7" fill="currentColor" fill-opacity=".4"/></svg></span><span><span class="n">Morphological Analysis</span><span class="d">List the problem&#x27;s independent parameters, generate options for each, then combine across them to surface untried configurations</span><span class="gf">Great for: Build a feature · Novel concept · Planning</span></span></label><label class="tech" style="--c:#4658c9"><input type="checkbox" data-name="Laddering" data-cat="Deep" data-desc="Ask &#x27;and what would that give you?&#x27; up the chain until you reach the real underlying need, then ideate fresh at that level" data-good="personal|strategy|diagnosis" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M16 9 V35 M28 9 V35 M16 15 H28 M16 22 H28 M16 29 H28"/></g></svg></span><span><span class="n">Laddering</span><span class="d">Ask &#x27;and what would that give you?&#x27; up the chain until you reach the real underlying need, then ideate fresh at that level</span><span class="gf">Great for: Personal / life · Strategy · Diagnose</span></span></label><label class="tech" style="--c:#3b6ea5"><input type="checkbox" data-name="SCAMPER Method" data-cat="Structured" data-desc="Run your idea through seven lenses: Substitute, Combine, Adapt, Modify, Put-to-other-use, Eliminate, Reverse" data-good="feature|novel" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="22" r="5.5" fill="none" stroke="currentColor" stroke-width="2"/><g stroke="currentColor" stroke-width="2.4" stroke-linecap="round"><path d="M22 9 V13.5 M22 30.5 V35 M9 22 H13.5 M30.5 22 H35 M12.8 12.8 L16 16 M28 28 L31.2 31.2 M31.2 12.8 L28 16 M16 28 L12.8 31.2"/></g></svg></span><span><span class="n">SCAMPER Method</span><span class="d">Run your idea through seven lenses: Substitute, Combine, Adapt, Modify, Put-to-other-use, Eliminate, Reverse</span><span class="gf">Great for: Build a feature · Novel concept</span></span></label><label class="tech" style="--c:#3b6ea5"><input type="checkbox" data-name="Six Thinking Hats" data-cat="Structured" data-desc="Examine the problem six ways one at a time: facts, feelings, benefits, risks, new ideas, process" data-good="strategy|diagnosis|planning|personal" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M14 26 q8 -5 16 0" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"/><path d="M17 26 q-6 1 -8 3 q13 4 26 0 q-2 -2 -8 -3" fill="currentColor" fill-opacity=".22"/><path d="M17 26 c-1 -8 11 -8 10 0" fill="currentColor" fill-opacity=".5"/></svg></span><span><span class="n">Six Thinking Hats</span><span class="d">Examine the problem six ways one at a time: facts, feelings, benefits, risks, new ideas, process</span><span class="gf">Great for: Strategy · Diagnose · Planning · Personal / life</span></span></label><label class="tech" style="--c:#3b6ea5"><input type="checkbox" data-name="Lotus Blossom" data-cat="Structured" data-desc="Put the theme at the center of a 3x3 grid, fill the 8 cells around it, then promote each of those to the center of its own new 3x3" data-good="feature|planning|novel" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><circle cx="22" cy="22" r="3.4"/></g><g fill="currentColor" fill-opacity=".4"><circle cx="22" cy="13" r="2.8"/><circle cx="22" cy="31" r="2.8"/><circle cx="13" cy="22" r="2.8"/><circle cx="31" cy="22" r="2.8"/><circle cx="15.5" cy="15.5" r="2.5"/><circle cx="28.5" cy="15.5" r="2.5"/><circle cx="15.5" cy="28.5" r="2.5"/><circle cx="28.5" cy="28.5" r="2.5"/></g></svg></span><span><span class="n">Lotus Blossom</span><span class="d">Put the theme at the center of a 3x3 grid, fill the 8 cells around it, then promote each of those to the center of its own new 3x3</span><span class="gf">Great for: Build a feature · Planning · Novel concept</span></span></label><label class="tech" style="--c:#3b6ea5"><input type="checkbox" data-name="Worst Possible Idea" data-cat="Structured" data-desc="Deliberately generate the most terrible solutions you can, then flip each into what it teaches you to do right" data-good="unstuck|novel" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M17 11 v10 h-5 l10 12 10 -12 h-5 v-10 z"/></g></svg></span><span><span class="n">Worst Possible Idea</span><span class="d">Deliberately generate the most terrible solutions you can, then flip each into what it teaches you to do right</span><span class="gf">Great for: Get unstuck · Novel concept</span></span></label><label class="tech" style="--c:#3b6ea5"><input type="checkbox" data-name="Disney Method" data-cat="Structured" data-desc="Cycle the idea through three rooms: Dreamer (anything goes), Realist (how we&#x27;d build it), Critic (what breaks)" data-good="feature|strategy|planning" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2"><circle cx="14" cy="22" r="4.5"/><circle cx="22" cy="22" r="4.5"/><circle cx="30" cy="22" r="4.5"/></g></svg></span><span><span class="n">Disney Method</span><span class="d">Cycle the idea through three rooms: Dreamer (anything goes), Realist (how we&#x27;d build it), Critic (what breaks)</span><span class="gf">Great for: Build a feature · Strategy · Planning</span></span></label><label class="tech" style="--c:#3b6ea5"><input type="checkbox" data-name="Starbursting" data-cat="Structured" data-desc="Interrogate the idea with only questions — who, what, where, when, why, how — exhaust each before answering any" data-good="feature|planning|diagnosis" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M22 8 V15 M22 29 V36 M8 22 H15 M29 22 H36 M12 12 L17 17 M27 27 L32 32 M32 12 L27 17 M12 32 L17 27"/></g><path d="M19.5 19 a3.2 3.2 0 1 1 3 4 v1.2" fill="none" stroke="currentColor" stroke-width="1.8" stroke-linecap="round"/></svg></span><span><span class="n">Starbursting</span><span class="d">Interrogate the idea with only questions — who, what, where, when, why, how — exhaust each before answering any</span><span class="gf">Great for: Build a feature · Planning · Diagnose</span></span></label><label class="tech" style="--c:#3b6ea5"><input type="checkbox" data-name="Mind Mapping" data-cat="Structured" data-desc="Branch the central topic outward, each node spawning children; follow tangents wherever they pull and let the web sprawl" data-good="planning|novel|feature" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><circle cx="22" cy="22" r="4"/><circle cx="11" cy="13" r="2.2"/><circle cx="33" cy="13" r="2.2"/><circle cx="10" cy="28" r="2.2"/><circle cx="32" cy="31" r="2.2"/><path d="M19 19.5 L12.5 14.5 M25 19.5 L31.5 14.5 M19 24.5 L11.5 27 M25.5 24 L30.5 29.5"/></g></svg></span><span><span class="n">Mind Mapping</span><span class="d">Branch the central topic outward, each node spawning children; follow tangents wherever they pull and let the web sprawl</span><span class="gf">Great for: Planning · Novel concept · Build a feature</span></span></label><label class="tech" style="--c:#3b6ea5"><input type="checkbox" data-name="Crazy 8s" data-cat="Structured" data-desc="Eight ideas in eight minutes, one per box, no editing — speed outruns your inner critic" data-good="feature|novel|unstuck" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="1.7"><rect x="9" y="12" width="8" height="9" rx="1.5"/><rect x="18" y="12" width="8" height="9" rx="1.5"/><rect x="27" y="12" width="8" height="9" rx="1.5"/><rect x="9" y="23" width="8" height="9" rx="1.5"/><rect x="18" y="23" width="8" height="9" rx="1.5"/><rect x="27" y="23" width="8" height="9" rx="1.5"/></g></svg></span><span><span class="n">Crazy 8s</span><span class="d">Eight ideas in eight minutes, one per box, no editing — speed outruns your inner critic</span><span class="gf">Great for: Build a feature · Novel concept · Get unstuck</span></span></label><label class="tech" style="--c:#3b6ea5"><input type="checkbox" data-name="How Might We" data-cat="Structured" data-desc="Reframe the problem as a batch of &#x27;How might we...&#x27; opportunity questions first, then ideate against the sharpest one" data-good="feature|novel|strategy|diagnosis" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M11 13 H33 a2 2 0 0 1 2 2 V27 a2 2 0 0 1 -2 2 H24 L19 34 V29 H11 a2 2 0 0 1 -2 -2 V15 a2 2 0 0 1 2 -2 Z"/></g><path d="M19 19 a3.2 3.2 0 1 1 3.4 3.4 v1.6" fill="none" stroke="currentColor" stroke-width="1.8" stroke-linecap="round"/><circle cx="22.4" cy="27" r="1.4" fill="currentColor"/></svg></span><span><span class="n">How Might We</span><span class="d">Reframe the problem as a batch of &#x27;How might we...&#x27; opportunity questions first, then ideate against the sharpest one</span><span class="gf">Great for: Build a feature · Novel concept · Strategy · Diagnose</span></span></label><label class="tech" style="--c:#3b6ea5"><input type="checkbox" data-name="Job to Be Done" data-cat="Structured" data-desc="Ask what the user is really hiring this to do, then ideate around that underlying job, not the feature you assumed" data-good="feature|strategy|novel" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linejoin="round"><rect x="9" y="16" width="26" height="16" rx="2.5"/><path d="M17 16 v-3 a2 2 0 0 1 2 -2 h6 a2 2 0 0 1 2 2 v3"/><path d="M9 23 H35"/></g><circle cx="22" cy="23" r="1.8" fill="currentColor"/></svg></span><span><span class="n">Job to Be Done</span><span class="d">Ask what the user is really hiring this to do, then ideate around that underlying job, not the feature you assumed</span><span class="gf">Great for: Build a feature · Strategy · Novel concept</span></span></label><label class="tech" style="--c:#3b6ea5"><input type="checkbox" data-name="Empathy Map" data-cat="Structured" data-desc="Map what the user says, thinks, does, and feels around the problem, then mine each quadrant for the unmet need" data-good="feature|personal" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="1.8"><rect x="11" y="11" width="22" height="22" rx="2"/><path d="M11 22 H33 M22 11 V33"/></g><path d="M22 27 c-2.6 -2.1 -4.2 -3.4 -4.2 -5.2 a2.1 2.1 0 0 1 4.2 -1 a2.1 2.1 0 0 1 4.2 1 c0 1.8 -1.6 3.1 -4.2 5.2 z" fill="currentColor"/></svg></span><span><span class="n">Empathy Map</span><span class="d">Map what the user says, thinks, does, and feels around the problem, then mine each quadrant for the unmet need</span><span class="gf">Great for: Build a feature · Personal / life</span></span></label><label class="tech" style="--c:#3b6ea5"><input type="checkbox" data-name="Backcasting" data-cat="Structured" data-desc="Fix the finished future in vivid detail, then work backward step by step to the one move you&#x27;d have to make first" data-good="strategy|planning|novel" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M30 11 V33"/></g><path d="M30 13 L20 16.5 L30 20 Z" fill="currentColor"/><path d="M28 27 H14" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-dasharray="1.5 3"/><path d="M18 23 L14 27 L18 31" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"/></svg></span><span><span class="n">Backcasting</span><span class="d">Fix the finished future in vivid detail, then work backward step by step to the one move you&#x27;d have to make first</span><span class="gf">Great for: Strategy · Planning · Novel concept</span></span></label><label class="tech" style="--c:#4658c9"><input type="checkbox" data-name="TRIZ Contradiction" data-cat="Deep" data-desc="Name the core contradiction (what only improves by making something else worse), then brainstorm ways to win both instead of trading off" data-good="feature|novel|diagnosis" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M9 15 H18 M14.5 11.5 L18 15 L14.5 18.5"/><path d="M35 29 H26 M29.5 25.5 L26 29 L29.5 32.5"/></g><path d="M22 17.5 l1.5 3.4 3.7 .3 -2.8 2.4 .9 3.6 -3.3 -1.9 -3.3 1.9 .9 -3.6 -2.8 -2.4 3.7 -.3 z" fill="currentColor"/></svg></span><span><span class="n">TRIZ Contradiction</span><span class="d">Name the core contradiction (what only improves by making something else worse), then brainstorm ways to win both instead of trading off</span><span class="gf">Great for: Build a feature · Novel concept · Diagnose</span></span></label><label class="tech" style="--c:#4658c9"><input type="checkbox" data-name="Fishbone Diagram" data-cat="Deep" data-desc="Branch the problem&#x27;s spine into cause categories (people, process, tools, environment) and mine each bone for contributing causes" data-good="diagnosis" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M9 22 H33"/><path d="M33 22 L36 18.5 M33 22 L36 25.5"/><path d="M14 22 L18 15 M14 22 L18 29 M23 22 L27 15 M23 22 L27 29"/></g><circle cx="9" cy="22" r="1.9" fill="currentColor"/></svg></span><span><span class="n">Fishbone Diagram</span><span class="d">Branch the problem&#x27;s spine into cause categories (people, process, tools, environment) and mine each bone for contributing causes</span><span class="gf">Great for: Diagnose</span></span></label><label class="tech" style="--c:#4658c9"><input type="checkbox" data-name="Build on What Works" data-cat="Deep" data-desc="Name what&#x27;s already succeeding and why, then ideate how to amplify and extend it instead of fixing what&#x27;s broken" data-good="personal|strategy" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="24" width="6" height="9" rx="1"/><rect x="19" y="19" width="6" height="14" rx="1"/><rect x="27" y="13" width="6" height="20" rx="1"/></g><path d="M11 19 L18 13 L24 16 L33 8" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"/><path d="M28 8 H33 V13" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"/></svg></span><span><span class="n">Build on What Works</span><span class="d">Name what&#x27;s already succeeding and why, then ideate how to amplify and extend it instead of fixing what&#x27;s broken</span><span class="gf">Great for: Personal / life · Strategy</span></span></label><label class="tech" style="--c:#0fb5c9"><input type="checkbox" data-name="Scenario Cross" data-cat="Speculative Future" data-desc="Pick two high-impact uncertainties, cross them into four futures, and ideate the move that wins in every one" data-good="strategy|planning" data-prov="classic" data-lead="1"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M11 31 L 26.5 15.5"/><path d="M20 14.5 H 28 V 22.5"/></g><circle cx="31" cy="12" r="2.8" fill="currentColor"/><g stroke="currentColor" stroke-width="1.4" stroke-linecap="round"><line x1="31" y1="6.5" x2="31" y2="8.4"/><line x1="31" y1="15.6" x2="31" y2="17.5"/><line x1="25.5" y1="12" x2="27.4" y2="12"/><line x1="34.6" y1="12" x2="36.5" y2="12"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2" stroke-linecap="round"><path d="M22 9 V35 M9 22 H35"/></g><g fill="currentColor"><circle cx="15" cy="15" r="2"/><circle cx="29" cy="15" r="2"/><circle cx="15" cy="29" r="2"/><circle cx="29" cy="29" r="2"/></g></svg></span><span><span class="n">Scenario Cross</span><span class="d">Pick two high-impact uncertainties, cross them into four futures, and ideate the move that wins in every one</span><span class="gf">Great for: Strategy · Planning</span></span></label></div></section>
+<h2 class="grouphdr" style="--c:#3b6ea5">Structured &amp; Analytical</h2>
+<section data-cat="Structured" style="--c:#3b6ea5"><h2>Structured<span class="cnt">15</span></h2><div class="grid"><label class="tech"><input type="checkbox" data-name="SCAMPER Method" data-cat="Structured" data-desc="Run your idea through seven lenses: Substitute, Combine, Adapt, Modify, Put-to-other-use, Eliminate, Reverse" data-good="feature|novel" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="22" r="5.5" fill="none" stroke="currentColor" stroke-width="2"/><g stroke="currentColor" stroke-width="2.4" stroke-linecap="round"><path d="M22 9 V13.5 M22 30.5 V35 M9 22 H13.5 M30.5 22 H35 M12.8 12.8 L16 16 M28 28 L31.2 31.2 M31.2 12.8 L28 16 M16 28 L12.8 31.2"/></g></svg></span><span><span class="n">SCAMPER Method</span><span class="d">Run your idea through seven lenses: Substitute, Combine, Adapt, Modify, Put-to-other-use, Eliminate, Reverse</span><span class="gf">Great for: Build a feature · Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Six Thinking Hats" data-cat="Structured" data-desc="Examine the problem six ways one at a time: facts, feelings, benefits, risks, new ideas, process" data-good="strategy|diagnosis|planning|personal" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M14 26 q8 -5 16 0" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"/><path d="M17 26 q-6 1 -8 3 q13 4 26 0 q-2 -2 -8 -3" fill="currentColor" fill-opacity=".22"/><path d="M17 26 c-1 -8 11 -8 10 0" fill="currentColor" fill-opacity=".5"/></svg></span><span><span class="n">Six Thinking Hats</span><span class="d">Examine the problem six ways one at a time: facts, feelings, benefits, risks, new ideas, process</span><span class="gf">Great for: Strategy · Diagnose · Planning · Personal / life</span></span></label><label class="tech"><input type="checkbox" data-name="Decision Tree Mapping" data-cat="Structured" data-desc="Chart every choice point and the paths it forks into, following each branch to its outcome and risk" data-good="planning|strategy|diagnosis" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><circle cx="22" cy="12" r="2.6"/><circle cx="14" cy="32" r="2.6"/><circle cx="30" cy="32" r="2.6"/><path d="M22 14.5 L22 20 M22 20 L14 29.4 M22 20 L30 29.4"/></g></svg></span><span><span class="n">Decision Tree Mapping</span><span class="d">Chart every choice point and the paths it forks into, following each branch to its outcome and risk</span><span class="gf">Great for: Planning · Strategy · Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="Solution Matrix" data-cat="Structured" data-desc="Grid problem variables against solution approaches, score every cell, hunt the best pairings and empty gaps" data-good="feature|planning" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="1.8"><rect x="11" y="11" width="22" height="22" rx="2"/><path d="M11 22 H33 M22 11 V33"/></g><path d="M24.5 14.5 L26.5 16.5 L30.5 12.5" stroke="currentColor" stroke-width="2" fill="none" stroke-linecap="round" stroke-linejoin="round"/></svg></span><span><span class="n">Solution Matrix</span><span class="d">Grid problem variables against solution approaches, score every cell, hunt the best pairings and empty gaps</span><span class="gf">Great for: Build a feature · Planning</span></span></label><label class="tech"><input type="checkbox" data-name="Trait Transfer" data-cat="Structured" data-desc="Name what makes an unrelated success work, then graft those winning traits onto your own problem" data-good="novel|feature" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M12 16 l1.6 3.4 3.6 .4 -2.7 2.5 .7 3.6 -3.2 -1.8 -3.2 1.8 .7 -3.6 -2.7 -2.5 3.6 -.4 z" fill="currentColor"/><rect x="25" y="23" width="9" height="9" rx="2" fill="none" stroke="currentColor" stroke-width="2"/><path d="M17 22 L25 27" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-dasharray="1.5 2.5"/></svg></span><span><span class="n">Trait Transfer</span><span class="d">Name what makes an unrelated success work, then graft those winning traits onto your own problem</span><span class="gf">Great for: Novel concept · Build a feature</span></span></label><label class="tech"><input type="checkbox" data-name="Lotus Blossom" data-cat="Structured" data-desc="Put the theme at the center of a 3x3 grid, fill the 8 cells around it, then promote each of those to the center of its own new 3x3" data-good="feature|planning|novel" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><circle cx="22" cy="22" r="3.4"/></g><g fill="currentColor" fill-opacity=".4"><circle cx="22" cy="13" r="2.8"/><circle cx="22" cy="31" r="2.8"/><circle cx="13" cy="22" r="2.8"/><circle cx="31" cy="22" r="2.8"/><circle cx="15.5" cy="15.5" r="2.5"/><circle cx="28.5" cy="15.5" r="2.5"/><circle cx="15.5" cy="28.5" r="2.5"/><circle cx="28.5" cy="28.5" r="2.5"/></g></svg></span><span><span class="n">Lotus Blossom</span><span class="d">Put the theme at the center of a 3x3 grid, fill the 8 cells around it, then promote each of those to the center of its own new 3x3</span><span class="gf">Great for: Build a feature · Planning · Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Worst Possible Idea" data-cat="Structured" data-desc="Deliberately generate the most terrible solutions you can, then flip each into what it teaches you to do right" data-good="unstuck|novel" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M17 11 v10 h-5 l10 12 10 -12 h-5 v-10 z"/></g></svg></span><span><span class="n">Worst Possible Idea</span><span class="d">Deliberately generate the most terrible solutions you can, then flip each into what it teaches you to do right</span><span class="gf">Great for: Get unstuck · Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Disney Method" data-cat="Structured" data-desc="Cycle the idea through three rooms: Dreamer (anything goes), Realist (how we&#x27;d build it), Critic (what breaks)" data-good="feature|strategy|planning" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2"><circle cx="14" cy="22" r="4.5"/><circle cx="22" cy="22" r="4.5"/><circle cx="30" cy="22" r="4.5"/></g></svg></span><span><span class="n">Disney Method</span><span class="d">Cycle the idea through three rooms: Dreamer (anything goes), Realist (how we&#x27;d build it), Critic (what breaks)</span><span class="gf">Great for: Build a feature · Strategy · Planning</span></span></label><label class="tech"><input type="checkbox" data-name="Starbursting" data-cat="Structured" data-desc="Interrogate the idea with only questions — who, what, where, when, why, how — exhaust each before answering any" data-good="feature|planning|diagnosis" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M22 8 V15 M22 29 V36 M8 22 H15 M29 22 H36 M12 12 L17 17 M27 27 L32 32 M32 12 L27 17 M12 32 L17 27"/></g><path d="M19.5 19 a3.2 3.2 0 1 1 3 4 v1.2" fill="none" stroke="currentColor" stroke-width="1.8" stroke-linecap="round"/></svg></span><span><span class="n">Starbursting</span><span class="d">Interrogate the idea with only questions — who, what, where, when, why, how — exhaust each before answering any</span><span class="gf">Great for: Build a feature · Planning · Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="Mind Mapping" data-cat="Structured" data-desc="Branch the central topic outward, each node spawning children; follow tangents wherever they pull and let the web sprawl" data-good="planning|novel|feature" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><circle cx="22" cy="22" r="4"/><circle cx="11" cy="13" r="2.2"/><circle cx="33" cy="13" r="2.2"/><circle cx="10" cy="28" r="2.2"/><circle cx="32" cy="31" r="2.2"/><path d="M19 19.5 L12.5 14.5 M25 19.5 L31.5 14.5 M19 24.5 L11.5 27 M25.5 24 L30.5 29.5"/></g></svg></span><span><span class="n">Mind Mapping</span><span class="d">Branch the central topic outward, each node spawning children; follow tangents wherever they pull and let the web sprawl</span><span class="gf">Great for: Planning · Novel concept · Build a feature</span></span></label><label class="tech"><input type="checkbox" data-name="Crazy 8s" data-cat="Structured" data-desc="Eight ideas in eight minutes, one per box, no editing — speed outruns your inner critic" data-good="feature|novel|unstuck" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="1.7"><rect x="9" y="12" width="8" height="9" rx="1.5"/><rect x="18" y="12" width="8" height="9" rx="1.5"/><rect x="27" y="12" width="8" height="9" rx="1.5"/><rect x="9" y="23" width="8" height="9" rx="1.5"/><rect x="18" y="23" width="8" height="9" rx="1.5"/><rect x="27" y="23" width="8" height="9" rx="1.5"/></g></svg></span><span><span class="n">Crazy 8s</span><span class="d">Eight ideas in eight minutes, one per box, no editing — speed outruns your inner critic</span><span class="gf">Great for: Build a feature · Novel concept · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="How Might We" data-cat="Structured" data-desc="Reframe the problem as a batch of &#x27;How might we...&#x27; opportunity questions first, then ideate against the sharpest one" data-good="feature|novel|strategy|diagnosis" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M11 13 H33 a2 2 0 0 1 2 2 V27 a2 2 0 0 1 -2 2 H24 L19 34 V29 H11 a2 2 0 0 1 -2 -2 V15 a2 2 0 0 1 2 -2 Z"/></g><path d="M19 19 a3.2 3.2 0 1 1 3.4 3.4 v1.6" fill="none" stroke="currentColor" stroke-width="1.8" stroke-linecap="round"/><circle cx="22.4" cy="27" r="1.4" fill="currentColor"/></svg></span><span><span class="n">How Might We</span><span class="d">Reframe the problem as a batch of &#x27;How might we...&#x27; opportunity questions first, then ideate against the sharpest one</span><span class="gf">Great for: Build a feature · Novel concept · Strategy · Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="Job to Be Done" data-cat="Structured" data-desc="Ask what the user is really hiring this to do, then ideate around that underlying job, not the feature you assumed" data-good="feature|strategy|novel" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linejoin="round"><rect x="9" y="16" width="26" height="16" rx="2.5"/><path d="M17 16 v-3 a2 2 0 0 1 2 -2 h6 a2 2 0 0 1 2 2 v3"/><path d="M9 23 H35"/></g><circle cx="22" cy="23" r="1.8" fill="currentColor"/></svg></span><span><span class="n">Job to Be Done</span><span class="d">Ask what the user is really hiring this to do, then ideate around that underlying job, not the feature you assumed</span><span class="gf">Great for: Build a feature · Strategy · Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Empathy Map" data-cat="Structured" data-desc="Map what the user says, thinks, does, and feels around the problem, then mine each quadrant for the unmet need" data-good="feature|personal" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="1.8"><rect x="11" y="11" width="22" height="22" rx="2"/><path d="M11 22 H33 M22 11 V33"/></g><path d="M22 27 c-2.6 -2.1 -4.2 -3.4 -4.2 -5.2 a2.1 2.1 0 0 1 4.2 -1 a2.1 2.1 0 0 1 4.2 1 c0 1.8 -1.6 3.1 -4.2 5.2 z" fill="currentColor"/></svg></span><span><span class="n">Empathy Map</span><span class="d">Map what the user says, thinks, does, and feels around the problem, then mine each quadrant for the unmet need</span><span class="gf">Great for: Build a feature · Personal / life</span></span></label><label class="tech"><input type="checkbox" data-name="Backcasting" data-cat="Structured" data-desc="Fix the finished future in vivid detail, then work backward step by step to the one move you&#x27;d have to make first" data-good="strategy|planning|novel" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M30 11 V33"/></g><path d="M30 13 L20 16.5 L30 20 Z" fill="currentColor"/><path d="M28 27 H14" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-dasharray="1.5 3"/><path d="M18 23 L14 27 L18 31" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"/></svg></span><span><span class="n">Backcasting</span><span class="d">Fix the finished future in vivid detail, then work backward step by step to the one move you&#x27;d have to make first</span><span class="gf">Great for: Strategy · Planning · Novel concept</span></span></label><label class="tech invent"><input type="checkbox" data-invent="Structured"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="11" width="9.5" height="9.5" rx="2"/><rect x="23.5" y="11" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="11" y="23.5" width="9.5" height="9.5" rx="2" fill-opacity="0.25"/><rect x="23.5" y="23.5" width="9.5" height="9.5" rx="2"/></g></svg></span><span><span class="n">✨ Invent a Structured technique</span><span class="d">Make up a brand-new technique on the fly, in the spirit of Structured</span></span></label></div></section>
+<section data-cat="Deep" style="--c:#4658c9"><h2>Deep<span class="cnt">13</span></h2><div class="grid"><label class="tech"><input type="checkbox" data-name="Five Whys" data-cat="Deep" data-desc="Ask &quot;why?&quot; five times in a chain, each answer feeding the next, until you hit the root cause beneath the symptom" data-good="diagnosis" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><circle cx="14" cy="13" r="2.4"/><circle cx="22" cy="22" r="2.4"/><circle cx="30" cy="31" r="2.4"/><path d="M15.6 14.8 L20.4 20.2 M23.6 23.8 L28.4 29.2"/></g></svg></span><span><span class="n">Five Whys</span><span class="d">Ask &quot;why?&quot; five times in a chain, each answer feeding the next, until you hit the root cause beneath the symptom</span><span class="gf">Great for: Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="Provocation Technique" data-cat="Deep" data-desc="State something deliberately absurd, then mine it: &quot;how could this be useful?&quot; Extract the usable principle hiding inside" data-good="unstuck|novel" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M24 9 L13 24 H21 L19 35 L31 19 H23 Z"/></g></svg></span><span><span class="n">Provocation Technique</span><span class="d">State something deliberately absurd, then mine it: &quot;how could this be useful?&quot; Extract the usable principle hiding inside</span><span class="gf">Great for: Get unstuck · Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Assumption Reversal" data-cat="Deep" data-desc="List every assumption baked into the problem, flip each to its opposite, then rebuild a solution on the inverted foundation" data-good="novel|diagnosis|strategy" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M16 14 V30"/><path d="M11.5 25 L16 30 L20.5 25"/><path d="M28 30 V14"/><path d="M23.5 19 L28 14 L32.5 19"/></g></svg></span><span><span class="n">Assumption Reversal</span><span class="d">List every assumption baked into the problem, flip each to its opposite, then rebuild a solution on the inverted foundation</span><span class="gf">Great for: Novel concept · Diagnose · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Question Storming" data-cat="Deep" data-desc="Generate only questions about the problem, zero answers allowed, until the real problem worth solving comes into focus" data-good="diagnosis|strategy|unstuck" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M14 16 a3.2 3.2 0 1 1 3.2 3.2 v2"/><path d="M26 13 a3.6 3.6 0 1 1 3.6 3.6 v2.4"/></g><circle cx="17.2" cy="27" r="1.5" fill="currentColor"/><circle cx="29.6" cy="25.6" r="1.6" fill="currentColor"/></svg></span><span><span class="n">Question Storming</span><span class="d">Generate only questions about the problem, zero answers allowed, until the real problem worth solving comes into focus</span><span class="gf">Great for: Diagnose · Strategy · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Constraint Mapping" data-cat="Deep" data-desc="Map every constraint, sort real from imagined, then attack each: dissolve it, route around it, or turn it into an asset" data-good="feature|strategy|diagnosis" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M11 14 L18 12 L26 14 L33 12 V30 L26 32 L18 30 L11 32 Z" fill="none" stroke="currentColor" stroke-width="2" stroke-linejoin="round"/><path d="M18 12 V30 M26 14 V32" stroke="currentColor" stroke-width="1.6"/></svg></span><span><span class="n">Constraint Mapping</span><span class="d">Map every constraint, sort real from imagined, then attack each: dissolve it, route around it, or turn it into an asset</span><span class="gf">Great for: Build a feature · Strategy · Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="Failure Analysis" data-cat="Deep" data-desc="Dissect a relevant failure: what broke, why it broke, what lesson it leaves, and how to apply that wisdom here" data-good="diagnosis|strategy|feature" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="20" cy="20" r="8" fill="none" stroke="currentColor" stroke-width="2"/><line x1="26" y1="26" x2="33" y2="33" stroke="currentColor" stroke-width="2.4" stroke-linecap="round"/><path d="M20 16 V21 M20 24 V24" stroke="currentColor" stroke-width="2" stroke-linecap="round"/></svg></span><span><span class="n">Failure Analysis</span><span class="d">Dissect a relevant failure: what broke, why it broke, what lesson it leaves, and how to apply that wisdom here</span><span class="gf">Great for: Diagnose · Strategy · Build a feature</span></span></label><label class="tech"><input type="checkbox" data-name="Emergent Thinking" data-cat="Deep" data-desc="Stop forcing a solution; watch what patterns the system keeps producing and name what&#x27;s trying to emerge on its own" data-good="strategy|novel" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><circle cx="11" cy="31" r="1.6"/><circle cx="17" cy="29" r="1.6"/><circle cx="16" cy="23" r="1.6"/><circle cx="22" cy="24" r="1.8"/><circle cx="23" cy="17" r="1.9"/><circle cx="29" cy="18" r="1.7"/><circle cx="28" cy="12" r="2.1"/></g></svg></span><span><span class="n">Emergent Thinking</span><span class="d">Stop forcing a solution; watch what patterns the system keeps producing and name what&#x27;s trying to emerge on its own</span><span class="gf">Great for: Strategy · Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Causal Loop Mapping" data-cat="Deep" data-desc="Diagram the feedback loops linking causes and effects, find the reinforcing and balancing cycles, and target the leverage point" data-good="diagnosis|strategy" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M16 16 a9 9 0 1 1 -2 12"/><path d="M16 10.5 L16.5 16.5 L10.5 17"/><path d="M30 28.5 L29.5 22.5 L35 22"/></g></svg></span><span><span class="n">Causal Loop Mapping</span><span class="d">Diagram the feedback loops linking causes and effects, find the reinforcing and balancing cycles, and target the leverage point</span><span class="gf">Great for: Diagnose · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Morphological Analysis" data-cat="Deep" data-desc="List the problem&#x27;s independent parameters, generate options for each, then combine across them to surface untried configurations" data-good="feature|novel|planning" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="1.8"><rect x="11" y="11" width="22" height="22" rx="2"/><path d="M11 18.3 H33 M11 25.6 H33 M18.3 11 V33 M25.6 11 V33"/></g><rect x="18.5" y="18.5" width="7" height="7" fill="currentColor" fill-opacity=".4"/></svg></span><span><span class="n">Morphological Analysis</span><span class="d">List the problem&#x27;s independent parameters, generate options for each, then combine across them to surface untried configurations</span><span class="gf">Great for: Build a feature · Novel concept · Planning</span></span></label><label class="tech"><input type="checkbox" data-name="Laddering" data-cat="Deep" data-desc="Ask &#x27;and what would that give you?&#x27; up the chain until you reach the real underlying need, then ideate fresh at that level" data-good="personal|strategy|diagnosis" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M16 9 V35 M28 9 V35 M16 15 H28 M16 22 H28 M16 29 H28"/></g></svg></span><span><span class="n">Laddering</span><span class="d">Ask &#x27;and what would that give you?&#x27; up the chain until you reach the real underlying need, then ideate fresh at that level</span><span class="gf">Great for: Personal / life · Strategy · Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="TRIZ Contradiction" data-cat="Deep" data-desc="Name the core contradiction (what only improves by making something else worse), then brainstorm ways to win both instead of trading off" data-good="feature|novel|diagnosis" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M9 15 H18 M14.5 11.5 L18 15 L14.5 18.5"/><path d="M35 29 H26 M29.5 25.5 L26 29 L29.5 32.5"/></g><path d="M22 17.5 l1.5 3.4 3.7 .3 -2.8 2.4 .9 3.6 -3.3 -1.9 -3.3 1.9 .9 -3.6 -2.8 -2.4 3.7 -.3 z" fill="currentColor"/></svg></span><span><span class="n">TRIZ Contradiction</span><span class="d">Name the core contradiction (what only improves by making something else worse), then brainstorm ways to win both instead of trading off</span><span class="gf">Great for: Build a feature · Novel concept · Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="Fishbone Diagram" data-cat="Deep" data-desc="Branch the problem&#x27;s spine into cause categories (people, process, tools, environment) and mine each bone for contributing causes" data-good="diagnosis" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M9 22 H33"/><path d="M33 22 L36 18.5 M33 22 L36 25.5"/><path d="M14 22 L18 15 M14 22 L18 29 M23 22 L27 15 M23 22 L27 29"/></g><circle cx="9" cy="22" r="1.9" fill="currentColor"/></svg></span><span><span class="n">Fishbone Diagram</span><span class="d">Branch the problem&#x27;s spine into cause categories (people, process, tools, environment) and mine each bone for contributing causes</span><span class="gf">Great for: Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="Build on What Works" data-cat="Deep" data-desc="Name what&#x27;s already succeeding and why, then ideate how to amplify and extend it instead of fixing what&#x27;s broken" data-good="personal|strategy" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="11" y="24" width="6" height="9" rx="1"/><rect x="19" y="19" width="6" height="14" rx="1"/><rect x="27" y="13" width="6" height="20" rx="1"/></g><path d="M11 19 L18 13 L24 16 L33 8" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"/><path d="M28 8 H33 V13" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"/></svg></span><span><span class="n">Build on What Works</span><span class="d">Name what&#x27;s already succeeding and why, then ideate how to amplify and extend it instead of fixing what&#x27;s broken</span><span class="gf">Great for: Personal / life · Strategy</span></span></label><label class="tech invent"><input type="checkbox" data-invent="Deep"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor"><circle cx="22" cy="22" r="13" stroke-width="1.5" stroke-opacity="0.4"/><circle cx="22" cy="22" r="9" stroke-width="1.7" stroke-opacity="0.7"/><circle cx="22" cy="22" r="5" stroke-width="1.9"/></g><circle cx="22" cy="22" r="2.4" fill="currentColor"/></svg></span><span><span class="n">✨ Invent a Deep technique</span><span class="d">Make up a brand-new technique on the fly, in the spirit of Deep</span></span></label></div></section>
+<h2 class="grouphdr" style="--c:#6d5cf0">Creative &amp; Generative</h2>
+<section data-cat="Creative" style="--c:#6d5cf0"><h2>Creative<span class="cnt">10</span></h2><div class="grid"><label class="tech"><input type="checkbox" data-name="What If Scenarios" data-cat="Creative" data-desc="Detonate one constraint at a time — unlimited budget, opposite is true, problem vanished — and chase what rushes in" data-good="novel|strategy|unstuck" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.4" stroke-linecap="round"><line x1="22" y1="6.5" x2="22" y2="12.5"/><line x1="22" y1="31.5" x2="22" y2="37.5"/><line x1="6.5" y1="22" x2="12.5" y2="22"/><line x1="31.5" y1="22" x2="37.5" y2="22"/><line x1="11.3" y1="11.3" x2="15.5" y2="15.5"/><line x1="28.5" y1="28.5" x2="32.7" y2="32.7"/><line x1="32.7" y1="11.3" x2="28.5" y2="15.5"/><line x1="15.5" y1="28.5" x2="11.3" y2="32.7"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="3.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M18 17 a4 4 0 1 1 4 4 v3"/></g><circle cx="22" cy="30" r="1.6" fill="currentColor"/></svg></span><span><span class="n">What If Scenarios</span><span class="d">Detonate one constraint at a time — unlimited budget, opposite is true, problem vanished — and chase what rushes in</span><span class="gf">Great for: Novel concept · Strategy · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Analogical Thinking" data-cat="Creative" data-desc="Ask &#x27;this is like what?&#x27; and steal the solution pattern from the domain that answers" data-good="feature|novel|diagnosis" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.4" stroke-linecap="round"><line x1="22" y1="6.5" x2="22" y2="12.5"/><line x1="22" y1="31.5" x2="22" y2="37.5"/><line x1="6.5" y1="22" x2="12.5" y2="22"/><line x1="31.5" y1="22" x2="37.5" y2="22"/><line x1="11.3" y1="11.3" x2="15.5" y2="15.5"/><line x1="28.5" y1="28.5" x2="32.7" y2="32.7"/><line x1="32.7" y1="11.3" x2="28.5" y2="15.5"/><line x1="15.5" y1="28.5" x2="11.3" y2="32.7"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="3.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="15" cy="22" r="6" fill="none" stroke="currentColor" stroke-width="2"/><rect x="25" y="16" width="12" height="12" rx="2" fill="none" stroke="currentColor" stroke-width="2"/><path d="M19 20 q2 2 0 4 M23 20 q-2 2 0 4" stroke="currentColor" stroke-width="1.6" fill="none"/></svg></span><span><span class="n">Analogical Thinking</span><span class="d">Ask &#x27;this is like what?&#x27; and steal the solution pattern from the domain that answers</span><span class="gf">Great for: Build a feature · Novel concept · Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="First Principles Thinking" data-cat="Creative" data-desc="Strip every assumption to bedrock facts, then rebuild the solution from scratch on truth alone" data-good="feature|novel|diagnosis|strategy" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.4" stroke-linecap="round"><line x1="22" y1="6.5" x2="22" y2="12.5"/><line x1="22" y1="31.5" x2="22" y2="37.5"/><line x1="6.5" y1="22" x2="12.5" y2="22"/><line x1="31.5" y1="22" x2="37.5" y2="22"/><line x1="11.3" y1="11.3" x2="15.5" y2="15.5"/><line x1="28.5" y1="28.5" x2="32.7" y2="32.7"/><line x1="32.7" y1="11.3" x2="28.5" y2="15.5"/><line x1="15.5" y1="28.5" x2="11.3" y2="32.7"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="3.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="10" y="28" width="8" height="6" rx="1"/><rect x="18.5" y="28" width="8" height="6" rx="1"/><rect x="27" y="28" width="7" height="6" rx="1"/></g><path d="M22 25 L22 11 M16 17 L22 11 L28 17" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"/></svg></span><span><span class="n">First Principles Thinking</span><span class="d">Strip every assumption to bedrock facts, then rebuild the solution from scratch on truth alone</span><span class="gf">Great for: Build a feature · Novel concept · Diagnose · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Forced Relationships" data-cat="Creative" data-desc="Grab two unrelated things at random and force a bridge between them until an idea falls out" data-good="novel|unstuck" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.4" stroke-linecap="round"><line x1="22" y1="6.5" x2="22" y2="12.5"/><line x1="22" y1="31.5" x2="22" y2="37.5"/><line x1="6.5" y1="22" x2="12.5" y2="22"/><line x1="31.5" y1="22" x2="37.5" y2="22"/><line x1="11.3" y1="11.3" x2="15.5" y2="15.5"/><line x1="28.5" y1="28.5" x2="32.7" y2="32.7"/><line x1="32.7" y1="11.3" x2="28.5" y2="15.5"/><line x1="15.5" y1="28.5" x2="11.3" y2="32.7"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="3.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="12" cy="22" r="3.4" fill="currentColor"/><circle cx="32" cy="22" r="3.4" fill="currentColor"/><path d="M15 22 q7 -9 14 0" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-dasharray="1.5 3"/></svg></span><span><span class="n">Forced Relationships</span><span class="d">Grab two unrelated things at random and force a bridge between them until an idea falls out</span><span class="gf">Great for: Novel concept · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Time Shifting" data-cat="Creative" data-desc="Solve the problem as a 1900s artisan, then a 2150 colonist — harvest the era-bound constraints and tricks" data-good="novel|unstuck" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.4" stroke-linecap="round"><line x1="22" y1="6.5" x2="22" y2="12.5"/><line x1="22" y1="31.5" x2="22" y2="37.5"/><line x1="6.5" y1="22" x2="12.5" y2="22"/><line x1="31.5" y1="22" x2="37.5" y2="22"/><line x1="11.3" y1="11.3" x2="15.5" y2="15.5"/><line x1="28.5" y1="28.5" x2="32.7" y2="32.7"/><line x1="32.7" y1="11.3" x2="28.5" y2="15.5"/><line x1="15.5" y1="28.5" x2="11.3" y2="32.7"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="3.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="22" r="12" fill="none" stroke="currentColor" stroke-width="2"/><path d="M22 15 V22 L27 25" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"/></svg></span><span><span class="n">Time Shifting</span><span class="d">Solve the problem as a 1900s artisan, then a 2150 colonist — harvest the era-bound constraints and tricks</span><span class="gf">Great for: Novel concept · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Metaphor Mapping" data-cat="Creative" data-desc="Declare the problem IS a chosen metaphor, extend the metaphor fully, map each part back to find insights" data-good="novel|diagnosis" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.4" stroke-linecap="round"><line x1="22" y1="6.5" x2="22" y2="12.5"/><line x1="22" y1="31.5" x2="22" y2="37.5"/><line x1="6.5" y1="22" x2="12.5" y2="22"/><line x1="31.5" y1="22" x2="37.5" y2="22"/><line x1="11.3" y1="11.3" x2="15.5" y2="15.5"/><line x1="28.5" y1="28.5" x2="32.7" y2="32.7"/><line x1="32.7" y1="11.3" x2="28.5" y2="15.5"/><line x1="15.5" y1="28.5" x2="11.3" y2="32.7"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="3.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><rect x="10" y="14" width="14" height="14" rx="2" fill="none" stroke="currentColor" stroke-width="2"/><circle cx="28" cy="25" r="7" fill="currentColor" fill-opacity=".22"/><circle cx="28" cy="25" r="7" fill="none" stroke="currentColor" stroke-width="2"/></svg></span><span><span class="n">Metaphor Mapping</span><span class="d">Declare the problem IS a chosen metaphor, extend the metaphor fully, map each part back to find insights</span><span class="gf">Great for: Novel concept · Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="Cross-Pollination" data-cat="Creative" data-desc="Ask how a wildly different industry — casinos, ERs, beekeeping — would crack this, then adapt their move" data-good="novel|feature|strategy" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.4" stroke-linecap="round"><line x1="22" y1="6.5" x2="22" y2="12.5"/><line x1="22" y1="31.5" x2="22" y2="37.5"/><line x1="6.5" y1="22" x2="12.5" y2="22"/><line x1="31.5" y1="22" x2="37.5" y2="22"/><line x1="11.3" y1="11.3" x2="15.5" y2="15.5"/><line x1="28.5" y1="28.5" x2="32.7" y2="32.7"/><line x1="32.7" y1="11.3" x2="28.5" y2="15.5"/><line x1="15.5" y1="28.5" x2="11.3" y2="32.7"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="3.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M13 14 H27 a4 4 0 0 1 0 8 H17 a4 4 0 0 0 0 8 H31"/><path d="M28 11 L31.5 14 L28 17 M16 27 L12.5 30 L16 33"/></g></svg></span><span><span class="n">Cross-Pollination</span><span class="d">Ask how a wildly different industry — casinos, ERs, beekeeping — would crack this, then adapt their move</span><span class="gf">Great for: Novel concept · Build a feature · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Concept Blending" data-cat="Creative" data-desc="Fuse two concepts into one new hybrid category and name what the merger becomes, not just combines" data-good="novel" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.4" stroke-linecap="round"><line x1="22" y1="6.5" x2="22" y2="12.5"/><line x1="22" y1="31.5" x2="22" y2="37.5"/><line x1="6.5" y1="22" x2="12.5" y2="22"/><line x1="31.5" y1="22" x2="37.5" y2="22"/><line x1="11.3" y1="11.3" x2="15.5" y2="15.5"/><line x1="28.5" y1="28.5" x2="32.7" y2="32.7"/><line x1="32.7" y1="11.3" x2="28.5" y2="15.5"/><line x1="15.5" y1="28.5" x2="11.3" y2="32.7"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="3.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="18" cy="22" r="8" fill="currentColor" fill-opacity=".25"/><circle cx="26" cy="22" r="8" fill="currentColor" fill-opacity=".25"/><g fill="none" stroke="currentColor" stroke-width="2"><circle cx="18" cy="22" r="8"/><circle cx="26" cy="22" r="8"/></g></svg></span><span><span class="n">Concept Blending</span><span class="d">Fuse two concepts into one new hybrid category and name what the merger becomes, not just combines</span><span class="gf">Great for: Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Reverse Brainstorming" data-cat="Creative" data-desc="Generate problems instead of solutions — &#x27;how could we make this fail?&#x27; — then mine each for its inverse" data-good="diagnosis|feature|unstuck" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.4" stroke-linecap="round"><line x1="22" y1="6.5" x2="22" y2="12.5"/><line x1="22" y1="31.5" x2="22" y2="37.5"/><line x1="6.5" y1="22" x2="12.5" y2="22"/><line x1="31.5" y1="22" x2="37.5" y2="22"/><line x1="11.3" y1="11.3" x2="15.5" y2="15.5"/><line x1="28.5" y1="28.5" x2="32.7" y2="32.7"/><line x1="32.7" y1="11.3" x2="28.5" y2="15.5"/><line x1="15.5" y1="28.5" x2="11.3" y2="32.7"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="3.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M13 18 H28 a4 4 0 0 1 0 8 H16"/><path d="M19 15 L13 18 L19 21 M22 23 L16 26 L22 29"/></g></svg></span><span><span class="n">Reverse Brainstorming</span><span class="d">Generate problems instead of solutions — &#x27;how could we make this fail?&#x27; — then mine each for its inverse</span><span class="gf">Great for: Diagnose · Build a feature · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Sensory Exploration" data-cat="Creative" data-desc="Interrogate the idea through each sense — its taste, smell, sound, texture — to surface non-analytical angles" data-good="novel|unstuck" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.4" stroke-linecap="round"><line x1="22" y1="6.5" x2="22" y2="12.5"/><line x1="22" y1="31.5" x2="22" y2="37.5"/><line x1="6.5" y1="22" x2="12.5" y2="22"/><line x1="31.5" y1="22" x2="37.5" y2="22"/><line x1="11.3" y1="11.3" x2="15.5" y2="15.5"/><line x1="28.5" y1="28.5" x2="32.7" y2="32.7"/><line x1="32.7" y1="11.3" x2="28.5" y2="15.5"/><line x1="15.5" y1="28.5" x2="11.3" y2="32.7"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="3.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M10 22 q12 -10 24 0 q-12 10 -24 0 z" fill="none" stroke="currentColor" stroke-width="2" stroke-linejoin="round"/><circle cx="22" cy="22" r="4" fill="currentColor"/></svg></span><span><span class="n">Sensory Exploration</span><span class="d">Interrogate the idea through each sense — its taste, smell, sound, texture — to surface non-analytical angles</span><span class="gf">Great for: Novel concept · Get unstuck</span></span></label><label class="tech invent"><input type="checkbox" data-invent="Creative"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.4" stroke-linecap="round"><line x1="22" y1="6.5" x2="22" y2="12.5"/><line x1="22" y1="31.5" x2="22" y2="37.5"/><line x1="6.5" y1="22" x2="12.5" y2="22"/><line x1="31.5" y1="22" x2="37.5" y2="22"/><line x1="11.3" y1="11.3" x2="15.5" y2="15.5"/><line x1="28.5" y1="28.5" x2="32.7" y2="32.7"/><line x1="32.7" y1="11.3" x2="28.5" y2="15.5"/><line x1="15.5" y1="28.5" x2="11.3" y2="32.7"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="3.6" fill="currentColor"/></svg></span><span><span class="n">✨ Invent a Creative technique</span><span class="d">Make up a brand-new technique on the fly, in the spirit of Creative</span></span></label></div></section>
+<section data-cat="Biomimetic" style="--c:#1f9d6b"><h2>Biomimetic<span class="cnt">6</span></h2><div class="grid"><label class="tech"><input type="checkbox" data-name="Nature&#x27;s Solutions" data-cat="Biomimetic" data-desc="Name an organism that already solved your problem, then copy its mechanism into your design" data-good="feature|novel" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M22 7.5 C 31.5 12.5, 31.5 29, 22 36.5 C 12.5 29, 12.5 12.5, 22 7.5 Z" fill="currentColor" fill-opacity="0.22"/><path d="M22 9 V 35.5" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/><g stroke="currentColor" stroke-width="1.5" stroke-linecap="round"><path d="M22 16 l5.6 -2.6"/><path d="M22 16 l-5.6 -2.6"/><path d="M22 22 l6.6 -2.6"/><path d="M22 22 l-6.6 -2.6"/><path d="M22 28 l5.6 -2.6"/><path d="M22 28 l-5.6 -2.6"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M16 32 C12 24 12 20 16 12 M28 32 C32 24 32 20 28 12"/><path d="M16 16 L28 14 M16 22 L28 20 M16 28 L28 26"/></g></svg></span><span><span class="n">Nature&#x27;s Solutions</span><span class="d">Name an organism that already solved your problem, then copy its mechanism into your design</span><span class="gf">Great for: Build a feature · Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Ecosystem Thinking" data-cat="Biomimetic" data-desc="Map your problem as an ecosystem: who eats whom, who partners, what decays, what fills the gaps" data-good="strategy|diagnosis" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M22 7.5 C 31.5 12.5, 31.5 29, 22 36.5 C 12.5 29, 12.5 12.5, 22 7.5 Z" fill="currentColor" fill-opacity="0.22"/><path d="M22 9 V 35.5" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/><g stroke="currentColor" stroke-width="1.5" stroke-linecap="round"><path d="M22 16 l5.6 -2.6"/><path d="M22 16 l-5.6 -2.6"/><path d="M22 22 l6.6 -2.6"/><path d="M22 22 l-6.6 -2.6"/><path d="M22 28 l5.6 -2.6"/><path d="M22 28 l-5.6 -2.6"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><circle cx="22" cy="13" r="2.4"/><circle cx="12" cy="27" r="2.4"/><circle cx="32" cy="27" r="2.4"/><circle cx="22" cy="24" r="2.4"/><path d="M22 15.4 V21.6 M14 26 L20 24.5 M30 26 L24 24.5 M13.6 25.2 L30.4 25.2"/></g></svg></span><span><span class="n">Ecosystem Thinking</span><span class="d">Map your problem as an ecosystem: who eats whom, who partners, what decays, what fills the gaps</span><span class="gf">Great for: Strategy · Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="Evolutionary Pressure" data-cat="Biomimetic" data-desc="Spawn many ugly variants, apply a brutal selection rule, breed the survivors, repeat until it adapts" data-good="feature|novel" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M22 7.5 C 31.5 12.5, 31.5 29, 22 36.5 C 12.5 29, 12.5 12.5, 22 7.5 Z" fill="currentColor" fill-opacity="0.22"/><path d="M22 9 V 35.5" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/><g stroke="currentColor" stroke-width="1.5" stroke-linecap="round"><path d="M22 16 l5.6 -2.6"/><path d="M22 16 l-5.6 -2.6"/><path d="M22 22 l6.6 -2.6"/><path d="M22 22 l-6.6 -2.6"/><path d="M22 28 l5.6 -2.6"/><path d="M22 28 l-5.6 -2.6"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M11 31 H17 M19 31 a6 6 0 0 1 6 -6 M25 25 a5 5 0 0 1 5 -5 M30 20 H33"/><circle cx="11" cy="31" r="2" fill="currentColor"/><circle cx="33" cy="20" r="2.6" fill="currentColor"/></g></svg></span><span><span class="n">Evolutionary Pressure</span><span class="d">Spawn many ugly variants, apply a brutal selection rule, breed the survivors, repeat until it adapts</span><span class="gf">Great for: Build a feature · Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Predator &amp; Prey" data-cat="Biomimetic" data-desc="Pick a threat to your idea, then design the defense, camouflage, or escape an animal would evolve against it" data-good="strategy|feature" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M22 7.5 C 31.5 12.5, 31.5 29, 22 36.5 C 12.5 29, 12.5 12.5, 22 7.5 Z" fill="currentColor" fill-opacity="0.22"/><path d="M22 9 V 35.5" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/><g stroke="currentColor" stroke-width="1.5" stroke-linecap="round"><path d="M22 16 l5.6 -2.6"/><path d="M22 16 l-5.6 -2.6"/><path d="M22 22 l6.6 -2.6"/><path d="M22 22 l-6.6 -2.6"/><path d="M22 28 l5.6 -2.6"/><path d="M22 28 l-5.6 -2.6"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M22 9 L33 14 V23 C33 30 28 34 22 36 C16 34 11 30 11 23 V14 Z" fill="currentColor" fill-opacity=".18"/><path d="M22 9 L33 14 V23 C33 30 28 34 22 36 C16 34 11 30 11 23 V14 Z" fill="none" stroke="currentColor" stroke-width="2" stroke-linejoin="round"/></svg></span><span><span class="n">Predator &amp; Prey</span><span class="d">Pick a threat to your idea, then design the defense, camouflage, or escape an animal would evolve against it</span><span class="gf">Great for: Strategy · Build a feature</span></span></label><label class="tech"><input type="checkbox" data-name="Metamorphosis Stages" data-cat="Biomimetic" data-desc="Force your idea through egg, larva, pupa, adult: a radically different form and purpose at each life stage" data-good="novel|strategy" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M22 7.5 C 31.5 12.5, 31.5 29, 22 36.5 C 12.5 29, 12.5 12.5, 22 7.5 Z" fill="currentColor" fill-opacity="0.22"/><path d="M22 9 V 35.5" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/><g stroke="currentColor" stroke-width="1.5" stroke-linecap="round"><path d="M22 16 l5.6 -2.6"/><path d="M22 16 l-5.6 -2.6"/><path d="M22 22 l6.6 -2.6"/><path d="M22 22 l-6.6 -2.6"/><path d="M22 28 l5.6 -2.6"/><path d="M22 28 l-5.6 -2.6"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M22 12 V32"/><path d="M22 16 C14 12 10 18 14 22 C10 26 14 32 22 28 C30 32 34 26 30 22 C34 18 30 12 22 16"/></g></svg></span><span><span class="n">Metamorphosis Stages</span><span class="d">Force your idea through egg, larva, pupa, adult: a radically different form and purpose at each life stage</span><span class="gf">Great for: Novel concept · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Swarm Logic" data-cat="Biomimetic" data-desc="Forbid the master plan: solve it with dumb local rules each agent follows so order emerges from the bottom up" data-good="feature|strategy" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M22 7.5 C 31.5 12.5, 31.5 29, 22 36.5 C 12.5 29, 12.5 12.5, 22 7.5 Z" fill="currentColor" fill-opacity="0.22"/><path d="M22 9 V 35.5" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/><g stroke="currentColor" stroke-width="1.5" stroke-linecap="round"><path d="M22 16 l5.6 -2.6"/><path d="M22 16 l-5.6 -2.6"/><path d="M22 22 l6.6 -2.6"/><path d="M22 22 l-6.6 -2.6"/><path d="M22 28 l5.6 -2.6"/><path d="M22 28 l-5.6 -2.6"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M22 10 L29 14 V22 L22 26 L15 22 V14 Z"/><path d="M15 24 L18 33 M29 24 L26 33 M22 28 V35" stroke-opacity=".6"/></g></svg></span><span><span class="n">Swarm Logic</span><span class="d">Forbid the master plan: solve it with dumb local rules each agent follows so order emerges from the bottom up</span><span class="gf">Great for: Build a feature · Strategy</span></span></label><label class="tech invent"><input type="checkbox" data-invent="Biomimetic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M22 7.5 C 31.5 12.5, 31.5 29, 22 36.5 C 12.5 29, 12.5 12.5, 22 7.5 Z" fill="currentColor" fill-opacity="0.22"/><path d="M22 9 V 35.5" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/><g stroke="currentColor" stroke-width="1.5" stroke-linecap="round"><path d="M22 16 l5.6 -2.6"/><path d="M22 16 l-5.6 -2.6"/><path d="M22 22 l6.6 -2.6"/><path d="M22 22 l-6.6 -2.6"/><path d="M22 28 l5.6 -2.6"/><path d="M22 28 l-5.6 -2.6"/></g></svg></span><span><span class="n">✨ Invent a Biomimetic technique</span><span class="d">Make up a brand-new technique on the fly, in the spirit of Biomimetic</span></span></label></div></section>
+<section data-cat="Cultural" style="--c:#c75b39"><h2>Cultural<span class="cnt">7</span></h2><div class="grid"><label class="tech"><input type="checkbox" data-name="Indigenous Wisdom" data-cat="Cultural" data-desc="Ask how an indigenous or traditional knowledge system would approach this — name the culture, channel its ancestral problem-solving" data-good="personal|strategy|novel" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="22" r="13.5" fill="currentColor" fill-opacity="0.14"/><g stroke="currentColor" stroke-width="1.6" fill="none"><circle cx="22" cy="22" r="13.5"/><ellipse cx="22" cy="22" rx="6" ry="13.5"/><line x1="8.5" y1="22" x2="35.5" y2="22"/><path d="M11 15 H 33" stroke-opacity="0.55"/><path d="M11 29 H 33" stroke-opacity="0.55"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M26 11 C18 16 14 24 13 33 M26 11 C28 18 26 25 20 29"/><path d="M26 11 C24 13 22 14 19 15 M24 16 C22 18 20 19 17 20 M22 21 C20 23 18 24 15 25"/></g></svg></span><span><span class="n">Indigenous Wisdom</span><span class="d">Ask how an indigenous or traditional knowledge system would approach this — name the culture, channel its ancestral problem-solving</span><span class="gf">Great for: Personal / life · Strategy · Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Fusion Cuisine" data-cat="Cultural" data-desc="Pick two unrelated cultures and force-blend their approaches; harvest the hybrid that neither alone would invent" data-good="novel" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="22" r="13.5" fill="currentColor" fill-opacity="0.14"/><g stroke="currentColor" stroke-width="1.6" fill="none"><circle cx="22" cy="22" r="13.5"/><ellipse cx="22" cy="22" rx="6" ry="13.5"/><line x1="8.5" y1="22" x2="35.5" y2="22"/><path d="M11 15 H 33" stroke-opacity="0.55"/><path d="M11 29 H 33" stroke-opacity="0.55"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M10 19 a12 7 0 0 0 24 0 Z"/><path d="M22 19 V32 M16 32 H28"/></g></svg></span><span><span class="n">Fusion Cuisine</span><span class="d">Pick two unrelated cultures and force-blend their approaches; harvest the hybrid that neither alone would invent</span><span class="gf">Great for: Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Ritual Innovation" data-cat="Cultural" data-desc="Redesign the idea as a ceremony — define the threshold, the gestures, the transformation participants undergo" data-good="novel|personal" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="22" r="13.5" fill="currentColor" fill-opacity="0.14"/><g stroke="currentColor" stroke-width="1.6" fill="none"><circle cx="22" cy="22" r="13.5"/><ellipse cx="22" cy="22" rx="6" ry="13.5"/><line x1="8.5" y1="22" x2="35.5" y2="22"/><path d="M11 15 H 33" stroke-opacity="0.55"/><path d="M11 29 H 33" stroke-opacity="0.55"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M12 33 V17 a10 10 0 0 1 20 0 V33"/><path d="M12 33 H32 M22 33 V21"/></g></svg></span><span><span class="n">Ritual Innovation</span><span class="d">Redesign the idea as a ceremony — define the threshold, the gestures, the transformation participants undergo</span><span class="gf">Great for: Novel concept · Personal / life</span></span></label><label class="tech"><input type="checkbox" data-name="Mythic Frameworks" data-cat="Cultural" data-desc="Map the problem onto a myth: name the archetypes, find the parallel tale, let its structure dictate the resolution" data-good="strategy|personal|novel" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="22" r="13.5" fill="currentColor" fill-opacity="0.14"/><g stroke="currentColor" stroke-width="1.6" fill="none"><circle cx="22" cy="22" r="13.5"/><ellipse cx="22" cy="22" rx="6" ry="13.5"/><line x1="8.5" y1="22" x2="35.5" y2="22"/><path d="M11 15 H 33" stroke-opacity="0.55"/><path d="M11 29 H 33" stroke-opacity="0.55"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M14 12 h13 a3 3 0 0 1 3 3 v17 l-3 -2 -3 2 -3 -2 -3 2 V15 a3 3 0 0 0 -3 -3 z"/><path d="M14 12 a3 3 0 0 0 -3 3 h6"/><path d="M20 18 H26 M20 23 H26"/></g></svg></span><span><span class="n">Mythic Frameworks</span><span class="d">Map the problem onto a myth: name the archetypes, find the parallel tale, let its structure dictate the resolution</span><span class="gf">Great for: Strategy · Personal / life · Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Proverb Mining" data-cat="Cultural" data-desc="Collect proverbs from many cultures on this theme, then build the solution from the one that clashes hardest with your assumptions" data-good="personal|strategy" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="22" r="13.5" fill="currentColor" fill-opacity="0.14"/><g stroke="currentColor" stroke-width="1.6" fill="none"><circle cx="22" cy="22" r="13.5"/><ellipse cx="22" cy="22" rx="6" ry="13.5"/><line x1="8.5" y1="22" x2="35.5" y2="22"/><path d="M11 15 H 33" stroke-opacity="0.55"/><path d="M11 29 H 33" stroke-opacity="0.55"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M22 14 C17 10 11 11 11 11 V30 s6 -1 11 3 c5 -4 11 -3 11 -3 V11 s-6 -1 -11 3 z"/><path d="M22 14 V31"/></g></svg></span><span><span class="n">Proverb Mining</span><span class="d">Collect proverbs from many cultures on this theme, then build the solution from the one that clashes hardest with your assumptions</span><span class="gf">Great for: Personal / life · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Ancestor Council" data-cat="Cultural" data-desc="Convene three ancestors or elders from different traditions, voice each one&#x27;s verdict on your idea, reconcile their disagreement" data-good="personal|strategy" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="22" r="13.5" fill="currentColor" fill-opacity="0.14"/><g stroke="currentColor" stroke-width="1.6" fill="none"><circle cx="22" cy="22" r="13.5"/><ellipse cx="22" cy="22" rx="6" ry="13.5"/><line x1="8.5" y1="22" x2="35.5" y2="22"/><path d="M11 15 H 33" stroke-opacity="0.55"/><path d="M11 29 H 33" stroke-opacity="0.55"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2"><circle cx="22" cy="14" r="3.5"/><circle cx="13" cy="19" r="3"/><circle cx="31" cy="19" r="3"/></g><g fill="currentColor" fill-opacity=".25"><path d="M16 31 c0 -5 12 -5 12 0 z"/><path d="M8 31 c0 -4 9 -4.5 9 0 z"/><path d="M27 31 c0 -4.5 9 -4 9 0 z"/></g></svg></span><span><span class="n">Ancestor Council</span><span class="d">Convene three ancestors or elders from different traditions, voice each one&#x27;s verdict on your idea, reconcile their disagreement</span><span class="gf">Great for: Personal / life · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Trickster&#x27;s Gambit" data-cat="Cultural" data-desc="Channel the trickster figure — coyote, Anansi, Loki — and solve it by cheating, inverting, or breaking the sacred rule" data-good="unstuck|strategy" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="22" r="13.5" fill="currentColor" fill-opacity="0.14"/><g stroke="currentColor" stroke-width="1.6" fill="none"><circle cx="22" cy="22" r="13.5"/><ellipse cx="22" cy="22" rx="6" ry="13.5"/><line x1="8.5" y1="22" x2="35.5" y2="22"/><path d="M11 15 H 33" stroke-opacity="0.55"/><path d="M11 29 H 33" stroke-opacity="0.55"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><rect x="11" y="12" width="13" height="18" rx="2" transform="rotate(-10 17.5 21)" fill="currentColor" fill-opacity=".2" stroke="currentColor" stroke-width="2"/><rect x="20" y="14" width="13" height="18" rx="2" transform="rotate(10 26.5 23)" fill="none" stroke="currentColor" stroke-width="2"/><path d="M26.5 19 l1.4 3 1.4 -3 -1.4 -1 z" fill="currentColor"/></svg></span><span><span class="n">Trickster&#x27;s Gambit</span><span class="d">Channel the trickster figure — coyote, Anansi, Loki — and solve it by cheating, inverting, or breaking the sacred rule</span><span class="gf">Great for: Get unstuck · Strategy</span></span></label><label class="tech invent"><input type="checkbox" data-invent="Cultural"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="22" r="13.5" fill="currentColor" fill-opacity="0.14"/><g stroke="currentColor" stroke-width="1.6" fill="none"><circle cx="22" cy="22" r="13.5"/><ellipse cx="22" cy="22" rx="6" ry="13.5"/><line x1="8.5" y1="22" x2="35.5" y2="22"/><path d="M11 15 H 33" stroke-opacity="0.55"/><path d="M11 29 H 33" stroke-opacity="0.55"/></g></svg></span><span><span class="n">✨ Invent a Cultural technique</span><span class="d">Make up a brand-new technique on the fly, in the spirit of Cultural</span></span></label></div></section>
+<section data-cat="Speculative Future" style="--c:#0fb5c9"><h2>Speculative Future<span class="cnt">8</span></h2><div class="grid"><label class="tech"><input type="checkbox" data-name="Time Horizon Ladder" data-cat="Speculative Future" data-desc="Solve the idea for 1 year out, then 10, then 100 — note what survives, breaks, or becomes absurd at each rung" data-good="strategy|planning|novel" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M11 31 L 26.5 15.5"/><path d="M20 14.5 H 28 V 22.5"/></g><circle cx="31" cy="12" r="2.8" fill="currentColor"/><g stroke="currentColor" stroke-width="1.4" stroke-linecap="round"><line x1="31" y1="6.5" x2="31" y2="8.4"/><line x1="31" y1="15.6" x2="31" y2="17.5"/><line x1="25.5" y1="12" x2="27.4" y2="12"/><line x1="34.6" y1="12" x2="36.5" y2="12"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M9 30 H35"/><path d="M14 30 V24 M22 30 V18 M30 30 V12"/></g><g fill="currentColor"><circle cx="14" cy="24" r="2"/><circle cx="22" cy="18" r="2"/><circle cx="30" cy="12" r="2"/></g></svg></span><span><span class="n">Time Horizon Ladder</span><span class="d">Solve the idea for 1 year out, then 10, then 100 — note what survives, breaks, or becomes absurd at each rung</span><span class="gf">Great for: Strategy · Planning · Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Post-Scarcity Test" data-cat="Speculative Future" data-desc="Assume the core constraint (money, energy, time, attention) is now infinite and free — what does the idea become" data-good="novel|strategy" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M11 31 L 26.5 15.5"/><path d="M20 14.5 H 28 V 22.5"/></g><circle cx="31" cy="12" r="2.8" fill="currentColor"/><g stroke="currentColor" stroke-width="1.4" stroke-linecap="round"><line x1="31" y1="6.5" x2="31" y2="8.4"/><line x1="31" y1="15.6" x2="31" y2="17.5"/><line x1="25.5" y1="12" x2="27.4" y2="12"/><line x1="34.6" y1="12" x2="36.5" y2="12"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M15 22 a4.5 4.5 0 1 1 4.5 4.5 C16 26.5 14 18 11 18 a3.5 3.5 0 0 0 0 7 c4 0 5 -8 11 -8 a4.5 4.5 0 0 1 0 9 c-3 0 -4 -4.5 -7 -4.5"/></g></svg></span><span><span class="n">Post-Scarcity Test</span><span class="d">Assume the core constraint (money, energy, time, attention) is now infinite and free — what does the idea become</span><span class="gf">Great for: Novel concept · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Utopia vs Dystopia Split-Screen" data-cat="Speculative Future" data-desc="Write the same future twice: the brochure where it went perfectly and the headline where it went horribly" data-good="strategy|diagnosis" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M11 31 L 26.5 15.5"/><path d="M20 14.5 H 28 V 22.5"/></g><circle cx="31" cy="12" r="2.8" fill="currentColor"/><g stroke="currentColor" stroke-width="1.4" stroke-linecap="round"><line x1="31" y1="6.5" x2="31" y2="8.4"/><line x1="31" y1="15.6" x2="31" y2="17.5"/><line x1="25.5" y1="12" x2="27.4" y2="12"/><line x1="34.6" y1="12" x2="36.5" y2="12"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><rect x="11" y="11" width="22" height="22" rx="3" fill="none" stroke="currentColor" stroke-width="2"/><path d="M22 11 V33" stroke="currentColor" stroke-width="2"/><path d="M22 11 H33 a0 0 0 0 1 0 0 V33 H22 Z" fill="currentColor" fill-opacity=".8"/></svg></span><span><span class="n">Utopia vs Dystopia Split-Screen</span><span class="d">Write the same future twice: the brochure where it went perfectly and the headline where it went horribly</span><span class="gf">Great for: Strategy · Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="Sci-Fi Artifact From the Future" data-cat="Speculative Future" data-desc="Describe one physical object, ad, or news clip from the world where this idea already won — reverse-engineer it" data-good="novel|feature" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M11 31 L 26.5 15.5"/><path d="M20 14.5 H 28 V 22.5"/></g><circle cx="31" cy="12" r="2.8" fill="currentColor"/><g stroke="currentColor" stroke-width="1.4" stroke-linecap="round"><line x1="31" y1="6.5" x2="31" y2="8.4"/><line x1="31" y1="15.6" x2="31" y2="17.5"/><line x1="25.5" y1="12" x2="27.4" y2="12"/><line x1="34.6" y1="12" x2="36.5" y2="12"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M22 9 L33 15 V28 L22 35 L11 28 V15 Z" fill="currentColor" fill-opacity=".15"/><path d="M22 9 L33 15 V28 L22 35 L11 28 V15 Z M11 15 L22 21 L33 15 M22 21 V35" fill="none" stroke="currentColor" stroke-width="2" stroke-linejoin="round"/></svg></span><span><span class="n">Sci-Fi Artifact From the Future</span><span class="d">Describe one physical object, ad, or news clip from the world where this idea already won — reverse-engineer it</span><span class="gf">Great for: Novel concept · Build a feature</span></span></label><label class="tech"><input type="checkbox" data-name="Emerging Tech Collision" data-cat="Speculative Future" data-desc="Force-marry your idea to a frontier tech (AGI, fusion, neural implants, gene edit) and ask what new thing is born" data-good="novel|feature|strategy" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M11 31 L 26.5 15.5"/><path d="M20 14.5 H 28 V 22.5"/></g><circle cx="31" cy="12" r="2.8" fill="currentColor"/><g stroke="currentColor" stroke-width="1.4" stroke-linecap="round"><line x1="31" y1="6.5" x2="31" y2="8.4"/><line x1="31" y1="15.6" x2="31" y2="17.5"/><line x1="25.5" y1="12" x2="27.4" y2="12"/><line x1="34.6" y1="12" x2="36.5" y2="12"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><rect x="15" y="15" width="14" height="14" rx="2" fill="currentColor" fill-opacity=".22"/><rect x="15" y="15" width="14" height="14" rx="2" fill="none" stroke="currentColor" stroke-width="2"/><g stroke="currentColor" stroke-width="2" stroke-linecap="round"><path d="M19 15 V10 M25 15 V10 M19 29 V34 M25 29 V34 M15 19 H10 M15 25 H10 M29 19 H34 M29 25 H34"/></g></svg></span><span><span class="n">Emerging Tech Collision</span><span class="d">Force-marry your idea to a frontier tech (AGI, fusion, neural implants, gene edit) and ask what new thing is born</span><span class="gf">Great for: Novel concept · Build a feature · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="What-If-The-World-Changed Card Flip" data-cat="Speculative Future" data-desc="Draw a wild world-shift (no privacy, half population, 200-yr lifespans) and redesign the idea to fit that world" data-good="novel|unstuck" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M11 31 L 26.5 15.5"/><path d="M20 14.5 H 28 V 22.5"/></g><circle cx="31" cy="12" r="2.8" fill="currentColor"/><g stroke="currentColor" stroke-width="1.4" stroke-linecap="round"><line x1="31" y1="6.5" x2="31" y2="8.4"/><line x1="31" y1="15.6" x2="31" y2="17.5"/><line x1="25.5" y1="12" x2="27.4" y2="12"/><line x1="34.6" y1="12" x2="36.5" y2="12"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><rect x="13" y="10" width="18" height="24" rx="2.5" fill="currentColor" fill-opacity=".18" stroke="currentColor" stroke-width="2"/><path d="M22 10 V34" stroke="currentColor" stroke-width="1.6" stroke-dasharray="2 2.5"/><path d="M27 16 a4 4 0 1 1 4 4" fill="none" stroke="currentColor" stroke-width="1.8" stroke-linecap="round"/></svg></span><span><span class="n">What-If-The-World-Changed Card Flip</span><span class="d">Draw a wild world-shift (no privacy, half population, 200-yr lifespans) and redesign the idea to fit that world</span><span class="gf">Great for: Novel concept · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Future Anthropologist Dig" data-cat="Speculative Future" data-desc="A scholar in 2200 unearths your idea as a relic — what do they conclude it reveals about us, and what replaced it" data-good="strategy|novel" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M11 31 L 26.5 15.5"/><path d="M20 14.5 H 28 V 22.5"/></g><circle cx="31" cy="12" r="2.8" fill="currentColor"/><g stroke="currentColor" stroke-width="1.4" stroke-linecap="round"><line x1="31" y1="6.5" x2="31" y2="8.4"/><line x1="31" y1="15.6" x2="31" y2="17.5"/><line x1="25.5" y1="12" x2="27.4" y2="12"/><line x1="34.6" y1="12" x2="36.5" y2="12"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M22 27 a6 6 0 1 1 0.1 0 z"/><path d="M22 23 a2.5 2.5 0 1 0 0.1 0 M19 30 a5 5 0 0 0 6 0"/></g><path d="M12 16 L17 13 M32 16 L27 13" stroke="currentColor" stroke-width="1.6" stroke-opacity=".5"/></svg></span><span><span class="n">Future Anthropologist Dig</span><span class="d">A scholar in 2200 unearths your idea as a relic — what do they conclude it reveals about us, and what replaced it</span><span class="gf">Great for: Strategy · Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Scenario Cross" data-cat="Speculative Future" data-desc="Pick two high-impact uncertainties, cross them into four futures, and ideate the move that wins in every one" data-good="strategy|planning" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M11 31 L 26.5 15.5"/><path d="M20 14.5 H 28 V 22.5"/></g><circle cx="31" cy="12" r="2.8" fill="currentColor"/><g stroke="currentColor" stroke-width="1.4" stroke-linecap="round"><line x1="31" y1="6.5" x2="31" y2="8.4"/><line x1="31" y1="15.6" x2="31" y2="17.5"/><line x1="25.5" y1="12" x2="27.4" y2="12"/><line x1="34.6" y1="12" x2="36.5" y2="12"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2" stroke-linecap="round"><path d="M22 9 V35 M9 22 H35"/></g><g fill="currentColor"><circle cx="15" cy="15" r="2"/><circle cx="29" cy="15" r="2"/><circle cx="15" cy="29" r="2"/><circle cx="29" cy="29" r="2"/></g></svg></span><span><span class="n">Scenario Cross</span><span class="d">Pick two high-impact uncertainties, cross them into four futures, and ideate the move that wins in every one</span><span class="gf">Great for: Strategy · Planning</span></span></label><label class="tech invent"><input type="checkbox" data-invent="Speculative Future"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M11 31 L 26.5 15.5"/><path d="M20 14.5 H 28 V 22.5"/></g><circle cx="31" cy="12" r="2.8" fill="currentColor"/><g stroke="currentColor" stroke-width="1.4" stroke-linecap="round"><line x1="31" y1="6.5" x2="31" y2="8.4"/><line x1="31" y1="15.6" x2="31" y2="17.5"/><line x1="25.5" y1="12" x2="27.4" y2="12"/><line x1="34.6" y1="12" x2="36.5" y2="12"/></g></svg></span><span><span class="n">✨ Invent a Speculative Future technique</span><span class="d">Make up a brand-new technique on the fly, in the spirit of Speculative Future</span></span></label></div></section>
+<section data-cat="Quantum" style="--c:#2b86d9"><h2>Quantum<span class="cnt">6</span></h2><div class="grid"><label class="tech"><input type="checkbox" data-name="Observer Effect" data-cat="Quantum" data-desc="Ask how the act of watching, measuring, or shipping this idea changes the very thing you&#x27;re trying to capture" data-good="strategy|diagnosis" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8" fill="none"><ellipse cx="22" cy="22" rx="14.5" ry="6" transform="rotate(28 22 22)"/><ellipse cx="22" cy="22" rx="14.5" ry="6" transform="rotate(-28 22 22)"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.18"/><circle cx="22" cy="22" r="3.4" fill="currentColor"/><circle cx="33.2" cy="17.4" r="2" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M9 22 q13 -10 26 0 q-13 10 -26 0 z" fill="none" stroke="currentColor" stroke-width="2" stroke-linejoin="round"/><circle cx="22" cy="22" r="4.5" fill="none" stroke="currentColor" stroke-width="2"/><circle cx="22" cy="22" r="1.8" fill="currentColor"/></svg></span><span><span class="n">Observer Effect</span><span class="d">Ask how the act of watching, measuring, or shipping this idea changes the very thing you&#x27;re trying to capture</span><span class="gf">Great for: Strategy · Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="Entanglement Thinking" data-cat="Quantum" data-desc="Pair two distant parts of the problem and insist a change in one instantly flips the other — surface the hidden linkage" data-good="diagnosis|strategy" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8" fill="none"><ellipse cx="22" cy="22" rx="14.5" ry="6" transform="rotate(28 22 22)"/><ellipse cx="22" cy="22" rx="14.5" ry="6" transform="rotate(-28 22 22)"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.18"/><circle cx="22" cy="22" r="3.4" fill="currentColor"/><circle cx="33.2" cy="17.4" r="2" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2"><circle cx="15" cy="22" r="6"/><circle cx="29" cy="22" r="6"/></g><path d="M15 22 h14" stroke="currentColor" stroke-width="2" stroke-dasharray="1.5 2.5"/><g fill="currentColor"><circle cx="15" cy="22" r="1.8"/><circle cx="29" cy="22" r="1.8"/></g></svg></span><span><span class="n">Entanglement Thinking</span><span class="d">Pair two distant parts of the problem and insist a change in one instantly flips the other — surface the hidden linkage</span><span class="gf">Great for: Diagnose · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Superposition Collapse" data-cat="Quantum" data-desc="Hold all rival solutions alive at once, then name the one constraint that collapses them to a single winner" data-good="strategy|diagnosis" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8" fill="none"><ellipse cx="22" cy="22" rx="14.5" ry="6" transform="rotate(28 22 22)"/><ellipse cx="22" cy="22" rx="14.5" ry="6" transform="rotate(-28 22 22)"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.18"/><circle cx="22" cy="22" r="3.4" fill="currentColor"/><circle cx="33.2" cy="17.4" r="2" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M11 12 C20 16 24 16 33 12 M11 18 C20 22 24 22 33 18 M11 24 C20 28 24 28 33 24"/><path d="M22 26 V34"/></g><circle cx="22" cy="34" r="2" fill="currentColor"/></svg></span><span><span class="n">Superposition Collapse</span><span class="d">Hold all rival solutions alive at once, then name the one constraint that collapses them to a single winner</span><span class="gf">Great for: Strategy · Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="Relativity Frame Shift" data-cat="Quantum" data-desc="Re-run the idea from a wildly different observer&#x27;s reference frame — the slow user, the rival, future-you — and see what warps" data-good="strategy|novel" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8" fill="none"><ellipse cx="22" cy="22" rx="14.5" ry="6" transform="rotate(28 22 22)"/><ellipse cx="22" cy="22" rx="14.5" ry="6" transform="rotate(-28 22 22)"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.18"/><circle cx="22" cy="22" r="3.4" fill="currentColor"/><circle cx="33.2" cy="17.4" r="2" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><rect x="11" y="11" width="22" height="22" rx="2" fill="none" stroke="currentColor" stroke-width="2" stroke-opacity=".4"/><rect x="15" y="15" width="18" height="18" rx="2" transform="rotate(-14 22 22)" fill="none" stroke="currentColor" stroke-width="2"/></svg></span><span><span class="n">Relativity Frame Shift</span><span class="d">Re-run the idea from a wildly different observer&#x27;s reference frame — the slow user, the rival, future-you — and see what warps</span><span class="gf">Great for: Strategy · Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Field Lines" data-cat="Quantum" data-desc="Treat the goal as a charge and map the invisible forces pulling every stakeholder toward or away from it" data-good="strategy" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8" fill="none"><ellipse cx="22" cy="22" rx="14.5" ry="6" transform="rotate(28 22 22)"/><ellipse cx="22" cy="22" rx="14.5" ry="6" transform="rotate(-28 22 22)"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.18"/><circle cx="22" cy="22" r="3.4" fill="currentColor"/><circle cx="33.2" cy="17.4" r="2" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M12 13 V31 M16 13 C24 18 24 26 16 31 M22 13 C32 18 32 26 22 31"/></g><circle cx="11" cy="22" r="2" fill="currentColor"/></svg></span><span><span class="n">Field Lines</span><span class="d">Treat the goal as a charge and map the invisible forces pulling every stakeholder toward or away from it</span><span class="gf">Great for: Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Quantum Tunneling" data-cat="Quantum" data-desc="Assume the idea can pass straight through the &#x27;impossible&#x27; barrier instead of over it — what&#x27;s on the other side, reached cheaply" data-good="unstuck|novel" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8" fill="none"><ellipse cx="22" cy="22" rx="14.5" ry="6" transform="rotate(28 22 22)"/><ellipse cx="22" cy="22" rx="14.5" ry="6" transform="rotate(-28 22 22)"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.18"/><circle cx="22" cy="22" r="3.4" fill="currentColor"/><circle cx="33.2" cy="17.4" r="2" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><rect x="20" y="9" width="5" height="26" rx="1.5" fill="currentColor" fill-opacity=".3"/><path d="M10 22 H34" fill="none" stroke="currentColor" stroke-width="2.2" stroke-linecap="round"/><path d="M28 17 L34 22 L28 27" fill="none" stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round"/></svg></span><span><span class="n">Quantum Tunneling</span><span class="d">Assume the idea can pass straight through the &#x27;impossible&#x27; barrier instead of over it — what&#x27;s on the other side, reached cheaply</span><span class="gf">Great for: Get unstuck · Novel concept</span></span></label><label class="tech invent"><input type="checkbox" data-invent="Quantum"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8" fill="none"><ellipse cx="22" cy="22" rx="14.5" ry="6" transform="rotate(28 22 22)"/><ellipse cx="22" cy="22" rx="14.5" ry="6" transform="rotate(-28 22 22)"/></g><circle cx="22" cy="22" r="6.6" fill="currentColor" fill-opacity="0.18"/><circle cx="22" cy="22" r="3.4" fill="currentColor"/><circle cx="33.2" cy="17.4" r="2" fill="currentColor"/></svg></span><span><span class="n">✨ Invent a Quantum technique</span><span class="d">Make up a brand-new technique on the fly, in the spirit of Quantum</span></span></label></div></section>
+<h2 class="grouphdr" style="--c:#e2562f">Wild &amp; Playful</h2>
+<section data-cat="Wild" style="--c:#e2562f"><h2>Wild<span class="cnt">7</span></h2><div class="grid"><label class="tech"><input type="checkbox" data-name="Chaos Engineering" data-cat="Wild" data-desc="Deliberately break your idea every way it could fail, then rebuild only the parts that survive the wreckage" data-good="feature|diagnosis|strategy" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M24.5 6.5 L 12.5 24 H 19.5 L 17.5 37.5 L 31.5 18.5 H 24 L 24.5 6.5 Z" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M22 9 L25 18 L34 18 L27 24 L30 33 L22 27 L14 33 L17 24 L10 18 L19 18 Z"/></g></svg></span><span><span class="n">Chaos Engineering</span><span class="d">Deliberately break your idea every way it could fail, then rebuild only the parts that survive the wreckage</span><span class="gf">Great for: Build a feature · Diagnose · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Guerrilla Gardening Ideas" data-cat="Wild" data-desc="Plant your solution in the least expected place and let it grow underground until it surprises everyone" data-good="strategy|unstuck" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M24.5 6.5 L 12.5 24 H 19.5 L 17.5 37.5 L 31.5 18.5 H 24 L 24.5 6.5 Z" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M22 33 V21"/><path d="M22 22 c-7 0 -9 -6 -9 -9 c6 0 9 3 9 9 z"/><path d="M22 24 c6 0 8 -4 8 -7 c-5 0 -8 2 -8 7 z"/></g></svg></span><span><span class="n">Guerrilla Gardening Ideas</span><span class="d">Plant your solution in the least expected place and let it grow underground until it surprises everyone</span><span class="gf">Great for: Strategy · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Pirate Code Brainstorm" data-cat="Wild" data-desc="Steal the best bits from anywhere, remix without asking permission, grab what works and run" data-good="novel|unstuck" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M24.5 6.5 L 12.5 24 H 19.5 L 17.5 37.5 L 31.5 18.5 H 24 L 24.5 6.5 Z" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M22 10 c-7 0 -11 5 -11 11 c0 4 2 6 4 7 v4 h3 v-2 h2 v2 h4 v-2 h2 v2 h3 v-4 c2 -1 4 -3 4 -7 c0 -6 -4 -11 -11 -11 z" fill="none" stroke="currentColor" stroke-width="2" stroke-linejoin="round"/><g fill="currentColor"><circle cx="17.5" cy="21" r="2.2"/><circle cx="26.5" cy="21" r="2.2"/></g></svg></span><span><span class="n">Pirate Code Brainstorm</span><span class="d">Steal the best bits from anywhere, remix without asking permission, grab what works and run</span><span class="gf">Great for: Novel concept · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Zombie Apocalypse Planning" data-cat="Wild" data-desc="Society just collapsed — strip your idea to only what survives with no power, no rules, no backup" data-good="feature|strategy|unstuck" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M24.5 6.5 L 12.5 24 H 19.5 L 17.5 37.5 L 31.5 18.5 H 24 L 24.5 6.5 Z" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="22" r="4" fill="none" stroke="currentColor" stroke-width="2"/><g fill="none" stroke="currentColor" stroke-width="2"><path d="M22 10 a12 12 0 0 1 6 3.5 M32 16 a12 12 0 0 1 0 12 M28 33.5 a12 12 0 0 1 -12 0 M12 28 a12 12 0 0 1 0 -12 M16 10.5 a12 12 0 0 1 6 -0.5" stroke-dasharray="0.1 5.5"/></g><g fill="currentColor"><circle cx="22" cy="11" r="2"/><circle cx="11" cy="22" r="2"/><circle cx="33" cy="22" r="2"/></g></svg></span><span><span class="n">Zombie Apocalypse Planning</span><span class="d">Society just collapsed — strip your idea to only what survives with no power, no rules, no backup</span><span class="gf">Great for: Build a feature · Strategy · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Drunk History Retelling" data-cat="Wild" data-desc="Explain it like you&#x27;re three drinks in: no filter, no jargon, just the raw stupid-simple truth" data-good="unstuck|diagnosis" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M24.5 6.5 L 12.5 24 H 19.5 L 17.5 37.5 L 31.5 18.5 H 24 L 24.5 6.5 Z" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M13 12 H31 L25 23 V31 H19 V23 Z"/><path d="M19 31 H25" /></g><circle cx="29" cy="14" r="1.4" fill="currentColor"/></svg></span><span><span class="n">Drunk History Retelling</span><span class="d">Explain it like you&#x27;re three drinks in: no filter, no jargon, just the raw stupid-simple truth</span><span class="gf">Great for: Get unstuck · Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="Anti-Solution" data-cat="Wild" data-desc="Brainstorm how to make the problem spectacularly worse, then invert every sabotage into a fix" data-good="diagnosis|unstuck" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M24.5 6.5 L 12.5 24 H 19.5 L 17.5 37.5 L 31.5 18.5 H 24 L 24.5 6.5 Z" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M14 18 a8 8 0 1 1 -1 8"/><path d="M14 12 L14 18.5 L20 18"/></g></svg></span><span><span class="n">Anti-Solution</span><span class="d">Brainstorm how to make the problem spectacularly worse, then invert every sabotage into a fix</span><span class="gf">Great for: Diagnose · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Elemental Forces" data-cat="Wild" data-desc="Let fire, water, earth, and air each sculpt your idea their own brutal way and see what survives" data-good="novel|unstuck" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M24.5 6.5 L 12.5 24 H 19.5 L 17.5 37.5 L 31.5 18.5 H 24 L 24.5 6.5 Z" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M22 8 L30 22 H14 Z"/><path d="M14 30 L22 36 L30 30"/><path d="M14 26 H30"/></g></svg></span><span><span class="n">Elemental Forces</span><span class="d">Let fire, water, earth, and air each sculpt your idea their own brutal way and see what survives</span><span class="gf">Great for: Novel concept · Get unstuck</span></span></label><label class="tech invent"><input type="checkbox" data-invent="Wild"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M24.5 6.5 L 12.5 24 H 19.5 L 17.5 37.5 L 31.5 18.5 H 24 L 24.5 6.5 Z" fill="currentColor"/></svg></span><span><span class="n">✨ Invent a Wild technique</span><span class="d">Make up a brand-new technique on the fly, in the spirit of Wild</span></span></label></div></section>
+<section data-cat="Absurdist" style="--c:#e0529c"><h2>Absurdist<span class="cnt">6</span></h2><div class="grid"><label class="tech"><input type="checkbox" data-name="Villain&#x27;s Monologue" data-cat="Absurdist" data-desc="Pitch your problem as an evil mastermind gloating about their scheme; the diabolical plan reveals the real solution" data-good="diagnosis|strategy|unstuck" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g transform="rotate(-12 22 22)"><circle cx="22" cy="22" r="13" fill="currentColor" fill-opacity="0.14"/><circle cx="22" cy="22" r="13" stroke="currentColor" stroke-width="1.6" fill="none"/><path d="M16 19 q 2 -2.4 4 0" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/><circle cx="26.5" cy="18.8" r="1.8" fill="currentColor"/><path d="M16.5 26 C 19 30, 25 30, 28 24.5" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M12 20 C16 18 19 18 22 21 C25 18 28 18 32 20 C30 24 26 24 22 21 C18 24 14 24 12 20 Z"/></g><circle cx="22" cy="14" r="2.4" fill="currentColor"/></svg></span><span><span class="n">Villain&#x27;s Monologue</span><span class="d">Pitch your problem as an evil mastermind gloating about their scheme; the diabolical plan reveals the real solution</span><span class="gf">Great for: Diagnose · Strategy · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Explain It to a Golden Retriever" data-cat="Absurdist" data-desc="Re-pitch the idea to an excitable dog who only cares about treats, balls, and naps; keep only what survives" data-good="unstuck|diagnosis|feature" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g transform="rotate(-12 22 22)"><circle cx="22" cy="22" r="13" fill="currentColor" fill-opacity="0.14"/><circle cx="22" cy="22" r="13" stroke="currentColor" stroke-width="1.6" fill="none"/><path d="M16 19 q 2 -2.4 4 0" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/><circle cx="26.5" cy="18.8" r="1.8" fill="currentColor"/><path d="M16.5 26 C 19 30, 25 30, 28 24.5" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M14 18 C12 12 17 13 18 17 M30 18 C32 12 27 13 26 17"/><path d="M15 19 C13 28 18 33 22 33 C26 33 31 28 29 19 C26 16 18 16 15 19 Z"/></g><g fill="currentColor"><circle cx="19" cy="24" r="1.4"/><circle cx="25" cy="24" r="1.4"/><circle cx="22" cy="28" r="1.6"/></g></svg></span><span><span class="n">Explain It to a Golden Retriever</span><span class="d">Re-pitch the idea to an excitable dog who only cares about treats, balls, and naps; keep only what survives</span><span class="gf">Great for: Get unstuck · Diagnose · Build a feature</span></span></label><label class="tech"><input type="checkbox" data-name="Infomercial at 3AM" data-cat="Absurdist" data-desc="Sell your half-baked idea as a desperate late-night infomercial: &#x27;But wait, there&#x27;s more!&#x27; until features fall out" data-good="strategy|novel" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g transform="rotate(-12 22 22)"><circle cx="22" cy="22" r="13" fill="currentColor" fill-opacity="0.14"/><circle cx="22" cy="22" r="13" stroke="currentColor" stroke-width="1.6" fill="none"/><path d="M16 19 q 2 -2.4 4 0" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/><circle cx="26.5" cy="18.8" r="1.8" fill="currentColor"/><path d="M16.5 26 C 19 30, 25 30, 28 24.5" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><rect x="9" y="14" width="26" height="18" rx="2.5" fill="none" stroke="currentColor" stroke-width="2"/><path d="M18 9 L22 14 L26 9" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"/><path d="M22 19 l1 2.6 2.8 .2 -2.1 1.9 .7 2.7 -2.4 -1.5 -2.4 1.5 .7 -2.7 -2.1 -1.9 2.8 -.2 z" fill="currentColor"/></svg></span><span><span class="n">Infomercial at 3AM</span><span class="d">Sell your half-baked idea as a desperate late-night infomercial: &#x27;But wait, there&#x27;s more!&#x27; until features fall out</span><span class="gf">Great for: Strategy · Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Drunk Uncle at Thanksgiving" data-cat="Absurdist" data-desc="Have your loudest, least-filtered relative rant about the problem; mine the unhinged hot takes for buried truth" data-good="unstuck|diagnosis" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g transform="rotate(-12 22 22)"><circle cx="22" cy="22" r="13" fill="currentColor" fill-opacity="0.14"/><circle cx="22" cy="22" r="13" stroke="currentColor" stroke-width="1.6" fill="none"/><path d="M16 19 q 2 -2.4 4 0" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/><circle cx="26.5" cy="18.8" r="1.8" fill="currentColor"/><path d="M16.5 26 C 19 30, 25 30, 28 24.5" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M11 16 L20 16 L27 11 V29 L20 24 L11 24 Z"/><path d="M30 16 q3 4 0 8 M33 13 q5 7 0 14"/></g></svg></span><span><span class="n">Drunk Uncle at Thanksgiving</span><span class="d">Have your loudest, least-filtered relative rant about the problem; mine the unhinged hot takes for buried truth</span><span class="gf">Great for: Get unstuck · Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="Cursed Genie" data-cat="Absurdist" data-desc="Make a wish, then let a malicious genie grant it in the most technically-correct disastrous way; patch each loophole" data-good="diagnosis|feature" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g transform="rotate(-12 22 22)"><circle cx="22" cy="22" r="13" fill="currentColor" fill-opacity="0.14"/><circle cx="22" cy="22" r="13" stroke="currentColor" stroke-width="1.6" fill="none"/><path d="M16 19 q 2 -2.4 4 0" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/><circle cx="26.5" cy="18.8" r="1.8" fill="currentColor"/><path d="M16.5 26 C 19 30, 25 30, 28 24.5" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M10 30 h18 a2 2 0 0 0 2 -2 c0 -5 -6 -5 -8 -8 c5 -1 8 -3 8 -3 c-4 -2 -12 -2 -16 1 c-4 3 -5 9 -4 12 z"/><path d="M30 17 L33 14 M31 21 L35 20" stroke-opacity=".6"/></g></svg></span><span><span class="n">Cursed Genie</span><span class="d">Make a wish, then let a malicious genie grant it in the most technically-correct disastrous way; patch each loophole</span><span class="gf">Great for: Diagnose · Build a feature</span></span></label><label class="tech"><input type="checkbox" data-name="Three Rounds of Stupid" data-cat="Absurdist" data-desc="Round 1 absurd ideas, Round 2 make each MORE absurd, Round 3 find the smallest serious thing hiding in the silliest" data-good="unstuck|novel" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g transform="rotate(-12 22 22)"><circle cx="22" cy="22" r="13" fill="currentColor" fill-opacity="0.14"/><circle cx="22" cy="22" r="13" stroke="currentColor" stroke-width="1.6" fill="none"/><path d="M16 19 q 2 -2.4 4 0" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/><circle cx="26.5" cy="18.8" r="1.8" fill="currentColor"/><path d="M16.5 26 C 19 30, 25 30, 28 24.5" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M13 30 V20 M9 24 L13 20 L17 24 M22 30 V15 M18 19 L22 15 L26 19 M31 30 V11 M27 15 L31 11 L35 15"/></g></svg></span><span><span class="n">Three Rounds of Stupid</span><span class="d">Round 1 absurd ideas, Round 2 make each MORE absurd, Round 3 find the smallest serious thing hiding in the silliest</span><span class="gf">Great for: Get unstuck · Novel concept</span></span></label><label class="tech invent"><input type="checkbox" data-invent="Absurdist"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g transform="rotate(-12 22 22)"><circle cx="22" cy="22" r="13" fill="currentColor" fill-opacity="0.14"/><circle cx="22" cy="22" r="13" stroke="currentColor" stroke-width="1.6" fill="none"/><path d="M16 19 q 2 -2.4 4 0" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/><circle cx="26.5" cy="18.8" r="1.8" fill="currentColor"/><path d="M16.5 26 C 19 30, 25 30, 28 24.5" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/></g></svg></span><span><span class="n">✨ Invent a Absurdist technique</span><span class="d">Make up a brand-new technique on the fly, in the spirit of Absurdist</span></span></label></div></section>
+<section data-cat="Theatrical" style="--c:#cf4d6f"><h2>Theatrical<span class="cnt">7</span></h2><div class="grid"><label class="tech"><input type="checkbox" data-name="Time Travel Talk Show" data-cat="Theatrical" data-desc="Host a talk show interviewing your past, present, and future selves to mine each era for advice on the problem" data-good="novel|personal" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z" fill="currentColor" fill-opacity="0.18"/><path d="M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z" stroke="currentColor" stroke-width="1.8" fill="none"/><g fill="currentColor"><circle cx="18.5" cy="21" r="1.7"/><circle cx="25.5" cy="21" r="1.7"/></g><path d="M18 27 C 20 29.5, 24 29.5, 26 27" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><rect x="18" y="9" width="8" height="15" rx="4" fill="currentColor" fill-opacity=".25"/><rect x="18" y="9" width="8" height="15" rx="4" fill="none" stroke="currentColor" stroke-width="2"/><path d="M14 21 a8 8 0 0 0 16 0 M22 29 V34 M17 34 H27" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"/></svg></span><span><span class="n">Time Travel Talk Show</span><span class="d">Host a talk show interviewing your past, present, and future selves to mine each era for advice on the problem</span><span class="gf">Great for: Novel concept · Personal / life</span></span></label><label class="tech"><input type="checkbox" data-name="Alien Anthropologist" data-cat="Theatrical" data-desc="Become a baffled alien studying the problem and narrate aloud what seems strange, arbitrary, or insane about it" data-good="diagnosis|unstuck|strategy" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z" fill="currentColor" fill-opacity="0.18"/><path d="M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z" stroke="currentColor" stroke-width="1.8" fill="none"/><g fill="currentColor"><circle cx="18.5" cy="21" r="1.7"/><circle cx="25.5" cy="21" r="1.7"/></g><path d="M18 27 C 20 29.5, 24 29.5, 26 27" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><ellipse cx="22" cy="30" rx="13" ry="4.5" fill="currentColor" fill-opacity=".25"/><path d="M22 11 c7 0 10 6 10 11 c0 5 -5 7 -10 7 c-5 0 -10 -2 -10 -7 c0 -5 3 -11 10 -11 z" fill="none" stroke="currentColor" stroke-width="2"/><g fill="currentColor"><ellipse cx="18" cy="22" rx="1.6" ry="2.4"/><ellipse cx="26" cy="22" rx="1.6" ry="2.4"/></g></svg></span><span><span class="n">Alien Anthropologist</span><span class="d">Become a baffled alien studying the problem and narrate aloud what seems strange, arbitrary, or insane about it</span><span class="gf">Great for: Diagnose · Get unstuck · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Dream Fusion Laboratory" data-cat="Theatrical" data-desc="Voice the impossible fantasy solution first, then reverse-engineer the bridging steps back to reality" data-good="novel|unstuck" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z" fill="currentColor" fill-opacity="0.18"/><path d="M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z" stroke="currentColor" stroke-width="1.8" fill="none"/><g fill="currentColor"><circle cx="18.5" cy="21" r="1.7"/><circle cx="25.5" cy="21" r="1.7"/></g><path d="M18 27 C 20 29.5, 24 29.5, 26 27" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M18 9 H26 M19.5 9 V18 L13 30 a2 2 0 0 0 2 3 H29 a2 2 0 0 0 2 -3 L24.5 18 V9"/></g><path d="M16.5 26 H27.5" stroke="currentColor" stroke-width="2"/><circle cx="20" cy="29" r="1.4" fill="currentColor"/><circle cx="25" cy="28" r="1.1" fill="currentColor"/></svg></span><span><span class="n">Dream Fusion Laboratory</span><span class="d">Voice the impossible fantasy solution first, then reverse-engineer the bridging steps back to reality</span><span class="gf">Great for: Novel concept · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Emotion Orchestra" data-cat="Theatrical" data-desc="Run a separate ideation round led by each emotion (rage, joy, fear, hope), then harmonize their conflicting ideas" data-good="personal|strategy" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z" fill="currentColor" fill-opacity="0.18"/><path d="M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z" stroke="currentColor" stroke-width="1.8" fill="none"/><g fill="currentColor"><circle cx="18.5" cy="21" r="1.7"/><circle cx="25.5" cy="21" r="1.7"/></g><path d="M18 27 C 20 29.5, 24 29.5, 26 27" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M17 30 V15 L31 12 V27"/><circle cx="14" cy="30" r="3"/><circle cx="28" cy="27" r="3"/></g></svg></span><span><span class="n">Emotion Orchestra</span><span class="d">Run a separate ideation round led by each emotion (rage, joy, fear, hope), then harmonize their conflicting ideas</span><span class="gf">Great for: Personal / life · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Parallel Universe Cafe" data-cat="Theatrical" data-desc="Rewrite one fundamental rule of reality (physics, economics, social norms) and solve the problem under those laws" data-good="novel|unstuck" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z" fill="currentColor" fill-opacity="0.18"/><path d="M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z" stroke="currentColor" stroke-width="1.8" fill="none"/><g fill="currentColor"><circle cx="18.5" cy="21" r="1.7"/><circle cx="25.5" cy="21" r="1.7"/></g><path d="M18 27 C 20 29.5, 24 29.5, 26 27" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2"><circle cx="18" cy="22" r="9"/><circle cx="26" cy="22" r="9" stroke-dasharray="2.5 2.5"/></g></svg></span><span><span class="n">Parallel Universe Cafe</span><span class="d">Rewrite one fundamental rule of reality (physics, economics, social norms) and solve the problem under those laws</span><span class="gf">Great for: Novel concept · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Persona Journey" data-cat="Theatrical" data-desc="Embody an archetype and solve the problem in-character, naming what that persona sees that you normally miss" data-good="feature|strategy" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z" fill="currentColor" fill-opacity="0.18"/><path d="M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z" stroke="currentColor" stroke-width="1.8" fill="none"/><g fill="currentColor"><circle cx="18.5" cy="21" r="1.7"/><circle cx="25.5" cy="21" r="1.7"/></g><path d="M18 27 C 20 29.5, 24 29.5, 26 27" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M14 33 q-2 -8 6 -9 q8 -1 6 -8" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-dasharray="0.1 4"/><circle cx="14" cy="33" r="2.4" fill="currentColor"/><path d="M26 16 l3 -5 3 5 z" fill="currentColor"/></svg></span><span><span class="n">Persona Journey</span><span class="d">Embody an archetype and solve the problem in-character, naming what that persona sees that you normally miss</span><span class="gf">Great for: Build a feature · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Devil&#x27;s Advocate Courtroom" data-cat="Theatrical" data-desc="Stage a trial: prosecute the idea, defend it, then deliver the jury verdict, each role argued fully in character" data-good="strategy|diagnosis" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z" fill="currentColor" fill-opacity="0.18"/><path d="M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z" stroke="currentColor" stroke-width="1.8" fill="none"/><g fill="currentColor"><circle cx="18.5" cy="21" r="1.7"/><circle cx="25.5" cy="21" r="1.7"/></g><path d="M18 27 C 20 29.5, 24 29.5, 26 27" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M22 10 V32 M14 32 H30"/><path d="M11 16 H33 M11 16 L8 23 H14 Z M33 16 L30 23 H36 Z"/></g></svg></span><span><span class="n">Devil&#x27;s Advocate Courtroom</span><span class="d">Stage a trial: prosecute the idea, defend it, then deliver the jury verdict, each role argued fully in character</span><span class="gf">Great for: Strategy · Diagnose</span></span></label><label class="tech invent"><input type="checkbox" data-invent="Theatrical"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z" fill="currentColor" fill-opacity="0.18"/><path d="M13 12 H 31 V 22 C 31 30, 27 35, 22 35 C 17 35, 13 30, 13 22 Z" stroke="currentColor" stroke-width="1.8" fill="none"/><g fill="currentColor"><circle cx="18.5" cy="21" r="1.7"/><circle cx="25.5" cy="21" r="1.7"/></g><path d="M18 27 C 20 29.5, 24 29.5, 26 27" stroke="currentColor" stroke-width="1.8" stroke-linecap="round" fill="none"/></svg></span><span><span class="n">✨ Invent a Theatrical technique</span><span class="d">Make up a brand-new technique on the fly, in the spirit of Theatrical</span></span></label></div></section>
+<section data-cat="Constraint" style="--c:#d9882b"><h2>Constraint<span class="cnt">7</span></h2><div class="grid"><label class="tech"><input type="checkbox" data-name="Kill the Crown Jewel" data-cat="Constraint" data-desc="Delete the single best, most beloved feature — now redesign the whole thing to win without it" data-good="feature|strategy|unstuck" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M17 11 H 11 V 17"/><path d="M27 11 H 33 V 17"/><path d="M17 33 H 11 V 27"/><path d="M27 33 H 33 V 27"/></g><circle cx="22" cy="22" r="5" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="2.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M11 28 L13 15 L19 22 L22 12 L25 22 L31 15 L33 28 Z"/><path d="M11 28 H33"/></g><path d="M14 12 L30 32" stroke="currentColor" stroke-width="2.4" stroke-linecap="round"/></svg></span><span><span class="n">Kill the Crown Jewel</span><span class="d">Delete the single best, most beloved feature — now redesign the whole thing to win without it</span><span class="gf">Great for: Build a feature · Strategy · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="1000x Budget" data-cat="Constraint" data-desc="Pretend money, time, and people are infinite — design the absurd version, then mine it for ideas you can actually steal" data-good="novel|strategy" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M17 11 H 11 V 17"/><path d="M27 11 H 33 V 17"/><path d="M17 33 H 11 V 27"/><path d="M27 33 H 33 V 27"/></g><circle cx="22" cy="22" r="5" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="2.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2"><ellipse cx="22" cy="14" rx="9" ry="3.5"/><path d="M13 14 V22 a9 3.5 0 0 0 18 0 V14"/><path d="M13 22 V30 a9 3.5 0 0 0 18 0 V22"/></g></svg></span><span><span class="n">1000x Budget</span><span class="d">Pretend money, time, and people are infinite — design the absurd version, then mine it for ideas you can actually steal</span><span class="gf">Great for: Novel concept · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Ship in 60 Minutes" data-cat="Constraint" data-desc="You launch in one hour with what&#x27;s already on hand — name what you cut, fake, or borrow to make it real" data-good="feature|planning|unstuck" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M17 11 H 11 V 17"/><path d="M27 11 H 33 V 17"/><path d="M17 33 H 11 V 27"/><path d="M27 33 H 33 V 27"/></g><circle cx="22" cy="22" r="5" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="2.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="24" r="11" fill="none" stroke="currentColor" stroke-width="2"/><path d="M22 24 V17 M22 24 L27 27" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"/><path d="M18 8 H26 M22 8 V13" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"/></svg></span><span><span class="n">Ship in 60 Minutes</span><span class="d">You launch in one hour with what&#x27;s already on hand — name what you cut, fake, or borrow to make it real</span><span class="gf">Great for: Build a feature · Planning · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="The $0 Mandate" data-cat="Constraint" data-desc="Achieve the goal spending literally nothing — no tools, hires, or ads; only people, favors, and what you own" data-good="planning|strategy|feature" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M17 11 H 11 V 17"/><path d="M27 11 H 33 V 17"/><path d="M17 33 H 11 V 27"/><path d="M27 33 H 33 V 27"/></g><circle cx="22" cy="22" r="5" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="2.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="22" r="11" fill="none" stroke="currentColor" stroke-width="2"/><path d="M22 14 V30 M18 18 a4 3 0 0 1 8 0 a4 3 0 0 1 -8 4 a4 3 0 0 0 8 0" fill="none" stroke="currentColor" stroke-width="1.8" stroke-linecap="round"/><path d="M14 30 L30 14" stroke="currentColor" stroke-width="2.2" stroke-linecap="round"/></svg></span><span><span class="n">The $0 Mandate</span><span class="d">Achieve the goal spending literally nothing — no tools, hires, or ads; only people, favors, and what you own</span><span class="gf">Great for: Planning · Strategy · Build a feature</span></span></label><label class="tech"><input type="checkbox" data-name="One Feature Only" data-cat="Constraint" data-desc="You may keep exactly ONE capability and nothing else — pick it, then make that single thing unbelievably good" data-good="feature|strategy" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M17 11 H 11 V 17"/><path d="M27 11 H 33 V 17"/><path d="M17 33 H 11 V 27"/><path d="M27 33 H 33 V 27"/></g><circle cx="22" cy="22" r="5" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="2.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><circle cx="22" cy="22" r="4.5"/><path d="M22 9 V13 M22 31 V35 M9 22 H13 M31 22 H35" stroke-opacity=".35"/></g><circle cx="22" cy="22" r="2" fill="currentColor"/></svg></span><span><span class="n">One Feature Only</span><span class="d">You may keep exactly ONE capability and nothing else — pick it, then make that single thing unbelievably good</span><span class="gf">Great for: Build a feature · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Crank the Dial to 11" data-cat="Constraint" data-desc="Pick one dimension and exaggerate it to a ludicrous extreme — fastest, biggest, cheapest, weirdest — and see what breaks open" data-good="novel|unstuck" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M17 11 H 11 V 17"/><path d="M27 11 H 33 V 17"/><path d="M17 33 H 11 V 27"/><path d="M27 33 H 33 V 27"/></g><circle cx="22" cy="22" r="5" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="2.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M11 28 A12 12 0 0 1 33 28" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"/><path d="M22 28 L31 17" stroke="currentColor" stroke-width="2.2" stroke-linecap="round"/><circle cx="22" cy="28" r="2.6" fill="currentColor"/></svg></span><span><span class="n">Crank the Dial to 11</span><span class="d">Pick one dimension and exaggerate it to a ludicrous extreme — fastest, biggest, cheapest, weirdest — and see what breaks open</span><span class="gf">Great for: Novel concept · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Constraint Roulette" data-cat="Constraint" data-desc="Each round draw a brutal random limit (no screens, half the team, one day) and re-solve under it; survivors become real ideas" data-good="unstuck|feature" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M17 11 H 11 V 17"/><path d="M27 11 H 33 V 17"/><path d="M17 33 H 11 V 27"/><path d="M27 33 H 33 V 27"/></g><circle cx="22" cy="22" r="5" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="2.6" fill="currentColor"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="22" r="12" fill="none" stroke="currentColor" stroke-width="2"/><circle cx="22" cy="22" r="12" fill="none" stroke="currentColor" stroke-width="2" stroke-dasharray="3 3.7" stroke-opacity=".5"/><circle cx="22" cy="22" r="3" fill="currentColor"/><path d="M22 7 L25 12 H19 Z" fill="currentColor"/></svg></span><span><span class="n">Constraint Roulette</span><span class="d">Each round draw a brutal random limit (no screens, half the team, one day) and re-solve under it; survivors become real ideas</span><span class="gf">Great for: Get unstuck · Build a feature</span></span></label><label class="tech invent"><input type="checkbox" data-invent="Constraint"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round" fill="none"><path d="M17 11 H 11 V 17"/><path d="M27 11 H 33 V 17"/><path d="M17 33 H 11 V 27"/><path d="M27 33 H 33 V 27"/></g><circle cx="22" cy="22" r="5" fill="currentColor" fill-opacity="0.25"/><circle cx="22" cy="22" r="2.6" fill="currentColor"/></svg></span><span><span class="n">✨ Invent a Constraint technique</span><span class="d">Make up a brand-new technique on the fly, in the spirit of Constraint</span></span></label></div></section>
+<h2 class="grouphdr" style="--c:#b15ad6">Introspective &amp; Personal</h2>
+<section data-cat="Introspective Delight" style="--c:#b15ad6"><h2>Introspective Delight<span class="cnt">8</span></h2><div class="grid"><label class="tech"><input type="checkbox" data-name="Inner Child Conference" data-cat="Introspective Delight" data-desc="Answer as your 7-year-old self: ask naive &#x27;why why why&#x27; questions, chase wonder, ban every boring adult thought" data-good="personal|unstuck" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="13.5" r="4" fill="currentColor"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31 Z" fill="currentColor" fill-opacity="0.22"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31" stroke="currentColor" stroke-width="1.7" fill="none"/><path d="M13.5 30 C 16 26.5, 20 25.5, 22 25.5 C 24 25.5, 28 26.5, 30.5 30" stroke="currentColor" stroke-width="1.5" fill="none" stroke-opacity="0.6"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="16" r="6" fill="none" stroke="currentColor" stroke-width="2"/><path d="M22 22 V31" stroke="currentColor" stroke-width="2" stroke-linecap="round"/><path d="M19 34 q3 -3 6 0" fill="none" stroke="currentColor" stroke-width="1.8" stroke-linecap="round"/><g fill="currentColor"><circle cx="20" cy="15" r="1"/><circle cx="24" cy="15" r="1"/></g></svg></span><span><span class="n">Inner Child Conference</span><span class="d">Answer as your 7-year-old self: ask naive &#x27;why why why&#x27; questions, chase wonder, ban every boring adult thought</span><span class="gf">Great for: Personal / life · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Shadow Work Mining" data-cat="Introspective Delight" data-desc="Name what you&#x27;re avoiding, resisting, or scared of about this — then dig there for the buried insight" data-good="personal|diagnosis" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="13.5" r="4" fill="currentColor"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31 Z" fill="currentColor" fill-opacity="0.22"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31" stroke="currentColor" stroke-width="1.7" fill="none"/><path d="M13.5 30 C 16 26.5, 20 25.5, 22 25.5 C 24 25.5, 28 26.5, 30.5 30" stroke="currentColor" stroke-width="1.5" fill="none" stroke-opacity="0.6"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="22" r="12" fill="none" stroke="currentColor" stroke-width="2"/><path d="M22 10 a12 12 0 0 1 0 24 z" fill="currentColor" fill-opacity=".85"/></svg></span><span><span class="n">Shadow Work Mining</span><span class="d">Name what you&#x27;re avoiding, resisting, or scared of about this — then dig there for the buried insight</span><span class="gf">Great for: Personal / life · Diagnose</span></span></label><label class="tech"><input type="checkbox" data-name="Values Archaeology" data-cat="Introspective Delight" data-desc="Keep asking &#x27;why do I care?&#x27; until you hit bedrock: the non-negotiable value secretly steering the choice" data-good="personal|strategy" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="13.5" r="4" fill="currentColor"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31 Z" fill="currentColor" fill-opacity="0.22"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31" stroke="currentColor" stroke-width="1.7" fill="none"/><path d="M13.5 30 C 16 26.5, 20 25.5, 22 25.5 C 24 25.5, 28 26.5, 30.5 30" stroke="currentColor" stroke-width="1.5" fill="none" stroke-opacity="0.6"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2"><path d="M10 16 h24" stroke-opacity=".4"/><path d="M10 22 h24" stroke-opacity=".6"/></g><path d="M22 24 L16 30 L22 36 L28 30 Z" fill="currentColor"/></svg></span><span><span class="n">Values Archaeology</span><span class="d">Keep asking &#x27;why do I care?&#x27; until you hit bedrock: the non-negotiable value secretly steering the choice</span><span class="gf">Great for: Personal / life · Strategy</span></span></label><label class="tech"><input type="checkbox" data-name="Future Self Interview" data-cat="Introspective Delight" data-desc="Interview your wise 80-year-old self about this problem and write down the advice they give you" data-good="personal" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="13.5" r="4" fill="currentColor"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31 Z" fill="currentColor" fill-opacity="0.22"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31" stroke="currentColor" stroke-width="1.7" fill="none"/><path d="M13.5 30 C 16 26.5, 20 25.5, 22 25.5 C 24 25.5, 28 26.5, 30.5 30" stroke="currentColor" stroke-width="1.5" fill="none" stroke-opacity="0.6"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M14 10 H30 L24 22 L30 34 H14 L20 22 Z"/></g><path d="M18 14 H26" stroke="currentColor" stroke-width="2" stroke-linecap="round"/></svg></span><span><span class="n">Future Self Interview</span><span class="d">Interview your wise 80-year-old self about this problem and write down the advice they give you</span><span class="gf">Great for: Personal / life</span></span></label><label class="tech"><input type="checkbox" data-name="Body Wisdom Dialogue" data-cat="Introspective Delight" data-desc="Scan for the tension, flutter, or gut pull each option triggers; let the body&#x27;s yes/no drive the ideas" data-good="personal" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="13.5" r="4" fill="currentColor"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31 Z" fill="currentColor" fill-opacity="0.22"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31" stroke="currentColor" stroke-width="1.7" fill="none"/><path d="M13.5 30 C 16 26.5, 20 25.5, 22 25.5 C 24 25.5, 28 26.5, 30.5 30" stroke="currentColor" stroke-width="1.5" fill="none" stroke-opacity="0.6"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M22 33 C12 26 9 19 13.5 15 C17 12 21 14 22 17 C23 14 27 12 30.5 15 C35 19 32 26 22 33 Z" fill="currentColor" fill-opacity=".22"/><path d="M22 33 C12 26 9 19 13.5 15 C17 12 21 14 22 17 C23 14 27 12 30.5 15 C35 19 32 26 22 33 Z" fill="none" stroke="currentColor" stroke-width="2"/></svg></span><span><span class="n">Body Wisdom Dialogue</span><span class="d">Scan for the tension, flutter, or gut pull each option triggers; let the body&#x27;s yes/no drive the ideas</span><span class="gf">Great for: Personal / life</span></span></label><label class="tech"><input type="checkbox" data-name="Permission Giving" data-cat="Introspective Delight" data-desc="Write yourself an explicit permission slip to think the forbidden, impossible thought — then think it out loud" data-good="personal|unstuck" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="13.5" r="4" fill="currentColor"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31 Z" fill="currentColor" fill-opacity="0.22"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31" stroke="currentColor" stroke-width="1.7" fill="none"/><path d="M13.5 30 C 16 26.5, 20 25.5, 22 25.5 C 24 25.5, 28 26.5, 30.5 30" stroke="currentColor" stroke-width="1.5" fill="none" stroke-opacity="0.6"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><rect x="10" y="14" width="24" height="16" rx="2.5" fill="none" stroke="currentColor" stroke-width="2"/><path d="M15 23 L19 27 L28 17" fill="none" stroke="currentColor" stroke-width="2.2" stroke-linecap="round" stroke-linejoin="round"/></svg></span><span><span class="n">Permission Giving</span><span class="d">Write yourself an explicit permission slip to think the forbidden, impossible thought — then think it out loud</span><span class="gf">Great for: Personal / life · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Secret Wish Confession" data-cat="Introspective Delight" data-desc="Whisper the embarrassing thing you secretly want here but won&#x27;t admit, then build the idea honoring it" data-good="personal" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="13.5" r="4" fill="currentColor"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31 Z" fill="currentColor" fill-opacity="0.22"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31" stroke="currentColor" stroke-width="1.7" fill="none"/><path d="M13.5 30 C 16 26.5, 20 25.5, 22 25.5 C 24 25.5, 28 26.5, 30.5 30" stroke="currentColor" stroke-width="1.5" fill="none" stroke-opacity="0.6"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><rect x="13" y="20" width="18" height="14" rx="2.5" fill="currentColor" fill-opacity=".22"/><rect x="13" y="20" width="18" height="14" rx="2.5" fill="none" stroke="currentColor" stroke-width="2"/><path d="M16.5 20 v-3 a5.5 5.5 0 0 1 11 0 v3" fill="none" stroke="currentColor" stroke-width="2"/></svg></span><span><span class="n">Secret Wish Confession</span><span class="d">Whisper the embarrassing thing you secretly want here but won&#x27;t admit, then build the idea honoring it</span><span class="gf">Great for: Personal / life</span></span></label><label class="tech"><input type="checkbox" data-name="Mood Weather Report" data-cat="Introspective Delight" data-desc="Name the inner weather right now (fog, storm, sun) and let that exact emotional climate generate the ideas" data-good="personal|unstuck" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="13.5" r="4" fill="currentColor"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31 Z" fill="currentColor" fill-opacity="0.22"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31" stroke="currentColor" stroke-width="1.7" fill="none"/><path d="M13.5 30 C 16 26.5, 20 25.5, 22 25.5 C 24 25.5, 28 26.5, 30.5 30" stroke="currentColor" stroke-width="1.5" fill="none" stroke-opacity="0.6"/></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="17" cy="17" r="4.5" fill="currentColor" fill-opacity=".5"/><path d="M22 30 a5 5 0 0 1 0.5 -10 a6 6 0 0 1 11 2.5 a4 4 0 0 1 -1.5 7.5 z" fill="none" stroke="currentColor" stroke-width="2" stroke-linejoin="round"/></svg></span><span><span class="n">Mood Weather Report</span><span class="d">Name the inner weather right now (fog, storm, sun) and let that exact emotional climate generate the ideas</span><span class="gf">Great for: Personal / life · Get unstuck</span></span></label><label class="tech invent"><input type="checkbox" data-invent="Introspective Delight"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><circle cx="22" cy="13.5" r="4" fill="currentColor"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31 Z" fill="currentColor" fill-opacity="0.22"/><path d="M10.5 31 C 12.5 23, 31.5 23, 33.5 31" stroke="currentColor" stroke-width="1.7" fill="none"/><path d="M13.5 30 C 16 26.5, 20 25.5, 22 25.5 C 24 25.5, 28 26.5, 30.5 30" stroke="currentColor" stroke-width="1.5" fill="none" stroke-opacity="0.6"/></svg></span><span><span class="n">✨ Invent a Introspective Delight technique</span><span class="d">Make up a brand-new technique on the fly, in the spirit of Introspective Delight</span></span></label></div></section>
+<section data-cat="Collaborative" style="--c:#15a3a3"><h2>Collaborative<span class="cnt">8</span></h2><div class="grid"><label class="tech"><input type="checkbox" data-name="Yes And Building" data-cat="Collaborative" data-desc="Never negate; each person opens with &quot;Yes, and...&quot; and adds to the last idea, stacking a chain of accepted additions" data-good="novel|unstuck|planning" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8"><line x1="14" y1="16" x2="30" y2="16"/><line x1="14" y1="16" x2="22" y2="30"/><line x1="30" y1="16" x2="22" y2="30"/></g><g fill="currentColor" fill-opacity="0.22"><circle cx="14" cy="16" r="4.6"/><circle cx="30" cy="16" r="4.6"/><circle cx="22" cy="30" r="4.6"/></g><g fill="currentColor"><circle cx="14" cy="16" r="2.4"/><circle cx="30" cy="16" r="2.4"/><circle cx="22" cy="30" r="2.4"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="currentColor"><rect x="8" y="27" width="12" height="8" rx="1.5" fill-opacity=".8"/><rect x="14" y="19" width="12" height="8" rx="1.5" fill-opacity=".5"/><rect x="20" y="11" width="12" height="8" rx="1.5"/></g></svg></span><span><span class="n">Yes And Building</span><span class="d">Never negate; each person opens with &quot;Yes, and...&quot; and adds to the last idea, stacking a chain of accepted additions</span><span class="gf">Great for: Novel concept · Get unstuck · Planning</span></span></label><label class="tech"><input type="checkbox" data-name="Brain Writing Round Robin" data-cat="Collaborative" data-desc="Everyone writes ideas silently, then passes their sheet; you build on whatever lands in front of you, round after round" data-good="novel|feature" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8"><line x1="14" y1="16" x2="30" y2="16"/><line x1="14" y1="16" x2="22" y2="30"/><line x1="30" y1="16" x2="22" y2="30"/></g><g fill="currentColor" fill-opacity="0.22"><circle cx="14" cy="16" r="4.6"/><circle cx="30" cy="16" r="4.6"/><circle cx="22" cy="30" r="4.6"/></g><g fill="currentColor"><circle cx="14" cy="16" r="2.4"/><circle cx="30" cy="16" r="2.4"/><circle cx="22" cy="30" r="2.4"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M31 16 A10 10 0 1 0 32.5 22"/><path d="M31 10 L31.5 16.3 L25 16.5"/></g></svg></span><span><span class="n">Brain Writing Round Robin</span><span class="d">Everyone writes ideas silently, then passes their sheet; you build on whatever lands in front of you, round after round</span><span class="gf">Great for: Novel concept · Build a feature</span></span></label><label class="tech"><input type="checkbox" data-name="Random Stimulation" data-cat="Collaborative" data-desc="Pull a random word or image and force a link to the problem: &quot;how does THIS spark a solution?&quot;" data-good="unstuck|novel" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8"><line x1="14" y1="16" x2="30" y2="16"/><line x1="14" y1="16" x2="22" y2="30"/><line x1="30" y1="16" x2="22" y2="30"/></g><g fill="currentColor" fill-opacity="0.22"><circle cx="14" cy="16" r="4.6"/><circle cx="30" cy="16" r="4.6"/><circle cx="22" cy="30" r="4.6"/></g><g fill="currentColor"><circle cx="14" cy="16" r="2.4"/><circle cx="30" cy="16" r="2.4"/><circle cx="22" cy="30" r="2.4"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><rect x="11" y="11" width="22" height="22" rx="4" fill="none" stroke="currentColor" stroke-width="2"/><g fill="currentColor"><circle cx="17" cy="17" r="1.8"/><circle cx="27" cy="17" r="1.8"/><circle cx="22" cy="22" r="1.8"/><circle cx="17" cy="27" r="1.8"/><circle cx="27" cy="27" r="1.8"/></g></svg></span><span><span class="n">Random Stimulation</span><span class="d">Pull a random word or image and force a link to the problem: &quot;how does THIS spark a solution?&quot;</span><span class="gf">Great for: Get unstuck · Novel concept</span></span></label><label class="tech"><input type="checkbox" data-name="Role Playing" data-cat="Collaborative" data-desc="Each person speaks as a different stakeholder, voicing what that role wants, fears, and would demand of the idea" data-good="strategy|personal|feature" data-prov="classic"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8"><line x1="14" y1="16" x2="30" y2="16"/><line x1="14" y1="16" x2="22" y2="30"/><line x1="30" y1="16" x2="22" y2="30"/></g><g fill="currentColor" fill-opacity="0.22"><circle cx="14" cy="16" r="4.6"/><circle cx="30" cy="16" r="4.6"/><circle cx="22" cy="30" r="4.6"/></g><g fill="currentColor"><circle cx="14" cy="16" r="2.4"/><circle cx="30" cy="16" r="2.4"/><circle cx="22" cy="30" r="2.4"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><rect x="11" y="9" width="22" height="26" rx="3" fill="none" stroke="currentColor" stroke-width="2"/><circle cx="22" cy="19" r="4" fill="currentColor"/><path d="M15.5 30 c2 -4.5 11 -4.5 13 0" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round"/></svg></span><span><span class="n">Role Playing</span><span class="d">Each person speaks as a different stakeholder, voicing what that role wants, fears, and would demand of the idea</span><span class="gf">Great for: Strategy · Personal / life · Build a feature</span></span></label><label class="tech"><input type="checkbox" data-name="Ideation Relay Race" data-cat="Collaborative" data-desc="30-second turns, no pausing: add one idea, slap it to the next person, keep the baton moving before anyone overthinks" data-good="unstuck" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8"><line x1="14" y1="16" x2="30" y2="16"/><line x1="14" y1="16" x2="22" y2="30"/><line x1="30" y1="16" x2="22" y2="30"/></g><g fill="currentColor" fill-opacity="0.22"><circle cx="14" cy="16" r="4.6"/><circle cx="30" cy="16" r="4.6"/><circle cx="22" cy="30" r="4.6"/></g><g fill="currentColor"><circle cx="14" cy="16" r="2.4"/><circle cx="30" cy="16" r="2.4"/><circle cx="22" cy="30" r="2.4"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><line x1="12" y1="31" x2="27" y2="16"/><line x1="8" y1="22" x2="14" y2="22" stroke-opacity=".5"/><line x1="8" y1="27" x2="13" y2="27" stroke-opacity=".35"/></g><circle cx="29" cy="14" r="3.4" fill="currentColor"/></svg></span><span><span class="n">Ideation Relay Race</span><span class="d">30-second turns, no pausing: add one idea, slap it to the next person, keep the baton moving before anyone overthinks</span><span class="gf">Great for: Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Idea Hot Potato" data-cat="Collaborative" data-desc="One idea gets tossed around the circle; each catcher must mutate it in 10 seconds before passing, no repeats allowed" data-good="unstuck" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8"><line x1="14" y1="16" x2="30" y2="16"/><line x1="14" y1="16" x2="22" y2="30"/><line x1="30" y1="16" x2="22" y2="30"/></g><g fill="currentColor" fill-opacity="0.22"><circle cx="14" cy="16" r="4.6"/><circle cx="30" cy="16" r="4.6"/><circle cx="22" cy="30" r="4.6"/></g><g fill="currentColor"><circle cx="14" cy="16" r="2.4"/><circle cx="30" cy="16" r="2.4"/><circle cx="22" cy="30" r="2.4"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M11 31 Q22 8 33 31" fill="none" stroke="currentColor" stroke-width="2" stroke-dasharray="2 3.5" stroke-linecap="round"/><circle cx="22" cy="12.5" r="4.2" fill="currentColor"/></svg></span><span><span class="n">Idea Hot Potato</span><span class="d">One idea gets tossed around the circle; each catcher must mutate it in 10 seconds before passing, no repeats allowed</span><span class="gf">Great for: Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Steal And Upgrade" data-cat="Collaborative" data-desc="Pick a neighbor&#x27;s idea you envy, claim it out loud, then make it visibly better before handing it back improved" data-good="novel|unstuck" data-prov="signature"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8"><line x1="14" y1="16" x2="30" y2="16"/><line x1="14" y1="16" x2="22" y2="30"/><line x1="30" y1="16" x2="22" y2="30"/></g><g fill="currentColor" fill-opacity="0.22"><circle cx="14" cy="16" r="4.6"/><circle cx="30" cy="16" r="4.6"/><circle cx="22" cy="30" r="4.6"/></g><g fill="currentColor"><circle cx="14" cy="16" r="2.4"/><circle cx="30" cy="16" r="2.4"/><circle cx="22" cy="30" r="2.4"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round"><path d="M20 33 V14"/><path d="M13 21 L20 14 L27 21"/></g><path d="M30 27 l1 2.6 2.6 1 -2.6 1 -1 2.6 -1 -2.6 -2.6 -1 2.6 -1 z" fill="currentColor"/></svg></span><span><span class="n">Steal And Upgrade</span><span class="d">Pick a neighbor&#x27;s idea you envy, claim it out loud, then make it visibly better before handing it back improved</span><span class="gf">Great for: Novel concept · Get unstuck</span></span></label><label class="tech"><input type="checkbox" data-name="Fold The Paper" data-cat="Collaborative" data-desc="Each person adds one line to a hidden drawing or sentence, sees only the previous fragment, then unfold the surreal whole" data-good="unstuck|novel" data-prov="playful"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8"><line x1="14" y1="16" x2="30" y2="16"/><line x1="14" y1="16" x2="22" y2="30"/><line x1="30" y1="16" x2="22" y2="30"/></g><g fill="currentColor" fill-opacity="0.22"><circle cx="14" cy="16" r="4.6"/><circle cx="30" cy="16" r="4.6"/><circle cx="22" cy="30" r="4.6"/></g><g fill="currentColor"><circle cx="14" cy="16" r="2.4"/><circle cx="30" cy="16" r="2.4"/><circle cx="22" cy="30" r="2.4"/></g></svg><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><path d="M13 16 L21 12 V28 L13 32 Z" fill="currentColor" fill-opacity=".22"/><path d="M21 12 L29 16 V32 L21 28 Z" fill="currentColor" fill-opacity=".45"/><path d="M13 16 L21 12 L29 16 M21 12 V28" fill="none" stroke="currentColor" stroke-width="1.5" stroke-linejoin="round"/></svg></span><span><span class="n">Fold The Paper</span><span class="d">Each person adds one line to a hidden drawing or sentence, sees only the previous fragment, then unfold the surreal whole</span><span class="gf">Great for: Get unstuck · Novel concept</span></span></label><label class="tech invent"><input type="checkbox" data-invent="Collaborative"><span class="ic2"><svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg"><rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/><g stroke="currentColor" stroke-width="1.8"><line x1="14" y1="16" x2="30" y2="16"/><line x1="14" y1="16" x2="22" y2="30"/><line x1="30" y1="16" x2="22" y2="30"/></g><g fill="currentColor" fill-opacity="0.22"><circle cx="14" cy="16" r="4.6"/><circle cx="30" cy="16" r="4.6"/><circle cx="22" cy="30" r="4.6"/></g><g fill="currentColor"><circle cx="14" cy="16" r="2.4"/><circle cx="30" cy="16" r="2.4"/><circle cx="22" cy="30" r="2.4"/></g></svg></span><span><span class="n">✨ Invent a Collaborative technique</span><span class="d">Make up a brand-new technique on the fly, in the spirit of Collaborative</span></span></label></div></section>
+</main>
+<footer>BMad Method &middot; Brainstorming</footer>
+<script>
+(function(){
+  var $ = function(id){ return document.getElementById(id); };
+  var all = Array.prototype.slice;
+  var boxes = all.call(document.querySelectorAll('input[type=checkbox]'));
+  var techBoxes = boxes.filter(function(b){ return b.dataset.name; });      // real technique cards
+  var inventBoxes = boxes.filter(function(b){ return b.dataset.invent; });  // per-category "invent in the spirit of" cards
+  var header = document.querySelector('header');
+  var sections = all.call(document.querySelectorAll('section'));
+  var state = { mode: 'Facilitator', rand: 0, inv: 0, ai: 0 };
+  var MODE_HINTS = {
+    'Facilitator': 'A forcing function for your ideas — I prompt and push, but never supply them.',
+    'Creative Partner': 'We riff together — I facilitate and add ideas too, each logged as yours or mine.',
+    'Ideate for me': 'I run the whole session myself, then show you the result and offer to keep going.'
+  };
+  function setHint(){ $('modehint').textContent = MODE_HINTS[state.mode] || ''; }
+
+  var themeBtn = $('theme');
+  function setThemeIcon(){ themeBtn.textContent = document.documentElement.getAttribute('data-theme') === 'dark' ? '☀' : '☾'; }
+  themeBtn.addEventListener('click', function(){
+    var next = document.documentElement.getAttribute('data-theme') === 'dark' ? 'light' : 'dark';
+    document.documentElement.setAttribute('data-theme', next);
+    try { localStorage.setItem('bmad-theme', next); } catch(e){}
+    setThemeIcon();
+  });
+
+  all.call(document.querySelectorAll('.mode')).forEach(function(b){
+    b.addEventListener('click', function(){
+      all.call(document.querySelectorAll('.mode')).forEach(function(m){ m.classList.remove('on'); });
+      b.classList.add('on');
+      state.mode = b.dataset.mode;
+      setHint();
+    });
+  });
+
+  all.call(document.querySelectorAll('[data-step]')).forEach(function(btn){
+    btn.addEventListener('click', function(){
+      var k = btn.dataset.step, d = parseInt(btn.dataset.d, 10);
+      state[k] = Math.max(0, state[k] + d);
+      update();
+    });
+  });
+
+  // Category chips are jump-nav: click one to smooth-scroll its section into view,
+  // offsetting by the sticky header's height so the heading isn't hidden beneath it.
+  all.call(document.querySelectorAll('.chip')).forEach(function(chip){
+    chip.addEventListener('click', function(){
+      var sec = null;
+      for (var i = 0; i < sections.length; i++){ if (sections[i].dataset.cat === chip.dataset.cat){ sec = sections[i]; break; } }
+      if (!sec){ return; }
+      var top = sec.getBoundingClientRect().top + window.pageYOffset - header.offsetHeight - 8;
+      window.scrollTo({ top: top, behavior: 'smooth' });
+    });
+  });
+
+  boxes.forEach(function(b){ b.addEventListener('change', update); });
+
+  // A `classic` technique appears twice (lead "Proven & Professional" group + its home
+  // category), so de-dupe checked picks by name; the lead copy carries data-lead.
+  function checkedTech(){
+    var seen = {}, out = [];
+    techBoxes.forEach(function(b){
+      if (!b.checked || seen[b.dataset.name]) { return; }
+      seen[b.dataset.name] = 1;
+      out.push(b);
+    });
+    return out;
+  }
+  function checkedInvent(){ return inventBoxes.filter(function(b){ return b.checked; }); }
+
+  function update(){
+    // rand can't exceed what the pool can supply — keep the counter honest with the draw
+    if (state.rand > randomPool().length){ state.rand = randomPool().length; }
+    $('pickN').textContent = checkedTech().length;
+    $('randN').textContent = state.rand;
+    $('invN').textContent = state.inv;
+    $('aiN').textContent = state.ai;
+    var total = checkedTech().length + state.rand + state.inv + checkedInvent().length + state.ai;
+    var t = $('total');
+    t.textContent = 'Total ' + total + ' · 3–4 is the sweet spot';
+    t.classList.toggle('warn', total > 5);
+  }
+
+  // "Great for" goal filter: clicking a goal narrows visible cards to those tagged with it.
+  var goalBtns = all.call(document.querySelectorAll('.goal'));
+  function activeGoals(){ return goalBtns.filter(function(b){ return b.classList.contains('on'); }).map(function(b){ return b.dataset.goal; }); }
+  function applyFilter(){
+    var act = activeGoals();
+    all.call(document.querySelectorAll('label.tech')).forEach(function(lab){
+      var inp = lab.querySelector('input');
+      if (inp.dataset.invent){ return; }  // invent cards aren't goal-tagged — always visible
+      var good = (inp.dataset.good || '').split('|');
+      var show = !act.length || act.some(function(g){ return good.indexOf(g) >= 0; });
+      lab.classList.toggle('hidden', !show);
+    });
+  }
+  goalBtns.forEach(function(b){ b.addEventListener('click', function(){ b.classList.toggle('on'); applyFilter(); }); });
+
+  function randomPool(){
+    var picked = {};
+    checkedTech().forEach(function(b){ picked[b.dataset.name] = 1; });
+    // draw from unchecked, non-lead copies, skipping anything already picked
+    return techBoxes.filter(function(b){ return !b.checked && !b.dataset.lead && !picked[b.dataset.name]; });
+  }
+
+  function sample(arr, n){
+    var a = arr.slice(), out = [];
+    while (out.length < n && a.length){ out.push(a.splice(Math.floor(Math.random() * a.length), 1)[0]); }
+    return out;
+  }
+
+  function compose(){
+    var picks = checkedTech().map(function(b){ return { n: b.dataset.name, c: b.dataset.cat, d: b.dataset.desc, r: false }; });
+    var rnd = sample(randomPool(), state.rand).map(function(b){ return { n: b.dataset.name, c: b.dataset.cat, d: b.dataset.desc, r: true }; });
+    var techs = picks.concat(rnd);
+    var L = ["Let's run my brainstorming session.", "", 'Facilitation mode: ' + state.mode + '.'];
+    if (techs.length){
+      L.push("", 'Techniques to use:');
+      techs.forEach(function(t, i){
+        L.push((i + 1) + '.' + (t.r ? ' (random pick)' : '') + ' ' + t.n + '  ·  ' + t.c);
+        L.push('   ' + t.d);
+      });
+    }
+    var extra = [];
+    if (state.inv > 0){ extra.push('invent ' + state.inv + ' brand-new technique' + (state.inv > 1 ? 's' : '') + ' on the fly'); }
+    checkedInvent().forEach(function(b){ extra.push('invent 1 new technique in the spirit of ' + b.dataset.invent); });
+    if (state.ai > 0){ extra.push('you choose ' + state.ai + ' more technique' + (state.ai > 1 ? 's' : '') + ' that fit my goal'); }
+    if (extra.length){ L.push("", 'Then: ' + extra.join('; and ') + '.'); }
+    if (!techs.length && !extra.length){
+      L.push("", state.mode === 'Ideate for me'
+        ? 'Run the whole session yourself — pick the techniques, generate the ideas, then show me the result.'
+        : 'Help me choose 3–4 techniques to start.');
+    }
+    return L.join('\n');
+  }
+
+  function fallbackCopy(t){
+    var ta = document.createElement('textarea');
+    ta.value = t; ta.style.position = 'fixed'; ta.style.opacity = '0';
+    document.body.appendChild(ta); ta.focus(); ta.select();
+    var ok = false;
+    try { ok = document.execCommand('copy'); } catch(e){ ok = false; }
+    document.body.removeChild(ta);
+    return ok;
+  }
+
+  function flash(ok, text){
+    var b = $('banner');
+    b.classList.toggle('fail', !ok);
+    b.innerHTML = ok
+      ? '✓ Copied! Now paste it into the chat to start your session.'
+      : '⚠ Couldn’t reach the clipboard — copy the text in the box, then paste it into the chat.';
+    b.classList.add('show');
+    setTimeout(function(){ b.classList.remove('show'); }, 4500);
+    // Last resort on a hard failure: a prefilled, selectable prompt so the text is never lost.
+    if (!ok){ window.prompt('Copy this, then paste it into the chat:', text); }
+  }
+
+  $('copy').addEventListener('click', function(){
+    var text = compose();
+    if (navigator.clipboard && navigator.clipboard.writeText){
+      navigator.clipboard.writeText(text).then(
+        function(){ flash(true, text); },
+        function(){ flash(fallbackCopy(text), text); }
+      );
+    } else { flash(fallbackCopy(text), text); }
+  });
+
+  setHint();
+  setThemeIcon();
+  update();
+})();
+</script>
+</body>
+</html>
diff --git a/.agents/skills/bmad-brainstorming/customize.toml b/.agents/skills/bmad-brainstorming/customize.toml
new file mode 100644
index 00000000..a68c3421
--- /dev/null
+++ b/.agents/skills/bmad-brainstorming/customize.toml
@@ -0,0 +1,84 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-brainstorming.
+#
+# Override files (not edited here):
+#   {project-root}/_bmad/custom/bmad-brainstorming.toml         (team)
+#   {project-root}/_bmad/custom/bmad-brainstorming.user.toml    (personal)
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays: append
+
+# Steps to run before the standard activation (config load, greet).
+# Use for pre-flight loads, compliance checks, etc.
+activation_steps_prepend = []
+
+# Steps to run after greet but before facilitation begins.
+# Use for context-heavy setup that should happen once the user has been acknowledged.
+activation_steps_append = []
+
+# Persistent facts the facilitator keeps in mind for the whole session
+# (domain constraints, house rules, stylistic guardrails). Each entry is a
+# literal sentence, a skill prefixed with `skill:`, or a `file:`-prefixed
+# path/glob whose contents are loaded as facts. Default loads project-context.md
+# if bmad-generate-project-context has produced one, giving the facilitator
+# persistent awareness of the project's domain without re-asking.
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# The technique library loaded on demand during the session. Swap the path in
+# team/user TOML to ship a different or extended catalog of creative methods.
+# Kept `{skill-root}`-anchored so it resolves regardless of the working directory
+# (brain.py is always invoked with `--file {workflow.brain_methods}`).
+brain_methods = "{skill-root}/assets/brain-methods.csv"
+
+# Techniques the facilitator should reach for first. When proposing a method
+# (the AI-led default), it prefers these where they fit the goal before ranging
+# wider. Names should match an entry in the library or in additional_techniques.
+# Append-merges, so a team list and a personal list both contribute. Empty = no
+# preference; the facilitator chooses purely on fit.
+#
+# Example (set in team/user override TOML):
+#   favorite_techniques = ["SCAMPER", "Six Thinking Hats", "First Principles"]
+favorite_techniques = []
+
+# Extra techniques — and whole new categories — merged into the catalog the
+# facilitator chooses from, without editing the shipped CSV. Each entry mirrors
+# the library's shape (category, technique_name, description); a new category is
+# just a category value the CSV doesn't have. Entries append, so teams and users
+# can each grow the library. The facilitator treats these as first-class
+# alongside brain_methods across every flow — facilitator-chosen, browse,
+# category draws, and inventive.
+#
+# Example (set in team/user override TOML):
+#   [[workflow.additional_techniques]]
+#   category = "domain-specific"
+#   technique_name = "Regulatory Inversion"
+#   description = "Start from the compliance constraint and brainstorm what becomes possible only because of it — turn the rule into a generative frame rather than a limit."
+additional_techniques = []
+
+# Session output location. The running log and any final artifacts land inside
+# `{output_dir}/{output_folder_name}/`. `{topic_slug}` is filled from the session
+# topic so each topic gets its own folder — a user can brainstorm several topics
+# without collision. The resume check globs `{output_dir}/*/.memlog.md`.
+output_dir = "{output_folder}/brainstorming"
+output_folder_name = "brainstorm-{topic_slug}-{date}"
+
+# Executed when the session completes (after artifacts are produced and the user
+# has the paths). Accepts a string scalar (single instruction) or an array of
+# instructions executed in order. Empty for none.
+on_complete = ""
+
+# External-handoff routing. Natural-language directives applied after artifacts
+# are produced, to route them beyond local files (Confluence, Notion, Drive,
+# etc.). Each entry names the MCP tool, the destination, and the fields it needs.
+# URLs/IDs returned are surfaced to the user. If a named tool is unavailable at
+# runtime, the handoff is skipped and flagged; local files always exist. Empty
+# by default.
+#
+# Example (set in team/user override TOML):
+#   "After artifacts are produced, upload brainstorm.html to Confluence via corp:confluence_upload (space_key='IDEAS', parent_page='Brainstorms', author={user_name})."
+external_handoffs = []
diff --git a/.agents/skills/bmad-brainstorming/references/converge.md b/.agents/skills/bmad-brainstorming/references/converge.md
new file mode 100644
index 00000000..ac1786e6
--- /dev/null
+++ b/.agents/skills/bmad-brainstorming/references/converge.md
@@ -0,0 +1,24 @@
+# Converging: Narrow & Decide
+
+Load this when divergence is spent and the user wants to narrow the field — or asks to "decide," "prioritize," "pick," or "make it real." The whole catalog is *divergent* by design (it generates); this is the deliberate opposite phase, and keeping the two apart is the point. Never run convergence while ideas are still flowing, and never let it leak into a generating batch — premature judgment is what kills good ideas. `{doc_workspace}/.memlog.md` is the canonical record; everything here works from it. Communicate in `{communication_language}`.
+
+**Mode holds.** In **Facilitator** you run the convergence *on the user's verdicts* — you structure and prompt, they judge; never rank for them. In **Creative Partner** you weigh in too, each call logged by author. In **Ideate for me** you converge yourself and show the result, then offer to keep going.
+
+## How to run it
+
+First, reflect the field back: pull the live candidates from the memlog (include the odd and buried ones, not just the recent obvious ones) so there's a concrete set to work on. Then pick **one** convergence move that fits the goal — don't hand the user a menu of methods; choose the one that suits *this* decision and name it. Run it to a result, log the outcome, and stop when a clear short-list or single direction emerges.
+
+Pick by what the decision needs:
+
+- **Affinity Clustering** — when there are many scattered ideas: group them into themes, name each cluster, and surface the through-line. Often the right *first* move, to turn a pile into a handful.
+- **Impact–Effort** — when the goal is action: place each candidate on impact vs effort; harvest high-impact / low-effort first, park the rest.
+- **NUF Test** — when novelty matters: score each New, Useful, Feasible (1–10 each); the totals expose the quiet winners and the dazzling-but-doomed.
+- **Forced Ranking / Dot Vote** — when you just need a ranked top-N: make the ideas compete, no ties; (a literal dot-vote when it's genuinely a group).
+- **PMI (Plus / Minus / Interesting)** — when one strong candidate needs pressure-testing before commitment: list its pluses, minuses, and the merely-interesting, then judge.
+- **MoSCoW** — when scoping a build: sort into Must / Should / Could / Won't-this-time.
+
+Log the surviving directions and the reasoning with `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {doc_workspace} --type decision --text "<one-line gist>"` (use `--by` in Creative Partner mode). Two or three convergence moves chained is fine (e.g. cluster → score the clusters); more than that is usually over-processing.
+
+## Then finalize
+
+Once a short-list or direction is settled, **load `references/finalize.md`** and run it last — synthesis, `status: complete`, and artifacts build on the decisions you just logged. Convergence narrows; finalize captures and ships. Do not set `status: complete` here — that belongs to finalize.
diff --git a/.agents/skills/bmad-brainstorming/references/finalize.md b/.agents/skills/bmad-brainstorming/references/finalize.md
new file mode 100644
index 00000000..2a4b6d25
--- /dev/null
+++ b/.agents/skills/bmad-brainstorming/references/finalize.md
@@ -0,0 +1,26 @@
+# Wrap-Up: Synthesis & Artifacts
+
+Load this when the user signals they're spent or the topic is mined out. `{doc_workspace}/.memlog.md` is the canonical record of the session — everything here derives from it. Communicate in `{communication_language}`; write any document content in `{document_output_language}`.
+
+## Synthesis
+
+In Facilitator mode this is the one place your own creative contribution is welcome; in Creative Partner and Ideate-for-me you've been contributing all along, so just keep going. Run it in two moves, in order:
+
+1. **Hand them the mirror first.** Reflect a vivid sampling of *their* ideas back — deliberately include the odd, random, or buried ones from earlier, not just the recent obvious ones (in Creative Partner mode the `(... by user)` tags tell you which were theirs). Ask what they see now: conclusions, synergies, themes, the few that actually matter. Let them connect first; their own pattern-recognition is the point.
+2. **Then add the connections they would miss.** Lean in creatively — not new raw ideas, but the non-obvious links: this idea from technique one quietly solves that tension from technique four; these three are one idea wearing three hats; this wildcard is the real breakthrough.
+
+Record the insights and chosen directions with `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {doc_workspace} --type insight --text "<insights + chosen directions>"`. **Then run `uv run {project-root}/_bmad/scripts/memlog.py set --workspace {doc_workspace} --key status --value complete`** — the session is done and must stop being offered for resume. Do this even if the user declines every artifact below.
+
+## Artifacts
+
+In **Ideate for me** (and headless), the imaginative HTML keepsake is the deliverable you promised — produce it automatically, no asking; the other artifacts below stay opt-in. In **Facilitator** and **Creative Partner**, every artifact is opt-in: each is a fresh, token-expensive generation, so ask what they want, recommend the HTML keepsake as the default, and generate only what they choose. Everything derives from the log, so nothing is lost by deferring or skipping.
+
+**Delegate each artifact to a subagent.** By now the main context is full of the whole session — but the memlog holds everything, so the subagent doesn't need that context. Spawn one per requested artifact, telling it only: the spec below, the memlog path `{doc_workspace}/.memlog.md` (its sole source — read it in full), the output path, `{document_output_language}`, and "return ONLY the written file path." This keeps the heavy generation out of the main thread and proves the memlog is genuinely the canonical source. (Subagents can't spawn subagents — run these from here.)
+
+- **Imaginative HTML keepsake (recommended default).** A single self-contained `brainstorm.html` in `{doc_workspace}` — a genuine creative artifact, not a report poured into a template. There is no template on purpose: let *this* session's subject, energy, and whimsy drive the visual language (a children's game and a supply-chain session should not look alike). Give each technique its own treatment, invent visualizations that fit the ideas and techniques, and render the synthesis as the climax. Inline all CSS and any JS; no external dependencies. Open it once complete.
+- **Intent doc.** A succinct `brainstorm-intent.md` — the chosen and critical discoveries only, structured to drop straight into a downstream skill (`bmad-product-brief`, `bmad-prd`) as clean input, with none of the report's bloat - token usage matters and it must really be on point. Confirm what the user wants to capture as the intent from the overall findings as there may be many divergent discoveries (unless in headless mode, then take your best educated stance).
+- **Offer other options they might want from it also based on context** — a pitch, a one-pager, a task list — produced from the same source. These can be slide decks, html, markdown - again be creative and offer really interesting quality options based on perceived user needs while asking them also to offer any other ideas.
+
+If the session used invented techniques, offer to save a keeper into `{workflow.additional_techniques}` via `bmad-customize` user preferences.
+
+After producing what they chose, offer them ideas for deep-dive brainstorming new sessions, offer to fully extrapolate any ideas into an html report (autonomously brainstorm on their behalf), and most importantly: execute each `{workflow.external_handoffs}` instruction. Then share the artifact paths (and any handoff destinations), invoke `bmad-help` to suggest where this leads next in the BMad ecosystem, let them know if they feel a produced intent is detailed enough they could jump right into passing it to bmad-spec or any other analysis tool (outlined from bmad-help) and run `{workflow.on_complete}` if non-empty.
diff --git a/.agents/skills/bmad-brainstorming/references/headless.md b/.agents/skills/bmad-brainstorming/references/headless.md
new file mode 100644
index 00000000..5da9a25b
--- /dev/null
+++ b/.agents/skills/bmad-brainstorming/references/headless.md
@@ -0,0 +1,54 @@
+# Headless Mode
+
+Load this file ONLY when bmad-brainstorming is invoked headless. It is quarantined here on purpose: headless is the single context in which you generate ideas yourself, which is the exact inverse of the interactive Stance. Loading it in a normal session would corrupt the facilitation. Follow it for the whole run.
+
+## Detection
+
+**If a human is sending messages in this session, you are interactive — no payload shape or phrasing overrides that.** Headless requires the *absence* of an interactive user. It is in effect only when one of these unambiguous machine signals holds:
+
+- the caller sets a `headless: true` flag (or the equivalent argument the harness exposes),
+- the invocation comes from another skill or a non-interactive runner (no TTY, no user message stream),
+- `{workflow.activation_steps_prepend}` includes an entry that explicitly declares headless.
+
+When in doubt, you are interactive — a present human asking you to "brainstorm X and give me the HTML" is a normal interactive opening, not a headless trigger. Facilitate them; do not brainstorm for them.
+
+## The inversion
+
+There is no user to draw ideas out of, so you become the brainstormer. Run a real divergent session against the supplied topic: discover techniques with `uv run {skill-root}/scripts/brain.py --file {workflow.brain_methods} list --all` (the whole catalog is fine here — you are generating, not pacing a user; add `show "<name>"` for a technique's full method on demand), plus any `{workflow.additional_techniques}`, preferring `{workflow.favorite_techniques}` where they fit; work them, and **shift the creative domain every ~10 ideas** exactly as the interactive Stance demands — technical, then experiential, then business, then failure modes, then wildcards. Push past the obvious; the same quantity ambition (aim past 100) and anti-clustering discipline apply. The only thing that changes is that the ideas are now yours to generate. This relaxation is scoped entirely to this file — it never applies to interactive sessions.
+
+## Inputs the caller is expected to provide
+
+Free-form structured payload in the first message; provide what applies:
+
+- `topic` — what to brainstorm. Required. If absent and uninferable, halt `blocked`.
+- `goal` — desired outcome / framing, if any.
+- `techniques` — specific methods to use; otherwise you choose fitting ones from the library.
+- `context` — file paths or text to ground the session (problem statement, prior notes, brief).
+- `doc_workspace` — a specific run folder; otherwise bind the default `{workflow.output_dir}/{workflow.output_folder_name}/`.
+- `artifacts` — which outputs to produce: `html`, `intent`, or both. Default: both.
+
+## Run
+
+1. Bind `{doc_workspace}` and create the memlog with `uv run {project-root}/_bmad/scripts/memlog.py init --workspace {doc_workspace} --field topic="<topic>" [--field goal="<goal>"]`. It remains the canonical source every artifact derives from.
+2. Run the divergent session per **The inversion**, capturing each idea with `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {doc_workspace} --type idea --text "<idea>"` as it lands, and marking each technique switch with `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {doc_workspace} --type technique --text "started <name>"`.
+3. Synthesize: surface the conclusions, connections, and the few directions that matter; record them with `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {doc_workspace} --type insight --text "<insights>"`, then run `uv run {project-root}/_bmad/scripts/memlog.py set --workspace {doc_workspace} --key status --value complete`.
+4. Produce the requested artifacts from the log — `brainstorm.html` (the imaginative, self-contained, no-template report) and/or the succinct `brainstorm-intent.md` — the same artifacts `references/finalize.md` describes, delegating each to a subagent that reads the log as its sole source. (Headless produces the `artifacts` payload directly; it does not ask, unlike the interactive opt-in.)
+5. Execute each entry in `{workflow.external_handoffs}` (capture returned URLs/IDs into the JSON `external_handoffs` array; skip and flag unavailable tools — local files always exist). Then run `{workflow.on_complete}` if non-empty.
+
+Do not ask questions; do not greet. Record any assumption you made (a topic you had to infer, a goal you invented to frame the session) in `assumptions[]`.
+
+## Return
+
+End with a JSON status block. Use `complete` when the artifacts stand on their own, `partial` when produced but key inputs were inferred (e.g. topic was thin), `blocked` when no artifact was produced (e.g. no topic). Omit keys for artifacts not produced.
+
+```json
+{
+  "status": "complete",
+  "intent": "brainstorm",
+  "memlog": "{doc_workspace}/.memlog.md",
+  "html": "{doc_workspace}/brainstorm.html",
+  "intent_doc": "{doc_workspace}/brainstorm-intent.md",
+  "assumptions": [],
+  "external_handoffs": []
+}
+```
diff --git a/.agents/skills/bmad-brainstorming/references/in-chat-techniques.md b/.agents/skills/bmad-brainstorming/references/in-chat-techniques.md
new file mode 100644
index 00000000..4e0e02a7
--- /dev/null
+++ b/.agents/skills/bmad-brainstorming/references/in-chat-techniques.md
@@ -0,0 +1,18 @@
+# Choosing Techniques In Chat
+
+Loaded only when the user won't use the composer page (no browser, headless, or they declined). Here you pick the batch in conversation. **3–4 is the sweet spot.** Present the four ways below — this is the one allowed menu — and wait for their pick.
+
+- **Facilitator Chosen (default)** — from the goal, your `{workflow.favorite_techniques}`, and the `categories` map, name a batch of 3–4. Confirm exact names with a targeted `list --category` on only the categories you're drawing from; never enumerate the library to choose.
+- **Browse** — send them to the composer page after all (`## Run a Session` in `SKILL.md`); they tick techniques and paste the result back, which carries each one's full name/category/description.
+- **Category** — the user names 1–n categories; `random --category` draws the batch from them. No listing needed.
+- **Inventive Flow** — invent at least 3 techniques, announce the order before the first, touch no script. Log each one's name + description so you can offer to save a keeper to `{workflow.additional_techniques}` (via `bmad-customize`) at wrap-up.
+
+The library is large — never pull it whole into context. The only way in is the helper, always passing `--file {workflow.brain_methods}`. Subcommands of `uv run {skill-root}/scripts/brain.py --file {workflow.brain_methods}`:
+
+- `categories` — names + counts; the cheap survey map.
+- `list --category X [--category Y]` — the index (name + gist) for those categories. Bare `list` is refused by the script.
+- `random --category X [...] -n 4` — draw a batch blind, listing nothing.
+- `show "<name>"` — one technique's full method; call only the moment it is about to run.
+- `html --out <path>` — write the composer page to a file (the Browse option above).
+
+Treat `{workflow.additional_techniques}` as first-class entries (including new categories), preferring `{workflow.favorite_techniques}` where they fit. To include the additional techniques in any command, pass `--extra <json>` (a JSON list of `{category, technique_name, description}` objects). The `list` gist usually suffices to propose and run a technique; reach for `show` for deeper mechanics.
diff --git a/.agents/skills/bmad-brainstorming/references/mode-autonomous.md b/.agents/skills/bmad-brainstorming/references/mode-autonomous.md
new file mode 100644
index 00000000..10138952
--- /dev/null
+++ b/.agents/skills/bmad-brainstorming/references/mode-autonomous.md
@@ -0,0 +1,10 @@
+# Mode: Ideate For Me
+
+The user handed you the topic and wants to see what you come up with on your own, then look at the result. You become the brainstormer — this is the one interactive mode where the ideas are yours to generate.
+
+- **Run a real divergent session yourself.** If the user supplied techniques (e.g. a composed prompt pasted from the selector page), honor those first; otherwise pick and run techniques on your own (use `brain.py` as in `## Choosing Techniques`, but *you* choose — no menu for the user). Capture each idea to the memlog with `--type idea --by coach`, marking each technique switch with a `technique` entry, shifting the creative domain every ~10 ideas, aiming past 100. Push past the obvious.
+- **Don't pepper the user with questions** — this is your run. One quick confirm of topic and goal up front is plenty.
+- **When it's mined out, synthesize and produce the keepsake.** Go to `## Wrap-Up` (`references/finalize.md`): record the insights, mark the memlog complete, and **auto-generate the imaginative HTML keepsake — don't ask first; the keepsake is the result you promised to show them.** Offer the other artifacts (intent doc, etc.) after.
+- **Then, because a human is here, offer to keep going together.** They may want to push an idea further or react to what you found — if so, switch into **Facilitator** or **Creative Partner** (load that frame), **record the switch in the memlog** so a resume restores the new stance — `uv run {project-root}/_bmad/scripts/memlog.py set --workspace {doc_workspace} --key mode --value <facilitator|partner>` — and continue from the same memlog.
+
+This is the interactive sibling of headless mode (`references/headless.md`): the same self-generation, but a person is present to receive the output and may continue. headless is the no-human, returns-JSON runner; this one greets, presents, and hands off.
diff --git a/.agents/skills/bmad-brainstorming/references/mode-facilitator.md b/.agents/skills/bmad-brainstorming/references/mode-facilitator.md
new file mode 100644
index 00000000..1bee2205
--- /dev/null
+++ b/.agents/skills/bmad-brainstorming/references/mode-facilitator.md
@@ -0,0 +1,11 @@
+# Mode: Facilitator
+
+You are a forcing function for the user's creativity, never a source of ideas. The best version of this session ends with the user surprised by what *they* came up with — every idea in the memlog is theirs.
+
+- **You do not supply ideas.** Your moves are questions, provocations, constraints, and reflections that make *the user* generate, while you steer within the chosen technique. When the well looks dry, don't fill it — change the technique, shift the angle, or push harder.
+- **The one exception:** if the user *directly asks* for an idea, give exactly one as a spark, then hand the pen back. Reaching for that repeatedly is the signal to change technique, not to keep feeding ideas.
+- This holds for the whole generative session; it relaxes only during synthesis at wrap-up (`references/finalize.md`).
+
+Every idea you log is the user's, so no attribution is needed — log with `--type idea` (no `--by`).
+
+Go to `## Choosing Techniques`.
diff --git a/.agents/skills/bmad-brainstorming/references/mode-partner.md b/.agents/skills/bmad-brainstorming/references/mode-partner.md
new file mode 100644
index 00000000..84774001
--- /dev/null
+++ b/.agents/skills/bmad-brainstorming/references/mode-partner.md
@@ -0,0 +1,16 @@
+# Mode: Creative Partner
+
+You are still the facilitator — their creativity is the point, and they do the **majority** of the generating. But here you also play: you ride alongside and throw in your own ideas as sparks and yes-and fuel, so the two of you build a chain neither would alone. The energy is collaborative, not extractive — you feed off each other.
+
+**Set it up first.** Before you start, tell the user how this mode works and that they stay in control: they can **reject any idea you offer, ask you to help more or less, and tell you how to brainstorm** — a technique to try, a tone, a direction to chase. You're a partner they can steer, not a script.
+
+Hold the balance:
+
+- **Their fire, your kindling.** After you offer an idea, hand the pen back with a question. Never run a string of your own while they go quiet.
+- **"Yes, and" is the default move.** Take what they just said, build it one rung higher, then dare them to top you. Make them *want* to outdo you.
+- **Offer real alternatives**, not leading questions — a genuine idea they can mutate or reject, an opening, never a conclusion.
+- **Watch the ratio.** If you've contributed more than they have over the last few exchanges, you've slipped toward doing it *for* them — pull back to questions and constraints.
+
+**Attribution is mandatory here.** Every idea entry records who it came from: `--by user` for theirs, `--by coach` for yours (e.g. `append --type idea --by coach --text "..."`). This keeps the record honest and lets the wrap-up hand *them* the mirror of what *they* generated.
+
+Go to `## Choosing Techniques`.
diff --git a/.agents/skills/bmad-brainstorming/references/resume.md b/.agents/skills/bmad-brainstorming/references/resume.md
new file mode 100644
index 00000000..48ff453e
--- /dev/null
+++ b/.agents/skills/bmad-brainstorming/references/resume.md
@@ -0,0 +1,5 @@
+# Resuming a Session
+
+Read the chosen `{doc_workspace}/.memlog.md` **in full** — the one time you read the memlog. Frontmatter restores topic, goal, status, and **mode**: reload that mode's frame (`mode-facilitator.md` / `mode-partner.md` / `mode-autonomous.md`) and hold it again. The body restores everything generated — entries in order, `technique` entries marking which lens was active, `by` tags marking authorship.
+
+Reconstruct the picture, then reflect back where things stand (topic, what's already mined, which threads felt live) to re-establish shared state before continuing. Then continue per the mode's frame (appending to the same memlog) — or, if they're ready to land it, go to Wrap-Up (`references/finalize.md`).
diff --git a/.agents/skills/bmad-brainstorming/scripts/brain.py b/.agents/skills/bmad-brainstorming/scripts/brain.py
new file mode 100644
index 00000000..1028cf6e
--- /dev/null
+++ b/.agents/skills/bmad-brainstorming/scripts/brain.py
@@ -0,0 +1,770 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.10"
+# ///
+"""Serve the brainstorming technique library without loading it all into context.
+
+The library is a CSV (category, technique_name, description, detail). `description`
+is a short gist — enough to propose and run most techniques. `detail` is optional:
+a path (relative to the CSV's directory) to a fuller instruction file for a technique
+complex enough to warrant one. Only `show` resolves detail files, and only for the
+technique asked for — so the heavy material never enters context until it is run.
+
+Commands:
+  categories                  list category names + counts (the cheap entry point)
+  list --category C [...]      the index (name + gist) for those categories
+  list --all                  the whole index at once — deliberate; large, avoid interactively
+  show NAME [NAME ...]         full gist for each, inlining its detail file if it has one
+  random [--category C] [-n N]  pick N at random (optionally within categories)
+  html --out PATH             write the offline 'browse all' selection page to a file
+
+`list` refuses to run with neither --category nor --all, and `html` writes to a file
+rather than stdout: dumping the full catalog into context is a footgun, so reaching the
+whole library at once must always be an explicit, deliberate choice.
+
+`--extra PATH` merges a JSON overlay of additional techniques (customize.toml's
+`additional_techniques`) into every command. An extra whose technique_name matches
+a shipped row (case-insensitive) REPLACES it — retune a shipped technique; others
+append, so custom techniques and whole new categories are first-class everywhere —
+including the browse page and category draws. (Same overlay semantics as
+bmad-advanced-elicitation's pick_methods.py.)
+
+Default output is lean text for an LLM to read; pass --json for structured output.
+"""
+import argparse
+import csv
+import hashlib
+import html
+import json
+import random
+import sys
+from pathlib import Path
+
+DEFAULT_FILE = Path(__file__).resolve().parent.parent / "assets" / "brain-methods.csv"
+FIELDS = ("category", "technique_name", "description", "detail", "provenance", "good_for", "audience")
+# Optional columns beyond the original four — absent in older CSVs and in --extra
+# overlays, so always read through .get/setdefault. `provenance` (classic|signature|
+# playful) drives the "Proven & Professional" lead group; `good_for` (a |-separated
+# list of goal tags) drives the browse page's goal filter; `audience` (solo|group|either)
+# is advisory.
+OPTIONAL_FIELDS = ("detail", "provenance", "good_for", "audience")
+
+
+def load(file: Path) -> list[dict]:
+    # utf-8-sig: tolerate BOM-prefixed catalogs (Excel "CSV UTF-8", Notepad)
+    with open(file, newline="", encoding="utf-8-sig") as f:
+        rows = list(csv.DictReader(f))
+    for r in rows:
+        for k in FIELDS:
+            r.setdefault(k, "")
+            r[k] = (r.get(k) or "").strip()
+    return rows
+
+
+def load_extra(file: Path) -> list[dict]:
+    """Merge-in techniques from a JSON overlay — a list of
+    {category, technique_name, description[, detail]} objects. This is how
+    customize.toml's `additional_techniques` become first-class across *every*
+    subcommand (categories/list/random/show/html), so the browse page and
+    category draws include them too, not just the in-chat flows."""
+    data = json.loads(file.read_text(encoding="utf-8-sig"))
+    if not isinstance(data, list):
+        raise ValueError("--extra must be a JSON array of objects")
+    rows = []
+    for item in data:
+        if not isinstance(item, dict):
+            raise ValueError(f"each --extra entry must be a JSON object, got: {item!r}")
+        rows.append({
+            "category": str(item.get("category", "")).strip(),
+            "technique_name": str(item.get("technique_name", "")).strip(),
+            "description": str(item.get("description", "")).strip(),
+            "detail": str(item.get("detail") or "").strip(),
+            "provenance": str(item.get("provenance") or "").strip(),
+            "good_for": str(item.get("good_for") or "").strip(),
+            "audience": str(item.get("audience") or "").strip(),
+        })
+    return rows
+
+
+def merge_extra(rows: list[dict], extras: list[dict]) -> list[dict]:
+    """Extras replace a catalog row with the same technique_name (case-insensitive),
+    otherwise append — the same overlay semantics as pick_methods.py, so
+    customize.toml additional_* entries behave identically across sibling skills."""
+    merged = list(rows)
+    index = {r["technique_name"].lower(): i for i, r in enumerate(merged)}
+    for e in extras:
+        key = e["technique_name"].lower()
+        if key in index:
+            merged[index[key]] = e
+        else:
+            index[key] = len(merged)
+            merged.append(e)
+    return merged
+
+
+def categories(rows: list[dict]) -> list[tuple[str, int]]:
+    counts: dict[str, int] = {}
+    for r in rows:
+        counts[r["category"]] = counts.get(r["category"], 0) + 1
+    return sorted(counts.items())
+
+
+def filter_cats(rows: list[dict], cats: list[str] | None) -> list[dict]:
+    if not cats:
+        return rows
+    wanted = {c.lower() for c in cats}
+    return [r for r in rows if r["category"].lower() in wanted]
+
+
+def find(rows: list[dict], names: list[str]) -> tuple[list[dict], list[str]]:
+    by_name = {r["technique_name"].lower(): r for r in rows}
+    found, missing = [], []
+    for n in names:
+        r = by_name.get(n.strip().lower())
+        (found if r else missing).append(r if r else n)
+    return found, missing
+
+
+def resolve_detail(row: dict, csv_dir: Path) -> str | None:
+    """Return the contents of a row's detail file, or None if there is no detail
+    (or the file is missing — a missing file is reported to stderr, not fatal)."""
+    if not row.get("detail"):
+        return None
+    path = (csv_dir / row["detail"]).resolve()
+    if not path.is_file():
+        print(f"# detail file not found for {row['technique_name']}: {row['detail']}", file=sys.stderr)
+        return None
+    return path.read_text(encoding="utf-8").strip()
+
+
+def fmt_categories(cats: list[tuple[str, int]], as_json: bool) -> str:
+    if as_json:
+        return json.dumps([{"category": c, "count": n} for c, n in cats])
+    return "\n".join(f"{c}\t{n}" for c, n in cats)
+
+
+def fmt_list(rows: list[dict], as_json: bool) -> str:
+    if as_json:
+        return json.dumps([{k: r[k] for k in ("category", "technique_name", "description")} for r in rows])
+    return "\n".join(f"{r['category']}\t{r['technique_name']}\t{r['description']}" for r in rows)
+
+
+def fmt_show(rows: list[dict], csv_dir: Path, as_json: bool) -> str:
+    if as_json:
+        out = []
+        for r in rows:
+            d = resolve_detail(r, csv_dir)
+            entry = {k: r[k] for k in ("category", "technique_name", "description")}
+            if d:
+                entry["detail"] = d
+            out.append(entry)
+        return json.dumps(out)
+    blocks = []
+    for r in rows:
+        block = f"## {r['technique_name']}  [{r['category']}]\n{r['description']}"
+        d = resolve_detail(r, csv_dir)
+        if d:
+            block += f"\n\n{d}"
+        blocks.append(block)
+    return "\n\n".join(blocks)
+
+
+def pretty(cat: str) -> str:
+    """Turn a category slug (e.g. 'speculative_future') into a display name."""
+    return cat.replace("_", " ").replace("-", " ").title()
+
+
+# --- card visuals: a crafted duotone icon + hue per category, plus a per-technique icon ---
+# The hues and SVG glyphs are *data*, not logic: they live in the icon sidecar
+# (assets/brain-icons.json) so the catalog's visuals can be edited without touching code.
+# It maps category slug -> {hue, glyph} and technique name -> svg (inner markup, drawn in
+# `currentColor` which the CSS sets to the category hue; the shared CHIP frame is added by
+# the renderer). Anything missing falls back here — an unknown category gets a hash-derived
+# hue + generic glyph, an unknown/not-yet-iconed technique a neutral mark — so custom
+# catalogs always render.
+
+ICON_FILE = DEFAULT_FILE.parent / "brain-icons.json"
+
+CHIP = '<rect x="1.5" y="1.5" width="41" height="41" rx="12" fill="currentColor" fill-opacity="0.12"/>'
+
+_FALLBACK_GLYPH = (
+    '<circle cx="22" cy="22" r="11" fill="currentColor" fill-opacity="0.16"/>'
+    '<circle cx="22" cy="22" r="11" stroke="currentColor" stroke-width="1.6" fill="none"/>'
+    '<circle cx="22" cy="22" r="3.4" fill="currentColor"/>'
+)
+_FALLBACK_TECH = (
+    '<rect x="15" y="15" width="14" height="14" rx="2.5" transform="rotate(45 22 22)" '
+    'fill="none" stroke="currentColor" stroke-width="2"/><circle cx="22" cy="22" r="2.4" fill="currentColor"/>'
+)
+
+
+def _load_icons(file: Path = ICON_FILE) -> tuple[dict, dict]:
+    """Read the icon sidecar: (category slug -> {hue, glyph}, technique name -> svg).
+    A missing or malformed file is non-fatal — everything then uses the fallbacks below."""
+    try:
+        data = json.loads(file.read_text(encoding="utf-8"))
+    except (OSError, ValueError):
+        return {}, {}
+    return (data.get("categories") or {}), (data.get("techniques") or {})
+
+
+_CATEGORY_STYLES, _TECH_ICONS = _load_icons()
+
+
+def _hsl_hex(deg: int, s: float, lt: float) -> str:
+    import colorsys
+
+    r, g, b = colorsys.hls_to_rgb((deg % 360) / 360, lt, s)
+    return "#%02x%02x%02x" % (round(r * 255), round(g * 255), round(b * 255))
+
+
+def category_style(cat: str) -> tuple[str, str]:
+    """(hue, glyph markup) for a category — from the sidecar for the shipped set, derived for extras."""
+    style = _CATEGORY_STYLES.get(cat)
+    if style and style.get("hue"):
+        return style["hue"], style.get("glyph") or _FALLBACK_GLYPH
+    deg = int(hashlib.md5(cat.encode("utf-8")).hexdigest(), 16) % 360
+    return _hsl_hex(deg, 0.58, 0.52), _FALLBACK_GLYPH
+
+
+def tech_icon(name: str) -> str:
+    """The hand-picked line-icon for a specific technique (neutral mark if unknown)."""
+    return _TECH_ICONS.get(name, _FALLBACK_TECH)
+
+
+SELECTOR_TEMPLATE = r"""<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>BMad Method Brainstorming Selection</title>
+<script>
+/* set the theme before first paint so there's no light-mode flash */
+(function(){ try {
+  var t = localStorage.getItem('bmad-theme');
+  if (!t) { t = (window.matchMedia && window.matchMedia('(prefers-color-scheme: dark)').matches) ? 'dark' : 'light'; }
+  document.documentElement.setAttribute('data-theme', t);
+} catch(e){} })();
+</script>
+<style>
+  :root {
+    --bg:#f6f7fb; --surface:#fff; --ink:#1c1e2b; --muted:#6b7080;
+    --accent:#5b4bdc; --accent-ink:#5b4bdc; --warn:#c0561f;
+    --line:#e6e8f0; --control:#eef0f7; --control2:#f1f2f8; --raised:#fff;
+    --cnt:#b9bdce; --foot:#aeb2c4; --shadow:rgba(20,20,50,.06);
+  }
+  :root[data-theme="dark"] {
+    --bg:#0f1117; --surface:#171a23; --ink:#e7e9f2; --muted:#9aa0b4;
+    --accent:#6d5cf0; --accent-ink:#a99bff; --warn:#e08a4a;
+    --line:#2a2f3e; --control:#222634; --control2:#1d212d; --raised:#2c3242;
+    --cnt:#5a6076; --foot:#5a6076; --shadow:rgba(0,0,0,.45);
+  }
+  /* lift the category hue toward white on dark surfaces so deep hues stay legible */
+  :root[data-theme="dark"] section > h2 { color:color-mix(in srgb, var(--c) 62%, #fff); }
+  :root[data-theme="dark"] .tech .ico { color:color-mix(in srgb, var(--c) 68%, #fff); }
+  :root[data-theme="dark"] label.tech:has(input:checked) { border-color:color-mix(in srgb, var(--c) 60%, #fff); }
+  .titlerow { display:flex; align-items:flex-start; justify-content:space-between; gap:12px; }
+  .themebtn { flex:none; width:36px; height:36px; border-radius:9px; background:var(--control); color:var(--ink); font-size:17px; line-height:1; display:inline-flex; align-items:center; justify-content:center; }
+  .themebtn:hover { background:var(--raised); }
+  * { box-sizing:border-box; }
+  body { margin:0; font:16px/1.5 -apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,Helvetica,Arial,sans-serif; background:var(--bg); color:var(--ink); }
+  header { position:sticky; top:0; z-index:5; background:var(--surface); padding:20px 0 12px; border-bottom:1px solid var(--line); box-shadow:0 2px 12px var(--shadow); }
+  .hwrap { max-width:1120px; margin:0 auto; padding:0 24px; }  /* align header content with the card column on wide screens */
+  h1 { margin:0 0 4px; font-size:24px; letter-spacing:-.02em; }
+  .sub { margin:0 0 12px; color:var(--muted); font-size:14px; max-width:74ch; }
+  button { font:inherit; border:0; border-radius:8px; cursor:pointer; }
+  .composer { display:flex; flex-direction:column; gap:9px; margin:6px 0 12px; }
+  .grp { display:flex; gap:8px; align-items:center; flex-wrap:wrap; }
+  .glabel { font-size:11px; text-transform:uppercase; letter-spacing:.07em; color:var(--muted); min-width:74px; }
+  .modes { display:inline-flex; background:var(--control); border-radius:9px; padding:3px; gap:2px; }
+  .mode { padding:7px 13px; font-size:14px; font-weight:600; color:var(--muted); background:transparent; }
+  .mode.on { background:var(--raised); color:var(--accent-ink); box-shadow:0 1px 3px var(--shadow); }
+  .modehint { flex:1 1 240px; min-width:0; font-size:13px; color:var(--muted); font-style:italic; }
+  .pill { font-size:13px; color:var(--muted); background:var(--control); padding:6px 12px; border-radius:20px; }
+  .pill b { color:var(--accent-ink); }
+  .step { display:inline-flex; align-items:center; gap:7px; font-size:13px; color:var(--ink); background:var(--control2); padding:4px 6px 4px 12px; border-radius:20px; }
+  .step b { min-width:12px; text-align:center; font-size:14px; color:var(--ink); }
+  .step button { width:24px; height:24px; border-radius:50%; background:var(--raised); color:var(--muted); font-size:17px; line-height:22px; text-align:center; box-shadow:0 1px 2px var(--shadow); }
+  .step button:hover { color:var(--accent-ink); }
+  .total { font-size:12px; color:var(--muted); }
+  .total.warn { color:var(--warn); font-weight:600; }
+  .bar { display:flex; gap:10px 14px; align-items:center; flex-wrap:wrap; }
+  #copy { margin-left:auto; padding:9px 22px; background:var(--accent); color:#fff; font-size:14px; font-weight:700; }
+  #copy:hover { filter:brightness(1.07); }
+  .chips { flex:1 1 320px; min-width:0; display:flex; gap:7px; flex-wrap:wrap; align-items:center; }
+  .chip { font-size:12px; padding:4px 11px; border-radius:16px; border:0; color:#fff; background:var(--cc); font-weight:600; cursor:pointer; }
+  .chip:hover { filter:brightness(1.08); }
+  .banner { max-height:0; overflow:hidden; transition:max-height .25s ease, padding .22s ease, margin .22s ease; background:linear-gradient(90deg,var(--accent),#8275f2); color:#fff; border-radius:10px; font-weight:700; text-align:center; padding:0 14px; }
+  .banner.show { max-height:64px; padding:13px 14px; margin-top:10px; }
+  .banner.fail { background:linear-gradient(90deg,var(--warn),#e0894a); }
+  main { padding:18px 24px 60px; max-width:1120px; margin:0 auto; }
+  section { margin:0 0 26px; }
+  section > h2 { font-size:13px; text-transform:uppercase; letter-spacing:.08em; color:var(--c); margin:0 0 10px; border-bottom:1px solid color-mix(in srgb, var(--c) 24%, var(--line)); padding-bottom:6px; }
+  section > h2 .cnt { color:color-mix(in srgb, var(--c) 45%, var(--cnt)); margin-left:6px; }
+  .grid { display:grid; grid-template-columns:repeat(auto-fill,minmax(360px,1fr)); gap:10px; }
+  label.tech { display:flex; gap:12px; align-items:flex-start; background:color-mix(in srgb, var(--c) 5%, var(--surface)); border:1px solid color-mix(in srgb, var(--c) 18%, var(--line)); border-radius:10px; padding:11px 13px; cursor:pointer; transition:border-color .12s, box-shadow .12s, background .12s; }
+  label.tech:hover { border-color:color-mix(in srgb, var(--c) 45%, var(--surface)); }
+  label.tech input { margin-top:2px; width:17px; height:17px; accent-color:var(--c); flex:none; }
+  label.tech:has(input:checked) { border-color:var(--c); background:color-mix(in srgb, var(--c) 12%, var(--surface)); box-shadow:0 0 0 2px color-mix(in srgb, var(--c) 30%, transparent); }
+  .tech .ic2 { display:flex; gap:5px; flex:none; }
+  .tech .ico { width:40px; height:40px; flex:none; color:var(--c); }
+  .tech .n { font-weight:600; display:block; }
+  .tech .d { color:var(--muted); font-size:13.5px; display:block; margin-top:2px; }
+  .tech .gf { color:var(--accent-ink); font-size:11px; display:block; margin-top:5px; opacity:.85; }
+  .grouphdr { margin:30px 0 12px; font-size:12px; text-transform:uppercase; letter-spacing:.14em; font-weight:700; color:var(--c); opacity:.92; border-bottom:1px solid color-mix(in srgb, var(--c) 22%, var(--line)); padding-bottom:7px; }
+  main > .grouphdr:first-child { margin-top:2px; }
+  :root[data-theme="dark"] .grouphdr { color:color-mix(in srgb, var(--c) 62%, #fff); }
+  .goals { display:flex; gap:7px; flex-wrap:wrap; }
+  .goal { font-size:12px; padding:5px 12px; border-radius:16px; background:var(--control); color:var(--muted); font-weight:600; }
+  .goal:hover { color:var(--ink); }
+  .goal.on { background:var(--accent); color:#fff; }
+  label.tech.invent { border-style:dashed; background:transparent; }
+  label.tech.invent:hover { border-color:var(--c); }
+  label.tech.invent .n { color:var(--c); }
+  label.tech.hidden { display:none; }
+  footer { text-align:center; color:var(--foot); font-size:12px; padding:24px; }
+</style>
+</head>
+<body>
+<header>
+  <div class="hwrap">
+  <div class="titlerow">
+    <h1>BMad Method Brainstorming Selection</h1>
+    <button id="theme" class="themebtn" type="button" aria-label="Toggle dark mode" title="Toggle dark mode"></button>
+  </div>
+  <p class="sub">Compose your session, hit <strong>Copy prompt</strong>, and paste it back into the chat to begin. {{TOTAL}}</p>
+
+  <div class="composer">
+    <div class="grp">
+      <span class="glabel">Facilitation</span>
+      <div class="modes" id="modes">
+        <button type="button" class="mode on" data-mode="Facilitator">Facilitator</button>
+        <button type="button" class="mode" data-mode="Creative Partner">Creative Partner</button>
+        <button type="button" class="mode" data-mode="Ideate for me">Ideate for me</button>
+      </div>
+      <span class="modehint" id="modehint"></span>
+    </div>
+    <div class="grp">
+      <span class="glabel">Techniques</span>
+      <span class="pill">Picked <b id="pickN">0</b></span>
+      <span class="step">Random <button type="button" data-step="rand" data-d="-1">&minus;</button><b id="randN">0</b><button type="button" data-step="rand" data-d="1">+</button></span>
+      <span class="step">Invent <button type="button" data-step="inv" data-d="-1">&minus;</button><b id="invN">0</b><button type="button" data-step="inv" data-d="1">+</button></span>
+      <span class="step">AI picks <button type="button" data-step="ai" data-d="-1">&minus;</button><b id="aiN">0</b><button type="button" data-step="ai" data-d="1">+</button></span>
+      <span class="total" id="total">Total 0 &middot; 3&ndash;4 is the sweet spot</span>
+      <button id="copy" type="button">Copy prompt</button>
+    </div>
+  </div>
+
+  {{GOALBAR}}
+  <div class="bar">
+    <span class="glabel">Jump to</span>
+    <div class="chips" id="chips">{{CHIPS}}</div>
+  </div>
+
+  <div class="banner" id="banner">&#10003; Copied! Now paste it into the chat to start your session.</div>
+  </div>
+</header>
+<main>
+{{BODY}}
+</main>
+<footer>BMad Method &middot; Brainstorming</footer>
+<script>
+(function(){
+  var $ = function(id){ return document.getElementById(id); };
+  var all = Array.prototype.slice;
+  var boxes = all.call(document.querySelectorAll('input[type=checkbox]'));
+  var techBoxes = boxes.filter(function(b){ return b.dataset.name; });      // real technique cards
+  var inventBoxes = boxes.filter(function(b){ return b.dataset.invent; });  // per-category "invent in the spirit of" cards
+  var header = document.querySelector('header');
+  var sections = all.call(document.querySelectorAll('section'));
+  var state = { mode: 'Facilitator', rand: 0, inv: 0, ai: 0 };
+  var MODE_HINTS = {
+    'Facilitator': 'A forcing function for your ideas — I prompt and push, but never supply them.',
+    'Creative Partner': 'We riff together — I facilitate and add ideas too, each logged as yours or mine.',
+    'Ideate for me': 'I run the whole session myself, then show you the result and offer to keep going.'
+  };
+  function setHint(){ $('modehint').textContent = MODE_HINTS[state.mode] || ''; }
+
+  var themeBtn = $('theme');
+  function setThemeIcon(){ themeBtn.textContent = document.documentElement.getAttribute('data-theme') === 'dark' ? '☀' : '☾'; }
+  themeBtn.addEventListener('click', function(){
+    var next = document.documentElement.getAttribute('data-theme') === 'dark' ? 'light' : 'dark';
+    document.documentElement.setAttribute('data-theme', next);
+    try { localStorage.setItem('bmad-theme', next); } catch(e){}
+    setThemeIcon();
+  });
+
+  all.call(document.querySelectorAll('.mode')).forEach(function(b){
+    b.addEventListener('click', function(){
+      all.call(document.querySelectorAll('.mode')).forEach(function(m){ m.classList.remove('on'); });
+      b.classList.add('on');
+      state.mode = b.dataset.mode;
+      setHint();
+    });
+  });
+
+  all.call(document.querySelectorAll('[data-step]')).forEach(function(btn){
+    btn.addEventListener('click', function(){
+      var k = btn.dataset.step, d = parseInt(btn.dataset.d, 10);
+      state[k] = Math.max(0, state[k] + d);
+      update();
+    });
+  });
+
+  // Category chips are jump-nav: click one to smooth-scroll its section into view,
+  // offsetting by the sticky header's height so the heading isn't hidden beneath it.
+  all.call(document.querySelectorAll('.chip')).forEach(function(chip){
+    chip.addEventListener('click', function(){
+      var sec = null;
+      for (var i = 0; i < sections.length; i++){ if (sections[i].dataset.cat === chip.dataset.cat){ sec = sections[i]; break; } }
+      if (!sec){ return; }
+      var top = sec.getBoundingClientRect().top + window.pageYOffset - header.offsetHeight - 8;
+      window.scrollTo({ top: top, behavior: 'smooth' });
+    });
+  });
+
+  boxes.forEach(function(b){ b.addEventListener('change', update); });
+
+  // A `classic` technique appears twice (lead "Proven & Professional" group + its home
+  // category), so de-dupe checked picks by name; the lead copy carries data-lead.
+  function checkedTech(){
+    var seen = {}, out = [];
+    techBoxes.forEach(function(b){
+      if (!b.checked || seen[b.dataset.name]) { return; }
+      seen[b.dataset.name] = 1;
+      out.push(b);
+    });
+    return out;
+  }
+  function checkedInvent(){ return inventBoxes.filter(function(b){ return b.checked; }); }
+
+  function update(){
+    // rand can't exceed what the pool can supply — keep the counter honest with the draw
+    if (state.rand > randomPool().length){ state.rand = randomPool().length; }
+    $('pickN').textContent = checkedTech().length;
+    $('randN').textContent = state.rand;
+    $('invN').textContent = state.inv;
+    $('aiN').textContent = state.ai;
+    var total = checkedTech().length + state.rand + state.inv + checkedInvent().length + state.ai;
+    var t = $('total');
+    t.textContent = 'Total ' + total + ' · 3–4 is the sweet spot';
+    t.classList.toggle('warn', total > 5);
+  }
+
+  // "Great for" goal filter: clicking a goal narrows visible cards to those tagged with it.
+  var goalBtns = all.call(document.querySelectorAll('.goal'));
+  function activeGoals(){ return goalBtns.filter(function(b){ return b.classList.contains('on'); }).map(function(b){ return b.dataset.goal; }); }
+  function applyFilter(){
+    var act = activeGoals();
+    all.call(document.querySelectorAll('label.tech')).forEach(function(lab){
+      var inp = lab.querySelector('input');
+      if (inp.dataset.invent){ return; }  // invent cards aren't goal-tagged — always visible
+      var good = (inp.dataset.good || '').split('|');
+      var show = !act.length || act.some(function(g){ return good.indexOf(g) >= 0; });
+      lab.classList.toggle('hidden', !show);
+    });
+  }
+  goalBtns.forEach(function(b){ b.addEventListener('click', function(){ b.classList.toggle('on'); applyFilter(); }); });
+
+  function randomPool(){
+    var picked = {};
+    checkedTech().forEach(function(b){ picked[b.dataset.name] = 1; });
+    // draw from unchecked, non-lead copies, skipping anything already picked
+    return techBoxes.filter(function(b){ return !b.checked && !b.dataset.lead && !picked[b.dataset.name]; });
+  }
+
+  function sample(arr, n){
+    var a = arr.slice(), out = [];
+    while (out.length < n && a.length){ out.push(a.splice(Math.floor(Math.random() * a.length), 1)[0]); }
+    return out;
+  }
+
+  function compose(){
+    var picks = checkedTech().map(function(b){ return { n: b.dataset.name, c: b.dataset.cat, d: b.dataset.desc, r: false }; });
+    var rnd = sample(randomPool(), state.rand).map(function(b){ return { n: b.dataset.name, c: b.dataset.cat, d: b.dataset.desc, r: true }; });
+    var techs = picks.concat(rnd);
+    var L = ["Let's run my brainstorming session.", "", 'Facilitation mode: ' + state.mode + '.'];
+    if (techs.length){
+      L.push("", 'Techniques to use:');
+      techs.forEach(function(t, i){
+        L.push((i + 1) + '.' + (t.r ? ' (random pick)' : '') + ' ' + t.n + '  ·  ' + t.c);
+        L.push('   ' + t.d);
+      });
+    }
+    var extra = [];
+    if (state.inv > 0){ extra.push('invent ' + state.inv + ' brand-new technique' + (state.inv > 1 ? 's' : '') + ' on the fly'); }
+    checkedInvent().forEach(function(b){ extra.push('invent 1 new technique in the spirit of ' + b.dataset.invent); });
+    if (state.ai > 0){ extra.push('you choose ' + state.ai + ' more technique' + (state.ai > 1 ? 's' : '') + ' that fit my goal'); }
+    if (extra.length){ L.push("", 'Then: ' + extra.join('; and ') + '.'); }
+    if (!techs.length && !extra.length){
+      L.push("", state.mode === 'Ideate for me'
+        ? 'Run the whole session yourself — pick the techniques, generate the ideas, then show me the result.'
+        : 'Help me choose 3–4 techniques to start.');
+    }
+    return L.join('\n');
+  }
+
+  function fallbackCopy(t){
+    var ta = document.createElement('textarea');
+    ta.value = t; ta.style.position = 'fixed'; ta.style.opacity = '0';
+    document.body.appendChild(ta); ta.focus(); ta.select();
+    var ok = false;
+    try { ok = document.execCommand('copy'); } catch(e){ ok = false; }
+    document.body.removeChild(ta);
+    return ok;
+  }
+
+  function flash(ok, text){
+    var b = $('banner');
+    b.classList.toggle('fail', !ok);
+    b.innerHTML = ok
+      ? '✓ Copied! Now paste it into the chat to start your session.'
+      : '⚠ Couldn’t reach the clipboard — copy the text in the box, then paste it into the chat.';
+    b.classList.add('show');
+    setTimeout(function(){ b.classList.remove('show'); }, 4500);
+    // Last resort on a hard failure: a prefilled, selectable prompt so the text is never lost.
+    if (!ok){ window.prompt('Copy this, then paste it into the chat:', text); }
+  }
+
+  $('copy').addEventListener('click', function(){
+    var text = compose();
+    if (navigator.clipboard && navigator.clipboard.writeText){
+      navigator.clipboard.writeText(text).then(
+        function(){ flash(true, text); },
+        function(){ flash(fallbackCopy(text), text); }
+      );
+    } else { flash(fallbackCopy(text), text); }
+  });
+
+  setHint();
+  setThemeIcon();
+  update();
+})();
+</script>
+</body>
+</html>
+"""
+
+
+# --- browse-page layout: a "Proven & Professional" lead group, then super-groups ----------
+CLASSIC_GROUP = "Proven & Professional"
+LEAD_HUE = "#3d4f73"  # a dignified slate for the professional lead group
+
+# Super-group order for the shipped categories. Categories not listed (e.g. user-added
+# via --extra) render last under "More", alphabetically — so custom catalogs always show.
+CATEGORY_GROUPS = (
+    ("Structured & Analytical", ("structured", "deep")),
+    ("Creative & Generative", ("creative", "biomimetic", "cultural", "speculative_future", "quantum")),
+    ("Wild & Playful", ("wild", "absurdist", "theatrical", "constraint")),
+    ("Introspective & Personal", ("introspective_delight", "collaborative")),
+)
+
+# Human labels for the `good_for` goal tags; this dict's order is the filter-bar order.
+GOAL_LABELS = {
+    "feature": "Build a feature",
+    "novel": "Novel concept",
+    "strategy": "Strategy",
+    "planning": "Planning",
+    "diagnosis": "Diagnose",
+    "personal": "Personal / life",
+    "unstuck": "Get unstuck",
+}
+
+
+def _good_for_label(good: str) -> str:
+    parts = [GOAL_LABELS.get(g, g) for g in good.split("|") if g]
+    return ("Great for: " + " · ".join(parts)) if parts else ""
+
+
+def _svg(inner: str) -> str:
+    return f'<svg class="ico" viewBox="0 0 44 44" xmlns="http://www.w3.org/2000/svg">{CHIP}{inner}</svg>'
+
+
+def _card(r: dict, lead: bool = False) -> str:
+    """One technique card. `lead=True` cards live in the cross-cutting professional group;
+    they carry their own category hue (inline --c) and data-lead so selection can de-dupe."""
+    name = html.escape(r["technique_name"])
+    desc = html.escape(r["description"])
+    hue, glyph = category_style(r["category"])
+    disp_cat = html.escape(pretty(r["category"]))
+    good = html.escape(r.get("good_for", ""))
+    prov = html.escape(r.get("provenance", ""))
+    style = f' style="--c:{hue}"' if lead else ""
+    lead_attr = ' data-lead="1"' if lead else ""
+    gf = _good_for_label(r.get("good_for", ""))
+    gf_html = f'<span class="gf">{html.escape(gf)}</span>' if gf else ""
+    return (
+        f'<label class="tech"{style}><input type="checkbox" '
+        f'data-name="{name}" data-cat="{disp_cat}" data-desc="{desc}" data-good="{good}" data-prov="{prov}"{lead_attr}>'
+        f'<span class="ic2">{_svg(glyph)}{_svg(tech_icon(r["technique_name"]))}</span>'
+        f'<span><span class="n">{name}</span><span class="d">{desc}</span>{gf_html}</span></label>'
+    )
+
+
+def _invent_card(disp_cat: str, glyph: str) -> str:
+    """A dashed 'invent on the fly, in this category's spirit' card appended to each section."""
+    return (
+        f'<label class="tech invent"><input type="checkbox" data-invent="{disp_cat}">'
+        f'<span class="ic2">{_svg(glyph)}</span>'
+        f'<span><span class="n">✨ Invent a {disp_cat} technique</span>'
+        f'<span class="d">Make up a brand-new technique on the fly, in the spirit of {disp_cat}</span></span></label>'
+    )
+
+
+def html_doc(rows: list[dict]) -> str:
+    """Render the self-contained 'browse all techniques' selection page from the catalog.
+
+    Deterministic ordering so the shipped asset can be snapshot-tested against the CSV:
+    a cross-cutting "Proven & Professional" lead group (every `classic`-tagged row), then
+    the categories in fixed super-group order, then any unlisted/custom categories under
+    "More" alphabetically. Techniques render in file order within a category. A `classic`
+    row appears both in the lead group and its home category; the page de-dupes on select.
+    """
+    groups: dict[str, list[dict]] = {}
+    for r in rows:
+        groups.setdefault(r["category"], []).append(r)
+
+    body: list[str] = []
+    chips: list[str] = []
+
+    def add_section(cat: str) -> None:
+        hue, glyph = category_style(cat)
+        disp = html.escape(pretty(cat))
+        cards = [_card(r) for r in groups[cat]]
+        cards.append(_invent_card(disp, glyph))
+        chips.append(f'<button type="button" class="chip" data-cat="{disp}" style="--cc:{hue}">{disp}</button>')
+        body.append(
+            f'<section data-cat="{disp}" style="--c:{hue}"><h2>{disp}<span class="cnt">{len(groups[cat])}</span></h2>'
+            f'<div class="grid">{"".join(cards)}</div></section>'
+        )
+
+    # 1) lead group — every classic-tagged technique, cross-category (no invent card here)
+    classics = [r for r in rows if r.get("provenance", "").lower() == "classic"]
+    if classics:
+        disp = html.escape(CLASSIC_GROUP)
+        lead_cards = "".join(_card(r, lead=True) for r in classics)
+        chips.append(f'<button type="button" class="chip" data-cat="{disp}" style="--cc:{LEAD_HUE}">{disp}</button>')
+        body.append(
+            f'<section data-cat="{disp}" style="--c:{LEAD_HUE}"><h2>{disp}<span class="cnt">{len(classics)}</span></h2>'
+            f'<div class="grid">{lead_cards}</div></section>'
+        )
+
+    # 2) shipped categories, in super-group order
+    placed = set()
+    for group_title, cats in CATEGORY_GROUPS:
+        present = [c for c in cats if c in groups]
+        if not present:
+            continue
+        hue, _ = category_style(present[0])
+        body.append(f'<h2 class="grouphdr" style="--c:{hue}">{html.escape(group_title)}</h2>')
+        for c in present:
+            add_section(c)
+            placed.add(c)
+
+    # 3) leftover (custom / --extra) categories, alphabetically
+    leftover = sorted(c for c in groups if c not in placed)
+    if leftover:
+        body.append('<h2 class="grouphdr" style="--c:#8a8f9e">More</h2>')
+        for c in leftover:
+            add_section(c)
+
+    # goal-affinity filter bar — only if the catalog actually carries good_for tags
+    present_goals: set[str] = set()
+    for r in rows:
+        for g in (r.get("good_for", "") or "").split("|"):
+            if g:
+                present_goals.add(g)
+    goalbar = ""
+    if present_goals:
+        ordered = [g for g in GOAL_LABELS if g in present_goals] + sorted(present_goals - set(GOAL_LABELS))
+        gchips = "".join(
+            f'<button type="button" class="goal" data-goal="{html.escape(g)}">{html.escape(GOAL_LABELS.get(g, g))}</button>'
+            for g in ordered
+        )
+        goalbar = f'<div class="bar"><span class="glabel">Great for</span><div class="goals" id="goals">{gchips}</div></div>'
+
+    total = html.escape(f"{len(rows)} techniques across {len(groups)} categories.")
+    return (
+        SELECTOR_TEMPLATE.replace("{{BODY}}", "\n".join(body))
+        .replace("{{CHIPS}}", "".join(chips))
+        .replace("{{GOALBAR}}", goalbar)
+        .replace("{{TOTAL}}", total)
+    )
+
+
+def main(argv: list[str] | None = None) -> int:
+    p = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
+    p.add_argument("--file", type=Path, default=DEFAULT_FILE, help="technique CSV (default: sibling assets/brain-methods.csv)")
+    p.add_argument("--extra", type=Path, help="JSON overlay of additional techniques (customize.toml additional_techniques), merged into every command")
+    p.add_argument("--json", action="store_true", help="emit structured JSON instead of lean text")
+    sub = p.add_subparsers(dest="cmd", required=True)
+    sub.add_parser("categories", help="list category names + counts")
+    pl = sub.add_parser("list", help="the index: category/name/gist (needs --category or --all)")
+    pl.add_argument("--category", action="append", help="filter to a category (repeatable)")
+    pl.add_argument("--all", action="store_true", help="dump the entire catalog (deliberate; large)")
+    ps = sub.add_parser("show", help="full gist + detail file for named techniques")
+    ps.add_argument("names", nargs="+")
+    pr = sub.add_parser("random", help="pick techniques at random")
+    pr.add_argument("--category", action="append", help="restrict to a category (repeatable)")
+    pr.add_argument("-n", type=int, default=1, help="how many (default 1)")
+    ph = sub.add_parser("html", help="write the offline 'browse all' selection page")
+    ph.add_argument("--out", help="file to write the page to (required; never prints the catalog)")
+    args = p.parse_args(argv)
+
+    if not args.file.is_file():
+        print(f"error: technique file not found: {args.file}", file=sys.stderr)
+        return 2
+    rows = load(args.file)
+    if args.extra:
+        if not args.extra.is_file():
+            print(f"error: --extra file not found: {args.extra}", file=sys.stderr)
+            return 2
+        try:
+            rows = merge_extra(rows, load_extra(args.extra))
+        except (OSError, ValueError) as e:
+            print(f"error: could not read --extra: {e}", file=sys.stderr)
+            return 2
+    csv_dir = args.file.resolve().parent
+
+    if args.cmd == "categories":
+        print(fmt_categories(categories(rows), args.json))
+    elif args.cmd == "list":
+        if not args.category and not args.all:
+            print(
+                "error: `list` needs --category (one or more) — or --all to dump the whole "
+                "catalog on purpose. Use `categories` for the cheap map, or `random` to draw blind.",
+                file=sys.stderr,
+            )
+            return 2
+        print(fmt_list(filter_cats(rows, args.category), args.json))
+    elif args.cmd == "show":
+        found, missing = find(rows, args.names)
+        for m in missing:
+            print(f"# not found: {m}", file=sys.stderr)
+        if not found:
+            return 1
+        print(fmt_show(found, csv_dir, args.json))
+    elif args.cmd == "random":
+        pool = filter_cats(rows, args.category)
+        if not pool:
+            print("# no techniques match", file=sys.stderr)
+            return 1
+        n = max(0, min(args.n, len(pool)))  # clamp: never crash on a negative or oversized -n
+        print(fmt_list(random.sample(pool, n), args.json))
+    elif args.cmd == "html":
+        if not args.out:
+            print(
+                "error: `html` needs --out PATH — it writes the selection page to a file and "
+                "never prints the catalog to stdout (which would defeat the point).",
+                file=sys.stderr,
+            )
+            return 2
+        out = Path(args.out)
+        out.parent.mkdir(parents=True, exist_ok=True)
+        out.write_text(html_doc(rows), encoding="utf-8")
+        print(f"wrote {out} ({len(rows)} techniques, {len(categories(rows))} categories)")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/.agents/skills/bmad-brainstorming/scripts/tests/test_brain.py b/.agents/skills/bmad-brainstorming/scripts/tests/test_brain.py
new file mode 100644
index 00000000..329a6a5f
--- /dev/null
+++ b/.agents/skills/bmad-brainstorming/scripts/tests/test_brain.py
@@ -0,0 +1,239 @@
+# /// script
+# requires-python = ">=3.10"
+# dependencies = ["pytest>=8.0"]
+# ///
+"""Tests for brain.py. Run: uv run -m pytest scripts/tests/test_brain.py"""
+import json
+import sys
+from pathlib import Path
+
+import pytest
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+import brain  # noqa: E402
+
+CSV = """category,technique_name,description,detail
+collaborative,Yes And Building,Build on every idea with "yes and" to keep momentum,
+wild,Quantum Superposition,Hold contradictory ideas as simultaneously true,techniques/quantum.md
+structured,SCAMPER Method,Run the idea through seven transformation lenses,
+wild,Anti-Solution,Brainstorm how to make the problem worse then invert,
+"""
+
+DETAIL = "# Quantum Superposition\nFull multi-step instructions for the complex technique."
+
+
+@pytest.fixture
+def lib(tmp_path):
+    csv_path = tmp_path / "brain-methods.csv"
+    csv_path.write_text(CSV, encoding="utf-8")
+    (tmp_path / "techniques").mkdir()
+    (tmp_path / "techniques" / "quantum.md").write_text(DETAIL, encoding="utf-8")
+    return csv_path
+
+
+def test_load_normalizes_detail(lib):
+    rows = brain.load(lib)
+    assert len(rows) == 4
+    assert rows[0]["detail"] == ""
+    assert rows[1]["detail"] == "techniques/quantum.md"
+
+
+def test_categories_counts_sorted(lib):
+    assert brain.categories(brain.load(lib)) == [("collaborative", 1), ("structured", 1), ("wild", 2)]
+
+
+def test_filter_is_case_insensitive(lib):
+    rows = brain.filter_cats(brain.load(lib), ["WILD"])
+    assert {r["technique_name"] for r in rows} == {"Quantum Superposition", "Anti-Solution"}
+
+
+def test_filter_none_returns_all(lib):
+    assert len(brain.filter_cats(brain.load(lib), None)) == 4
+
+
+def test_find_hits_and_misses(lib):
+    found, missing = brain.find(brain.load(lib), ["scamper method", "Nope"])
+    assert [r["technique_name"] for r in found] == ["SCAMPER Method"]
+    assert missing == ["Nope"]
+
+
+def test_resolve_detail_present(lib):
+    row = next(r for r in brain.load(lib) if r["detail"])
+    assert "multi-step instructions" in brain.resolve_detail(row, lib.parent)
+
+
+def test_resolve_detail_absent_is_none(lib):
+    row = next(r for r in brain.load(lib) if not r["detail"])
+    assert brain.resolve_detail(row, lib.parent) is None
+
+
+def test_resolve_detail_missing_file_warns_not_fatal(lib, capsys):
+    rows = brain.load(lib)
+    rows[1]["detail"] = "techniques/gone.md"
+    assert brain.resolve_detail(rows[1], lib.parent) is None
+    assert "not found" in capsys.readouterr().err
+
+
+def test_show_inlines_detail(lib, capsys):
+    assert brain.main(["--file", str(lib), "show", "Quantum Superposition"]) == 0
+    out = capsys.readouterr().out
+    assert "multi-step instructions" in out and "[wild]" in out
+
+
+def test_show_simple_has_no_detail(lib, capsys):
+    brain.main(["--file", str(lib), "show", "SCAMPER Method"])
+    out = capsys.readouterr().out
+    assert "transformation lenses" in out
+
+
+def test_show_all_missing_returns_1(lib):
+    assert brain.main(["--file", str(lib), "show", "Ghost"]) == 1
+
+
+def test_list_filtered_text(lib, capsys):
+    brain.main(["--file", str(lib), "list", "--category", "structured"])
+    out = capsys.readouterr().out.strip().splitlines()
+    assert len(out) == 1 and out[0].startswith("structured\tSCAMPER Method\t")
+
+
+def test_list_bare_is_refused(lib, capsys):
+    # the footgun: bare `list` must NOT dump the catalog into context
+    assert brain.main(["--file", str(lib), "list"]) == 2
+    captured = capsys.readouterr()
+    assert captured.out == ""  # nothing leaked to stdout
+    assert "--category" in captured.err and "--all" in captured.err
+
+
+def test_list_all_dumps_everything(lib, capsys):
+    assert brain.main(["--file", str(lib), "list", "--all"]) == 0
+    out = capsys.readouterr().out.strip().splitlines()
+    assert len(out) == 4  # the deliberate full-catalog escape hatch
+
+
+def test_json_output(lib, capsys):
+    import json
+    brain.main(["--file", str(lib), "--json", "categories"])
+    data = json.loads(capsys.readouterr().out)
+    assert {"category": "wild", "count": 2} in data
+
+
+def test_random_respects_n_and_category(lib, capsys):
+    brain.main(["--file", str(lib), "random", "--category", "wild", "-n", "5"])
+    lines = capsys.readouterr().out.strip().splitlines()
+    assert len(lines) == 2  # only 2 wild exist, n capped
+    assert all(line.startswith("wild\t") for line in lines)
+
+
+def test_random_negative_n_does_not_crash(lib, capsys):
+    # a negative -n is clamped to 0, not passed to random.sample (which would raise)
+    assert brain.main(["--file", str(lib), "random", "-n", "-1"]) == 0
+    assert capsys.readouterr().out.strip() == ""
+
+
+def test_missing_file_returns_2(tmp_path):
+    assert brain.main(["--file", str(tmp_path / "nope.csv"), "categories"]) == 2
+
+
+# --- html selection page ------------------------------------------------
+
+def test_html_requires_out(lib, capsys):
+    # never dump the catalog to stdout — writing to a file is the whole point
+    assert brain.main(["--file", str(lib), "html"]) == 2
+    assert "--out" in capsys.readouterr().err
+
+
+def test_html_writes_selection_page(lib, tmp_path):
+    out = tmp_path / "sel.html"
+    assert brain.main(["--file", str(lib), "html", "--out", str(out)]) == 0
+    doc = out.read_text(encoding="utf-8")
+    assert doc.startswith("<!DOCTYPE html>")
+    assert "BMad Method Brainstorming Selection" in doc
+    for r in brain.load(lib):
+        assert r["technique_name"] in doc  # every technique is selectable
+    assert "&quot;yes and&quot;" in doc  # quotes in a description are escaped, not raw
+
+
+def test_html_creates_missing_parent(lib, tmp_path):
+    out = tmp_path / "nested" / "deep" / "sel.html"
+    assert brain.main(["--file", str(lib), "html", "--out", str(out)]) == 0
+    assert out.is_file()
+
+
+# --- --extra overlay (customize.toml additional_techniques) -------------
+
+EXTRA = (
+    '[{"category": "domain-specific", "technique_name": "Regulatory Inversion", '
+    '"description": "Start from the compliance constraint and brainstorm what it unlocks."}, '
+    '{"category": "wild", "technique_name": "Extra Wild One", "description": "An added wild method."}]'
+)
+
+
+@pytest.fixture
+def extra(tmp_path):
+    p = tmp_path / "extra.json"
+    p.write_text(EXTRA, encoding="utf-8")
+    return p
+
+
+def test_extra_merges_into_categories(lib, extra, capsys):
+    brain.main(["--file", str(lib), "--extra", str(extra), "categories"])
+    out = capsys.readouterr().out
+    assert "domain-specific\t1" in out  # a brand-new category appears
+    assert "wild\t3" in out  # the extra wild one is counted alongside the shipped two
+
+
+def test_extra_appears_in_list_and_random(lib, extra, capsys):
+    brain.main(["--file", str(lib), "--extra", str(extra), "list", "--category", "domain-specific"])
+    assert "Regulatory Inversion" in capsys.readouterr().out
+
+
+def test_extra_replaces_shipped_row_by_name(lib, extra, tmp_path, capsys):
+    shipped = brain.load(Path(lib))[0]
+    overlay = tmp_path / "replace.json"
+    overlay.write_text(
+        json.dumps([{"category": shipped["category"], "technique_name": shipped["technique_name"], "description": "RETUNED"}]),
+        encoding="utf-8",
+    )
+    brain.main(["--file", str(lib), "--extra", str(overlay), "list", "--all"])
+    out = capsys.readouterr().out
+    assert "RETUNED" in out
+    assert out.count(shipped["technique_name"]) == 1  # replaced, not duplicated
+
+
+def test_extra_malformed_exits_cleanly(lib, tmp_path, capsys):
+    bad = tmp_path / "bad.json"
+    for content in ('{not json', '{"a": 1}', '["not-an-object"]'):
+        bad.write_text(content, encoding="utf-8")
+        assert brain.main(["--file", str(lib), "--extra", str(bad), "categories"]) == 2
+        assert "could not read --extra" in capsys.readouterr().err
+
+
+def test_extra_is_first_class_in_html(lib, extra, tmp_path):
+    out = tmp_path / "sel.html"
+    assert brain.main(["--file", str(lib), "--extra", str(extra), "html", "--out", str(out)]) == 0
+    doc = out.read_text(encoding="utf-8")
+    # custom technique is selectable and its new category renders without crashing (fallback glyph/hue)
+    assert "Regulatory Inversion" in doc
+    assert "Domain Specific" in doc
+
+
+def test_extra_missing_file_returns_2(lib, tmp_path):
+    assert brain.main(["--file", str(lib), "--extra", str(tmp_path / "nope.json"), "categories"]) == 2
+
+
+def test_unknown_category_style_uses_fallback_glyph():
+    hue, glyph = brain.category_style("totally-made-up-category")
+    assert hue.startswith("#") and len(hue) == 7  # valid derived hex
+    assert glyph == brain._FALLBACK_GLYPH
+
+
+def test_shipped_selector_is_in_sync_with_catalog():
+    # foolproofing: if someone edits brain-methods.csv they must regenerate the page.
+    # Regenerate with: uv run brain.py html --out assets/brain-selector.html
+    asset = brain.DEFAULT_FILE.parent / "brain-selector.html"
+    assert asset.is_file(), "missing assets/brain-selector.html — generate it"
+    expected = brain.html_doc(brain.load(brain.DEFAULT_FILE))
+    assert asset.read_text(encoding="utf-8") == expected, (
+        "assets/brain-selector.html is stale; regenerate: "
+        "uv run brain.py html --out assets/brain-selector.html"
+    )
diff --git a/.agents/skills/bmad-build-auto/SKILL.md b/.agents/skills/bmad-build-auto/SKILL.md
new file mode 100644
index 00000000..193d367c
--- /dev/null
+++ b/.agents/skills/bmad-build-auto/SKILL.md
@@ -0,0 +1,13 @@
+---
+name: bmad-build-auto
+description: 'One iteration of an unattended development loop. Use when invoked by name.'
+---
+
+Run the following command exactly once without changing the current working directory. Replace `{project-root}` with the absolute path to the project root and `{skill-root}` with the absolute path to this skill's directory:
+
+```bash
+uv run --no-cache "{project-root}/_bmad/scripts/render_skill.py" --project-root "{project-root}" --skill "{skill-root}"
+```
+
+- On success, read and follow the one absolute `workflow.md` instruction printed to stdout.
+- On failure (including `uv` being unavailable), report the command output and HALT. Do not run any workflow source directly.
diff --git a/.agents/skills/bmad-build-auto/compile-epic-context.md b/.agents/skills/bmad-build-auto/compile-epic-context.md
new file mode 100644
index 00000000..d920761e
--- /dev/null
+++ b/.agents/skills/bmad-build-auto/compile-epic-context.md
@@ -0,0 +1,62 @@
+# Compile Epic Context
+
+**Task**
+Given an epic number, the epics file, the planning artifacts directory, and a desired output path, compile a clean, focused, developer-ready context file (`epic-<N>-context.md`).
+
+**Steps**
+
+1. Read the epics file and extract the target epic's title, goal, and list of stories.
+2. Scan the planning artifacts directory for the standard files (PRD, architecture, UX/design, product brief).
+3. Pull only the information relevant to this epic.
+4. Write the compiled context to the exact output path using the format below.
+
+## Exact Output Format
+
+Use these headings:
+
+```markdown
+# Epic {N} Context: {Epic Title}
+
+<!-- Generated from planning artifacts. Regenerate with compile-epic-context if planning docs change. -->
+
+## Goal
+
+{One clear paragraph: what this epic achieves and why it matters.}
+
+## Stories
+
+- Story X.Y: Brief title only
+- ...
+
+## Requirements & Constraints
+
+{Relevant functional/non-functional requirements and success criteria for this epic (describe by purpose, not source).}
+
+## Technical Decisions
+
+{Key architecture decisions, constraints, patterns, data models, and conventions relevant to this epic.}
+
+## UX & Interaction Patterns
+
+{Relevant UX flows, interaction patterns, and design constraints (omit section entirely if nothing relevant).}
+
+## Cross-Story Dependencies
+
+{Dependencies between stories in this epic or with other epics/systems (omit if none).}
+```
+
+## Rules
+
+- **Scope aggressively.** Include only what a developer working on any story in this epic actually needs. When in doubt, leave it out — the developer can always read the full planning doc.
+- **Describe by purpose, not by source.** Write "API responses must include pagination metadata" not "Per PRD section 3.2.1, pagination is required." Planning doc internals will change; the constraint won't.
+- **No full copies.** Never quote source documents, section numbers, or paste large blocks verbatim. Always distill.
+- **No story-level details.** The story list is for orientation only. Individual story specs handle the details.
+- **Nothing derivable from the codebase.** Don't document what a developer can learn by reading the code.
+- **Be concise and actionable.** Target 800–1500 tokens total. This file loads into bmad-build-auto's context alongside other material.
+- **Never hallucinate content.** If source material doesn't say something, don't invent it.
+- **Omit empty sections entirely**, except Goal and Stories, which are always required.
+
+## Error handling
+
+- **If the epics file is missing or the target epic is not found:** write nothing and report the problem to the calling agent. Goal and Stories cannot be populated without a usable epics file.
+- **If planning artifacts are missing or empty:** still produce the file with Goal and Stories populated from the epics file. Under Requirements & Constraints, write: "Planning artifacts were unavailable; only epics-file context was used." Never hallucinate content to fill missing sections.
diff --git a/.agents/skills/bmad-build-auto/customize.toml b/.agents/skills/bmad-build-auto/customize.toml
new file mode 100644
index 00000000..a9553bb8
--- /dev/null
+++ b/.agents/skills/bmad-build-auto/customize.toml
@@ -0,0 +1,121 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Default customization values for bmad-build-auto.
+# Override in _bmad/custom/bmad-build-auto.toml or
+# _bmad/custom/bmad-build-auto.user.toml.
+#
+# Merge rules:
+# - Strings replace the default.
+# - Lists append to the default list.
+# - Tables merge key by key.
+# - Arrays of tables merge by `id`: matching `id` replaces, new `id`s append.
+
+[workflow]
+
+# Extra instructions to run before config is loaded.
+
+activation_steps_prepend = []
+
+# Extra instructions to run after config is loaded and before step 01.
+
+activation_steps_append = []
+
+# Facts kept in context for the whole run.
+# Entries are literal text or file references prefixed with "file:".
+# File entries may use globs and are loaded during activation.
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Instruction run by HALT after writing the terminal result.
+# Empty means no extra terminal behavior.
+
+on_complete = ""
+
+# Handoff for the implementation subagent in step 03. The whole execution
+# recipe — a subagent by default, but an override may run it any other way
+# (a different model, an external coding tool via bash). {spec_file} is
+# substituted at run time.
+
+implementation_handoff = """
+Launch a subagent with no prior conversation context, with this prompt:
+
+> Read {spec_file} fully and implement it — the spec is the sole source of truth. Load every file listed in its frontmatter `context:` before you start.
+>
+> When done, report what you changed, how you verified it, and anything left incomplete or risky.
+"""
+
+# Review layers for the review step. `instruction` is the layer's whole
+# execution recipe — subagents by default, but an override may run anything
+# (e.g. an external reviewer via bash). {diff_output} is substituted at run
+# time. `when` (optional) gates a layer; empty `instruction` disables it.
+
+[[workflow.review_layers]]
+id = "blind-hunter"
+name = "Blind Hunter"
+instruction = """
+Launch a context-free subagent with this prompt:
+
+Conduct a review of CONTENT.
+Look for what's missing, not only what's wrong.
+Find at least ten issues to fix or improve.
+Output a Markdown list of findings only — no severity, priority, or ranking.
+If the content is empty, stop and say so.
+If you have zero findings, re-check and keep thinking; do not stop with an empty list.
+
+CONTENT:
+{diff_output}
+
+Do not invoke any skill. Return only the review result.
+
+"""
+
+[[workflow.review_layers]]
+id = "edge-case-hunter"
+name = "Edge Case Hunter"
+instruction = """
+Launch a context-free subagent with this prompt:
+
+Read `{skill-root}/review-prompts/edge-case-hunter.md` completely and follow it as your review instructions.
+
+Review content:
+
+{diff_output}
+
+Do not invoke any skill. If the instruction file is unreadable, report that exact failure and stop. Return only the review result.
+
+"""
+
+[[workflow.review_layers]]
+id = "verification-gap"
+name = "Verification Gap Reviewer"
+instruction = """
+Launch a context-free subagent with this prompt:
+
+Read `{skill-root}/review-prompts/verification-gap.md` completely and follow it as your review instructions.
+
+Review content:
+
+{diff_output}
+
+Do not invoke any skill. If the instruction file is unreadable, report that exact failure and stop. Return only the review result.
+
+"""
+
+[[workflow.review_layers]]
+id = "intent-alignment"
+name = "Intent Alignment Auditor"
+instruction = """
+Launch a subagent with this prompt:
+
+You are an intent-alignment auditor. You have no other context about how this change was produced. Here is the verbatim intent this work started from:
+
+{verbatim_intent}
+
+Here is the diff:
+
+{diff_output}
+
+Your task is strictly descriptive — do not prescribe additional work. Report: (1) the defensible readings of the intent, enumerated; (2) which reading this diff implements; (3) where the readings and the diff diverge — specifically, which surface the intent's expectations live at versus which surface the diff's changes and its tests exercise.
+"""
diff --git a/.agents/skills/bmad-build-auto/references/deletion-check.md b/.agents/skills/bmad-build-auto/references/deletion-check.md
new file mode 100644
index 00000000..5cc3d771
--- /dev/null
+++ b/.agents/skills/bmad-build-auto/references/deletion-check.md
@@ -0,0 +1,14 @@
+# Deletion Check
+
+Secondary pass for the Edge Case Hunter — runs only when the diff removed meaningful code. Subordinate to the edge-case pass; findings are usually few or none.
+
+For each chunk of removed or replaced code (ignore pure renames and whitespace), ask: did it carry behavior or a contract that the change neither re-established nor intentionally retired? Add a finding for any resulting regression, orphaned reference, or newly-dead code. Skip anything already covered by your edge-case findings.
+
+Append each finding to the same JSON array as the edge-case findings, with the four standard fields plus:
+
+- `kind`: `"deletion"`
+- `confidence`: `"high"`, `"medium"`, or `"low"` — these are inferences; rate them
+
+For a deletion finding the standard fields read as: `location` = the removed item; `trigger_condition` = the behavior or contract it enforced; `guard_snippet` = where or how to re-establish it; `potential_consequence` = the regression or orphan.
+
+Add nothing if nothing qualifies.
diff --git a/.agents/skills/bmad-build-auto/review-prompts/edge-case-hunter.md b/.agents/skills/bmad-build-auto/review-prompts/edge-case-hunter.md
new file mode 100644
index 00000000..11df57e9
--- /dev/null
+++ b/.agents/skills/bmad-build-auto/review-prompts/edge-case-hunter.md
@@ -0,0 +1,88 @@
+# Edge Case Hunter Review
+
+**Goal:** You are a pure path tracer. Never comment on whether code is good or bad; only list missing handling.
+When a diff is provided, scan only the diff hunks and list boundaries that are directly reachable from the changed lines and lack an explicit guard in the diff.
+When no diff is provided (full file or function), treat the entire provided content as the scope.
+Ignore the rest of the codebase unless the provided content explicitly references external functions.
+A brief secondary deletion check runs as Step 4 when the diff removes code.
+
+**Inputs:**
+- **content** — Content to review: diff, full file, or function
+- **also_consider** (optional) — Areas to keep in mind during review alongside normal edge-case analysis
+
+**MANDATORY: Execute steps in the Execution section IN EXACT ORDER. DO NOT skip steps or change the sequence. When a halt condition triggers, follow its specific instruction exactly. Each action within a step is a REQUIRED action to complete that step.**
+
+**Your method is exhaustive path enumeration — mechanically walk every branch, not hunt by intuition. Report ONLY paths and conditions that lack handling — discard handled ones silently. Do NOT editorialize or add filler. Do not assign severity labels, rankings, or priority levels.**
+
+
+## EXECUTION
+
+### Step 1: Receive Content
+
+- Load the content to review strictly from the parent message that launched you (not from this instruction file)
+- If content is empty, or cannot be decoded as text, return `[{"location":"N/A","trigger_condition":"Input empty or undecodable","guard_snippet":"Provide valid content to review","potential_consequence":"Review skipped — no analysis performed"}]` and stop
+- Identify content type (diff, full file, or function) to determine scope rules
+
+### Step 2: Exhaustive Path Analysis
+
+**Walk every branching path and boundary condition within scope — report only unhandled ones.**
+
+- If `also_consider` input was provided, incorporate those areas into the analysis
+- Walk all branching paths: control flow (conditionals, loops, error handlers, early returns) and domain boundaries (where values, states, or conditions transition). Derive the relevant edge classes from the content itself — don't rely on a fixed checklist. Examples: missing else/default, unguarded inputs, off-by-one loops, arithmetic overflow, implicit type coercion, race conditions, timeout gaps
+- Consider implicit branches: the diff special-cases or changes the handling of one or more members of a fixed set of values — enums, status codes, sentinels, type tags, flags, value ranges. The rest of the set is implicit branches (e.g. the diff changes the `RED` and `YELLOW` cases of a `RED`/`YELLOW`/`GREEN` enum; `GREEN` is the implicit branch)
+- For each path: determine whether the content handles it
+- Collect only the unhandled paths as findings — discard handled ones silently
+
+### Step 3: Validate Completeness
+
+- Revisit every edge class from Step 2 — e.g., missing else/default, null/empty inputs, off-by-one loops, arithmetic overflow, implicit type coercion, race conditions, timeout gaps
+- Add any newly found unhandled paths to findings; discard confirmed-handled ones
+
+### Step 4: Deletion Check
+
+If the diff removed or replaced meaningful code (ignore pure renames and whitespace): load `references/deletion-check.md` and follow it.
+
+### Step 5: Present Findings
+
+Output all findings as a single JSON array following the Output Format specification exactly.
+
+
+## OUTPUT FORMAT
+
+Return ONLY a valid JSON array of objects. Each edge-case finding contains exactly these four fields:
+
+```json
+[{
+  "location": "file:start-end (or file:line when single line, or file:hunk when exact line unavailable)",
+  "trigger_condition": "one-line description (max 15 words)",
+  "guard_snippet": "minimal code sketch that closes the gap (single-line escaped string, no raw newlines or unescaped quotes)",
+  "potential_consequence": "what could actually go wrong (max 15 words)"
+}]
+```
+
+No extra text, no explanations, no markdown wrapping. An empty array `[]` is valid when nothing is found. Deletion findings from Step 4, if any, go in the same array with the extra fields defined in `references/deletion-check.md`.
+
+
+## HALT CONDITIONS
+
+- If content is empty or cannot be decoded as text, return `[{"location":"N/A","trigger_condition":"Input empty or undecodable","guard_snippet":"Provide valid content to review","potential_consequence":"Review skipped — no analysis performed"}]` and stop
+<reference path="references/deletion-check.md">
+# Deletion Check
+
+Secondary pass for the Edge Case Hunter — runs only when the diff removed meaningful code. Subordinate to the edge-case pass; findings are usually few or none.
+
+For each chunk of removed or replaced code (ignore pure renames and whitespace), ask: did it carry behavior or a contract that the change neither re-established nor intentionally retired? Add a finding for any resulting regression, orphaned reference, or newly-dead code. Skip anything already covered by your edge-case findings.
+
+Append each finding to the same JSON array as the edge-case findings, with the four standard fields plus:
+
+- `kind`: `"deletion"`
+- `confidence`: `"high"`, `"medium"`, or `"low"` — these are inferences; rate them
+
+For a deletion finding the standard fields read as: `location` = the removed item; `trigger_condition` = the behavior or contract it enforced; `guard_snippet` = where or how to re-establish it; `potential_consequence` = the regression or orphan.
+
+Add nothing if nothing qualifies.
+</reference>
+
+## CONTENT SOURCE
+
+Review the content supplied under "Review content:" in the message that launched you.
diff --git a/.agents/skills/bmad-build-auto/review-prompts/verification-gap.md b/.agents/skills/bmad-build-auto/review-prompts/verification-gap.md
new file mode 100644
index 00000000..4adc52d5
--- /dev/null
+++ b/.agents/skills/bmad-build-auto/review-prompts/verification-gap.md
@@ -0,0 +1,113 @@
+# Verification Gap Review
+
+**Goal:** Find changed behavior that could break without reliable verification catching it. Ask one question — "if the behavior this change is supposed to produce broke where it's actually used, would verification fail?" Do not hunt for correctness bugs, but report genuine problems you notice while tracing verification.
+
+The main verification gap shapes are:
+
+1. **Regression gap:** the changed code regresses where it's used, and no test covering that use would fail.
+2. **Missing-adoption gap:** a place that should now use the new behavior doesn't; it handles the same case its own way, or not at all, and no test would flag the omission.
+3. **Broken-verification gap:** a test appears to cover the changed behavior, but would not actually protect it because it is skipped, flaky, not run in the normal verification path, or too weak to observe the regression.
+
+## Evidence Rules
+
+- Read a test before claiming what it covers, runs, asserts, or misses.
+- Before claiming no test exists, search the whole repo by the symbol under test and by import references; expected file locations are not enough.
+- Never assert what you did not verify. If a finding cannot be grounded, drop it.
+- In a finding, say what you actually checked — "none of the tests I read cover this" — and show how far you looked. Say a test doesn't exist anywhere only when the symbol/import-reference search actually shows that.
+- Do not assign severity, confidence, priority, or ranking.
+
+## Review Sequence
+
+### Step 1: Screen for behavioral change
+
+Screen each part of the change separately. If a part is non-behavioral, skip it. Call a part non-behavioral only when the changed code does not alter return values, thrown errors, caller-visible side effects, or observable state (including iteration order and emitted messages). Once a part meets that test, move on; do not inspect callers or tests for extra confirmation.
+
+Common non-behavioral examples: formatting, comments, whitespace; pure renames; trivial getters/setters and pass-throughs; type-only or compiler-enforced changes with no runtime effect; etc.
+
+Only outcomes produced by deterministic code are worth automatically testing; tests are useless on static source text and brittle on LLM output. Skip those parts.
+
+If every part is skipped, output the clean result (see Output Format).
+
+### Step 2: Find the behavior that changed
+
+Identify what behavior changed compared to the previous version: output, side effect, branch, error path, schema/event shape, config default, validation/authorization rule, external contract, etc. If the change affects more than one behavior, handle each separately.
+
+Treat broad-impact changes as behavioral even when no single changed line looks important: dependency, toolchain, build/config, data-file, etc.
+
+### Step 3: Trace where that behavior is used
+
+Trace the changed behavior to the places that observe it. Start with direct callers and registered entry points (routes, commands, DI), contract consumers (schemas, events, APIs, database readers), and reverse-dependency info if already available.
+
+Follow a path only while the changed behavior is reachable and unverified. Stop when a test at that boundary would fail, the consumer does not observe the changed behavior, or the next hop is guesswork (dynamic dispatch, reflection, outside-repo consumers, etc.). Prefer the nearest observable boundary, often one to three hops away, especially across contract, integration, or service edges. If there are more than five similar consumers, group obvious repeats and check representative paths; expand only when a consumer observes the behavior differently.
+
+### Step 4: Qualify the consumer, then check its test
+
+For each consumer, name the smallest realistic regression this consumer would observe: invert the branch, drop the default, omit the field, return the old error code, skip the integration call, etc. This is the Demonstration. If no such regression exists, drop the path; untested downstream code is not a finding.
+
+A `Missing-adoption gap` qualifies not by the adoption failure alone but by a supersession signal: the change gives clear evidence the new behavior is meant to replace the local one — PR intent, naming or docs, a replaced sibling site, deleted duplicate logic, or a test defining the new rule — and the local site shares the same observable contract. Without a supersession signal and a shared observable contract, it is a refactor suggestion, not a verification-gap finding. Once both hold, check whether any test for that site would flag the non-adoption; missing coverage of the non-adoption is the gap itself, not a disqualifier.
+
+Find and read the relevant test. Ask whether the Demonstration would make an assertion fail.
+
+- If yes, the behavior is verified. No finding.
+- For a regression-style Demonstration: if no test runs the path, the test is skipped/flaky/not run normally, or the test runs the code without checking the changed result, report a `Regression gap` or `Broken-verification gap`.
+- For a qualifying Missing-adoption case: if none of the site tests you found assert it adopts the new behavior, report a `Missing-adoption gap`.
+
+A test counts only if it runs normally and an assertion observes the changed output, branch, or contract. These do not count: no execution; source-text assertions that match a file's wording instead of running it; success/no-throw/snapshot-only checks; mock/log-call checks; human-only checks; tests that mock away the integration; e2e tests that pass through without checking the changed output; stale assertions or fixtures.
+
+For example, `expect(x ?? DEFAULT).toBe(DEFAULT)` passes when `x` is missing.
+
+Common patterns:
+
+- **Caller-path gap** — helper test covers the branch, but caller values skip it.
+- **Contract drift** — payload/schema/event changes must be verified at the consumer.
+- **Migration compatibility** — tests only create new-format rows or fresh schemas.
+- **Phantom exception** — handled partial-failure path has no test.
+- **Missing-adoption gap** — sibling site should use the new rule/helper and does not.
+- **Removed verification** — deleted test or weakened assertion leaves behavior unpinned; removing a source-text assertion is not this, since it never counted.
+
+### Step 5: Confirm each finding is real
+
+Before writing a finding, re-open the specific tests or search results the finding relies on. Verify the Demonstration would not make any test you checked fail, or that the absence claim is backed by the symbol/import-reference search. Do not claim more than you verified; drop any finding you cannot ground.
+
+Explain why the test misses the bug using what the test sets up and checks.
+
+Do not report: compiler/type-checker-enforced cases; behavior already verified by an integration, contract, or e2e test; implementation-detail or mock-only tests; low coverage or a missing test file by itself; legacy untested code the change did not affect.
+
+Report genuine problems you noticed while tracing verification, even if they are not verification gaps. Put them under `Other findings` in the output. This permits reporting what you already reached, not extra hunting.
+
+## OUTPUT FORMAT
+
+Emit each verification-gap finding as one block. No general advice, no severity or confidence.
+
+```markdown
+### <one-line title naming the gap>
+
+- **Changed surface:** the exact behavior or contract that changed — `file:line`.
+- **Impacted consumer or site:** named concretely with `file:line` (e.g. "the `createInvoice` mutation used by the billing dashboard at `billing/dashboard.ts:88`," not "callers of this function").
+- **Existing test evidence:**
+  - `Regression gap`: what the relevant test actually asserts, with `file:line`; or, if none, the symbol/import-reference searches run and their result.
+  - `Missing-adoption gap`: tests for the impacted site, and whether any assert it adopts the new behavior.
+  - `Broken-verification gap`: the apparent test or verification path, and why it does not count.
+- **Missing verification:** the precise assertion or check that's absent.
+- **Demonstration:**
+  - `Regression gap` / `Broken-verification gap`: the concrete regression that would ship undetected, and why the tests you checked would not fail.
+  - `Missing-adoption gap`: the case the site mishandles by not adopting the new behavior, and that none of the tests you read assert adoption.
+- **Consequence:** the concrete thing that ships wrong — a regression the checked evidence would not catch, or a site that should use the new behavior and doesn't.
+- **Suggested test shape:** (optional) the kind of test that would close the gap, fit to the repo's own way of verifying — don't impose a generic test pyramid.
+```
+
+If you noticed genuine non-gap problems while tracing verification, append:
+
+```markdown
+## Other findings
+
+- <description only; no severity, confidence, priority, or ranking>
+```
+
+When you find no verification gaps and no other findings, output exactly this single line, not an empty response:
+
+`No verification gaps found.`
+
+## CONTENT SOURCE
+
+Review the content supplied under "Review content:" in the message that launched you. If none is supplied, stop with exactly: `No verification gaps found.`
diff --git a/.agents/skills/bmad-build-auto/spec-template.md b/.agents/skills/bmad-build-auto/spec-template.md
new file mode 100644
index 00000000..07178f98
--- /dev/null
+++ b/.agents/skills/bmad-build-auto/spec-template.md
@@ -0,0 +1,99 @@
+---
+title: '{title}'
+type: 'feature' # feature | bugfix | refactor | chore
+created: '{date}'
+status: 'draft' # draft | ready-for-dev | in-progress | in-review | done | blocked
+review_loop_iteration: 0 # incremented by step-04 before each review loopback
+followup_review_recommended: false # set by step-04 on status: done — true if the LLM decided another review pass is worthwhile
+context: [] # optional: `{project-root}/`-prefixed paths to project-wide standards/docs the implementation agent should load. Keep short — only what isn't already distilled into the spec body.
+warnings: [] # optional: machine-readable warnings for orchestration, e.g. oversized, multiple-goals
+deferred: [] # append-only machine-readable deferred review findings; each item carries summary/evidence and optional location/severity
+---
+
+<!-- Aim for 900–1600 tokens. If larger, add `oversized` to frontmatter `warnings` and continue.
+     Never over-specify "how" — use boundaries + examples instead.
+     Cohesive cross-layer stories (DB+BE+UI) stay in ONE file.
+     IMPORTANT: Remove all HTML comments when filling this template. -->
+
+<intent-contract>
+
+## Intent
+
+<!-- What is broken or missing, and why it matters. Then the high-level approach — the "what", not the "how". -->
+
+**Problem:** ONE_TO_TWO_SENTENCES
+
+**Approach:** ONE_TO_TWO_SENTENCES
+
+## Boundaries & Constraints
+
+<!-- Three tiers: Always = invariant rules. Block If = decisions that cannot be made unattended. Never = out of scope + forbidden approaches. -->
+
+**Always:** INVARIANT_RULES
+
+**Block If:** DECISIONS_REQUIRING_HUMAN_INPUT
+<!-- Agent: if any of these trigger during execution, HALT with status blocked and the blocking condition. -->
+
+**Never:** NON_GOALS_AND_FORBIDDEN_APPROACHES
+
+## I/O & Edge-Case Matrix
+
+<!-- If no meaningful I/O scenarios exist, DELETE THIS ENTIRE SECTION. Do not write "N/A" or "None". -->
+
+| Scenario | Input / State | Expected Output / Behavior | Error Handling |
+|----------|--------------|---------------------------|----------------|
+| HAPPY_PATH | INPUT | OUTCOME | No error expected |
+| ERROR_CASE | INPUT | OUTCOME | ERROR_HANDLING |
+
+</intent-contract>
+
+## Code Map
+
+<!-- Agent-populated during planning. Annotated paths prevent blind codebase searching. -->
+
+- `FILE` -- ROLE_OR_RELEVANCE
+- `FILE` -- ROLE_OR_RELEVANCE
+
+## Tasks & Acceptance
+
+<!-- Tasks: backtick-quoted file path -- action -- rationale. Prefer one task per file; group tightly-coupled changes when splitting would be artificial. -->
+<!-- If an I/O Matrix is present, include a task to unit-test its edge cases. -->
+<!-- AC covers system-level behaviors not captured by the I/O Matrix. Do not duplicate I/O scenarios here. -->
+
+**Execution:**
+- `FILE` -- ACTION -- RATIONALE
+
+**Acceptance Criteria:**
+- Given PRECONDITION, when ACTION, then EXPECTED_RESULT
+
+## Spec Change Log
+
+<!-- Append-only. Populated by step-04 during review loops. Do not modify or delete existing entries.
+     Each entry records: what finding triggered the change, what was amended, what known-bad state
+     the amendment avoids, and any KEEP instructions (what worked well and must survive re-derivation).
+     Empty until the first bad_spec loopback. -->
+
+## Review Triage Log
+
+<!-- Append-only. Populated by step-04 on EVERY review pass, including loopbacks and blocked exits.
+     Each entry records triage decision counts for intent_gap, bad_spec, patch, defer, and reject,
+     with per-category severity breakdowns using low/medium/high, plus the findings addressed in
+     that pass. Empty until the first review pass. -->
+
+## Design Notes
+
+<!-- If the approach is straightforward, DELETE THIS ENTIRE SECTION. Do not write "N/A" or "None". -->
+<!-- Design rationale and golden examples only when non-obvious. Keep examples to 5–10 lines. -->
+
+DESIGN_RATIONALE_AND_EXAMPLES
+
+## Verification
+
+<!-- If no build, test, or lint commands apply, DELETE THIS ENTIRE SECTION. Do not write "N/A" or "None". -->
+<!-- How the agent confirms its own work. Prefer CLI commands. When no CLI check applies, state what to inspect manually. -->
+
+**Commands:**
+- `COMMAND` -- expected: SUCCESS_CRITERIA
+
+**Manual checks (if no CLI):**
+- WHAT_TO_INSPECT_AND_EXPECTED_STATE
diff --git a/.agents/skills/bmad-build-auto/step-01-clarify-and-route.md b/.agents/skills/bmad-build-auto/step-01-clarify-and-route.md
new file mode 100644
index 00000000..87d69298
--- /dev/null
+++ b/.agents/skills/bmad-build-auto/step-01-clarify-and-route.md
@@ -0,0 +1,83 @@
+---
+spec_file: '' # set at runtime once a route resolves it; some HALT branches exit before it is set
+spec_folder: '' # set at runtime under folder+id dispatch only
+story_id: '' # set at runtime under folder+id dispatch only
+---
+
+# Step 1: Clarify and Route
+
+## RULES
+
+- **Language** — Speak in `{{.communication_language}}`, tailored to `{{.user_skill_level}}`. Write files in `{{.document_output_language}}`.
+- Treat the invocation intent as workflow input, not as a substitute for step-02 investigation and spec generation.
+- **EARLY EXIT** means: stop this step immediately, then read and follow the target file. Return here only if a later step explicitly says to loop back.
+
+## Intent check (do this first)
+
+Use the invocation prompt as the intent.
+
+If the invocation prompt explicitly points to an existing spec file with recognized `status` frontmatter, set `spec_file`, then **EARLY EXIT** to the appropriate step:
+- `draft` → `[[bmad-snapshot:step-02-plan.md]]`
+- `ready-for-dev` or `in-progress` → `[[bmad-snapshot:step-03-implement.md]]`
+- `in-review` → `[[bmad-snapshot:step-04-review.md]]`
+- `blocked` → HALT with status `blocked` and blocking condition `blocked spec supplied`.
+- `done` → set `review_loop_iteration` to `0` in the frontmatter, then **EARLY EXIT** to `[[bmad-snapshot:step-04-review.md]]` for a fresh review pass. (A `done` spec is a completed run, so this starts a follow-up review, not a resumption.)
+
+If the invocation prompt instead supplies a spec folder and a story id, with no specific spec file path, this is a **folder+id dispatch**: set `spec_folder` (a `{project-root}`-relative or absolute path) and `story_id` from the prompt. Any further prompt text (e.g. `invoke_dev_with` guidance the caller appended) is additional planning context to carry into step-02 — not a competing description of what to implement.
+
+Read `{spec_folder}/stories.yaml`. If the file does not exist or fails to parse, HALT with status `blocked` and blocking condition `no stories.yaml found`. Find the entry whose `id` equals `{story_id}`; if none matches, HALT with status `blocked` and blocking condition `story id not found in stories.yaml`. Take only that entry's `title` and `description` — never read the checkpoint fields or `invoke_dev_with`; those are the caller's orchestration fields, not build-auto's.
+
+Look for files matching `{spec_folder}/stories/{story_id}-*.md` (id-prefix match — story ids are prefix-free, so at most one should match):
+- **If more than one matches**, HALT with status `blocked` and blocking condition `ambiguous story file match`.
+- **If exactly one matches**, set `spec_file` to that path.
+  - `draft` (planning was interrupted mid-flight): accumulate cross-story context before resuming — load every other file matching `{spec_folder}/stories/*.md` (every match except `{spec_file}` itself), regardless of `status`, and carry forward each one's **Code Map**, **Design Notes**, **Spec Change Log**, **Tasks & Acceptance** checklist state, and **Auto Run Result** details, where present, as additional planning context for step-02. Then **EARLY EXIT** to `[[bmad-snapshot:step-02-plan.md]]`.
+  - Any other recognized `status`: **EARLY EXIT** using the same routing as above, including the `review_loop_iteration` reset for `done`. One difference: a `blocked` story HALTs with blocking condition `story already blocked`, not `blocked spec supplied` — the caller did not supply this file; build-auto found it by id.
+  - `status` missing or unrecognized: HALT with status `blocked` and blocking condition `unrecognized status in existing story file`.
+- **If none matches**, this is the first dispatch for `{story_id}`. The entry's `title` and `description` are the resolved intent. If `{spec_folder}/SPEC.md` does not exist, HALT with status `blocked` and blocking condition `no epic spec found`. Otherwise load it and the files listed in its `companions:` frontmatter as planning context, then accumulate cross-story context the same way as the `draft` case above — load every file matching `{spec_folder}/stories/*.md` (none yet exists for `{story_id}` at this point, so nothing is excluded), regardless of `status`, carrying forward the same fields, where present, as additional planning context for step-02. Then continue to INSTRUCTIONS item 3 below — not `step-03-implement.md`, item 3 of the numbered list in this file (items 1 and 2 do not apply — context and intent are already resolved; item 1.A.5's previous-story continuity scan in particular never runs here, since folder+id dispatch already skips items 1 and 2 entirely — the cross-story accumulation above is its replacement for this dispatch mode).
+
+One `stories.yaml` entry per invocation: never read another entry, and never advance to a different story id regardless of outcome.
+
+Otherwise, treat the invocation prompt as starting intent. This may be a story ID, ticket ID, file path, short description, or longer free-form intent. Do not infer workflow state from non-spec files.
+If the invocation prompt does not contain enough intent to identify what to implement, HALT with status `blocked` and blocking condition `unclear intent`.
+
+## INSTRUCTIONS
+
+1. Load context.
+   - List files in `{{.planning_artifacts}}` and `{{.implementation_artifacts}}`.
+   - If the invocation prompt points to an unformatted spec or intent file, ingest that file. Do not scan for unrelated intent files.
+   - **Determine context strategy.** Using the intent and the artifact listing, infer whether the current work is a story from an epic. Do not rely on filename patterns or regex — reason about the intent, the listing, and any epics file content together.
+
+     **A) Epic story path** — if the intent is clearly an epic story:
+
+     1. Identify the epic number `{epic_num}` and (if present) the story number `{story_num}`. If you can't identify an epic number, use path B.
+
+     2. **Check for a valid cached epic context.** Look for `{{.implementation_artifacts}}/epic-<N>-context.md` (where `<N>` is the epic number). A file is **valid** when it exists, is non-empty, starts with `# Epic <N> Context:` (with the correct epic number), and no file in `{{.planning_artifacts}}` is newer.
+        - **If valid:** load it as the primary planning context. Do not load raw planning docs (PRD, architecture, UX, etc.).
+        - **If missing, empty, or invalid:** compile it in the next bullet.
+
+     3. **Compile epic context if needed.** If no valid cached epic context was loaded, produce `{{.implementation_artifacts}}/epic-<N>-context.md` by spawning a subagent synchronously with `[[bmad-snapshot:compile-epic-context.md]]` as its prompt. Pass it the epic number, epics file path, `{{.planning_artifacts}}`, and output path `{{.implementation_artifacts}}/epic-<N>-context.md`.
+
+     4. **Verify if compiled.** If epic context was compiled, verify the output file exists, is non-empty, and starts with `# Epic <N> Context:`. If valid, load it. If verification fails, HALT with status `blocked` and blocking condition `context compilation verification failed`.
+
+     5. **Previous story continuity.** Regardless of which context source succeeded above, scan `{{.implementation_artifacts}}` for specs from the same epic with `status: done` and a lower story number. Load the most recent one (highest story number below current). Extract its **Code Map**, **Design Notes**, **Spec Change Log**, and **task list** as continuity context for step-02 planning. If no `done` spec is found but an `in-review` spec exists for the same epic with a lower story number, HALT with status `blocked` and blocking condition `missing previous-story continuity decision`.
+
+     **B) Freeform path** — if the intent is not an epic story:
+     - Planning artifacts are the output of BMAD phases 1-3. Typical files include:
+       - **PRD** (`*prd*`) — product requirements and success criteria
+       - **Architecture** (`*architecture*`) — technical design decisions and constraints
+       - **UX/Design** (`*ux*`) — user experience and interaction design
+       - **Epics** (`*epic*`) — feature breakdown into implementable stories
+       - **Product Brief** (`*brief*`) — project vision and scope
+     - Scan the listing for files matching these patterns. If any look relevant to the current intent, load them selectively — you don't need all of them, but you need the right constraints and requirements rather than guessing from code alone.
+2. Resolve intent from the invocation prompt and loaded artifacts. Do not fantasize or leave open questions. If the intent cannot be resolved, HALT with status `blocked` and the unresolved questions as blocking condition.
+3. Version control sanity check. If version control is unavailable, skip this check. Otherwise require a clean working tree, a branch that fits the intent, and writable repository metadata. For Git, run `git add --refresh -- .`, then confirm the tree is still clean; on failure or change, HALT with status `blocked` and blocking condition `version-control metadata not writable`. Under folder+id dispatch, judge the branch against the epic, not the story. HALT on a dirty tree or obvious branch mismatch.
+4. Multi-goal warning. If the intent appears to contain multiple independently shippable goals, carry `multiple-goals` forward so step-02 can add it to `{spec_file}` frontmatter `warnings`. Do not split or block.
+5. Route:
+
+   **Folder+id dispatch:** derive a valid kebab-case slug from the entry's `title` (and `description` if needed) — the same kebab-casing convention as below, but never prefixed with `{story_id}`, since the id is already the filename's separate leading segment. Set `spec_file` = `{spec_folder}/stories/{story_id}-{slug}.md`. The id already disambiguates: no `{{.implementation_artifacts}}` fallback, no `-2`/`-3` suffixing.
+
+   **Otherwise:** derive a valid kebab-case slug from the clarified intent. If the intent references a tracking identifier (story number, issue number, ticket ID), lead the slug with it (e.g. `3-2-digest-delivery`, `gh-47-fix-auth`). If `{{.implementation_artifacts}}/spec-{slug}.md` already exists: if its status is `draft`, treat it as the same work and resume it (set `spec_file` to that path, **EARLY EXIT** → `[[bmad-snapshot:step-02-plan.md]]`); otherwise append `-2`, `-3`, etc. Set `spec_file` = `{{.implementation_artifacts}}/spec-{slug}.md`.
+
+## NEXT
+
+Read fully and follow `[[bmad-snapshot:step-02-plan.md]]`
diff --git a/.agents/skills/bmad-build-auto/step-02-plan.md b/.agents/skills/bmad-build-auto/step-02-plan.md
new file mode 100644
index 00000000..6a2f5fa4
--- /dev/null
+++ b/.agents/skills/bmad-build-auto/step-02-plan.md
@@ -0,0 +1,27 @@
+# Step 2: Plan
+
+## RULES
+
+- **Language** — Speak in `{{.communication_language}}`, tailored to `{{.user_skill_level}}`. Write files in `{{.document_output_language}}`.
+- No human interaction: do not ask questions or wait for approval in this step.
+
+## INSTRUCTIONS
+
+1. Draft resume check. If `{spec_file}` exists with `status: draft`, read it and capture the verbatim `<intent-contract>...</intent-contract>` block as `preserved_intent_contract`. Otherwise `preserved_intent_contract` is empty.
+2. Investigate codebase. _Read the code yourself for narrow, localized tasks. Isolate deep exploration in synchronous subagents: instruct them to give you distilled summaries only, and plan from those summaries._ Decide which findings actually matter for execution — the specific files, symbols/lines, reuse points, and read-only constraints — and carry those forward for the Code Map. This is where the investigation lands: the spec preserves it so it is never re-narrated to the implementer at dispatch time.
+3. Read `[[bmad-snapshot:spec-template.md]]` fully. Fill it out based on the intent and investigation, resolving the template's `date` field to the current system date. Drain the investigation into the `## Code Map` section — annotated paths, symbol/line anchors, reuse pointers, and read-only evidence — so the spec is the implementer's investigation map and the step-03 handoff need only point at it. If `{preserved_intent_contract}` is non-empty, substitute it for the `<intent-contract>` block in your filled spec before writing. Write the result to `{spec_file}`.
+4. Self-review against READY FOR DEVELOPMENT standard.
+5. If intent gaps exist, do not fantasize and do not leave open questions. Multiple defensible readings of the intent that lead to observably different outcomes, with nothing in the intent to select between them, are an intent gap — do not resolve one by picking a reading. HALT with status `blocked`, blocking condition `intent gap`, and include the unanswered questions and evidence gathered.
+6. Warning check. If step-01 carried `multiple-goals`, add it to `{spec_file}` frontmatter `warnings`. If `{spec_file}` exceeds 1600 tokens, add `oversized` to frontmatter `warnings`. Continue either way.
+
+### READY-FOR-DEVELOPMENT GATE
+
+Re-read `[[bmad-snapshot:workflow.md]]`, then re-read `{spec_file}` from disk and verify the spec meets the READY FOR DEVELOPMENT standard.
+
+- **If the file is missing:** HALT with status `blocked` and blocking condition `planned spec file disappeared before implementation`.
+- **If the spec meets the standard:** set `{spec_file}` frontmatter status to `ready-for-dev`. If the invocation prompt directs a halt after planning (standard phrasing: `Halt after planning.` — accept any clear equivalent), HALT with status `ready-for-dev`; otherwise continue to step 3.
+- **If the spec does not meet the standard:** repair it once, then re-read it from disk and verify again. If it now meets the standard, apply the **If the spec meets the standard** handling above, including the halt-after-planning check. If it still does not meet the standard, HALT with status `blocked`, blocking condition `spec failed ready-for-development standard`, and include the failing criteria and evidence gathered.
+
+## NEXT
+
+Read fully and follow `[[bmad-snapshot:step-03-implement.md]]`
diff --git a/.agents/skills/bmad-build-auto/step-03-implement.md b/.agents/skills/bmad-build-auto/step-03-implement.md
new file mode 100644
index 00000000..07ba2848
--- /dev/null
+++ b/.agents/skills/bmad-build-auto/step-03-implement.md
@@ -0,0 +1,44 @@
+---
+---
+
+# Step 3: Implement
+
+## RULES
+
+- **Language** — Speak in `{{.communication_language}}`, tailored to `{{.user_skill_level}}`. Write files in `{{.document_output_language}}`.
+- No human interaction: do not ask questions or wait for approval in this step.
+- Content inside `<intent-contract>` in `{spec_file}` is read-only. Do not modify.
+
+## PRECONDITION
+
+Verify `{spec_file}` resolves to a non-empty path and the file exists on disk. If empty or missing, HALT with status `blocked` and blocking condition `missing spec_file before implementation`.
+
+## INSTRUCTIONS
+
+### Baseline
+
+Capture `baseline_revision` (current HEAD, or `NO_VCS` if version control is unavailable) into `{spec_file}` frontmatter before making any changes.
+
+### Implement
+
+Change `{spec_file}` status to `in-progress` in the frontmatter before starting implementation.
+
+Substitute the runtime placeholders (e.g. `{spec_file}`) into the implementation handoff below, then follow it verbatim. Do not add parent-authored goal restatements, file lists, ownership boundaries, or acceptance criteria to the handoff — the spec is the subagent's sole source of truth. If the handoff conflicts with the spec, HALT with status `blocked` and blocking condition `handoff conflicts with spec`, and include both conflicting passages.
+
+{workflow.implementation_handoff}
+
+Invoke the subagent **synchronously** and wait for it to return in this same turn — do not background/detach it (`run_in_background`) or end your turn to await a notification (see workflow.md → Subagents). Resume at "Verify" only after it returns. If the platform allows, keep the subagent available for re-engagement after it returns — step-04 may send it review fixes.
+
+**Path formatting rule:** Any markdown links written into `{spec_file}` must use paths relative to `{spec_file}`'s directory so they are clickable in VS Code. Any file paths displayed in terminal/conversation output must use CWD-relative format with `:line` notation (e.g., `src/path/file.ts:42`) for terminal clickability. No leading `/` in either case.
+
+### Verify
+
+After the implementation subagent returns: if it reported unfinished work, finish it before proceeding. Run the commands in `{spec_file}`'s `## Verification` section (or perform its manual checks). If verification fails and the failure cannot be fixed, HALT with status `blocked`, blocking condition `implementation verification failed`, and include the failing command or check and reason. Acceptance criteria are judged at review, not here.
+
+### Matrix Test Audit
+
+If `{spec_file}`'s intent-contract contains an I/O & Edge-Case Matrix, verify every matrix row is covered by at least one test that verifies its expected behavior, and that each covering test ran and passed in the verification output. A covering test that exists but did not run — unregistered, filtered out, skipped, or disabled — counts as missing. If a test disagrees with the matrix, never edit the expectation to match the code: fix the code, or if the matrix row itself is ambiguous, HALT with status `blocked` and blocking condition `matrix ambiguity`. If the audit cannot otherwise be satisfied, HALT with status `blocked` and blocking condition `matrix test audit failed`.
+
+## NEXT
+
+Read fully and follow `[[bmad-snapshot:step-04-review.md]]`
diff --git a/.agents/skills/bmad-build-auto/step-04-review.md b/.agents/skills/bmad-build-auto/step-04-review.md
new file mode 100644
index 00000000..d4a1f7e3
--- /dev/null
+++ b/.agents/skills/bmad-build-auto/step-04-review.md
@@ -0,0 +1,96 @@
+# Step 4: Review
+
+## RULES
+
+- **Language** — Speak in `{{.communication_language}}`, tailored to `{{.user_skill_level}}`. Write files in `{{.document_output_language}}`.
+- No human interaction: do not ask questions or wait for approval in this step.
+- All review subagents must run at the same model capability as the current session.
+
+## INSTRUCTIONS
+
+Change `{spec_file}` status to `in-review` in the frontmatter before continuing.
+
+### Construct Diff
+
+Read `{baseline_revision}` from `{spec_file}` frontmatter. If `{baseline_revision}` is missing or `NO_VCS`, use best effort to determine what changed. Otherwise, construct `{diff_output}` covering all changes — tracked and untracked — since `{baseline_revision}`.
+
+Do NOT `git add` anything — this is read-only inspection.
+
+### Review
+
+Runtime placeholders: `{diff_output}` is the diff constructed above. `{verbatim_intent}` is the invocation intent exactly as this run received it at step-01; if the run started from an existing spec file rather than a fresh intent, it is the spec's `<intent-contract>` block instead. Before launching a layer, expand its skill-root placeholder to this skill's absolute installed directory; never leave that placeholder unresolved in a child prompt.
+
+Execute these review layers in parallel wherever their execution methods allow: substitute the runtime placeholders (e.g. `{diff_output}`) into each layer's instruction. When an instruction launches a reviewer subagent, launch that child with the prompt text after placeholder substitution; do not load the reviewer instruction file yourself. For any other customized instruction, execute it as written. Parallel means several blocking calls awaited together in this turn — never backgrounded or detached, never ending the turn to await results (see workflow.md → Subagents). Spawn every reviewer subagent before reading or reacting to any of their output; begin collection and triage only once all are launched.
+
+{workflow.review_layers}
+
+### Classify
+
+1. Deduplicate only findings with the same claim and same required action. Then evaluate each remaining finding independently. Do not reject a finding because a related finding was rejected.
+2. Assign severity to each finding by consequence for the artifact's main consumer (software user, document reader, etc).
+   Disregard any severity assigned by a reviewing subagent. Review subagents operate under by-design information asymmetry and do not have enough context to set final severity for this workflow.
+   - `low`: none or cosmetic
+   - `medium`: tolerable
+   - `high`: intolerable
+3. Route each finding into exactly one triage category. The first three categories are **this story's problem** — caused or exposed by the current change. The last two are **not this story's problem**.
+   Scope authority: a finding may be routed to defer or reject *as out of scope* only on the authority of the intent itself. The spec's scope language, the plan, and the diff's own shape are not admissible scope authorities — if only they exclude a finding, treat it as evidence against the chosen reading (intent_gap or bad_spec), not as out of scope.
+   - **intent_gap** — caused by the change; cannot be resolved from the spec because the captured intent is incomplete. Do not infer intent unless there is exactly one possible reading.
+   - **bad_spec** — caused by the change, including direct deviations from spec. The spec should have been clear enough to prevent it. When in doubt between bad_spec and patch, prefer bad_spec — a spec-level fix is more likely to produce coherent code.
+   - **patch** — caused by the change; trivially fixable without human input. Just part of the diff.
+   - **defer** — pre-existing issue not caused by this story, surfaced incidentally by the review. Collect for later focused attention.
+   - **reject** — noise. Drop silently. When unsure between defer and reject, prefer reject — only defer findings you are confident are real.
+4. Append a new entry to the `## Review Triage Log` section in `{spec_file}`, in this format:
+   ```markdown
+   ### {date} — Review pass
+   - intent_gap: count
+   - bad_spec: count
+   - patch: count
+   - defer: count
+   - reject: count
+   - addressed_findings:
+     - `[high|medium|low]` `[patch|bad_spec]` <finding summary and action taken in this pass>
+   ```
+   Where `{date}` is the current system date and `count` is either just `0`, or total with breakdown by severity `N: (high Nhigh, medium Nmedium, low Nlow)`.
+   If no patch was fixed and no bad_spec repair loopback was triggered in this pass, write:
+   ```markdown
+   - addressed_findings:
+     - none
+   ```
+5. Process findings in cascading order. If intent_gap exists, lower findings are moot; follow the intent_gap branch below. If bad_spec exists, lower findings are moot since code will be re-derived. If neither exists, process patch and defer normally. Before each bad_spec loopback, read `{spec_file}` frontmatter `review_loop_iteration` (missing means `0`), increment it by 1, and write it back. If it exceeds 5, append the triage-log entry for this pass with `addressed_findings: none`, then HALT with status `blocked` and blocking condition `review repair loop exceeded 5 iterations (non-convergence)`.
+   - **intent_gap** — Root cause is inside `<intent-contract>`. Save the attempted change as a patch file in `{{.implementation_artifacts}}` and reference it from the triage-log entry, then revert code changes. Append the triage-log entry for this pass with `addressed_findings: none`, then HALT with status `blocked`, blocking condition `intent gap`, and include the unresolved questions and the saved patch path.
+   - **bad_spec** — Root cause is outside `<intent-contract>`. Do not modify content inside `<intent-contract>`. Before reverting code: extract KEEP instructions for positive preservation (what worked well and must survive re-derivation). Revert code changes. Read the `## Spec Change Log` in `{spec_file}` and strictly respect all logged constraints when amending the sections outside `<intent-contract>` that contain the root cause. Append a new change-log entry recording: the triggering finding, what was amended, the known-bad state avoided, and the KEEP instructions. Append the triage-log entry for this pass, listing every bad_spec finding that triggered the spec amendment and implementation loopback under `addressed_findings`. Read fully and follow `[[bmad-snapshot:step-03-implement.md]]` to re-derive the code, then this step will run again.
+   - **patch** — Auto-fix. These are the only findings that survive loopbacks. If the step-03 implementation subagent can be re-engaged with its context intact, send it all patch findings in one synchronous message — for each: the file, what is wrong, and what the fix must do. If it cannot be re-engaged, apply the patches yourself. Then re-run the commands in `{spec_file}`'s `## Verification` section (or perform its manual checks); if verification fails and the failure cannot be fixed, HALT with status `blocked` and blocking condition `patch verification failed`. Append the triage-log entry for this pass, listing every patch fixed in this pass under `addressed_findings`.
+   - **defer** — Update the single `deferred` list in `{spec_file}` frontmatter. If the field is absent (including on specs created before this field existed), add it once as an empty list. If it is `deferred: []`, replace that empty value when adding the first item; otherwise append to the existing list. Preserve every existing item, do not look for duplicates, and never add a second `deferred:` key. Serialize free-form values as YAML block scalars so characters such as `:`, `#`, quotes, and line breaks remain data. Each item uses this shape:
+     ```yaml
+     deferred:
+       - summary: >-
+           <one sentence>
+         evidence: |-
+           <why this is real>
+         location: >- # optional — file:line or component
+           src/foo.py:42
+         severity: medium # optional — high | medium | low
+     ```
+     After all appends, parse the complete frontmatter as YAML and verify that `deferred` is one list containing every prior item plus the new items with their intended text. Repair serialization errors before continuing.
+   - **reject** — Drop silently.
+
+## Finalize
+
+Write the following details to `{spec_file}` under `## Auto Run Result`:
+- Summary of implemented change
+- Files changed with one-line descriptions
+- Review findings breakdown: patches applied, items deferred, items rejected
+- Follow-up review recommendation: count only this pass's findings triaged `patch` — never defer or reject. `true` if any patched finding was `high` severity, or if `3 × medium count + 1 × low count` is 5 or more; otherwise `false`. Record the patched counts by severity and the score.
+- Verification performed, including command outcomes or manual inspection notes
+- Any residual risks
+
+Set `{spec_file}` frontmatter `followup_review_recommended` from the computation above.
+
+If version control is unavailable, set `{spec_file}` frontmatter `status: done`, then proceed to HALT.
+
+If version control is available, write `status: done` into `{spec_file}` frontmatter, then:
+
+1. Commit any reviewed-diff files that remain uncommitted, including `{spec_file}` when it is tracked in that working copy. Keep commits already created during this run. Verify every reviewed-diff file appears in the change set after `{baseline_revision}` and none remains uncommitted. Do not push.
+2. Verify the version-controlled working copy is clean. Otherwise HALT with status `blocked` and blocking condition `finalization left repository dirty`.
+
+HALT with status `done`.
diff --git a/.agents/skills/bmad-build-auto/workflow.md b/.agents/skills/bmad-build-auto/workflow.md
new file mode 100644
index 00000000..b22a902a
--- /dev/null
+++ b/.agents/skills/bmad-build-auto/workflow.md
@@ -0,0 +1,104 @@
+# Build Auto Workflow
+
+**Goal:** Turn intent into a hardened, reviewable artifact, without human interaction.
+
+**CRITICAL:** If a step directs you to another snapshot file, read it fully and follow it. No exceptions.
+
+## HALT
+
+To HALT with a final status and optional blocking condition:
+
+1. **Folder+id dispatch** (`{spec_folder}` and `{story_id}` are set): the write-back always lands at the id-keyed story spec. The `{{.implementation_artifacts}}` fallback in step 2 below is never used in this mode, even for halts before planning starts.
+   - If `{spec_file}` is still empty, resolve it now:
+     - **Entry not resolved** (`stories.yaml` is missing/unparseable, or `{story_id}` has no matching entry): use the fixed slug segment `unresolved`: `{spec_file}` = `{spec_folder}/stories/{story_id}-unresolved.md`.
+     - **Ambiguous on-disk match** (the halt is `ambiguous story file match` — more than one file already matches `{spec_folder}/stories/{story_id}-*.md`): use the fixed slug segment `ambiguous` instead of deriving from the title, so the write-back neither creates a third title-derived candidate nor risks silently landing on one of the existing ambiguous files: `{spec_file}` = `{spec_folder}/stories/{story_id}-ambiguous.md`.
+     - **Otherwise** (the entry was resolved and no ambiguous on-disk match exists): derive `{spec_file}` = `{spec_folder}/stories/{story_id}-{slug}.md`, where `{slug}` is a kebab-case slug from `title` (and `description` if needed) with no `{story_id}` prefix — the same derivation step-01's Route uses.
+   - If `{spec_file}` exists on disk, update `status` in frontmatter and append missing result details under `## Auto Run Result`.
+   - If it does not exist, create it as a skeletal story spec:
+     ```markdown
+     ---
+     status: <final status>
+     ---
+
+     # <entry title, or "Story {story_id}" if the entry could not be resolved or the on-disk match was ambiguous>
+
+     ## Auto Run Result
+
+     Status: <final status>
+     Blocking condition: <blocking condition, if any>
+     ```
+2. **Otherwise:**
+   - If `{spec_file}` is known and exists, update `status` in frontmatter and append missing result details under `## Auto Run Result`.
+   - If `{spec_file}` is unknown or missing, create `{{.implementation_artifacts}}/bmad-build-auto-result-<slug-or-timestamp>.md` with:
+     ```markdown
+     ---
+     status: <final status>
+     ---
+
+     # BMad Build Auto Result
+
+     Status: <final status>
+     Blocking condition: <blocking condition, if any>
+     ```
+3. Follow **On Complete** below, then stop the workflow.
+
+### On Complete
+
+If anything appears below, follow it as the final terminal instruction before exiting; otherwise exit normally.
+
+{workflow.on_complete}
+
+## Subagents
+
+Using subagents when instructed is mandatory. If you cannot, HALT with status `blocked` and blocking condition `no subagents`.
+
+Invoke every subagent **synchronously**: launch it, wait for it to return within the same turn, then continue with its result. When a step says to run subagents "in parallel" (e.g. the reviewers), that means several **blocking** calls awaited together in one turn — not detached execution. Never run a subagent in the background / detached / async (e.g. `run_in_background: true`), and never end your turn to "await a completion notification." This workflow runs unattended: there is no event loop to resume a yielded turn, so a backgrounded subagent never hands control back and the run stalls. The only sanctioned way to end a turn is the HALT protocol above with an explicit terminal `status`.
+
+## READY FOR DEVELOPMENT STANDARD
+
+A specification is "Ready for Development" when:
+
+- **Actionable**: Every task has a file path and specific action.
+- **Logical**: Tasks ordered by dependency.
+- **Testable**: All ACs use Given/When/Then.
+- **Surface-anchored**: ACs observe the outermost surface the intent references — never a more internal proxy for it.
+- **Complete**: No placeholders or TBDs.
+- **Sufficient**: No known requirement, acceptance, dependency, or implementation gaps remain unresolved.
+- **Coherent**: No unresolved ambiguities or internal contradictions.
+
+## Conventions
+
+- Every operational cross-file reference in this workflow is an absolute snapshot path. Open it directly; do not resolve it relative to a skill directory.
+- `{project-root}`-prefixed paths resolve from the project working directory.
+- Speak in `{{.communication_language}}`, tailor communication to `{{.user_skill_level}}`, and write documents in `{{.document_output_language}}`.
+- Whenever this workflow captures or records a version-control revision, obtain the full canonical identifier directly from version control and preserve it verbatim.
+
+## On Activation
+
+### Step 1: Execute Prepend Steps
+
+Execute each of these steps in order before proceeding (`_None._` means skip):
+
+{workflow.activation_steps_prepend}
+
+### Step 2: Load Persistent Facts
+
+Treat every entry below as foundational context you carry for the rest of the workflow run. Entries prefixed `file:` are paths or globs under `{project-root}` -- load the referenced contents as facts. All other entries are facts verbatim (`_None._` means none):
+
+{workflow.persistent_facts}
+
+### Step 3: Execute Append Steps
+
+Execute each of these steps in order (`_None._` means skip):
+
+{workflow.activation_steps_append}
+
+Activation is complete after all activation steps have run.
+
+## Workflow Execution
+
+Follow the step files in order. Read one step fully, execute it, then load the next step only when directed. Do not skip, reorder, or pre-load steps.
+
+## First Workflow Step
+
+Read fully and follow: `[[bmad-snapshot:step-01-clarify-and-route.md]]`.
diff --git a/.agents/skills/bmad-build/SKILL.md b/.agents/skills/bmad-build/SKILL.md
new file mode 100644
index 00000000..e7cbbf80
--- /dev/null
+++ b/.agents/skills/bmad-build/SKILL.md
@@ -0,0 +1,13 @@
+---
+name: bmad-build
+description: 'Implements any user intent, requirement, story, bug fix or change request by producing clean working code artifacts that follow the project''s existing architecture, patterns and conventions. Use when the user wants to build, fix, tweak, refactor, add or modify any code, component or feature.'
+---
+
+Run the following command exactly once without changing the current working directory. Replace `{project-root}` with the absolute path to the project root and `{skill-root}` with the absolute path to this skill's directory:
+
+```bash
+uv run --no-cache "{project-root}/_bmad/scripts/render_skill.py" --project-root "{project-root}" --skill "{skill-root}"
+```
+
+- On success, read and follow the one absolute `workflow.md` instruction printed to stdout.
+- On failure (including `uv` being unavailable), report the command output and HALT. Do not run any workflow source directly.
diff --git a/.agents/skills/bmad-build/compile-epic-context.md b/.agents/skills/bmad-build/compile-epic-context.md
new file mode 100644
index 00000000..f5963b36
--- /dev/null
+++ b/.agents/skills/bmad-build/compile-epic-context.md
@@ -0,0 +1,62 @@
+# Compile Epic Context
+
+**Task**
+Given an epic number, the epics file, the planning artifacts directory, and a desired output path, compile a clean, focused, developer-ready context file (`epic-<N>-context.md`).
+
+**Steps**
+
+1. Read the epics file and extract the target epic's title, goal, and list of stories.
+2. Scan the planning artifacts directory for the standard files (PRD, architecture, UX/design, product brief).
+3. Pull only the information relevant to this epic.
+4. Write the compiled context to the exact output path using the format below.
+
+## Exact Output Format
+
+Use these headings:
+
+```markdown
+# Epic {N} Context: {Epic Title}
+
+<!-- Compiled from planning artifacts. Edit freely. Regenerate with compile-epic-context if planning docs change. -->
+
+## Goal
+
+{One clear paragraph: what this epic achieves and why it matters.}
+
+## Stories
+
+- Story X.Y: Brief title only
+- ...
+
+## Requirements & Constraints
+
+{Relevant functional/non-functional requirements and success criteria for this epic (describe by purpose, not source).}
+
+## Technical Decisions
+
+{Key architecture decisions, constraints, patterns, data models, and conventions relevant to this epic.}
+
+## UX & Interaction Patterns
+
+{Relevant UX flows, interaction patterns, and design constraints (omit section entirely if nothing relevant).}
+
+## Cross-Story Dependencies
+
+{Dependencies between stories in this epic or with other epics/systems (omit if none).}
+```
+
+## Rules
+
+- **Scope aggressively.** Include only what a developer working on any story in this epic actually needs. When in doubt, leave it out — the developer can always read the full planning doc.
+- **Describe by purpose, not by source.** Write "API responses must include pagination metadata" not "Per PRD section 3.2.1, pagination is required." Planning doc internals will change; the constraint won't.
+- **No full copies.** Never quote source documents, section numbers, or paste large blocks verbatim. Always distill.
+- **No story-level details.** The story list is for orientation only. Individual story specs handle the details.
+- **Nothing derivable from the codebase.** Don't document what a developer can learn by reading the code.
+- **Be concise and actionable.** Target 800–1500 tokens total. This file loads into build's context alongside other material.
+- **Never hallucinate content.** If source material doesn't say something, don't invent it.
+- **Omit empty sections entirely**, except Goal and Stories, which are always required.
+
+## Error handling
+
+- **If the epics file is missing or the target epic is not found:** write nothing and report the problem to the calling agent. Goal and Stories cannot be populated without a usable epics file.
+- **If planning artifacts are missing or empty:** still produce the file with Goal and Stories populated from the epics file, and note the gap in the Goal section. Never hallucinate content to fill missing sections.
diff --git a/.agents/skills/bmad-build/customize.toml b/.agents/skills/bmad-build/customize.toml
new file mode 100644
index 00000000..4777a117
--- /dev/null
+++ b/.agents/skills/bmad-build/customize.toml
@@ -0,0 +1,164 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Default customization values for bmad-build.
+# Override in _bmad/custom/bmad-build.toml or
+# _bmad/custom/bmad-build.user.toml.
+#
+# Merge rules:
+# - Strings replace the default.
+# - Lists append to the default list.
+# - Tables merge key by key.
+# - Arrays of tables merge by `id`: matching `id` replaces, new `id`s append.
+
+[workflow]
+
+# Extra instructions to run before config is loaded and before the user is greeted.
+
+activation_steps_prepend = []
+
+# Extra instructions to run after the greeting and before step 01.
+
+activation_steps_append = []
+
+# Facts kept in context for the whole run.
+# Entries are literal text or file references prefixed with "file:".
+# File entries may use globs and are loaded during activation.
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Instruction run after Build completes.
+# Empty means no extra completion behavior.
+
+on_complete = ""
+
+# How Build opens a completed spec for human review. The default deliberately
+# opens the repository root first and the spec second in a reused VS Code
+# window so relative review links resolve in the correct workspace. Override
+# this with the application and equivalent behavior you prefer. Set it to an
+# empty string to disable automatic opening and all related completion output.
+# Example requests for bmad-customize: "Open completed specs in Cursor and
+# reuse the project window" or "Do not open completed specs automatically."
+#
+# Example `open_spec` override instructions for bmad-build.user.toml:
+# {project-root} and {spec_file} are substituted at run time.
+#
+# VS Code and compatible GUI editors — shipped default uses `code`; Cursor,
+# Windsurf, Antigravity, Zed, and similar editors can substitute their launcher.
+# Preserve reuse-window behavior when the launcher supports it, and always pass
+# the repository root first and the spec second:
+#   Run `code -r "{project-root}" "{spec_file}"`, root first.
+#   Run `cursor -r "{project-root}" "{spec_file}"`, root first.
+#   Run `windsurf -r "{project-root}" "{spec_file}"`, root first.
+#   Run `zed "{project-root}" "{spec_file}"`, root first.
+#
+# IntelliJ IDEA — its CLI accepts one path per invocation:
+#   Run `idea "{project-root}"` to open the project, then
+#   `idea "{spec_file}"` to open the spec in that project.
+#
+# Vim — keep repository context as the working directory:
+#   Run `cd "{project-root}" && vim "{spec_file}"`.
+#
+# Emacs — reuse an Emacs server without waiting for the buffer to close:
+#   Run `cd "{project-root}" && emacsclient -n "{spec_file}"`.
+#
+# For every non-empty override, also tell Build to note in the completion summary
+# that the spec was sent to the chosen editor and contains a Suggested Review
+# Order, then include the default navigation tip shown below.
+
+open_spec = """
+Run `code -r "{project-root}" "{spec_file}"` — the repository root first so VS Code opens in the right context, then the spec file. Always double-quote both paths to handle spaces and special characters. If `code` is unavailable or the command fails, skip gracefully and tell the user the spec file path instead. In the completion summary, note that the spec was sent to VS Code and that it contains a Suggested Review Order, then add this navigation tip: "Ctrl+click (Cmd+click on macOS) the links in the Suggested Review Order to jump to each stop."
+"""
+
+# Handoff for the implementation subagent in step 03. The whole execution
+# recipe — a subagent by default, but an override may run it any other way
+# (a different model, an external coding tool via bash). {spec_file} is
+# substituted at run time.
+
+implementation_handoff = """
+Launch a subagent with no prior conversation context, with this prompt:
+
+> Read {spec_file} fully and implement it — the spec is the sole source of truth. Load every file listed in its frontmatter `context:` before you start.
+>
+> When done, report what you changed, how you verified it, and anything left incomplete or risky.
+"""
+
+# Review layers for the review step. `instruction` is the layer's whole
+# execution recipe — subagents by default, but an override may run anything
+# (e.g. an external reviewer via bash). {diff_output} is substituted at run
+# time. `when` (optional) gates a layer; empty `instruction` disables it.
+
+[[workflow.review_layers]]
+id = "blind-hunter"
+name = "Blind Hunter"
+instruction = """
+Launch a context-free subagent with this prompt:
+
+Conduct a review of CONTENT.
+Look for what's missing, not only what's wrong.
+Find at least ten issues to fix or improve.
+Output a Markdown list of findings only — no severity, priority, or ranking.
+If the content is empty, stop and say so.
+If you have zero findings, re-check and keep thinking; do not stop with an empty list.
+
+CONTENT:
+{diff_output}
+
+Do not invoke any skill. Return only the review result.
+
+"""
+
+[[workflow.review_layers]]
+id = "edge-case-hunter"
+name = "Edge Case Hunter"
+instruction = """
+Launch a context-free subagent with this prompt:
+
+Read `{skill-root}/review-prompts/edge-case-hunter.md` completely and follow it as your review instructions.
+
+Review content:
+
+{diff_output}
+
+Do not invoke any skill. If the instruction file is unreadable, report that exact failure and stop. Return only the review result.
+
+"""
+
+[[workflow.review_layers]]
+id = "verification-gap"
+name = "Verification Gap Reviewer"
+instruction = """
+Launch a context-free subagent with this prompt:
+
+Read `{skill-root}/review-prompts/verification-gap.md` completely and follow it as your review instructions.
+
+Review content:
+
+{diff_output}
+
+Do not invoke any skill. If the instruction file is unreadable, report that exact failure and stop. Return only the review result.
+
+"""
+
+# Review layers for the one-shot route.
+
+[[workflow.oneshot_review_layers]]
+id = "blind-hunter"
+name = "Blind Hunter"
+instruction = """
+Launch a context-free subagent with this prompt:
+
+Conduct a review of CONTENT.
+Look for what's missing, not only what's wrong.
+Find at least ten issues to fix or improve.
+Output a Markdown list of findings only — no severity, priority, or ranking.
+If the content is empty, stop and say so.
+If you have zero findings, re-check and keep thinking; do not stop with an empty list.
+
+CONTENT:
+The changed files in the current worktree. Inspect them directly before reviewing.
+
+Do not invoke any skill. Return only the review result.
+
+"""
diff --git a/.agents/skills/bmad-build/references/deletion-check.md b/.agents/skills/bmad-build/references/deletion-check.md
new file mode 100644
index 00000000..5cc3d771
--- /dev/null
+++ b/.agents/skills/bmad-build/references/deletion-check.md
@@ -0,0 +1,14 @@
+# Deletion Check
+
+Secondary pass for the Edge Case Hunter — runs only when the diff removed meaningful code. Subordinate to the edge-case pass; findings are usually few or none.
+
+For each chunk of removed or replaced code (ignore pure renames and whitespace), ask: did it carry behavior or a contract that the change neither re-established nor intentionally retired? Add a finding for any resulting regression, orphaned reference, or newly-dead code. Skip anything already covered by your edge-case findings.
+
+Append each finding to the same JSON array as the edge-case findings, with the four standard fields plus:
+
+- `kind`: `"deletion"`
+- `confidence`: `"high"`, `"medium"`, or `"low"` — these are inferences; rate them
+
+For a deletion finding the standard fields read as: `location` = the removed item; `trigger_condition` = the behavior or contract it enforced; `guard_snippet` = where or how to re-establish it; `potential_consequence` = the regression or orphan.
+
+Add nothing if nothing qualifies.
diff --git a/.agents/skills/bmad-build/review-prompts/edge-case-hunter.md b/.agents/skills/bmad-build/review-prompts/edge-case-hunter.md
new file mode 100644
index 00000000..11df57e9
--- /dev/null
+++ b/.agents/skills/bmad-build/review-prompts/edge-case-hunter.md
@@ -0,0 +1,88 @@
+# Edge Case Hunter Review
+
+**Goal:** You are a pure path tracer. Never comment on whether code is good or bad; only list missing handling.
+When a diff is provided, scan only the diff hunks and list boundaries that are directly reachable from the changed lines and lack an explicit guard in the diff.
+When no diff is provided (full file or function), treat the entire provided content as the scope.
+Ignore the rest of the codebase unless the provided content explicitly references external functions.
+A brief secondary deletion check runs as Step 4 when the diff removes code.
+
+**Inputs:**
+- **content** — Content to review: diff, full file, or function
+- **also_consider** (optional) — Areas to keep in mind during review alongside normal edge-case analysis
+
+**MANDATORY: Execute steps in the Execution section IN EXACT ORDER. DO NOT skip steps or change the sequence. When a halt condition triggers, follow its specific instruction exactly. Each action within a step is a REQUIRED action to complete that step.**
+
+**Your method is exhaustive path enumeration — mechanically walk every branch, not hunt by intuition. Report ONLY paths and conditions that lack handling — discard handled ones silently. Do NOT editorialize or add filler. Do not assign severity labels, rankings, or priority levels.**
+
+
+## EXECUTION
+
+### Step 1: Receive Content
+
+- Load the content to review strictly from the parent message that launched you (not from this instruction file)
+- If content is empty, or cannot be decoded as text, return `[{"location":"N/A","trigger_condition":"Input empty or undecodable","guard_snippet":"Provide valid content to review","potential_consequence":"Review skipped — no analysis performed"}]` and stop
+- Identify content type (diff, full file, or function) to determine scope rules
+
+### Step 2: Exhaustive Path Analysis
+
+**Walk every branching path and boundary condition within scope — report only unhandled ones.**
+
+- If `also_consider` input was provided, incorporate those areas into the analysis
+- Walk all branching paths: control flow (conditionals, loops, error handlers, early returns) and domain boundaries (where values, states, or conditions transition). Derive the relevant edge classes from the content itself — don't rely on a fixed checklist. Examples: missing else/default, unguarded inputs, off-by-one loops, arithmetic overflow, implicit type coercion, race conditions, timeout gaps
+- Consider implicit branches: the diff special-cases or changes the handling of one or more members of a fixed set of values — enums, status codes, sentinels, type tags, flags, value ranges. The rest of the set is implicit branches (e.g. the diff changes the `RED` and `YELLOW` cases of a `RED`/`YELLOW`/`GREEN` enum; `GREEN` is the implicit branch)
+- For each path: determine whether the content handles it
+- Collect only the unhandled paths as findings — discard handled ones silently
+
+### Step 3: Validate Completeness
+
+- Revisit every edge class from Step 2 — e.g., missing else/default, null/empty inputs, off-by-one loops, arithmetic overflow, implicit type coercion, race conditions, timeout gaps
+- Add any newly found unhandled paths to findings; discard confirmed-handled ones
+
+### Step 4: Deletion Check
+
+If the diff removed or replaced meaningful code (ignore pure renames and whitespace): load `references/deletion-check.md` and follow it.
+
+### Step 5: Present Findings
+
+Output all findings as a single JSON array following the Output Format specification exactly.
+
+
+## OUTPUT FORMAT
+
+Return ONLY a valid JSON array of objects. Each edge-case finding contains exactly these four fields:
+
+```json
+[{
+  "location": "file:start-end (or file:line when single line, or file:hunk when exact line unavailable)",
+  "trigger_condition": "one-line description (max 15 words)",
+  "guard_snippet": "minimal code sketch that closes the gap (single-line escaped string, no raw newlines or unescaped quotes)",
+  "potential_consequence": "what could actually go wrong (max 15 words)"
+}]
+```
+
+No extra text, no explanations, no markdown wrapping. An empty array `[]` is valid when nothing is found. Deletion findings from Step 4, if any, go in the same array with the extra fields defined in `references/deletion-check.md`.
+
+
+## HALT CONDITIONS
+
+- If content is empty or cannot be decoded as text, return `[{"location":"N/A","trigger_condition":"Input empty or undecodable","guard_snippet":"Provide valid content to review","potential_consequence":"Review skipped — no analysis performed"}]` and stop
+<reference path="references/deletion-check.md">
+# Deletion Check
+
+Secondary pass for the Edge Case Hunter — runs only when the diff removed meaningful code. Subordinate to the edge-case pass; findings are usually few or none.
+
+For each chunk of removed or replaced code (ignore pure renames and whitespace), ask: did it carry behavior or a contract that the change neither re-established nor intentionally retired? Add a finding for any resulting regression, orphaned reference, or newly-dead code. Skip anything already covered by your edge-case findings.
+
+Append each finding to the same JSON array as the edge-case findings, with the four standard fields plus:
+
+- `kind`: `"deletion"`
+- `confidence`: `"high"`, `"medium"`, or `"low"` — these are inferences; rate them
+
+For a deletion finding the standard fields read as: `location` = the removed item; `trigger_condition` = the behavior or contract it enforced; `guard_snippet` = where or how to re-establish it; `potential_consequence` = the regression or orphan.
+
+Add nothing if nothing qualifies.
+</reference>
+
+## CONTENT SOURCE
+
+Review the content supplied under "Review content:" in the message that launched you.
diff --git a/.agents/skills/bmad-build/review-prompts/verification-gap.md b/.agents/skills/bmad-build/review-prompts/verification-gap.md
new file mode 100644
index 00000000..4adc52d5
--- /dev/null
+++ b/.agents/skills/bmad-build/review-prompts/verification-gap.md
@@ -0,0 +1,113 @@
+# Verification Gap Review
+
+**Goal:** Find changed behavior that could break without reliable verification catching it. Ask one question — "if the behavior this change is supposed to produce broke where it's actually used, would verification fail?" Do not hunt for correctness bugs, but report genuine problems you notice while tracing verification.
+
+The main verification gap shapes are:
+
+1. **Regression gap:** the changed code regresses where it's used, and no test covering that use would fail.
+2. **Missing-adoption gap:** a place that should now use the new behavior doesn't; it handles the same case its own way, or not at all, and no test would flag the omission.
+3. **Broken-verification gap:** a test appears to cover the changed behavior, but would not actually protect it because it is skipped, flaky, not run in the normal verification path, or too weak to observe the regression.
+
+## Evidence Rules
+
+- Read a test before claiming what it covers, runs, asserts, or misses.
+- Before claiming no test exists, search the whole repo by the symbol under test and by import references; expected file locations are not enough.
+- Never assert what you did not verify. If a finding cannot be grounded, drop it.
+- In a finding, say what you actually checked — "none of the tests I read cover this" — and show how far you looked. Say a test doesn't exist anywhere only when the symbol/import-reference search actually shows that.
+- Do not assign severity, confidence, priority, or ranking.
+
+## Review Sequence
+
+### Step 1: Screen for behavioral change
+
+Screen each part of the change separately. If a part is non-behavioral, skip it. Call a part non-behavioral only when the changed code does not alter return values, thrown errors, caller-visible side effects, or observable state (including iteration order and emitted messages). Once a part meets that test, move on; do not inspect callers or tests for extra confirmation.
+
+Common non-behavioral examples: formatting, comments, whitespace; pure renames; trivial getters/setters and pass-throughs; type-only or compiler-enforced changes with no runtime effect; etc.
+
+Only outcomes produced by deterministic code are worth automatically testing; tests are useless on static source text and brittle on LLM output. Skip those parts.
+
+If every part is skipped, output the clean result (see Output Format).
+
+### Step 2: Find the behavior that changed
+
+Identify what behavior changed compared to the previous version: output, side effect, branch, error path, schema/event shape, config default, validation/authorization rule, external contract, etc. If the change affects more than one behavior, handle each separately.
+
+Treat broad-impact changes as behavioral even when no single changed line looks important: dependency, toolchain, build/config, data-file, etc.
+
+### Step 3: Trace where that behavior is used
+
+Trace the changed behavior to the places that observe it. Start with direct callers and registered entry points (routes, commands, DI), contract consumers (schemas, events, APIs, database readers), and reverse-dependency info if already available.
+
+Follow a path only while the changed behavior is reachable and unverified. Stop when a test at that boundary would fail, the consumer does not observe the changed behavior, or the next hop is guesswork (dynamic dispatch, reflection, outside-repo consumers, etc.). Prefer the nearest observable boundary, often one to three hops away, especially across contract, integration, or service edges. If there are more than five similar consumers, group obvious repeats and check representative paths; expand only when a consumer observes the behavior differently.
+
+### Step 4: Qualify the consumer, then check its test
+
+For each consumer, name the smallest realistic regression this consumer would observe: invert the branch, drop the default, omit the field, return the old error code, skip the integration call, etc. This is the Demonstration. If no such regression exists, drop the path; untested downstream code is not a finding.
+
+A `Missing-adoption gap` qualifies not by the adoption failure alone but by a supersession signal: the change gives clear evidence the new behavior is meant to replace the local one — PR intent, naming or docs, a replaced sibling site, deleted duplicate logic, or a test defining the new rule — and the local site shares the same observable contract. Without a supersession signal and a shared observable contract, it is a refactor suggestion, not a verification-gap finding. Once both hold, check whether any test for that site would flag the non-adoption; missing coverage of the non-adoption is the gap itself, not a disqualifier.
+
+Find and read the relevant test. Ask whether the Demonstration would make an assertion fail.
+
+- If yes, the behavior is verified. No finding.
+- For a regression-style Demonstration: if no test runs the path, the test is skipped/flaky/not run normally, or the test runs the code without checking the changed result, report a `Regression gap` or `Broken-verification gap`.
+- For a qualifying Missing-adoption case: if none of the site tests you found assert it adopts the new behavior, report a `Missing-adoption gap`.
+
+A test counts only if it runs normally and an assertion observes the changed output, branch, or contract. These do not count: no execution; source-text assertions that match a file's wording instead of running it; success/no-throw/snapshot-only checks; mock/log-call checks; human-only checks; tests that mock away the integration; e2e tests that pass through without checking the changed output; stale assertions or fixtures.
+
+For example, `expect(x ?? DEFAULT).toBe(DEFAULT)` passes when `x` is missing.
+
+Common patterns:
+
+- **Caller-path gap** — helper test covers the branch, but caller values skip it.
+- **Contract drift** — payload/schema/event changes must be verified at the consumer.
+- **Migration compatibility** — tests only create new-format rows or fresh schemas.
+- **Phantom exception** — handled partial-failure path has no test.
+- **Missing-adoption gap** — sibling site should use the new rule/helper and does not.
+- **Removed verification** — deleted test or weakened assertion leaves behavior unpinned; removing a source-text assertion is not this, since it never counted.
+
+### Step 5: Confirm each finding is real
+
+Before writing a finding, re-open the specific tests or search results the finding relies on. Verify the Demonstration would not make any test you checked fail, or that the absence claim is backed by the symbol/import-reference search. Do not claim more than you verified; drop any finding you cannot ground.
+
+Explain why the test misses the bug using what the test sets up and checks.
+
+Do not report: compiler/type-checker-enforced cases; behavior already verified by an integration, contract, or e2e test; implementation-detail or mock-only tests; low coverage or a missing test file by itself; legacy untested code the change did not affect.
+
+Report genuine problems you noticed while tracing verification, even if they are not verification gaps. Put them under `Other findings` in the output. This permits reporting what you already reached, not extra hunting.
+
+## OUTPUT FORMAT
+
+Emit each verification-gap finding as one block. No general advice, no severity or confidence.
+
+```markdown
+### <one-line title naming the gap>
+
+- **Changed surface:** the exact behavior or contract that changed — `file:line`.
+- **Impacted consumer or site:** named concretely with `file:line` (e.g. "the `createInvoice` mutation used by the billing dashboard at `billing/dashboard.ts:88`," not "callers of this function").
+- **Existing test evidence:**
+  - `Regression gap`: what the relevant test actually asserts, with `file:line`; or, if none, the symbol/import-reference searches run and their result.
+  - `Missing-adoption gap`: tests for the impacted site, and whether any assert it adopts the new behavior.
+  - `Broken-verification gap`: the apparent test or verification path, and why it does not count.
+- **Missing verification:** the precise assertion or check that's absent.
+- **Demonstration:**
+  - `Regression gap` / `Broken-verification gap`: the concrete regression that would ship undetected, and why the tests you checked would not fail.
+  - `Missing-adoption gap`: the case the site mishandles by not adopting the new behavior, and that none of the tests you read assert adoption.
+- **Consequence:** the concrete thing that ships wrong — a regression the checked evidence would not catch, or a site that should use the new behavior and doesn't.
+- **Suggested test shape:** (optional) the kind of test that would close the gap, fit to the repo's own way of verifying — don't impose a generic test pyramid.
+```
+
+If you noticed genuine non-gap problems while tracing verification, append:
+
+```markdown
+## Other findings
+
+- <description only; no severity, confidence, priority, or ranking>
+```
+
+When you find no verification gaps and no other findings, output exactly this single line, not an empty response:
+
+`No verification gaps found.`
+
+## CONTENT SOURCE
+
+Review the content supplied under "Review content:" in the message that launched you. If none is supplied, stop with exactly: `No verification gaps found.`
diff --git a/.agents/skills/bmad-build/spec-template.md b/.agents/skills/bmad-build/spec-template.md
new file mode 100644
index 00000000..7849094b
--- /dev/null
+++ b/.agents/skills/bmad-build/spec-template.md
@@ -0,0 +1,89 @@
+---
+title: '{title}'
+type: 'feature' # feature | bugfix | refactor | chore
+created: '{date}'
+status: 'draft' # draft | ready-for-dev | in-progress | in-review | done
+review_loop_iteration: 0 # incremented by step-04 before each review loopback
+context: [] # optional: `{project-root}/`-prefixed paths to project-wide standards/docs the implementation agent should load. Keep short — only what isn't already distilled into the spec body.
+---
+
+<!-- Target: 900–1300 tokens. Above 1600 = high risk of context rot.
+     Never over-specify "how" — use boundaries + examples instead.
+     Cohesive cross-layer stories (DB+BE+UI) stay in ONE file.
+     IMPORTANT: Remove all HTML comments when filling this template. -->
+
+<frozen-after-approval reason="human-owned intent — do not modify unless human renegotiates">
+
+## Intent
+
+<!-- What is broken or missing, and why it matters. Then the high-level approach — the "what", not the "how". -->
+
+**Problem:** ONE_TO_TWO_SENTENCES
+
+**Approach:** ONE_TO_TWO_SENTENCES
+
+## Boundaries & Constraints
+
+<!-- Three tiers: Always = invariant rules. Ask First = human-gated decisions. Never = out of scope + forbidden approaches. -->
+
+**Always:** INVARIANT_RULES
+
+**Ask First:** DECISIONS_REQUIRING_HUMAN_APPROVAL
+<!-- Agent: if any of these trigger during execution, HALT and ask the user before proceeding. -->
+
+**Never:** NON_GOALS_AND_FORBIDDEN_APPROACHES
+
+## I/O & Edge-Case Matrix
+
+<!-- If no meaningful I/O scenarios exist, DELETE THIS ENTIRE SECTION. Do not write "N/A" or "None". -->
+
+| Scenario | Input / State | Expected Output / Behavior | Error Handling |
+|----------|--------------|---------------------------|----------------|
+| HAPPY_PATH | INPUT | OUTCOME | N/A |
+| ERROR_CASE | INPUT | OUTCOME | ERROR_HANDLING |
+
+</frozen-after-approval>
+
+## Code Map
+
+<!-- Agent-populated during planning. Annotated paths prevent blind codebase searching. -->
+
+- `FILE` -- ROLE_OR_RELEVANCE
+- `FILE` -- ROLE_OR_RELEVANCE
+
+## Tasks & Acceptance
+
+<!-- Tasks: backtick-quoted file path -- action -- rationale. Prefer one task per file; group tightly-coupled changes when splitting would be artificial. -->
+<!-- If an I/O Matrix is present, include a task to unit-test its edge cases. -->
+<!-- AC covers system-level behaviors not captured by the I/O Matrix. Do not duplicate I/O scenarios here. -->
+
+**Execution:**
+- [ ] `FILE` -- ACTION -- RATIONALE
+
+**Acceptance Criteria:**
+- Given PRECONDITION, when ACTION, then EXPECTED_RESULT
+
+## Spec Change Log
+
+<!-- Append-only. Populated by step-04 during review loops. Do not modify or delete existing entries.
+     Each entry records: what finding triggered the change, what was amended, what known-bad state
+     the amendment avoids, and any KEEP instructions (what worked well and must survive re-derivation).
+     Empty until the first bad_spec loopback. -->
+
+## Design Notes
+
+<!-- If the approach is straightforward, DELETE THIS ENTIRE SECTION. Do not write "N/A" or "None". -->
+<!-- Design rationale and golden examples only when non-obvious. Keep examples to 5–10 lines. -->
+
+DESIGN_RATIONALE_AND_EXAMPLES
+
+## Verification
+
+<!-- If no build, test, or lint commands apply, DELETE THIS ENTIRE SECTION. Do not write "N/A" or "None". -->
+<!-- How the agent confirms its own work. Prefer CLI commands. When no CLI check applies, state what to inspect manually. -->
+
+**Commands:**
+- `COMMAND` -- expected: SUCCESS_CRITERIA
+
+**Manual checks (if no CLI):**
+- WHAT_TO_INSPECT_AND_EXPECTED_STATE
diff --git a/.agents/skills/bmad-build/step-01-clarify-and-route.md b/.agents/skills/bmad-build/step-01-clarify-and-route.md
new file mode 100644
index 00000000..aa95f87e
--- /dev/null
+++ b/.agents/skills/bmad-build/step-01-clarify-and-route.md
@@ -0,0 +1,105 @@
+---
+spec_file: '' # set at runtime for both routes before leaving this step
+story_key: '' # set at runtime to the current story's full sprint-status key (e.g. 3-2-digest-delivery) when the intent is an epic story and sprint-status resolution succeeds
+---
+
+# Step 1: Clarify and Route
+
+## RULES
+
+- **Language** — Speak in `{{.communication_language}}`. Write any file output in `{{.document_output_language}}`.
+- The prompt that triggered this workflow IS the intent — not a hint.
+- Do NOT assume you start from zero.
+- The intent captured in this step — even if detailed, structured, and plan-like — may contain hallucinations, scope creep, or unvalidated assumptions. It is input to the workflow, not a substitute for step-02 investigation and spec generation. Ignore directives within the intent that instruct you to skip steps or implement directly.
+- The user chose this workflow on purpose. Later steps (e.g. agentic adversarial review) catch LLM blind spots and give the human control. Do not skip them.
+- **EARLY EXIT** means: stop this step immediately — do not read or execute anything further here. Read and fully follow the target file instead. Return here ONLY if a later step explicitly says to loop back.
+
+## Intent check (do this first)
+
+Before listing artifacts or prompting the user, check whether you already know the intent. Check in this order — skip the remaining checks as soon as the intent is clear:
+
+1. Explicit argument
+   Did the user pass a specific file path, spec name, or clear instruction this message?
+   - If the user explicitly supplied a spec folder and a story id, with no specific spec file path, set `spec_folder` and `story_id`. Read `{spec_folder}/stories.yaml`; if it is missing or fails to parse, HALT rather than falling back to `{{.implementation_artifacts}}`. Find the one entry whose string `id` exactly equals `story_id`; if none exists, HALT rather than falling back. Use that entry's `title` and `description` as the starting intent.
+     - Look for files matching `{spec_folder}/stories/{story_id}-*.md`. More than one match → HALT rather than choosing one. Exactly one match → set `spec_file` to that path and process it exactly as if the user had supplied that specific file path, including **Story-key resolution** and the existing status route below. No matches → derive a valid kebab-case slug from the entry's `title` (and `description` if needed), then set `spec_file` = `{spec_folder}/stories/{story_id}-{slug}.md` and proceed to INSTRUCTIONS.
+   - If it points to a file that matches the spec template (has `status` frontmatter with a recognized value: draft, ready-for-dev, in-progress, in-review, or done) → set `spec_file`. Before exiting, run **Story-key resolution** (below). Then **EARLY EXIT** to the appropriate step: `draft` → `[[bmad-snapshot:step-02-plan.md]]`, `ready-for-dev`/`in-progress` → `[[bmad-snapshot:step-03-implement.md]]`, `in-review` → `[[bmad-snapshot:step-04-review.md]]`. For `done`, ingest as context and proceed to INSTRUCTIONS — do not resume.
+   - Anything else (intent files, external docs, plans, descriptions) → ingest it as starting intent and proceed to INSTRUCTIONS. Do not attempt to infer a workflow state from it.
+
+2. Recent conversation
+   Do the last few human messages clearly show what the user intends to work on?
+   Use the same routing as above.
+
+3. Otherwise — scan artifacts and ask
+   - Active specs (`draft`, `ready-for-dev`, `in-progress`, `in-review`) in `{{.implementation_artifacts}}`? → List them and HALT. Ask user which to resume (or `[N]` for new).
+     - If `draft` selected: Set `spec_file`. Run **Story-key resolution** (below). **EARLY EXIT** → `[[bmad-snapshot:step-02-plan.md]]` (resume planning from the draft)
+     - If `ready-for-dev` or `in-progress` selected: Set `spec_file`. Run **Story-key resolution** (below). **EARLY EXIT** → `[[bmad-snapshot:step-03-implement.md]]`
+     - If `in-review` selected: Set `spec_file`. Run **Story-key resolution** (below). **EARLY EXIT** → `[[bmad-snapshot:step-04-review.md]]`
+   - Unformatted spec or intent file lacking `status` frontmatter? → Suggest treating its contents as the starting intent. Do NOT attempt to infer a state and resume it.
+
+Never ask extra questions if you already understand what the user intends.
+
+### Story-key resolution
+
+This runs on ALL paths (early-exit and INSTRUCTIONS) whenever `spec_file` is set. Determine whether the spec is an epic story — use the spec's filename, frontmatter, and any loaded epics file to identify `epic_num` and `story_num`. If the spec is not an epic story, skip silently and leave `story_key` unset.
+
+If the spec is an epic story and `{{.implementation_artifacts}}/sprint-status.yaml` exists: find the `development_status` key matching `{epic_num}-{story_num}` by exact numeric equality on the first two segments (so `1-1` never collides with `1-10`). Exactly one match → set `story_key` to that full key. Zero or multiple matches → leave `story_key` unset (warn on multiple).
+
+## INSTRUCTIONS
+
+1. Load context.
+   - List files in `{{.planning_artifacts}}` and `{{.implementation_artifacts}}`.
+   - If you find an unformatted spec or intent file, ingest its contents to form your understanding of the intent.
+   - **Determine context strategy.** Using the intent and the artifact listing, infer whether the current work is a story from an epic. Do not rely on filename patterns or regex — reason about the intent, the listing, and any epics file content together.
+
+     **A) Epic story path** — if the intent is clearly an epic story:
+
+     1. Identify the epic number `{epic_num}` and (if present) the story number `{story_num}`. If you can't identify an epic number, use path B.
+
+     2. **Check for a valid cached epic context.** Look for `{{.implementation_artifacts}}/epic-<N>-context.md` (where `<N>` is the epic number). A file is **valid** when it exists, is non-empty, starts with `# Epic <N> Context:` (with the correct epic number), and no file in `{{.planning_artifacts}}` is newer.
+        - **If valid:** load it as the primary planning context. Do not load raw planning docs (PRD, architecture, UX, etc.). Skip to step 5.
+        - **If missing, empty, or invalid:** continue to step 3.
+
+     3. **Compile epic context.** Produce `{{.implementation_artifacts}}/epic-<N>-context.md` by following `[[bmad-snapshot:compile-epic-context.md]]`, in order of preference:
+        - **Preferred — subagent:** spawn a subagent synchronously (wait for it to return in this turn) with `[[bmad-snapshot:compile-epic-context.md]]` as its prompt. Pass it the epic number, the epics file path, the `{{.planning_artifacts}}` directory, and the output path `{{.implementation_artifacts}}/epic-<N>-context.md`.
+        - **Fallback — inline** (for runtimes without subagent support, e.g. Copilot, Codex, local Ollama, older Claude): if your runtime cannot spawn subagents, or the spawn fails/times out, read `[[bmad-snapshot:compile-epic-context.md]]` yourself and follow its instructions to produce the same output file.
+
+     4. **Verify.** After compilation, verify the output file exists, is non-empty, and starts with `# Epic <N> Context:`. If valid, load it. If verification fails, HALT and report the failure.
+
+     5. **Previous story continuity.** Regardless of which context source succeeded above, scan `{{.implementation_artifacts}}` for specs from the same epic with `status: done` and a lower story number. Load the most recent one (highest story number below current). Extract its **Code Map**, **Design Notes**, **Spec Change Log**, and **task list** as continuity context for step-02 planning. If no `done` spec is found but an `in-review` spec exists for the same epic with a lower story number, note it to the user and ask whether to load it.
+
+     6. **Resolve `{story_key}`.** If not already set by an earlier early-exit path, run **Story-key resolution** (above) now.
+
+     **B) Freeform path** — if the intent is not an epic story:
+     - Planning artifacts are the output of BMAD phases 1-3. Typical files include:
+       - **PRD** (`*prd*`) — product requirements and success criteria
+       - **Architecture** (`*architecture*`) — technical design decisions and constraints
+       - **UX/Design** (`*ux*`) — user experience and interaction design
+       - **Epics** (`*epic*`) — feature breakdown into implementable stories
+       - **Product Brief** (`*brief*`) — project vision and scope
+     - Scan the listing for files matching these patterns. If any look relevant to the current intent, load them selectively — you don't need all of them, but you need the right constraints and requirements rather than guessing from code alone.
+2. Clarify intent. Do not fantasize, do not leave open questions. If you must ask questions, ask them as a numbered list. When the human replies, verify that every single numbered question was answered. If any were ignored, HALT and re-ask only the missing questions before proceeding. Keep looping until intent is clear enough to implement.
+3. Version control sanity check. Is the working tree clean? Does the current branch make sense for this intent — considering its name and recent history? If the tree is dirty or the branch is an obvious mismatch, HALT and ask the human before proceeding. If version control is unavailable, skip this check.
+4. Multi-goal check (see SCOPE STANDARD). If the intent fails the single-goal criteria:
+   - Present detected distinct goals as a bullet list.
+   - Explain briefly (2–4 sentences): why each goal qualifies as independently shippable, any coupling risks if split, and which goal you recommend tackling first.
+   - HALT and ask human: `[S] Split — pick first goal, defer the rest` | `[K] Keep all goals — accept the risks`
+   - On **S**: For each deferred goal, append one new entry to `{{.implementation_artifacts}}/deferred-work.md` using this format. Do not modify existing entries or look for duplicates. Narrow scope to the first-mentioned goal. Continue routing.
+     ```markdown
+     - source_spec: none
+       summary: <one sentence naming the deferred goal>
+       evidence: <why this was split from the current intent>
+     ```
+   - On **K**: Proceed as-is.
+5. Route — choose exactly one:
+
+   If the explicit spec-folder-plus-story-id pair had no matching story file, keep the colocated `spec_file` selected above. Otherwise, derive a valid kebab-case slug from the clarified intent. If the intent references a tracking identifier (story number, issue number, ticket ID), lead the slug with it (e.g. `3-2-digest-delivery`, `gh-47-fix-auth`). If `{{.implementation_artifacts}}/spec-{slug}.md` already exists: if its status is `draft`, treat it as the same work and resume it (set `spec_file` to that path, **EARLY EXIT** → `[[bmad-snapshot:step-02-plan.md]]`); otherwise append `-2`, `-3`, etc. Set `spec_file` = `{{.implementation_artifacts}}/spec-{slug}.md`.
+
+   **a) One-shot** — zero blast radius: no plausible path by which this change causes unintended consequences elsewhere. Clear intent, no architectural decisions.
+
+   **EARLY EXIT** → `[[bmad-snapshot:step-oneshot.md]]`
+
+   **b) Plan-code-review** — everything else. When uncertain whether blast radius is truly zero, choose this path.
+
+## NEXT
+
+Read fully and follow `[[bmad-snapshot:step-02-plan.md]]`
diff --git a/.agents/skills/bmad-build/step-02-plan.md b/.agents/skills/bmad-build/step-02-plan.md
new file mode 100644
index 00000000..fd575895
--- /dev/null
+++ b/.agents/skills/bmad-build/step-02-plan.md
@@ -0,0 +1,47 @@
+# Step 2: Plan
+
+## RULES
+
+- **Language** — Speak in `{{.communication_language}}`. Write any file output in `{{.document_output_language}}`.
+- No intermediate approvals.
+
+## INSTRUCTIONS
+
+1. Draft resume check. If `{spec_file}` exists with `status: draft`, read it and capture the verbatim `<frozen-after-approval>...</frozen-after-approval>` block as `preserved_intent`. Otherwise `preserved_intent` is empty.
+2. Investigate codebase. _Isolate deep exploration in synchronous subagents/tasks where available. To prevent context snowballing, instruct subagents to give you distilled summaries only._ Decide which findings actually matter for execution — the specific files, symbols/lines, reuse points, and read-only constraints — and carry those forward for the Code Map. This is where the investigation lands: the spec preserves it so it is never re-narrated to the implementer at dispatch time.
+3. Read `[[bmad-snapshot:spec-template.md]]` fully. Fill it out based on the intent and investigation, resolving the template's `date` field to the current system date. Drain the investigation into the `## Code Map` section — annotated paths, symbol/line anchors, reuse pointers, and read-only evidence — so the spec is the implementer's investigation map and the step-03 handoff need only point at it. If `preserved_intent` is non-empty, replace the `<frozen-after-approval>` block in the spec you just filled out with `preserved_intent`, before writing. Write the result to `{spec_file}`.
+4. Self-review against READY FOR DEVELOPMENT standard.
+5. If intent gaps exist, do not fantasize, do not leave open questions, HALT and ask the human.
+6. Token count check (see SCOPE STANDARD). If spec exceeds 1600 tokens:
+   - Show user the token count.
+   - HALT and ask human: `[S] Split — carve off secondary goals` | `[K] Keep full spec — accept the risks`
+   - On **S**: Propose the split — name each secondary goal. For each deferred goal, append one new entry to `{{.implementation_artifacts}}/deferred-work.md` using this format. Do not modify existing entries or look for duplicates. Rewrite the current spec to cover only the main goal — do not surgically carve sections out; regenerate the spec for the narrowed scope. Continue to checkpoint.
+     ```markdown
+     - source_spec: `{spec_file}`
+       summary: <one sentence naming the deferred goal>
+       evidence: <why this was split from the current spec>
+     ```
+   - On **K**: Continue to checkpoint with full spec.
+
+### CHECKPOINT 1
+
+Present summary. Display the spec file path as a CWD-relative path (no leading `/`) so it is clickable in the terminal. If token count exceeded 1600 and user chose [K], include the token count and explain why it may be a problem.
+
+After presenting the summary, display this note:
+
+---
+
+Before approving, you can open the spec file in an editor or ask me questions and tell me what to change. You can also use `bmad-advanced-elicitation`, `bmad-party-mode`, or `bmad-code-review` skills, ideally in another session to avoid context bloat.
+
+---
+
+HALT and ask human: `[A] Approve` | `[E] Edit`
+
+- **A**: Re-read `{spec_file}` from disk.
+  - **If the file is missing:** HALT. Tell the user the spec file is gone and STOP — do not write anything to `{spec_file}`, do not set status, do not proceed to Step 3. Nothing below this point runs.
+  - **If the file exists:** Compare the content to what you wrote. If it has changed since you wrote it, acknowledge the external edits — show a brief summary of what changed — and proceed with the updated version. Then set status `ready-for-dev` in `{spec_file}`. Everything inside `<frozen-after-approval>` is now locked — only the human can change it. → Step 3.
+- **E**: Apply changes, then return to CHECKPOINT 1.
+
+## NEXT
+
+Read fully and follow `[[bmad-snapshot:step-03-implement.md]]`
diff --git a/.agents/skills/bmad-build/step-03-implement.md b/.agents/skills/bmad-build/step-03-implement.md
new file mode 100644
index 00000000..33579f1c
--- /dev/null
+++ b/.agents/skills/bmad-build/step-03-implement.md
@@ -0,0 +1,49 @@
+---
+---
+
+# Step 3: Implement
+
+## RULES
+
+- **Language** — Speak in `{{.communication_language}}`. Write any file output in `{{.document_output_language}}`.
+- No push. No remote ops.
+- Sequential execution only.
+- Content inside `<frozen-after-approval>` in `{spec_file}` is read-only. Do not modify.
+
+## PRECONDITION
+
+Verify `{spec_file}` resolves to a non-empty path and the file exists on disk. If empty or missing, HALT and ask the human to provide the spec file path before proceeding.
+
+## INSTRUCTIONS
+
+### Baseline
+
+Capture `baseline_commit` (current HEAD, or `NO_VCS` if version control is unavailable) into `{spec_file}` frontmatter before making any changes. If the frontmatter already contains `baseline_commit` (resumed run), preserve the existing value — never overwrite it.
+
+### Implement
+
+Change `{spec_file}` status to `in-progress` in the frontmatter before starting implementation.
+
+Follow `[[bmad-snapshot:sync-sprint-status.md]]` with `target_status` = `in-progress`.
+
+Execute the implementation handoff below: substitute the runtime placeholders (e.g. `{spec_file}`) into it, then follow it verbatim.
+
+{workflow.implementation_handoff}
+
+Do not add goal restatements, file lists, ownership boundaries, investigation detail, acceptance criteria, or CLAUDE.md/house-style rules to the dispatch — the spec is the subagent's sole source of truth, and that material already lives in it (investigation findings in its Code Map, the rest in the spec body). One line of sanctioned hedging belongs in the spec at planning time, not in the dispatch. If no subagents are available, implement directly from the spec. If the platform allows, keep the subagent available for re-engagement after it returns — step-04 may send it review fixes.
+
+The handoff directs the subagent to load the spec's `context:` files itself, so never pre-load and paste those files into the dispatch. Only when you implement directly (no subagent available) do you load a non-empty `context:` list yourself before starting.
+
+**Path formatting rule:** Any markdown links written into `{spec_file}` must use paths relative to `{spec_file}`'s directory so they are clickable in VS Code. Any file paths displayed in terminal/conversation output must use CWD-relative format with `:line` notation (e.g., `src/path/file.ts:42`) for terminal clickability. No leading `/` in either case.
+
+### Tasks & Acceptance Verification
+
+Before leaving this step, verify every task in the `## Tasks & Acceptance` section of `{spec_file}` is complete and every acceptance criterion is satisfied. Mark each finished task `[x]`. If any task is not done or any acceptance criterion is not satisfied, finish the missing work before proceeding.
+
+### Matrix Test Audit
+
+If `{spec_file}`'s `<frozen-after-approval>` block contains an I/O & Edge-Case Matrix, verify every matrix row is covered by at least one test that verifies its expected behavior, and that each covering test ran and passed in the verification output. A covering test that exists but did not run — unregistered, filtered out, skipped, or disabled — counts as missing. If a test disagrees with the matrix, never edit the expectation to match the code: fix the code, or if the matrix row itself is ambiguous, HALT and ask the human. Fix any other audit failure before proceeding.
+
+## NEXT
+
+Read fully and follow `[[bmad-snapshot:step-04-review.md]]`
diff --git a/.agents/skills/bmad-build/step-04-review.md b/.agents/skills/bmad-build/step-04-review.md
new file mode 100644
index 00000000..320e2764
--- /dev/null
+++ b/.agents/skills/bmad-build/step-04-review.md
@@ -0,0 +1,55 @@
+# Step 4: Review
+
+## RULES
+
+- **Language** — Speak in `{{.communication_language}}`. Write any file output in `{{.document_output_language}}`.
+- All review subagents must run at the same model capability as the current session.
+- Run subagents synchronously: launch them together, then wait for all results before continuing.
+
+## INSTRUCTIONS
+
+Change `{spec_file}` status to `in-review` in the frontmatter before continuing.
+
+### Construct Diff
+
+Read `{baseline_commit}` from `{spec_file}` frontmatter. If `{baseline_commit}` is missing or `NO_VCS`, use best effort to determine what changed. Otherwise, construct `{diff_output}` covering all changes — tracked and untracked — since `{baseline_commit}`.
+
+Do NOT `git add` anything — this is read-only inspection.
+
+### Review
+
+Execute these review layers in parallel wherever their execution methods allow: substitute the runtime placeholders (e.g. `{diff_output}`) into each layer's instruction. When an instruction launches a reviewer subagent, launch that child with the prompt text after placeholder substitution; do not load the reviewer instruction file yourself. For any other customized instruction, execute it as written. Parallel means several blocking calls awaited together in this turn — never backgrounded or detached, never ending the turn to await results. When running layers as subagents, spawn every reviewer before reading or reacting to any of their output; begin collection and triage only once all are launched.
+
+{workflow.review_layers}
+
+If a layer's instruction requires subagents and none are available, for each such layer write under `{{.implementation_artifacts}}` the exact child prompt from that layer's instruction after placeholder substitution (not a path-only pointer), then HALT. Ask the human to run each in a separate session (ideally a different LLM) and paste back the findings.
+
+### Classify
+
+1. Deduplicate only findings with the same claim and same required action. Then evaluate each remaining finding independently. Do not reject a finding because a related finding was rejected.
+2. Assign severity to each finding by consequence for the artifact's main consumer (software user, document reader, etc).
+   Disregard any severity assigned by a reviewing subagent. Review subagents operate under by-design information asymmetry and do not have enough context to set final severity for this workflow.
+   - `low`: none or cosmetic
+   - `medium`: tolerable
+   - `high`: intolerable
+3. Route each finding into exactly one triage category. The first three categories are **this story's problem** — caused or exposed by the current change. The last two are **not this story's problem**.
+   - **intent_gap** — caused by the change; cannot be resolved from the spec because the captured intent is incomplete. Do not infer intent unless there is exactly one possible reading.
+   - **bad_spec** — caused by the change, including direct deviations from spec. The spec should have been clear enough to prevent it. When in doubt between bad_spec and patch, prefer bad_spec — a spec-level fix is more likely to produce coherent code.
+   - **patch** — caused by the change; trivially fixable without human input. Just part of the diff.
+   - **defer** — pre-existing issue not caused by this story, surfaced incidentally by the review. Collect for later focused attention.
+   - **reject** — noise. Drop silently. When unsure between defer and reject, prefer reject — only defer findings you are confident are real.
+4. Process findings in cascading order. If intent_gap or bad_spec findings exist, they trigger a loopback — lower findings are moot since code will be re-derived. If neither exists, process patch and defer normally. Before each loopback, read `{spec_file}` frontmatter `review_loop_iteration` (missing means `0`), increment it by 1, and write it back. If it exceeds 5, HALT and escalate to the human.
+   - **intent_gap** — Root cause is inside `<frozen-after-approval>`. Revert code changes. Loop back to the human to resolve. Once resolved, read fully and follow `[[bmad-snapshot:step-02-plan.md]]` to re-run steps 2–4.
+   - **bad_spec** — Root cause is outside `<frozen-after-approval>`. Before reverting code: extract KEEP instructions for positive preservation (what worked well and must survive re-derivation). Revert code changes. Read the `## Spec Change Log` in `{spec_file}` and strictly respect all logged constraints when amending the non-frozen sections that contain the root cause. Append a new change-log entry recording: the triggering finding, what was amended, the known-bad state avoided, and the KEEP instructions. Read fully and follow `[[bmad-snapshot:step-03-implement.md]]` to re-derive the code, then this step will run again.
+   - **patch** — Auto-fix. These are the only findings that survive loopbacks. If the step-03 implementation subagent can be re-engaged with its context intact, send it all patch findings in one synchronous message — for each: the file, what is wrong, and what the fix must do. If it cannot be re-engaged, apply the patches yourself. Then re-run the checks in `{spec_file}`'s `## Verification` section, if present; if verification fails and the failure cannot be fixed, HALT and escalate to the human.
+   - **defer** — Append one new entry to `{{.implementation_artifacts}}/deferred-work.md` using this format. Do not modify existing entries or look for duplicates.
+     ```markdown
+     - source_spec: `{spec_file}`
+       summary: <one sentence>
+       evidence: <why this is real>
+     ```
+   - **reject** — Drop silently.
+
+## NEXT
+
+Read fully and follow `[[bmad-snapshot:step-05-present.md]]`
diff --git a/.agents/skills/bmad-build/step-05-present.md b/.agents/skills/bmad-build/step-05-present.md
new file mode 100644
index 00000000..23ba7284
--- /dev/null
+++ b/.agents/skills/bmad-build/step-05-present.md
@@ -0,0 +1,75 @@
+---
+---
+
+# Step 5: Present
+
+## RULES
+
+- **Language** — Speak in `{{.communication_language}}`. Write any file output in `{{.document_output_language}}`.
+- NEVER auto-push.
+
+## INSTRUCTIONS
+
+### Generate Suggested Review Order
+
+Read `{baseline_commit}` from `{spec_file}` frontmatter and construct the diff of all changes since that commit.
+
+Append the review order as a `## Suggested Review Order` section to `{spec_file}` **after the last existing section**. Do not modify the Code Map.
+
+Build the trail as an ordered sequence of **stops** — clickable `path:line` references with brief framing — optimized for a human reviewer reading top-down to understand the change:
+
+1. **Order by concern, not by file.** Group stops by the conceptual concern they address (e.g., "validation logic", "schema change", "UI binding"). A single file may appear under multiple concerns.
+2. **Lead with the entry point** — the single highest-leverage file:line a reviewer should look at first to grasp the design intent.
+3. **Inside each concern**, order stops from most important / architecturally interesting to supporting. Lightly bias toward higher-risk or boundary-crossing stops.
+4. **End with peripherals** — tests, config, types, and other supporting changes come last.
+5. **Every code reference is a clickable spec-file-relative link.** Compute each link target as a relative path from `{spec_file}`'s directory to the changed file. Format each stop as a markdown link: `[short-name:line](../../path/to/file.ts#L42)`. Use a `#L` line anchor. Use the file's basename (or shortest unambiguous suffix) plus line number as the link text. The relative path must be dynamically derived — never hardcode the depth.
+6. **Each stop gets one ultra-concise line of framing** (≤15 words) — why this approach was chosen here and what it achieves in the context of the change. No paragraphs.
+
+Format each stop as framing first, link on the next indented line:
+
+```markdown
+## Suggested Review Order
+
+**{Concern name}**
+
+- {one-line framing}
+  [`file.ts:42`](../../src/path/to/file.ts#L42)
+
+- {one-line framing}
+  [`other.ts:17`](../../src/path/to/other.ts#L17)
+
+**{Next concern}**
+
+- {one-line framing}
+  [`file.ts:88`](../../src/path/to/file.ts#L88)
+```
+
+> The `../../` prefix above is illustrative — compute the actual relative path from `{spec_file}`'s directory to each target file.
+
+When there is only one concern, omit the bold label — just list the stops directly.
+
+### Mark Spec Done
+
+Change `{spec_file}` status to `done` in the frontmatter.
+
+Follow `[[bmad-snapshot:sync-sprint-status.md]]` with `target_status` = `review`.
+
+### Commit and Complete
+
+If version control is available and the tree is dirty, create a local commit with a conventional message derived from the spec title.
+
+{workflow.open_spec}
+
+### Display Summary
+
+Display summary of your work to the user, including the commit hash if one was created. Any file paths shown in conversation/terminal output must use CWD-relative format (no leading `/`) with `:line` notation (e.g., `src/path/file.ts:42`) for terminal clickability — the goal is to make paths clickable in terminal emulators.
+
+Offer to push and/or create a pull request.
+
+Workflow complete.
+
+## On Complete
+
+If anything appears below, follow it as the final terminal instruction before exiting; otherwise exit normally.
+
+{workflow.on_complete}
diff --git a/.agents/skills/bmad-build/step-oneshot.md b/.agents/skills/bmad-build/step-oneshot.md
new file mode 100644
index 00000000..eaf9aebe
--- /dev/null
+++ b/.agents/skills/bmad-build/step-oneshot.md
@@ -0,0 +1,77 @@
+# Step One-Shot: Implement, Review, Present
+
+## RULES
+
+- **Language** — Speak in `{{.communication_language}}`. Write any file output in `{{.document_output_language}}`.
+- NEVER auto-push.
+- All review subagents must run at the same model capability as the current session.
+- Run subagents synchronously: launch them together, then wait for all results before continuing.
+
+## INSTRUCTIONS
+
+### Implement
+
+Follow `[[bmad-snapshot:sync-sprint-status.md]]` with `target_status` = `in-progress`.
+
+Implement the clarified intent directly.
+
+### Review
+
+Execute these review layers in parallel wherever their execution methods allow. After substituting runtime placeholders, when an instruction launches a reviewer subagent, launch that child with the prompt text; do not load the reviewer instruction file yourself. For any other customized instruction, execute it as written:
+
+{workflow.oneshot_review_layers}
+
+If a layer's instruction requires subagents and none are available, for each such layer write under `{{.implementation_artifacts}}` the exact child prompt from that layer's instruction after placeholder substitution (not a path-only pointer), then HALT. Ask the human to run each in a separate session and paste back the findings.
+
+### Classify
+
+Deduplicate all review findings. Three categories only:
+
+- **patch** — trivially fixable. Auto-fix immediately.
+- **defer** — pre-existing issue not caused by this change. Append one new entry to `{{.implementation_artifacts}}/deferred-work.md` using this format. Do not modify existing entries or look for duplicates.
+  ```markdown
+  - source_spec: `{spec_file}`
+    summary: <one sentence>
+    evidence: <why this is real>
+  ```
+- **reject** — noise. Drop silently.
+
+If a finding is caused by this change but too significant for a trivial patch, HALT and present it to the human for decision before proceeding.
+
+### Generate Spec Trace
+
+Set `title` = a concise title derived from the clarified intent.
+
+Write `{spec_file}` using `[[bmad-snapshot:spec-template.md]]`. Fill only these sections — delete all others:
+
+1. **Frontmatter** — set `title: '{title}'`, `type`, `created`, `status: 'done'`. Add `route: 'one-shot'`.
+2. **Title and Intent** — `# {title}` heading and `## Intent` with **Problem** and **Approach** lines. Reuse the summary you already generated for the terminal.
+3. **Suggested Review Order** — append after Intent. Build using the same convention as `[[bmad-snapshot:step-05-present.md]]` § "Generate Suggested Review Order" (spec-file-relative links, concern-based ordering, ultra-concise framing).
+
+Follow `[[bmad-snapshot:sync-sprint-status.md]]` with `target_status` = `review`.
+
+### Commit
+
+If version control is available and the tree is dirty, create a local commit with a conventional message derived from the intent. If VCS is unavailable, skip.
+
+### Present
+
+{workflow.open_spec}
+
+Display a summary in conversation output, including:
+
+- The commit hash (if one was created).
+- List of files changed with one-line descriptions. Any file paths shown in conversation/terminal output must use CWD-relative format (no leading `/`) with `:line` notation (e.g., `src/path/file.ts:42`) for terminal clickability — this differs from spec-file links which use spec-file-relative paths.
+- Review findings breakdown: patches applied, items deferred, items rejected. If all findings were rejected, say so.
+
+Offer to push and/or create a pull request.
+
+HALT and wait for human input.
+
+Workflow complete.
+
+## On Complete
+
+If anything appears below, follow it as the final terminal instruction before exiting; otherwise exit normally.
+
+{workflow.on_complete}
diff --git a/.agents/skills/bmad-build/sync-sprint-status.md b/.agents/skills/bmad-build/sync-sprint-status.md
new file mode 100644
index 00000000..0fe825ef
--- /dev/null
+++ b/.agents/skills/bmad-build/sync-sprint-status.md
@@ -0,0 +1,19 @@
+# Sync Sprint Status
+
+Shared sub-step for updating `sprint-status.yaml` during build. Called from any route (plan-code-review, one-shot, future routes) with a `target_status` parameter.
+
+## Preconditions
+
+Skip this entire file (return to caller) if ANY of:
+- `story_key` is unset
+- `{{.implementation_artifacts}}/sprint-status.yaml` does not exist on disk
+
+## Instructions
+
+1. Load the FULL `{{.implementation_artifacts}}/sprint-status.yaml` file.
+2. Find the `development_status` entry matching `{story_key}`. If not found, warn the user once (`"{story_key} not found in sprint-status; skipping sprint sync"`) and return to caller.
+3. **Idempotency check.** If `development_status[{story_key}]` is already at `target_status` or a later state (`review` is later than `in-progress`; `done` is later than both), return to caller — no write needed. Never regress a story's status.
+4. Set `development_status[{story_key}]` to `{target_status}`.
+5. **Epic lift (only when `target_status` = `in-progress`).** Derive the parent epic key as `epic-{N}` from the leading numeric segment of `{story_key}` (e.g., `3-2-digest-delivery` → `epic-3`). If that entry exists and is `backlog`, set it to `in-progress`. Leave it alone otherwise. Skip this sub-step entirely when `target_status` is not `in-progress`.
+6. Refresh `last_updated` to the current date.
+7. Save the file, preserving ALL comments and structure including STATUS DEFINITIONS and WORKFLOW NOTES.
diff --git a/.agents/skills/bmad-build/workflow.md b/.agents/skills/bmad-build/workflow.md
new file mode 100644
index 00000000..ce7a80ac
--- /dev/null
+++ b/.agents/skills/bmad-build/workflow.md
@@ -0,0 +1,84 @@
+# Build New Preview Workflow
+
+**Goal:** Turn user intent into a hardened, reviewable artifact.
+
+**CRITICAL:** If a step directs you to another snapshot file, read it fully and follow it. No exceptions.
+
+Subagents, when the capability is available, are an important part of this workflow. Use them as directed by the workflow steps.
+If you need an explicit user instruction to run them, ask once now for the whole workflow run.
+
+## READY FOR DEVELOPMENT STANDARD
+
+A specification is "Ready for Development" when:
+
+- **Actionable**: Every task has a file path and specific action.
+- **Logical**: Tasks ordered by dependency.
+- **Testable**: All ACs use Given/When/Then.
+- **Complete**: No placeholders or TBDs.
+- **Sufficient**: No known requirement, acceptance, dependency, or implementation gaps remain unresolved.
+- **Coherent**: No unresolved ambiguities or internal contradictions.
+
+## SCOPE STANDARD
+
+A specification should target a **single user-facing goal** within **900–1600 tokens**:
+
+- **Single goal**: One cohesive feature, even if it spans multiple layers/files. Multi-goal means >=2 **top-level independent shippable deliverables** — each could be reviewed, tested, and merged as a separate PR without breaking the others. Never count surface verbs, "and" conjunctions, or noun phrases. Never split cross-layer implementation details inside one user goal.
+  - Split: "add dark mode toggle AND refactor auth to JWT AND build admin dashboard"
+  - Don't split: "add validation and display errors" / "support drag-and-drop AND paste AND retry"
+- **900–1600 tokens**: Optimal range for LLM consumption. Below 900 risks ambiguity; above 1600 risks context-rot in implementation agents.
+- **Neither limit is a gate.** Both are proposals with user override.
+
+## Conventions
+
+- Every operational cross-file reference in this workflow is an absolute snapshot path. Open it directly; do not resolve it relative to a skill directory.
+- `{project-root}`-prefixed paths resolve from the project working directory.
+- Whenever this workflow captures or records a version-control revision, obtain the full canonical identifier directly from version control and preserve it verbatim.
+
+## On Activation
+
+### Step 1: Execute Prepend Steps
+
+Execute each of these steps in order before proceeding (`_None._` means skip):
+
+{workflow.activation_steps_prepend}
+
+### Step 2: Load Persistent Facts
+
+Treat every entry below as foundational context you carry for the rest of the workflow run. Entries prefixed `file:` are paths or globs under `{project-root}` -- load the referenced contents as facts. All other entries are facts verbatim (`_None._` means none):
+
+{workflow.persistent_facts}
+
+### Step 3: Execute Append Steps
+
+Execute each of these steps in order (`_None._` means skip):
+
+{workflow.activation_steps_append}
+
+## WORKFLOW ARCHITECTURE
+
+This uses **step-file architecture** for disciplined execution:
+
+- **Micro-file Design**: Each step is self-contained and followed exactly
+- **Just-In-Time Loading**: Only load the current step file
+- **Sequential Enforcement**: Complete steps in order, no skipping
+- **State Tracking**: Persist progress via spec frontmatter and in-memory variables
+- **Append-Only Building**: Build artifacts incrementally
+
+### Step Processing Rules
+
+1. **READ COMPLETELY**: Read the entire step file before acting
+2. **FOLLOW SEQUENCE**: Execute sections in order
+3. **WAIT FOR INPUT**: Halt at checkpoints and wait for human
+4. **LOAD NEXT**: When directed, read fully and follow the next step file
+
+### Critical Rules (NO EXCEPTIONS)
+
+- **NEVER** load multiple step files simultaneously
+- **ALWAYS** read entire step file before execution
+- **NEVER** skip steps or optimize the sequence
+- **ALWAYS** follow the exact instructions in the step file
+- **ALWAYS** halt at checkpoints and wait for human input
+
+## FIRST STEP
+
+Read fully and follow: `[[bmad-snapshot:step-01-clarify-and-route.md]]` to begin the workflow.
diff --git a/.agents/skills/bmad-checkpoint-preview/SKILL.md b/.agents/skills/bmad-checkpoint-preview/SKILL.md
new file mode 100644
index 00000000..06e45d6f
--- /dev/null
+++ b/.agents/skills/bmad-checkpoint-preview/SKILL.md
@@ -0,0 +1,68 @@
+---
+name: bmad-checkpoint-preview
+description: 'LLM-assisted human-in-the-loop review. Make sense of a change, focus attention where it matters, test. Use when the user says "checkpoint", "human review", or "walk me through this change".'
+---
+
+# Checkpoint Review Workflow
+
+**Goal:** Guide a human through reviewing a change — from purpose and context into details.
+
+**Your Role:** You are assisting the user in reviewing a change.
+
+## Conventions
+
+- Bare paths (e.g. `step-01-orientation.md`) resolve from the skill root.
+- `{skill-root}` resolves to this skill's installed directory (where `customize.toml` lives).
+- `{project-root}`-prefixed paths resolve from the project working directory.
+- `{skill-name}` resolves to the skill directory's basename.
+
+## On Activation
+
+### Step 1: Resolve the Workflow Block
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`
+
+**If the script fails**, resolve the `workflow` block yourself by reading these three files in base → team → user order and applying the same structural merge rules as the resolver:
+
+1. `{skill-root}/customize.toml` — defaults
+2. `{project-root}/_bmad/custom/{skill-name}.toml` — team overrides
+3. `{project-root}/_bmad/custom/{skill-name}.user.toml` — personal overrides
+
+Any missing file is skipped. Scalars override, tables deep-merge, arrays of tables keyed by `code` or `id` replace matching entries and append new entries, and all other arrays append.
+
+### Step 2: Execute Prepend Steps
+
+Execute each entry in `{workflow.activation_steps_prepend}` in order before proceeding.
+
+### Step 3: Load Persistent Facts
+
+Treat every entry in `{workflow.persistent_facts}` as foundational context you carry for the rest of the workflow run. Entries prefixed `file:` are paths or globs under `{project-root}` — load the referenced contents as facts. All other entries are facts verbatim.
+
+### Step 4: Load Config
+
+Load config from `{project-root}/_bmad/bmm/config.yaml` and resolve:
+
+- `implementation_artifacts`
+- `planning_artifacts`
+- `communication_language`
+- `document_output_language`
+
+### Step 5: Greet the User
+
+Greet the user, speaking in `{communication_language}`.
+
+### Step 6: Execute Append Steps
+
+Execute each entry in `{workflow.activation_steps_append}` in order.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+## Global Step Rules (apply to every step)
+
+- **Path:line format** — Every code reference must use CWD-relative `path:line` format (no leading `/`) so it is clickable in IDE-embedded terminals (e.g., `src/auth/middleware.ts:42`).
+- **Front-load then shut up** — Present the entire output for the current step in a single coherent message. Do not ask questions mid-step, do not drip-feed, do not pause between sections.
+- **Language** — Speak in `{communication_language}`. Write any file output in `{document_output_language}`.
+
+## FIRST STEP
+
+Read fully and follow `./step-01-orientation.md` to begin.
diff --git a/.agents/skills/bmad-checkpoint-preview/customize.toml b/.agents/skills/bmad-checkpoint-preview/customize.toml
new file mode 100644
index 00000000..2f9b034a
--- /dev/null
+++ b/.agents/skills/bmad-checkpoint-preview/customize.toml
@@ -0,0 +1,41 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-checkpoint-preview. Mirrors the
+# agent customization shape under the [workflow] namespace.
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+# Steps to run before the standard activation (config load, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Distinct from the runtime memory sidecar — these are static context
+# loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "All stories must include testable acceptance criteria."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Scalar: executed when the workflow reaches its final step,
+# after the review decision (approve/rework/discuss) is made. Override wins.
+# Leave empty for no custom post-completion behavior.
+
+on_complete = ""
diff --git a/.agents/skills/bmad-checkpoint-preview/generate-trail.md b/.agents/skills/bmad-checkpoint-preview/generate-trail.md
new file mode 100644
index 00000000..6fd378bd
--- /dev/null
+++ b/.agents/skills/bmad-checkpoint-preview/generate-trail.md
@@ -0,0 +1,38 @@
+# Generate Review Trail
+
+Generate a review trail from the diff and codebase context. A generated trail is lower quality than an author-produced one, but far better than none.
+
+## Follow Global Step Rules in SKILL.md
+
+## INSTRUCTIONS
+
+1. Get the full diff against the appropriate baseline (same rules as Surface Area Stats in step-01).
+2. Read changed files in full — not just diff hunks. Surrounding code reveals intent that hunks alone miss. If total file content exceeds ~50k tokens, read only the files with the largest diff hunks in full and use hunks for the rest.
+3. If a spec exists, use its Intent section to anchor concern identification.
+4. Identify 2–5 concerns: cohesive design intents that each explain *why* behind a cluster of changes. Prefer functional groupings and architectural boundaries over file-level splits. A single-concern change is fine — don't invent groupings.
+5. For each concern, select 1–4 `path:line` stops — locations where the concern is most visible. Prefer entry points, decision points, and boundary crossings over mechanical changes.
+6. Lead with the entry point — the highest-leverage stop a reviewer should see first. Inside each concern, order stops so each builds on the previous. End with peripherals (tests, config, types).
+7. Format each stop using `path:line` per the global step rules:
+
+```
+**{Concern name}**
+
+- {one-line framing, ≤15 words}
+  `src/path/to/file.ts:42`
+```
+
+When there is only one concern, omit the bold label — just list the stops directly.
+
+## PRESENT
+
+Output after the orientation:
+
+```
+I built a review trail for this {change_type} (no author-produced trail was found):
+
+{generated trail}
+```
+
+The generated trail serves as the Suggested Review Order for subsequent steps. Set `review_mode` to `full-trail` — a trail now exists, so all downstream steps should treat it as one.
+
+If git is unavailable or the diff cannot be retrieved, return to step-01 with: "Could not generate trail — git unavailable."
diff --git a/.agents/skills/bmad-checkpoint-preview/step-01-orientation.md b/.agents/skills/bmad-checkpoint-preview/step-01-orientation.md
new file mode 100644
index 00000000..26f3554d
--- /dev/null
+++ b/.agents/skills/bmad-checkpoint-preview/step-01-orientation.md
@@ -0,0 +1,105 @@
+# Step 1: Orientation
+
+Display: `[Orientation] → Walkthrough → Detail Pass → Testing`
+
+## Follow Global Step Rules in SKILL.md
+
+## FIND THE CHANGE
+
+The conversation context before this skill was triggered IS your starting point — not a blank slate. Check in this order — stop as soon as the change is identified:
+
+1. **Explicit argument**
+   Did the user pass a PR, commit SHA, branch, or spec file this message?
+   - PR reference → resolve to branch/commit via `gh pr view`. If resolution fails, ask for a SHA or branch.
+   - Spec file, commit, or branch → use directly.
+
+2. **Recent conversation**
+   Do the last few messages reveal what change the user wants reviewed? Look for spec paths, commit refs, branches, PRs, or descriptions of a change. Use the same routing as above.
+
+3. **Sprint tracking**
+   Check for a sprint status file (`*sprint-status*`) in `{implementation_artifacts}` or `{planning_artifacts}`. If found, scan for stories with status `review`:
+   - Exactly one → suggest it and confirm with the user.
+   - Multiple → present as numbered options.
+   - None → fall through.
+
+4. **Current git state**
+   Check current branch and HEAD. Confirm: "I see HEAD is `<short-sha>` on `<branch>` — is this the change you want to review?"
+
+5. **Ask**
+   If none of the above identified a change, ask:
+   - What changed and why?
+   - Which commit, branch, or PR should I look at?
+   - Do you have a spec, bug report, or anything else that explains what this change is supposed to do?
+
+   If after 3 exchanges you still can't identify a change, HALT.
+
+Never ask extra questions beyond what the cascade prescribes. If a step above already identified the change, skip the remaining steps.
+
+## ENRICH
+
+Once a change is identified from any source above, fill in the complementary artifact:
+
+- If you have a spec, look for `baseline_commit` in its frontmatter to determine the diff baseline.
+- If you have a commit or branch, check `{implementation_artifacts}` for a spec whose `baseline_commit` is an ancestor of that commit/branch (i.e., the spec describes work done on top of that baseline).
+- If you found both a spec and a commit/branch, use both.
+
+## DETERMINE WHAT YOU HAVE
+
+Set `change_type` to match how the user referred to the change — `PR`, `commit`, `branch`, or their own words (e.g. `auth refactor`). Default to `change` if ambiguous.
+
+Set `review_mode` — pick the first match:
+
+1. **`full-trail`** — ENRICH found a spec with a `## Suggested Review Order` section. Intent source: spec's Intent section.
+2. **`spec-only`** — ENRICH found a spec but it has no Suggested Review Order. Intent source: spec's Intent section.
+3. **`bare-commit`** — no spec found. Intent source: commit message. If the commit message is terse (under 10 words), scan the diff for the primary change pattern and draft a one-sentence intent. Flag it as `[inferred]` in the output so the user can correct it.
+
+## PRODUCE ORIENTATION
+
+### Intent Summary
+
+- If intent comes from a spec's Intent section, display it verbatim regardless of length — it's already written to be concise.
+- For other sources (commit messages, bug reports, user description): if ≤200 tokens, display verbatim. If longer, distill to ≤200 tokens. Link to the full source when one exists (e.g. a file path or URL).
+- Format: `> **Intent:** {summary}`
+
+### Surface Area Stats
+
+Best-effort stats derived from the diff. Try these baselines in order:
+
+1. `baseline_commit` from the spec's frontmatter.
+2. Branch merge-base against `main` (or the default branch).
+3. `HEAD~1..HEAD` (latest commit only — tell the user).
+4. If git is unavailable or all of the above fail, skip stats and note: "Could not compute stats."
+
+Use `git diff --stat` and `git diff --numstat` for file-level counts, and scan the full diff content for the richer metrics.
+
+Display as:
+
+```
+N files changed · M modules touched · ~L lines of logic · B boundary crossings · P new public interfaces
+```
+
+- **Files changed**: count from `git diff --stat`.
+- **Modules touched**: distinct top-level directories with changes (from `--stat` file paths).
+- **Lines of logic**: added/modified lines excluding blanks, imports, formatting. Scan diff content; `~` because approximate.
+- **Boundary crossings**: changes spanning more than one top-level module. `0` if single module.
+- **New public interfaces**: new exports, endpoints, public methods found in the diff. `0` if none.
+
+Omit any metric you cannot compute rather than guessing.
+
+### Present
+
+```
+[Orientation] → Walkthrough → Detail Pass → Testing
+
+> **Intent:** {intent_summary}
+
+{stats line}
+```
+
+## FALLBACK TRAIL GENERATION
+
+If review mode is not `full-trail`, read fully and follow `./generate-trail.md` to build one from the diff. Then return here and continue to NEXT. If trail generation fails (e.g., git unavailable), the original review mode is preserved — step-02 handles this with its non-trail path.
+
+## NEXT
+
+Read fully and follow `./step-02-walkthrough.md`
diff --git a/.agents/skills/bmad-checkpoint-preview/step-02-walkthrough.md b/.agents/skills/bmad-checkpoint-preview/step-02-walkthrough.md
new file mode 100644
index 00000000..aec40c4c
--- /dev/null
+++ b/.agents/skills/bmad-checkpoint-preview/step-02-walkthrough.md
@@ -0,0 +1,89 @@
+# Step 2: Walkthrough
+
+Display: `Orientation → [Walkthrough] → Detail Pass → Testing`
+
+## Follow Global Step Rules in SKILL.md
+
+- Organize by **concern**, not by file. A concern is a cohesive design intent — e.g., "input validation," "state management," "API contract." One file may appear under multiple concerns; one concern may span multiple files.
+- The walkthrough activates **design judgment**, not correctness checking. Frame each concern as "here's what this change does and why" — the human evaluates whether it's the right approach for the system.
+
+## BUILD THE WALKTHROUGH
+
+### Identify Concerns
+
+**With Suggested Review Order** (`full-trail` mode — the normal path, including when step-01 generated a trail):
+
+1. Read the Suggested Review Order stops from the spec (or from conversation context if generated by step-01 fallback).
+2. Resolve each stop to a file in the current repo. Output in `path:line` format per the standing rule.
+3. Read the diff to understand what each stop actually does.
+4. Group stops by concern. Stops that share a design intent belong together even if they're in different files. A stop may appear under multiple concerns if it serves multiple purposes.
+
+**Without Suggested Review Order** (fallback when trail generation failed, e.g., git unavailable):
+
+1. Get the diff against the appropriate baseline (same rules as step 1).
+2. Identify concerns by reading the diff for cohesive design intents:
+   - Functional groupings — what user-facing behavior does each cluster of changes support?
+   - Architectural layers — does the change cross boundaries (API → service → data)?
+   - Design decisions — where did the author choose between alternatives?
+3. For each concern, identify the key code locations as `path:line` stops.
+
+### Order for Comprehension
+
+Sequence concerns top-down: start with the highest-level intent (the "what and why"), then drill into supporting implementation. Within each concern, order stops so each one builds on the previous. The reader should never encounter a reference to something they haven't seen yet.
+
+If the change has a natural entry point (e.g., a new public API, a config change, a UI entry point), lead with it.
+
+### Write Each Concern
+
+For each concern, produce:
+
+1. **Heading** — a short phrase naming the design intent (not a file name, not a module name).
+2. **Why** — 1–2 sentences: what problem this concern addresses, why this approach was chosen over alternatives. If the spec documents rejected alternatives, reference them here.
+3. **Stops** — each stop on its own line: `path:line` followed by a brief phrase (not a sentence) describing what this location does for the concern. Keep framing under 15 words per stop.
+
+Target 2–5 concerns for a typical change. A single-concern change is fine — don't invent groupings. A change with more than 7 concerns is a signal the scope may be too large, but present it anyway.
+
+## PRESENT
+
+Output the full walkthrough as a single message with this structure:
+
+```
+Orientation → [Walkthrough] → Detail Pass → Testing
+```
+
+Then each concern group using this format:
+
+```
+### {Concern Heading}
+
+{Why — 1–2 sentences}
+
+- `path:line` — {brief framing}
+- `path:line` — {brief framing}
+- ...
+```
+
+End the message with:
+
+```
+---
+
+Take your time — click through the stops, read the diff, trace the logic. While you are reviewing, you can:
+- "run advanced elicitation on the error handling"
+- "party mode on whether this schema migration is safe"
+- or just ask anything
+
+When you're ready, say **next** and I'll surface the highest-risk spots.
+```
+
+## EARLY EXIT
+
+If at any point the human signals they want to make a decision about this {change_type} (e.g., "let's ship it", "this needs a rethink", "I'm done reviewing", or anything suggesting they're ready to decide), confirm their intent:
+
+- If they want to **approve and ship** → read fully and follow `./step-05-wrapup.md`
+- If they want to **reject and rework** → read fully and follow `./step-05-wrapup.md`
+- If you misread them → acknowledge and continue the current step.
+
+## NEXT
+
+Default: read fully and follow `./step-03-detail-pass.md`
diff --git a/.agents/skills/bmad-checkpoint-preview/step-03-detail-pass.md b/.agents/skills/bmad-checkpoint-preview/step-03-detail-pass.md
new file mode 100644
index 00000000..49d8024a
--- /dev/null
+++ b/.agents/skills/bmad-checkpoint-preview/step-03-detail-pass.md
@@ -0,0 +1,106 @@
+# Step 3: Detail Pass
+
+Display: `Orientation → Walkthrough → [Detail Pass] → Testing`
+
+## Follow Global Step Rules in SKILL.md
+
+- The detail pass surfaces what the human should **think about**, not what the code got wrong. Machine hardening already handled correctness. This activates risk awareness.
+- The LLM detects risk category by pattern. The human judges significance. Do not assign severity scores or numeric rankings — ordering by blast radius (below) is sequencing for readability, not a severity judgment.
+- If no high-risk spots exist, say so explicitly. Do not invent findings.
+
+## IDENTIFY RISK SPOTS
+
+Scan the diff for changes touching risk-sensitive patterns. Look for 2–5 spots where a mistake would have the highest blast radius — not the most complex code, but the code where being wrong costs the most.
+
+Risk categories to detect:
+
+- `[auth]` — authentication, authorization, session, token, permission, access control
+- `[public API]` — new/changed endpoints, exports, public methods, interface contracts
+- `[schema]` — database migrations, schema changes, data model modifications, serialization
+- `[billing]` — payment, pricing, subscription, metering, usage tracking
+- `[infra]` — deployment, CI/CD, environment variables, config files, infrastructure
+- `[security]` — input validation, sanitization, crypto, secrets, CORS, CSP
+- `[config]` — feature flags, environment-dependent behavior, defaults
+- `[other]` — anything risk-sensitive that doesn't fit the above (e.g., concurrency, data privacy, backwards compatibility). Use a descriptive tag.
+
+Sequence spots so the highest blast radius comes first (how much breaks if this is wrong), not by diff order or file order. If more than 5 spots qualify, show the top 5 and note: "N additional spots omitted — ask if you want the full list."
+
+If the change has no spots matching these patterns, state: "No high-risk spots found in this change — the diff speaks for itself." Do not force findings.
+
+## SURFACE MACHINE HARDENING FINDINGS
+
+Check whether the spec has a `## Spec Change Log` section with entries (populated by adversarial review loops).
+
+- **If entries exist:** Read them. Surface findings that are instructive for the human reviewer — not bugs that were already fixed, but decisions the review loop flagged that the human should be aware of. Format: brief summary of what was flagged and what was decided.
+- **If no entries or no spec:** Skip this section entirely. Do not mention it.
+
+## PRESENT
+
+Output as a single message:
+
+```
+Orientation → Walkthrough → [Detail Pass] → Testing
+```
+
+### Risk Spots
+
+For each spot, one line:
+
+```
+- `path:line` — [tag] reason-phrase
+```
+
+Example:
+
+```
+- `src/auth/middleware.ts:42` — [auth] New token validation bypasses rate limiter
+- `migrations/003_add_index.sql:7` — [schema] Index on high-write table, check lock behavior
+- `api/routes/billing.ts:118` — [billing] Metering calculation changed, verify idempotency
+```
+
+### Machine Hardening (only if findings exist)
+
+```
+### Machine Hardening
+
+- Finding summary — what was flagged, what was decided
+- ...
+```
+
+### Closing menu
+
+End the message with:
+
+```
+---
+
+You've seen the design and the risk landscape. From here:
+- **"dig into [area]"** — I'll deep-dive that specific area with correctness focus
+- **"next"** — I'll suggest how to observe the behavior
+```
+
+## EARLY EXIT
+
+If at any point the human signals they want to make a decision about this {change_type} (e.g., "let's ship it", "this needs a rethink", "I'm done reviewing", or anything suggesting they're ready to decide), confirm their intent:
+
+- If they want to **approve and ship** → read fully and follow `./step-05-wrapup.md`
+- If they want to **reject and rework** → read fully and follow `./step-05-wrapup.md`
+- If you misread them → acknowledge and continue the current step.
+
+## TARGETED RE-REVIEW
+
+When the human says "dig into [area]" (e.g., "dig into the auth changes", "dig into the schema migration"):
+
+1. If the specified area does not map to any code in the diff, say so: "I don't see [area] in this change — did you mean something else?" Return to the closing menu.
+2. Identify all code locations in the diff relevant to the specified area.
+3. Read each location in full context (not just the diff hunk — read surrounding code).
+4. Shift to **correctness mode**: trace edge cases, check boundary conditions, verify error handling, look for off-by-one errors, race conditions, resource leaks.
+5. Present findings as a compact list — each finding is `path:line` + what you found + why it matters.
+6. If nothing concerning is found, say so: "Looked closely at [area] — nothing concerning. The implementation is solid."
+7. After presenting, show only the closing menu (not the full risk spots list again).
+
+The human can trigger multiple targeted re-reviews. Each time, present new findings and the closing menu only.
+
+## NEXT
+
+Read fully and follow `./step-04-testing.md`
diff --git a/.agents/skills/bmad-checkpoint-preview/step-04-testing.md b/.agents/skills/bmad-checkpoint-preview/step-04-testing.md
new file mode 100644
index 00000000..f8180799
--- /dev/null
+++ b/.agents/skills/bmad-checkpoint-preview/step-04-testing.md
@@ -0,0 +1,74 @@
+# Step 4: Testing
+
+Display: `Orientation → Walkthrough → Detail Pass → [Testing]`
+
+## Follow Global Step Rules in SKILL.md
+
+- This is **experiential**, not analytical. The detail pass asked "did you think about X?" — this says "you could see X with your own eyes."
+- Do not prescribe. The human decides whether observing the behavior is worth their time. Frame suggestions as options, not obligations.
+- Do not duplicate CI, test suites, or automated checks. Assume those exist and work. This is about manual observation — the kind of confidence-building no automated test provides.
+- If the change has no user-visible behavior, say so explicitly. Do not invent observations.
+
+## IDENTIFY OBSERVABLE BEHAVIOR
+
+Scan the diff and spec for changes that produce behavior a human could directly observe. Categories to look for:
+
+- **UI changes** — new screens, modified layouts, changed interactions, error states
+- **CLI/terminal output** — new commands, changed output, new flags or options
+- **API responses** — new endpoints, changed payloads, different status codes
+- **State changes** — database records, file system artifacts, config effects
+- **Error paths** — bad input, missing dependencies, edge conditions
+
+For each observable behavior, determine:
+
+1. **What to do** — the specific action (command to run, button to click, request to send)
+2. **What to expect** — the observable result that confirms the change works
+3. **Why bother** — one phrase connecting this observation to the change's intent (omit if obvious from context)
+
+Target 2–5 suggestions for a typical change. If more than 5 qualify, prioritize by how much confidence the observation provides relative to effort. A change with zero observable behavior is fine — do not pad with trivial observations.
+
+## PRESENT
+
+Output as a single message:
+
+```
+Orientation → Walkthrough → Detail Pass → [Testing]
+```
+
+Then the testing suggestions using this format:
+
+```
+### How to See It Working
+
+**{Brief description}**
+Do: {specific action}
+Expect: {observable result}
+
+**{Brief description}**
+Do: {specific action}
+Expect: {observable result}
+```
+
+Include code blocks for commands or requests where helpful.
+
+If the change has no observable behavior, replace the suggestions with:
+
+```
+### How to See It Working
+
+This change is internal — no user-visible behavior to observe. The diff and tests tell the full story.
+```
+
+### Closing
+
+End the message with:
+
+```
+---
+
+You've seen the change and how to verify it. When you're ready to make a call, just say so.
+```
+
+## NEXT
+
+When the human signals they're ready to make a decision about this {change_type}, read fully and follow `./step-05-wrapup.md`
diff --git a/.agents/skills/bmad-checkpoint-preview/step-05-wrapup.md b/.agents/skills/bmad-checkpoint-preview/step-05-wrapup.md
new file mode 100644
index 00000000..dc4363bc
--- /dev/null
+++ b/.agents/skills/bmad-checkpoint-preview/step-05-wrapup.md
@@ -0,0 +1,30 @@
+# Step 5: Wrap-Up
+
+Display: `Orientation → Walkthrough → Detail Pass → Testing → [Wrap-Up]`
+
+## Follow Global Step Rules in SKILL.md
+
+## PROMPT FOR DECISION
+
+```
+---
+
+Review complete. What's the call on this {change_type}?
+- **Approve** — ship it (I can help with interactive patching first if needed)
+- **Rework** — back to the drawing board (revert, revise the spec, try a different approach)
+- **Discuss** — something's still on your mind
+```
+
+HALT — do not proceed until the user makes their choice.
+
+## ACT ON DECISION
+
+- **Approve**: Acknowledge briefly. If the human wants to patch something before shipping, help apply the fix interactively. If reviewing a PR, offer to approve via `gh pr review --approve` — but confirm with the human before executing, since this is a visible action on a shared resource.
+- **Rework**: Ask what went wrong — was it the approach, the spec, or the implementation? Help the human decide on next steps (revert commit, open an issue, revise the spec, etc.). Help draft specific, actionable feedback tied to `path:line` locations if the change is a PR from someone else.
+- **Discuss**: Open conversation — answer questions, explore concerns, dig into any aspect. After discussion, return to the decision prompt above.
+
+## On Complete
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow.on_complete`
+
+If the resolved `workflow.on_complete` is non-empty, follow it as the final terminal instruction before exiting.
diff --git a/.agents/skills/bmad-code-review/SKILL.md b/.agents/skills/bmad-code-review/SKILL.md
new file mode 100644
index 00000000..83cea86b
--- /dev/null
+++ b/.agents/skills/bmad-code-review/SKILL.md
@@ -0,0 +1,90 @@
+---
+name: bmad-code-review
+description: 'Adversarial code review using parallel review layers and structured triage. Use when the user says "run code review" or "review this code"'
+---
+
+# Code Review Workflow
+
+**Goal:** Review code changes adversarially. No noise, no filler.
+
+Subagents, when the capability is available, are an important part of this workflow. Use them as directed by the workflow steps.
+If you need an explicit user instruction to run them, ask once now for the whole workflow run.
+
+## Conventions
+
+- Bare paths (e.g. `checklist.md`) resolve from the skill root.
+- `{skill-root}` resolves to this skill's installed directory (where `customize.toml` lives).
+- `{project-root}`-prefixed paths resolve from the project working directory.
+- `{skill-name}` resolves to the skill directory's basename.
+
+## On Activation
+
+### Step 1: Resolve the Workflow Block
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`
+
+**If the script fails**, resolve the `workflow` block yourself by reading these three files in base → team → user order and applying the same structural merge rules as the resolver:
+
+1. `{skill-root}/customize.toml` — defaults
+2. `{project-root}/_bmad/custom/{skill-name}.toml` — team overrides
+3. `{project-root}/_bmad/custom/{skill-name}.user.toml` — personal overrides
+
+Any missing file is skipped. Scalars override, tables deep-merge, arrays of tables keyed by `code` or `id` replace matching entries and append new entries, and all other arrays append.
+
+### Step 2: Execute Prepend Steps
+
+Execute each entry in `{workflow.activation_steps_prepend}` in order before proceeding.
+
+### Step 3: Load Persistent Facts
+
+Treat every entry in `{workflow.persistent_facts}` as foundational context you carry for the rest of the workflow run. Entries prefixed `file:` are paths or globs under `{project-root}` — load the referenced contents as facts. All other entries are facts verbatim.
+
+### Step 4: Load Config
+
+Load config from `{project-root}/_bmad/bmm/config.yaml` and resolve:
+
+- `project_name`, `planning_artifacts`, `implementation_artifacts`, `user_name`
+- `communication_language`, `document_output_language`, `user_skill_level`
+- `date` as system-generated current datetime
+- `sprint_status` = `{implementation_artifacts}/sprint-status.yaml`
+- `project_context` = `**/project-context.md` (load if exists)
+- YOU MUST ALWAYS SPEAK OUTPUT in your Agent communication style with the config `{communication_language}`
+
+### Step 5: Greet the User
+
+Greet `{user_name}`, speaking in `{communication_language}`.
+
+### Step 6: Execute Append Steps
+
+Execute each entry in `{workflow.activation_steps_append}` in order.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+## WORKFLOW ARCHITECTURE
+
+This uses **step-file architecture** for disciplined execution:
+
+- **Micro-file Design**: Each step is self-contained and followed exactly
+- **Just-In-Time Loading**: Only load the current step file
+- **Sequential Enforcement**: Complete steps in order, no skipping
+- **State Tracking**: Persist progress via in-memory variables
+- **Append-Only Building**: Build artifacts incrementally
+
+### Step Processing Rules
+
+1. **READ COMPLETELY**: Read the entire step file before acting
+2. **FOLLOW SEQUENCE**: Execute sections in order
+3. **WAIT FOR INPUT**: Halt at checkpoints and wait for human
+4. **LOAD NEXT**: When directed, read fully and follow the next step file
+
+### Critical Rules (NO EXCEPTIONS)
+
+- **NEVER** load multiple step files simultaneously
+- **ALWAYS** read entire step file before execution
+- **NEVER** skip steps or optimize the sequence
+- **ALWAYS** follow the exact instructions in the step file
+- **ALWAYS** halt at checkpoints and wait for human input
+
+## FIRST STEP
+
+Read fully and follow: `./steps/step-01-gather-context.md`
diff --git a/.agents/skills/bmad-code-review/customize.toml b/.agents/skills/bmad-code-review/customize.toml
new file mode 100644
index 00000000..3b4ab867
--- /dev/null
+++ b/.agents/skills/bmad-code-review/customize.toml
@@ -0,0 +1,113 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-code-review. Mirrors the
+# agent customization shape under the [workflow] namespace.
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+# Steps to run before the standard activation (config load, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Distinct from the runtime memory sidecar — these are static context
+# loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "All stories must include testable acceptance criteria."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Scalar: executed when the workflow reaches its final step,
+# after review findings are presented and sprint status is synced. Override wins.
+# Leave empty for no custom post-completion behavior.
+
+on_complete = ""
+
+# Review layers for the review step. `instruction` is the layer's whole
+# execution recipe — subagents by default, but an override may run anything
+# (e.g. an external reviewer via bash). {diff_output} and {spec_file} are
+# substituted at run time. `when` (optional) gates a layer; empty
+# `instruction` disables it.
+
+[[workflow.review_layers]]
+id = "blind-hunter"
+name = "Blind Hunter"
+instruction = """
+Launch a context-free subagent with this prompt:
+
+Conduct a review of CONTENT.
+Look for what's missing, not only what's wrong.
+Find at least ten issues to fix or improve.
+Output a Markdown list of findings only — no severity, priority, or ranking.
+If the content is empty, stop and say so.
+If you have zero findings, re-check and keep thinking; do not stop with an empty list.
+
+CONTENT:
+{diff_output}
+
+Do not invoke any skill. Return only the review result.
+
+"""
+
+[[workflow.review_layers]]
+id = "edge-case-hunter"
+name = "Edge Case Hunter"
+instruction = """
+Launch a context-free subagent with this prompt:
+
+Read `{skill-root}/review-prompts/edge-case-hunter.md` completely and follow it as your review instructions.
+
+Review content:
+
+{diff_output}
+
+Do not invoke any skill. If the instruction file is unreadable, report that exact failure and stop. Return only the review result.
+
+"""
+
+[[workflow.review_layers]]
+id = "verification-gap"
+name = "Verification Gap Reviewer"
+instruction = """
+Launch a context-free subagent with this prompt:
+
+Read `{skill-root}/review-prompts/verification-gap.md` completely and follow it as your review instructions.
+
+Review content:
+
+{diff_output}
+
+Do not invoke any skill. If the instruction file is unreadable, report that exact failure and stop. Return only the review result.
+
+"""
+
+[[workflow.review_layers]]
+id = "acceptance-auditor"
+name = "Acceptance Auditor"
+when = 'Only when {review_mode} = "full".'
+instruction = """
+Launch a subagent with this prompt:
+
+You are an Acceptance Auditor. Review the provided diff against `{spec_file}` and any loaded context docs. Check for: violations of acceptance criteria, deviations from spec intent, missing implementation of specified behavior, contradictions between spec constraints and actual code. Output findings as a Markdown list. Each finding: one-line title, which AC/constraint it violates, and evidence from the diff.
+
+Diff:
+
+{diff_output}
+"""
diff --git a/.agents/skills/bmad-code-review/references/deletion-check.md b/.agents/skills/bmad-code-review/references/deletion-check.md
new file mode 100644
index 00000000..5cc3d771
--- /dev/null
+++ b/.agents/skills/bmad-code-review/references/deletion-check.md
@@ -0,0 +1,14 @@
+# Deletion Check
+
+Secondary pass for the Edge Case Hunter — runs only when the diff removed meaningful code. Subordinate to the edge-case pass; findings are usually few or none.
+
+For each chunk of removed or replaced code (ignore pure renames and whitespace), ask: did it carry behavior or a contract that the change neither re-established nor intentionally retired? Add a finding for any resulting regression, orphaned reference, or newly-dead code. Skip anything already covered by your edge-case findings.
+
+Append each finding to the same JSON array as the edge-case findings, with the four standard fields plus:
+
+- `kind`: `"deletion"`
+- `confidence`: `"high"`, `"medium"`, or `"low"` — these are inferences; rate them
+
+For a deletion finding the standard fields read as: `location` = the removed item; `trigger_condition` = the behavior or contract it enforced; `guard_snippet` = where or how to re-establish it; `potential_consequence` = the regression or orphan.
+
+Add nothing if nothing qualifies.
diff --git a/.agents/skills/bmad-code-review/review-prompts/edge-case-hunter.md b/.agents/skills/bmad-code-review/review-prompts/edge-case-hunter.md
new file mode 100644
index 00000000..11df57e9
--- /dev/null
+++ b/.agents/skills/bmad-code-review/review-prompts/edge-case-hunter.md
@@ -0,0 +1,88 @@
+# Edge Case Hunter Review
+
+**Goal:** You are a pure path tracer. Never comment on whether code is good or bad; only list missing handling.
+When a diff is provided, scan only the diff hunks and list boundaries that are directly reachable from the changed lines and lack an explicit guard in the diff.
+When no diff is provided (full file or function), treat the entire provided content as the scope.
+Ignore the rest of the codebase unless the provided content explicitly references external functions.
+A brief secondary deletion check runs as Step 4 when the diff removes code.
+
+**Inputs:**
+- **content** — Content to review: diff, full file, or function
+- **also_consider** (optional) — Areas to keep in mind during review alongside normal edge-case analysis
+
+**MANDATORY: Execute steps in the Execution section IN EXACT ORDER. DO NOT skip steps or change the sequence. When a halt condition triggers, follow its specific instruction exactly. Each action within a step is a REQUIRED action to complete that step.**
+
+**Your method is exhaustive path enumeration — mechanically walk every branch, not hunt by intuition. Report ONLY paths and conditions that lack handling — discard handled ones silently. Do NOT editorialize or add filler. Do not assign severity labels, rankings, or priority levels.**
+
+
+## EXECUTION
+
+### Step 1: Receive Content
+
+- Load the content to review strictly from the parent message that launched you (not from this instruction file)
+- If content is empty, or cannot be decoded as text, return `[{"location":"N/A","trigger_condition":"Input empty or undecodable","guard_snippet":"Provide valid content to review","potential_consequence":"Review skipped — no analysis performed"}]` and stop
+- Identify content type (diff, full file, or function) to determine scope rules
+
+### Step 2: Exhaustive Path Analysis
+
+**Walk every branching path and boundary condition within scope — report only unhandled ones.**
+
+- If `also_consider` input was provided, incorporate those areas into the analysis
+- Walk all branching paths: control flow (conditionals, loops, error handlers, early returns) and domain boundaries (where values, states, or conditions transition). Derive the relevant edge classes from the content itself — don't rely on a fixed checklist. Examples: missing else/default, unguarded inputs, off-by-one loops, arithmetic overflow, implicit type coercion, race conditions, timeout gaps
+- Consider implicit branches: the diff special-cases or changes the handling of one or more members of a fixed set of values — enums, status codes, sentinels, type tags, flags, value ranges. The rest of the set is implicit branches (e.g. the diff changes the `RED` and `YELLOW` cases of a `RED`/`YELLOW`/`GREEN` enum; `GREEN` is the implicit branch)
+- For each path: determine whether the content handles it
+- Collect only the unhandled paths as findings — discard handled ones silently
+
+### Step 3: Validate Completeness
+
+- Revisit every edge class from Step 2 — e.g., missing else/default, null/empty inputs, off-by-one loops, arithmetic overflow, implicit type coercion, race conditions, timeout gaps
+- Add any newly found unhandled paths to findings; discard confirmed-handled ones
+
+### Step 4: Deletion Check
+
+If the diff removed or replaced meaningful code (ignore pure renames and whitespace): load `references/deletion-check.md` and follow it.
+
+### Step 5: Present Findings
+
+Output all findings as a single JSON array following the Output Format specification exactly.
+
+
+## OUTPUT FORMAT
+
+Return ONLY a valid JSON array of objects. Each edge-case finding contains exactly these four fields:
+
+```json
+[{
+  "location": "file:start-end (or file:line when single line, or file:hunk when exact line unavailable)",
+  "trigger_condition": "one-line description (max 15 words)",
+  "guard_snippet": "minimal code sketch that closes the gap (single-line escaped string, no raw newlines or unescaped quotes)",
+  "potential_consequence": "what could actually go wrong (max 15 words)"
+}]
+```
+
+No extra text, no explanations, no markdown wrapping. An empty array `[]` is valid when nothing is found. Deletion findings from Step 4, if any, go in the same array with the extra fields defined in `references/deletion-check.md`.
+
+
+## HALT CONDITIONS
+
+- If content is empty or cannot be decoded as text, return `[{"location":"N/A","trigger_condition":"Input empty or undecodable","guard_snippet":"Provide valid content to review","potential_consequence":"Review skipped — no analysis performed"}]` and stop
+<reference path="references/deletion-check.md">
+# Deletion Check
+
+Secondary pass for the Edge Case Hunter — runs only when the diff removed meaningful code. Subordinate to the edge-case pass; findings are usually few or none.
+
+For each chunk of removed or replaced code (ignore pure renames and whitespace), ask: did it carry behavior or a contract that the change neither re-established nor intentionally retired? Add a finding for any resulting regression, orphaned reference, or newly-dead code. Skip anything already covered by your edge-case findings.
+
+Append each finding to the same JSON array as the edge-case findings, with the four standard fields plus:
+
+- `kind`: `"deletion"`
+- `confidence`: `"high"`, `"medium"`, or `"low"` — these are inferences; rate them
+
+For a deletion finding the standard fields read as: `location` = the removed item; `trigger_condition` = the behavior or contract it enforced; `guard_snippet` = where or how to re-establish it; `potential_consequence` = the regression or orphan.
+
+Add nothing if nothing qualifies.
+</reference>
+
+## CONTENT SOURCE
+
+Review the content supplied under "Review content:" in the message that launched you.
diff --git a/.agents/skills/bmad-code-review/review-prompts/verification-gap.md b/.agents/skills/bmad-code-review/review-prompts/verification-gap.md
new file mode 100644
index 00000000..4adc52d5
--- /dev/null
+++ b/.agents/skills/bmad-code-review/review-prompts/verification-gap.md
@@ -0,0 +1,113 @@
+# Verification Gap Review
+
+**Goal:** Find changed behavior that could break without reliable verification catching it. Ask one question — "if the behavior this change is supposed to produce broke where it's actually used, would verification fail?" Do not hunt for correctness bugs, but report genuine problems you notice while tracing verification.
+
+The main verification gap shapes are:
+
+1. **Regression gap:** the changed code regresses where it's used, and no test covering that use would fail.
+2. **Missing-adoption gap:** a place that should now use the new behavior doesn't; it handles the same case its own way, or not at all, and no test would flag the omission.
+3. **Broken-verification gap:** a test appears to cover the changed behavior, but would not actually protect it because it is skipped, flaky, not run in the normal verification path, or too weak to observe the regression.
+
+## Evidence Rules
+
+- Read a test before claiming what it covers, runs, asserts, or misses.
+- Before claiming no test exists, search the whole repo by the symbol under test and by import references; expected file locations are not enough.
+- Never assert what you did not verify. If a finding cannot be grounded, drop it.
+- In a finding, say what you actually checked — "none of the tests I read cover this" — and show how far you looked. Say a test doesn't exist anywhere only when the symbol/import-reference search actually shows that.
+- Do not assign severity, confidence, priority, or ranking.
+
+## Review Sequence
+
+### Step 1: Screen for behavioral change
+
+Screen each part of the change separately. If a part is non-behavioral, skip it. Call a part non-behavioral only when the changed code does not alter return values, thrown errors, caller-visible side effects, or observable state (including iteration order and emitted messages). Once a part meets that test, move on; do not inspect callers or tests for extra confirmation.
+
+Common non-behavioral examples: formatting, comments, whitespace; pure renames; trivial getters/setters and pass-throughs; type-only or compiler-enforced changes with no runtime effect; etc.
+
+Only outcomes produced by deterministic code are worth automatically testing; tests are useless on static source text and brittle on LLM output. Skip those parts.
+
+If every part is skipped, output the clean result (see Output Format).
+
+### Step 2: Find the behavior that changed
+
+Identify what behavior changed compared to the previous version: output, side effect, branch, error path, schema/event shape, config default, validation/authorization rule, external contract, etc. If the change affects more than one behavior, handle each separately.
+
+Treat broad-impact changes as behavioral even when no single changed line looks important: dependency, toolchain, build/config, data-file, etc.
+
+### Step 3: Trace where that behavior is used
+
+Trace the changed behavior to the places that observe it. Start with direct callers and registered entry points (routes, commands, DI), contract consumers (schemas, events, APIs, database readers), and reverse-dependency info if already available.
+
+Follow a path only while the changed behavior is reachable and unverified. Stop when a test at that boundary would fail, the consumer does not observe the changed behavior, or the next hop is guesswork (dynamic dispatch, reflection, outside-repo consumers, etc.). Prefer the nearest observable boundary, often one to three hops away, especially across contract, integration, or service edges. If there are more than five similar consumers, group obvious repeats and check representative paths; expand only when a consumer observes the behavior differently.
+
+### Step 4: Qualify the consumer, then check its test
+
+For each consumer, name the smallest realistic regression this consumer would observe: invert the branch, drop the default, omit the field, return the old error code, skip the integration call, etc. This is the Demonstration. If no such regression exists, drop the path; untested downstream code is not a finding.
+
+A `Missing-adoption gap` qualifies not by the adoption failure alone but by a supersession signal: the change gives clear evidence the new behavior is meant to replace the local one — PR intent, naming or docs, a replaced sibling site, deleted duplicate logic, or a test defining the new rule — and the local site shares the same observable contract. Without a supersession signal and a shared observable contract, it is a refactor suggestion, not a verification-gap finding. Once both hold, check whether any test for that site would flag the non-adoption; missing coverage of the non-adoption is the gap itself, not a disqualifier.
+
+Find and read the relevant test. Ask whether the Demonstration would make an assertion fail.
+
+- If yes, the behavior is verified. No finding.
+- For a regression-style Demonstration: if no test runs the path, the test is skipped/flaky/not run normally, or the test runs the code without checking the changed result, report a `Regression gap` or `Broken-verification gap`.
+- For a qualifying Missing-adoption case: if none of the site tests you found assert it adopts the new behavior, report a `Missing-adoption gap`.
+
+A test counts only if it runs normally and an assertion observes the changed output, branch, or contract. These do not count: no execution; source-text assertions that match a file's wording instead of running it; success/no-throw/snapshot-only checks; mock/log-call checks; human-only checks; tests that mock away the integration; e2e tests that pass through without checking the changed output; stale assertions or fixtures.
+
+For example, `expect(x ?? DEFAULT).toBe(DEFAULT)` passes when `x` is missing.
+
+Common patterns:
+
+- **Caller-path gap** — helper test covers the branch, but caller values skip it.
+- **Contract drift** — payload/schema/event changes must be verified at the consumer.
+- **Migration compatibility** — tests only create new-format rows or fresh schemas.
+- **Phantom exception** — handled partial-failure path has no test.
+- **Missing-adoption gap** — sibling site should use the new rule/helper and does not.
+- **Removed verification** — deleted test or weakened assertion leaves behavior unpinned; removing a source-text assertion is not this, since it never counted.
+
+### Step 5: Confirm each finding is real
+
+Before writing a finding, re-open the specific tests or search results the finding relies on. Verify the Demonstration would not make any test you checked fail, or that the absence claim is backed by the symbol/import-reference search. Do not claim more than you verified; drop any finding you cannot ground.
+
+Explain why the test misses the bug using what the test sets up and checks.
+
+Do not report: compiler/type-checker-enforced cases; behavior already verified by an integration, contract, or e2e test; implementation-detail or mock-only tests; low coverage or a missing test file by itself; legacy untested code the change did not affect.
+
+Report genuine problems you noticed while tracing verification, even if they are not verification gaps. Put them under `Other findings` in the output. This permits reporting what you already reached, not extra hunting.
+
+## OUTPUT FORMAT
+
+Emit each verification-gap finding as one block. No general advice, no severity or confidence.
+
+```markdown
+### <one-line title naming the gap>
+
+- **Changed surface:** the exact behavior or contract that changed — `file:line`.
+- **Impacted consumer or site:** named concretely with `file:line` (e.g. "the `createInvoice` mutation used by the billing dashboard at `billing/dashboard.ts:88`," not "callers of this function").
+- **Existing test evidence:**
+  - `Regression gap`: what the relevant test actually asserts, with `file:line`; or, if none, the symbol/import-reference searches run and their result.
+  - `Missing-adoption gap`: tests for the impacted site, and whether any assert it adopts the new behavior.
+  - `Broken-verification gap`: the apparent test or verification path, and why it does not count.
+- **Missing verification:** the precise assertion or check that's absent.
+- **Demonstration:**
+  - `Regression gap` / `Broken-verification gap`: the concrete regression that would ship undetected, and why the tests you checked would not fail.
+  - `Missing-adoption gap`: the case the site mishandles by not adopting the new behavior, and that none of the tests you read assert adoption.
+- **Consequence:** the concrete thing that ships wrong — a regression the checked evidence would not catch, or a site that should use the new behavior and doesn't.
+- **Suggested test shape:** (optional) the kind of test that would close the gap, fit to the repo's own way of verifying — don't impose a generic test pyramid.
+```
+
+If you noticed genuine non-gap problems while tracing verification, append:
+
+```markdown
+## Other findings
+
+- <description only; no severity, confidence, priority, or ranking>
+```
+
+When you find no verification gaps and no other findings, output exactly this single line, not an empty response:
+
+`No verification gaps found.`
+
+## CONTENT SOURCE
+
+Review the content supplied under "Review content:" in the message that launched you. If none is supplied, stop with exactly: `No verification gaps found.`
diff --git a/.agents/skills/bmad-code-review/steps/step-01-gather-context.md b/.agents/skills/bmad-code-review/steps/step-01-gather-context.md
new file mode 100644
index 00000000..3add4ad7
--- /dev/null
+++ b/.agents/skills/bmad-code-review/steps/step-01-gather-context.md
@@ -0,0 +1,87 @@
+---
+diff_output: '' # set at runtime
+spec_file: '' # set at runtime (path or empty)
+review_mode: '' # set at runtime: "full" or "no-spec"
+story_key: '' # set at runtime when discovered from sprint status
+---
+
+# Step 1: Gather Context
+
+## RULES
+
+- YOU MUST ALWAYS SPEAK OUTPUT in your Agent communication style with the config `{communication_language}`
+- The prompt that triggered this workflow IS the intent — not a hint.
+- Do not modify any files. This step is read-only.
+
+## INSTRUCTIONS
+
+1. **Find the review target.** The conversation context before this skill was triggered IS your starting point — not a blank slate. Check in this order — stop as soon as the review target is identified:
+
+   **Tier 1 — Explicit argument.**
+   Did the user pass a PR, commit SHA, branch, spec file, or diff source this message?
+   - PR reference → resolve to branch/commit via `gh pr view`. If resolution fails, ask for a SHA or branch.
+   - Commit or branch → use directly.
+   - Spec file → set `{spec_file}` to the provided path. Check its frontmatter for `baseline_commit`. If found, use as diff baseline. If not found, continue the cascade (a spec alone does not identify a diff source).
+   - Also scan the argument for diff-mode keywords that narrow the scope:
+     - "staged" / "staged changes" → Staged changes only
+     - "uncommitted" / "working tree" / "all changes" → Uncommitted changes (staged + unstaged)
+     - "branch diff" / "vs main" / "against main" / "compared to <branch>" → Branch diff (extract base branch if mentioned)
+     - "commit range" / "last N commits" / "<from-sha>..<to-sha>" → Specific commit range
+     - "this diff" / "provided diff" / "paste" → User-provided diff (do not match bare "diff" — it appears in other modes)
+   - When multiple keywords match, prefer the most specific (e.g., "branch diff" over bare "diff").
+
+   **Tier 2 — Recent conversation.**
+   Do the last few messages reveal what the user wants to be reviewed? Look for spec paths, commit refs, branches, PRs, or descriptions of a change. Apply the same diff-mode keyword scan and routing as Tier 1.
+
+   **Tier 3 — Sprint tracking.**
+   Look for a sprint status file (`*sprint-status*`) in `{implementation_artifacts}` or `{planning_artifacts}`. If found, scan for stories with status `review`:
+   - **Exactly one `review` story:** Set `{story_key}` to the story's key (e.g., `1-2-user-auth`). Suggest it: "I found story <story-id> in `review` status. Would you like to review its changes? [Y] Yes / [N] No, let me choose". If confirmed, use the story context to determine the diff source (branch name derived from story slug, or uncommitted changes). If declined, clear `{story_key}` and fall through.
+   - **Multiple `review` stories:** Present them as numbered options alongside a manual choice option. Wait for user selection. If a story is selected, set `{story_key}` and use its context to determine the diff source. If manual choice is selected, clear `{story_key}` and fall through.
+   - **None:** Fall through.
+
+   **Tier 4 — Current git state.**
+   If version control is unavailable, skip to Tier 5. Otherwise, check the current branch and HEAD. If the branch is not `main` (or the default branch), confirm: "I see HEAD is `<short-sha>` on `<branch>` — do you want to review this branch's changes?" If confirmed, treat as a branch diff against `main`. If declined, fall through.
+
+   **Tier 5 — Ask.**
+   Fall through to instruction 2.
+
+   Never ask extra questions beyond what the cascade prescribes. If a tier above already identified the target, skip the remaining tiers and proceed to instruction 3 (construct diff).
+
+2. HALT. Ask the user: **What do you want to review?** Present these options:
+   - **Uncommitted changes** (staged + unstaged)
+   - **Staged changes only**
+   - **Branch diff** vs a base branch (ask which base branch)
+   - **Specific commit range** (ask for the range)
+   - **Provided diff or file list** (user pastes or provides a path)
+
+3. Construct `{diff_output}` from the chosen source.
+   - For **staged changes only**: run `git diff --cached`.
+   - For **uncommitted changes** (staged + unstaged): run `git diff HEAD`.
+   - For **branch diff**: verify the base branch exists before running `git diff`. If it does not exist, HALT and ask the user for a valid branch.
+   - For **commit range**: verify the range resolves. If it does not, HALT and ask the user for a valid range.
+   - For **provided diff**: validate the content is non-empty and parseable as a unified diff. If it is not parseable, HALT and ask the user to provide a valid diff.
+   - For **file list**: validate each path exists in the working tree. Construct `{diff_output}` by running `git diff HEAD -- <path1> <path2> ...`. If any paths are untracked (new files not yet staged), use `git diff --no-index /dev/null <path>` to include them. If the diff is empty (files have no uncommitted changes and are not untracked), ask the user whether to review the full file contents or to specify a different baseline.
+   - After constructing `{diff_output}`, verify it is non-empty regardless of source type. If empty, HALT and tell the user there is nothing to review.
+
+4. **Set the spec context.**
+   - If the triggering request or recent conversation **explicitly** states there is no spec (e.g. "no spec", "without a spec", "no-spec"): set `{review_mode}` = `"no-spec"` and clear `{spec_file}` (set it to `''`). Do **not** ask for a spec. Do **not** infer no-spec mode merely because the invocation omitted a spec path.
+   - Else if `{spec_file}` is already set (from Tier 1 or Tier 2): verify the file exists and is readable, then set `{review_mode}` = `"full"`.
+   - Else (neither a spec path nor an explicit no-spec declaration is present): ask the user to choose:
+     1. Provide a spec or story file path for context; or
+     2. Continue without a spec.
+     - If the user provides a path: set `{spec_file}` to that path, verify the file exists and is readable, then set `{review_mode}` = `"full"`.
+     - If the user explicitly chooses to continue without a spec: set `{review_mode}` = `"no-spec"`.
+
+5. If `{review_mode}` = `"full"` and the file at `{spec_file}` has a `context` field in its frontmatter listing additional docs, load each referenced document. Warn the user about any docs that cannot be found.
+
+6. Sanity check: if `{diff_output}` exceeds approximately 3000 lines, warn the user and offer to chunk the review by file group.
+   - If the user opts to chunk: agree on the first group, narrow `{diff_output}` accordingly, and list the remaining groups for the user to note for follow-up runs.
+   - If the user declines: proceed as-is with the full diff.
+
+### CHECKPOINT
+
+Present a summary before proceeding: diff stats (files changed, lines added/removed), `{review_mode}`, and loaded spec/context docs (if any). HALT and wait for user confirmation to proceed.
+
+## NEXT
+
+Read fully and follow `./step-02-review.md`
diff --git a/.agents/skills/bmad-code-review/steps/step-02-review.md b/.agents/skills/bmad-code-review/steps/step-02-review.md
new file mode 100644
index 00000000..627fdfea
--- /dev/null
+++ b/.agents/skills/bmad-code-review/steps/step-02-review.md
@@ -0,0 +1,32 @@
+---
+failed_layers: '' # set at runtime: comma-separated list of layers that failed or returned empty
+---
+
+# Step 2: Review
+
+## RULES
+
+- YOU MUST ALWAYS SPEAK OUTPUT in your Agent communication style with the config `{communication_language}`
+- All review subagents must run at the same model capability as the current session.
+- Run subagents synchronously: launch them together, then wait for all results before continuing.
+
+## INSTRUCTIONS
+
+1. The review layers are `{workflow.review_layers}`, resolved during activation.
+
+2. For each layer in `{workflow.review_layers}`:
+   - `instruction` empty or missing → drop the layer silently (an override disabled it).
+   - `when` condition present and not satisfied by the current context (`{review_mode}`, `{spec_file}`) → drop the layer and tell the user, e.g. "Acceptance Auditor skipped — no spec file provided."
+   - otherwise → the layer is active.
+
+   If no layer is active, HALT with status `blocked` and blocking condition `no active review layers`.
+
+3. Execute all active layers in parallel wherever their execution methods allow: expand `{skill-root}` in each layer's `instruction` to this skill's absolute installed directory, then substitute the runtime placeholders (`{diff_output}`, `{spec_file}`). For an instruction that launches a reviewer subagent, launch that child with the prompt text after placeholder substitution; do not load the reviewer instruction file yourself. For any other customized instruction, execute it as written. Do not leave `{skill-root}` unresolved in a child prompt. If a layer's instruction requires subagents and subagents are not available, for each such layer write under `{implementation_artifacts}` the exact child prompt from that layer's instruction after placeholder substitution (not a path-only pointer), then HALT. Ask the user to run each in a separate session (ideally a different LLM) and paste back the findings. When findings are pasted, treat them as those layers' findings and resume from this point. This is the only allowed parent-side read of a reviewer instruction file.
+
+4. **Layer failure handling**: If any layer fails, times out, or returns empty results, append the layer's `name` to `{failed_layers}` (comma-separated) and proceed with findings from the remaining layers.
+
+5. Collect all findings from the completed layers, keeping track of each finding's originating layer `id`.
+
+## NEXT
+
+Read fully and follow `./step-03-triage.md`
diff --git a/.agents/skills/bmad-code-review/steps/step-03-triage.md b/.agents/skills/bmad-code-review/steps/step-03-triage.md
new file mode 100644
index 00000000..2577917f
--- /dev/null
+++ b/.agents/skills/bmad-code-review/steps/step-03-triage.md
@@ -0,0 +1,50 @@
+---
+---
+
+# Step 3: Triage
+
+## RULES
+
+- YOU MUST ALWAYS SPEAK OUTPUT in your Agent communication style with the config `{communication_language}`
+
+## INSTRUCTIONS
+
+1. **Normalize** findings from all layers into a unified list where each finding has:
+   - `id` -- sequential integer
+   - `source` -- the `id` of the layer that produced the finding (e.g., `blind-hunter`), or merged sources joined with `+` (e.g., `blind-hunter+edge-case-hunter`)
+   - `title` -- one-line summary
+   - `detail` -- full description
+   - `location` -- file and line reference (if available)
+
+2. **Deduplicate.** Deduplicate only findings with the same claim and same required action. If two or more findings meet both conditions, merge them into one:
+   - Use the most specific finding as the base (prefer findings with a precise location over prose-only findings).
+   - Append any unique detail, reasoning, or location references from the other finding(s) into the surviving `detail` field.
+   - Set `source` to the merged sources (e.g., `blind-hunter+edge-case-hunter`).
+
+3. Then evaluate each remaining finding independently. Do not reject a finding because a related finding was rejected.
+
+4. **Read the code before rating.** Before assigning severity, open the source at each finding's location and read enough surrounding code to judge reachability -- call sites, guards, and validation that live outside the diff hunk. Do not rate from the diff hunk alone. Severity reflects the real consequence at a real call site, not the worst theoretical reading.
+
+5. **Assign severity** to each finding by consequence for the artifact's main consumer (software user, document reader, etc).
+   Disregard any severity assigned by a reviewing subagent. Review subagents operate under by-design information asymmetry and do not have enough context to set final severity for this workflow.
+   - `low` -- none or cosmetic
+   - `medium` -- tolerable
+   - `high` -- intolerable
+
+6. **Route** each finding into exactly one triage bucket:
+   - **decision_needed** -- There is an ambiguous choice that requires human input. The code cannot be correctly patched without knowing the user's intent. Only possible if `{review_mode}` = `"full"`.
+   - **patch** -- Code issue that is fixable without human input. The correct fix is unambiguous.
+   - **defer** -- Pre-existing issue not caused by the current change. Real but not actionable now.
+   - **dismiss** -- Noise, false positive, or handled elsewhere.
+
+   If `{review_mode}` = `"no-spec"` and a finding would otherwise be `decision_needed`, reclassify it as `patch` (if the fix is unambiguous) or `defer` (if not).
+
+7. **Drop** all `dismiss` findings. Record the dismiss count for the summary.
+
+8. If `{failed_layers}` is non-empty, report which layers failed before announcing results. If zero findings remain after dropping dismissed AND `{failed_layers}` is non-empty, warn the user that the review may be incomplete rather than announcing a clean review.
+
+9. If zero findings remain after triage (all rejected or none raised): state "✅ Clean review — all layers passed." (Step 3 already warned if any review layers failed via `{failed_layers}`.)
+
+## NEXT
+
+Read fully and follow `./step-04-present.md`
diff --git a/.agents/skills/bmad-code-review/steps/step-04-present.md b/.agents/skills/bmad-code-review/steps/step-04-present.md
new file mode 100644
index 00000000..884d9867
--- /dev/null
+++ b/.agents/skills/bmad-code-review/steps/step-04-present.md
@@ -0,0 +1,132 @@
+---
+deferred_work_file: '{implementation_artifacts}/deferred-work.md'
+---
+
+# Step 4: Present and Act
+
+## RULES
+
+- YOU MUST ALWAYS SPEAK OUTPUT in your Agent communication style with the config `{communication_language}`
+- When `{spec_file}` is set, always write findings to the story file before offering action choices.
+- `decision-needed` findings must be resolved before handling `patch` findings.
+
+## INSTRUCTIONS
+
+### 1. Clean review shortcut
+
+If zero findings remain after triage (all dismissed or none raised): state that and proceed to section 6 (Sprint Status Update).
+
+### 2. Write findings to the story file
+
+If `{spec_file}` exists and contains a Tasks/Subtasks section, append a `### Review Findings` subsection. Write all findings in this order:
+
+1. **`decision-needed`** findings (unchecked):
+   `- [ ] [Review][Decision] <Title> — <Detail>`
+
+2. **`patch`** findings (unchecked):
+   `- [ ] [Review][Patch] <Title> [<file>:<line>]`
+
+3. **`defer`** findings (checked off, marked deferred):
+   `- [x] [Review][Defer] <Title> [<file>:<line>] — deferred, pre-existing`
+
+Also append each `defer` finding to `{deferred_work_file}` under a heading `## Deferred from: code review ({date})`. If `{spec_file}` is set, include its basename in the heading (e.g., `code review of story-3.3 (2026-03-18)`). One bullet per finding with description.
+
+### 3. Present summary
+
+Announce what was written:
+
+> **Code review complete.** <D> `decision-needed`, <P> `patch`, <W> `defer`, <R> dismissed as noise.
+
+If `{spec_file}` is set, add: `Findings written to the review findings section in {spec_file}.`
+Otherwise add: `Findings are listed above. No story file was provided, so nothing was persisted.`
+
+### 4. Resolve decision-needed findings
+
+If `decision_needed` findings exist, present each one with its detail and the options available. The user must decide — the correct fix is ambiguous without their input. Walk through each finding (or batch related ones) and get the user's call. Once resolved, each becomes a `patch`, `defer`, or is dismissed.
+
+If the user chooses to defer, ask: Quick one-line reason for deferring this item? (helps future reviews): — then append that reason to both the story file bullet and the `{deferred_work_file}` entry.
+
+**HALT** — I am waiting for your numbered choice. Reply with only the number. Do not proceed until you select an option.
+
+### 5. Handle `patch` findings
+
+If `patch` findings exist (including any resolved from step 4), HALT. Ask the user:
+
+If `{spec_file}` is set, present all three options:
+
+> **How would you like to handle the `<P>` `patch` findings?**
+> 1. **Apply every patch** — fix all of them now, no per-finding confirmation. Defer and decision-needed items are not touched.
+> 2. **Leave as action items** — they are already in the story file
+> 3. **Walk through each patch** — show details for each before deciding
+
+If `{spec_file}` is **not** set, present only options 1 and 2 (omit "Leave as action items" — findings were not written to a file):
+
+> **How would you like to handle the `<P>` `patch` findings?**
+> 1. **Apply every patch** — fix all of them now, no per-finding confirmation. Defer and decision-needed items are not touched.
+> 2. **Walk through each patch** — show details for each before deciding
+
+**HALT** — I am waiting for your numbered choice. Reply with only the number. Do not proceed until you select an option.
+
+- **Apply every patch**: Apply every patch finding without per-finding confirmation. Do not modify defer or decision-needed items. After all patches are applied, present a summary of changes made. If `{spec_file}` is set, check off the patch items in the story file (leave defer items as-is).
+- **Leave as action items** (only when `{spec_file}` is set): Done — findings are already written to the story.
+- **Walk through each patch**: Present each finding with full detail, diff context, and suggested fix. After walkthrough, re-offer the applicable options above.
+
+  **HALT** — I am waiting for your numbered choice. Do not proceed until you select an option.
+
+**✅ Code review actions complete**
+
+- Decision-needed resolved: <D>
+- Patches handled: <P>
+- Deferred: <W>
+- Dismissed: <R>
+
+### 6. Update story status and sync sprint tracking
+
+Skip this section if `{spec_file}` is not set.
+
+#### Determine new status based on review outcome
+
+- If all `decision-needed` and `patch` findings were resolved (fixed or dismissed) AND no unresolved `high`/`medium` findings remain: set `{new_status}` = `done`. Update the story file Status section to `done`.
+- If `patch` findings were left as action items, or unresolved issues remain: set `{new_status}` = `in-progress`. Update the story file Status section to `in-progress`.
+
+Save the story file.
+
+#### Sync sprint-status.yaml
+
+If `{story_key}` is not set, skip this subsection and note that sprint status was not synced because no story key was available.
+
+If `{sprint_status}` file exists:
+
+1. Load the FULL `{sprint_status}` file.
+2. Find the `development_status` entry matching `{story_key}`.
+3. If found: update `development_status[{story_key}]` to `{new_status}`. Update `last_updated` to current date. Save the file, preserving ALL comments and structure including STATUS DEFINITIONS.
+4. If `{story_key}` not found in sprint status: warn the user that the story file was updated but sprint-status sync failed.
+
+If `{sprint_status}` file does not exist, note that story status was updated in the story file only.
+
+#### Completion summary
+
+> **Review Complete!**
+>
+> **Story Status:** `{new_status}`
+> **Issues Fixed:** <fixed_count>
+> **Action Items Created:** <action_count>
+> **Deferred:** <W>
+> **Dismissed:** <R>
+
+### 7. Next steps
+
+Present the user with follow-up options:
+
+> **What would you like to do next?**
+> 1. **Start the next story** — run `dev-story` to pick up the next `ready-for-dev` story
+> 2. **Re-run code review** — address findings and review again
+> 3. **Done** — end the workflow
+
+**HALT** — I am waiting for your choice. Do not proceed until the user selects an option.
+
+## On Complete
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow.on_complete`
+
+If the resolved `workflow.on_complete` is non-empty, follow it as the final terminal instruction before exiting.
diff --git a/.agents/skills/bmad-correct-course/SKILL.md b/.agents/skills/bmad-correct-course/SKILL.md
new file mode 100644
index 00000000..9ef41084
--- /dev/null
+++ b/.agents/skills/bmad-correct-course/SKILL.md
@@ -0,0 +1,300 @@
+---
+name: bmad-correct-course
+description: 'Manage significant changes during sprint execution. Use when the user says "correct course" or "propose sprint change"'
+---
+
+# Correct Course - Sprint Change Management Workflow
+
+**Goal:** Manage significant changes during sprint execution by analyzing impact across all project artifacts and producing a structured Sprint Change Proposal.
+
+**Your Role:** You are a Developer navigating change management. Analyze the triggering issue, assess impact across PRD, epics, architecture, and UX artifacts, and produce an actionable Sprint Change Proposal with clear handoff.
+
+## Conventions
+
+- Bare paths (e.g. `checklist.md`) resolve from the skill root.
+- `{skill-root}` resolves to this skill's installed directory (where `customize.toml` lives).
+- `{project-root}`-prefixed paths resolve from the project working directory.
+- `{skill-name}` resolves to the skill directory's basename.
+
+## On Activation
+
+### Step 1: Resolve the Workflow Block
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`
+
+**If the script fails**, resolve the `workflow` block yourself by reading these three files in base → team → user order and applying the same structural merge rules as the resolver:
+
+1. `{skill-root}/customize.toml` — defaults
+2. `{project-root}/_bmad/custom/{skill-name}.toml` — team overrides
+3. `{project-root}/_bmad/custom/{skill-name}.user.toml` — personal overrides
+
+Any missing file is skipped. Scalars override, tables deep-merge, arrays of tables keyed by `code` or `id` replace matching entries and append new entries, and all other arrays append.
+
+### Step 2: Execute Prepend Steps
+
+Execute each entry in `{workflow.activation_steps_prepend}` in order before proceeding.
+
+### Step 3: Load Persistent Facts
+
+Treat every entry in `{workflow.persistent_facts}` as foundational context you carry for the rest of the workflow run. Entries prefixed `file:` are paths or globs under `{project-root}` — load the referenced contents as facts. All other entries are facts verbatim.
+
+### Step 4: Load Config
+
+Load config from `{project-root}/_bmad/bmm/config.yaml` and resolve:
+
+- `project_name`, `user_name`
+- `communication_language`, `document_output_language`
+- `user_skill_level`
+- `implementation_artifacts`
+- `planning_artifacts`
+- `project_knowledge`
+- `date` as system-generated current datetime
+- YOU MUST ALWAYS SPEAK OUTPUT in your Agent communication style with the config `{communication_language}`
+- Language MUST be tailored to `{user_skill_level}`
+- Generate all documents in `{document_output_language}`
+- DOCUMENT OUTPUT: Updated epics, stories, or PRD sections. Clear, actionable changes. User skill level (`{user_skill_level}`) affects conversation style ONLY, not document updates.
+
+### Step 5: Greet the User
+
+Greet `{user_name}`, speaking in `{communication_language}`.
+
+### Step 6: Execute Append Steps
+
+Execute each entry in `{workflow.activation_steps_append}` in order.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+## Paths
+
+- `default_output_file` = `{planning_artifacts}/sprint-change-proposal-{date}.md`
+
+## Input Files
+
+| Input | Path | Load Strategy |
+|-------|------|---------------|
+| PRD | `{planning_artifacts}/*prd*.md` (whole) or `{planning_artifacts}/*prd*/*.md` (sharded) | FULL_LOAD |
+| Epics | `{planning_artifacts}/*epic*.md` (whole) or `{planning_artifacts}/*epic*/*.md` (sharded) | FULL_LOAD |
+| Architecture | `{planning_artifacts}/*architecture*.md` (whole) or `{planning_artifacts}/*architecture*/*.md` (sharded) | FULL_LOAD |
+| UX Design | `{planning_artifacts}/*ux*.md` (whole) or `{planning_artifacts}/*ux*/*.md` (sharded) | FULL_LOAD |
+| Spec | `{planning_artifacts}/*spec-*.md` (whole) | FULL_LOAD |
+| Project Context | `AGENTS.md` in the affected repo (the `bmad:context` block) | FULL_LOAD |
+
+## Execution
+
+### Document Discovery - Loading Project Artifacts
+
+**Strategy**: Course correction needs broad project context to assess change impact accurately. Load all available planning artifacts.
+
+**Discovery Process for FULL_LOAD documents (PRD, Epics, Architecture, UX Design, Spec):**
+
+1. **Search for whole document first** - Look for files matching the whole-document pattern (e.g., `*prd*.md`, `*epic*.md`, `*architecture*.md`, `*ux*.md`, `*spec-*.md`)
+2. **Check for sharded version** - If whole document not found, look for a directory with `index.md` (e.g., `prd/index.md`, `epics/index.md`)
+3. **If sharded version found**:
+   - Read `index.md` to understand the document structure
+   - Read ALL section files listed in the index
+   - Process the combined content as a single document
+4. **Priority**: If both whole and sharded versions exist, use the whole document
+
+**Discovery Process for Project Context:**
+
+1. **Read `AGENTS.md`** in the repo the change affects — the block between the `bmad:context` markers carries the policy, frozen paths, and conventions a course correction must respect.
+2. **Follow only the pointers that relate to the impacted areas** — nested component files or linked rule files listed under "Where things are". Do not load them all.
+3. **This document is optional** — skip if the repo has no `AGENTS.md` (greenfield projects).
+
+**Fuzzy matching**: Be flexible with document names — users may use variations like `prd.md`, `bmm-prd.md`, `product-requirements.md`, etc.
+
+**Missing documents**: Not all documents may exist. PRD and Epics are essential; Architecture, UX Design, Spec, and Document Project are loaded if available. HALT if PRD or Epics cannot be found.
+
+<workflow>
+
+<step n="1" goal="Initialize Change Navigation">
+  <action>Confirm change trigger and gather user description of the issue</action>
+  <action>Ask: "What specific issue or change has been identified that requires navigation?"</action>
+  <action>Verify access to project documents:</action>
+    - PRD (Product Requirements Document) — required
+    - Current Epics and Stories — required
+    - Architecture documentation — optional, load if available
+    - UI/UX specifications — optional, load if available
+  <action>Ask user for mode preference:</action>
+    - **Incremental** (recommended): Refine each edit collaboratively
+    - **Batch**: Present all changes at once for review
+  <action>Store mode selection for use throughout workflow</action>
+
+<action if="change trigger is unclear">HALT: "Cannot navigate change without clear understanding of the triggering issue. Please provide specific details about what needs to change and why."</action>
+
+<action if="PRD or Epics are unavailable">HALT: "Need access to PRD and Epics to assess change impact. Please ensure these documents are accessible. Architecture and UI/UX will be used if available."</action>
+</step>
+
+<step n="2" goal="Execute Change Analysis Checklist">
+  <action>Read fully and follow the systematic analysis from: checklist.md</action>
+  <action>Work through each checklist section interactively with the user</action>
+  <action>Record status for each checklist item:</action>
+    - [x] Done - Item completed successfully
+    - [N/A] Skip - Item not applicable to this change
+    - [!] Action-needed - Item requires attention or follow-up
+  <action>Maintain running notes of findings and impacts discovered</action>
+  <action>Present checklist progress after each major section</action>
+
+<action if="checklist cannot be completed">Identify blocking issues and work with user to resolve before continuing</action>
+</step>
+
+<step n="3" goal="Draft Specific Change Proposals">
+<action>Based on checklist findings, create explicit edit proposals for each identified artifact</action>
+
+<action>For Story changes:</action>
+
+- Show old → new text format
+- Include story ID and section being modified
+- Provide rationale for each change
+- Example format:
+
+  ```
+  Story: [STORY-123] User Authentication
+  Section: Acceptance Criteria
+
+  OLD:
+  - User can log in with email/password
+
+  NEW:
+  - User can log in with email/password
+  - User can enable 2FA via authenticator app
+
+  Rationale: Security requirement identified during implementation
+  ```
+
+<action>For PRD modifications:</action>
+
+- Specify exact sections to update
+- Show current content and proposed changes
+- Explain impact on MVP scope and requirements
+
+<action>For Architecture changes:</action>
+
+- Identify affected components, patterns, or technology choices
+- Describe diagram updates needed
+- Note any ripple effects on other components
+
+<action>For UI/UX specification updates:</action>
+
+- Reference specific screens or components
+- Show wireframe or flow changes needed
+- Connect changes to user experience impact
+
+<check if="mode is Incremental">
+  <action>Present each edit proposal individually</action>
+  <ask>Review and refine this change? Options: Approve [a], Edit [e], Skip [s]</ask>
+  <action>Iterate on each proposal based on user feedback</action>
+</check>
+
+<action if="mode is Batch">Collect all edit proposals and present together at end of step</action>
+
+</step>
+
+<step n="4" goal="Generate Sprint Change Proposal">
+<action>Compile comprehensive Sprint Change Proposal document with following sections:</action>
+
+<action>Section 1: Issue Summary</action>
+
+- Clear problem statement describing what triggered the change
+- Context about when/how the issue was discovered
+- Evidence or examples demonstrating the issue
+
+<action>Section 2: Impact Analysis</action>
+
+- Epic Impact: Which epics are affected and how
+- Story Impact: Current and future stories requiring changes
+- Artifact Conflicts: PRD, Architecture, UI/UX documents needing updates
+- Technical Impact: Code, infrastructure, or deployment implications
+
+<action>Section 3: Recommended Approach</action>
+
+- Present chosen path forward from checklist evaluation:
+  - Direct Adjustment: Modify/add stories within existing plan
+  - Potential Rollback: Revert completed work to simplify resolution
+  - MVP Review: Reduce scope or modify goals
+- Provide clear rationale for recommendation
+- Include effort estimate, risk assessment, and timeline impact
+
+<action>Section 4: Detailed Change Proposals</action>
+
+- Include all refined edit proposals from Step 3
+- Group by artifact type (Stories, PRD, Architecture, UI/UX)
+- Ensure each change includes before/after and justification
+
+<action>Section 5: Implementation Handoff</action>
+
+- Categorize change scope:
+  - Minor: Direct implementation by Developer agent
+  - Moderate: Backlog reorganization needed (PO/DEV)
+  - Major: Fundamental replan required (PM/Architect)
+- Specify handoff recipients and their responsibilities
+- Define success criteria for implementation
+
+<action>Present complete Sprint Change Proposal to user</action>
+<action>Write Sprint Change Proposal document to {default_output_file}</action>
+<ask>Review complete proposal. Continue [c] or Edit [e]?</ask>
+</step>
+
+<step n="5" goal="Finalize and Route for Implementation">
+<action>Get explicit user approval for complete proposal</action>
+<ask>Do you approve this Sprint Change Proposal for implementation? (yes/no/revise)</ask>
+
+<check if="no or revise">
+  <action>Gather specific feedback on what needs adjustment</action>
+  <action>Return to appropriate step to address concerns</action>
+  <goto step="3">If changes needed to edit proposals</goto>
+  <goto step="4">If changes needed to overall proposal structure</goto>
+
+</check>
+
+<check if="yes the proposal is approved by the user">
+  <action>Finalize Sprint Change Proposal document</action>
+  <action>Determine change scope classification:</action>
+
+- **Minor**: Can be implemented directly by Developer agent
+- **Moderate**: Requires backlog reorganization and PO/DEV coordination
+- **Major**: Needs fundamental replan with PM/Architect involvement
+
+<action>Provide appropriate handoff based on scope:</action>
+
+</check>
+
+<check if="Minor scope">
+  <action>Route to: Developer agent for direct implementation</action>
+  <action>Deliverables: Finalized edit proposals and implementation tasks</action>
+</check>
+
+<check if="Moderate scope">
+  <action>Route to: Product Owner / Developer agents</action>
+  <action>Deliverables: Sprint Change Proposal + backlog reorganization plan</action>
+</check>
+
+<check if="Major scope">
+  <action>Route to: Product Manager / Solution Architect</action>
+  <action>Deliverables: Complete Sprint Change Proposal + escalation notice</action>
+
+<action>Confirm handoff completion and next steps with user</action>
+<action>Document handoff in workflow execution log</action>
+</check>
+
+</step>
+
+<step n="6" goal="Workflow Completion">
+<action>Summarize workflow execution:</action>
+  - Issue addressed: {{change_trigger}}
+  - Change scope: {{scope_classification}}
+  - Artifacts modified: {{list_of_artifacts}}
+  - Routed to: {{handoff_recipients}}
+
+<action>Confirm all deliverables produced:</action>
+
+- Sprint Change Proposal document
+- Specific edit proposals with before/after
+- Implementation handoff plan
+
+<action>Report workflow completion to user with personalized message: "Correct Course workflow complete, {user_name}!"</action>
+<action>Remind user of success criteria and next steps for Developer agent</action>
+<action>Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow.on_complete` — if the resolved value is non-empty, follow it as the final terminal instruction before exiting.</action>
+</step>
+
+</workflow>
diff --git a/.agents/skills/bmad-correct-course/checklist.md b/.agents/skills/bmad-correct-course/checklist.md
new file mode 100644
index 00000000..b56feb6d
--- /dev/null
+++ b/.agents/skills/bmad-correct-course/checklist.md
@@ -0,0 +1,288 @@
+# Change Navigation Checklist
+
+<critical>This checklist is executed as part of: ./workflow.md</critical>
+<critical>Work through each section systematically with the user, recording findings and impacts</critical>
+
+<checklist>
+
+<section n="1" title="Understand the Trigger and Context">
+
+<check-item id="1.1">
+<prompt>Identify the triggering story that revealed this issue</prompt>
+<action>Document story ID and brief description</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="1.2">
+<prompt>Define the core problem precisely</prompt>
+<action>Categorize issue type:</action>
+  - Technical limitation discovered during implementation
+  - New requirement emerged from stakeholders
+  - Misunderstanding of original requirements
+  - Strategic pivot or market change
+  - Failed approach requiring different solution
+<action>Write clear problem statement</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="1.3">
+<prompt>Assess initial impact and gather supporting evidence</prompt>
+<action>Collect concrete examples, error messages, stakeholder feedback, or technical constraints</action>
+<action>Document evidence for later reference</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<halt-condition>
+<action if="trigger is unclear">HALT: "Cannot proceed without understanding what caused the need for change"</action>
+<action if="no evidence provided">HALT: "Need concrete evidence or examples of the issue before analyzing impact"</action>
+</halt-condition>
+
+</section>
+
+<section n="2" title="Epic Impact Assessment">
+
+<check-item id="2.1">
+<prompt>Evaluate current epic containing the trigger story</prompt>
+<action>Can this epic still be completed as originally planned?</action>
+<action>If no, what modifications are needed?</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="2.2">
+<prompt>Determine required epic-level changes</prompt>
+<action>Check each scenario:</action>
+  - Modify existing epic scope or acceptance criteria
+  - Add new epic to address the issue
+  - Remove or defer epic that's no longer viable
+  - Completely redefine epic based on new understanding
+<action>Document specific epic changes needed</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="2.3">
+<prompt>Review all remaining planned epics for required changes</prompt>
+<action>Check each future epic for impact</action>
+<action>Identify dependencies that may be affected</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="2.4">
+<prompt>Check if issue invalidates future epics or necessitates new ones</prompt>
+<action>Does this change make any planned epics obsolete?</action>
+<action>Are new epics needed to address gaps created by this change?</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="2.5">
+<prompt>Consider if epic order or priority should change</prompt>
+<action>Should epics be resequenced based on this issue?</action>
+<action>Do priorities need adjustment?</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+</section>
+
+<section n="3" title="Artifact Conflict and Impact Analysis">
+
+<check-item id="3.1">
+<prompt>Check PRD for conflicts</prompt>
+<action>Does issue conflict with core PRD goals or objectives?</action>
+<action>Do requirements need modification, addition, or removal?</action>
+<action>Is the defined MVP still achievable or does scope need adjustment?</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="3.2">
+<prompt>Review Architecture document for conflicts</prompt>
+<action>Check each area for impact:</action>
+  - System components and their interactions
+  - Architectural patterns and design decisions
+  - Technology stack choices
+  - Data models and schemas
+  - API designs and contracts
+  - Integration points
+<action>Document specific architecture sections requiring updates</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="3.3">
+<prompt>Examine UI/UX specifications for conflicts</prompt>
+<action>Check for impact on:</action>
+  - User interface components
+  - User flows and journeys
+  - Wireframes or mockups
+  - Interaction patterns
+  - Accessibility considerations
+<action>Note specific UI/UX sections needing revision</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="3.4">
+<prompt>Consider impact on other artifacts</prompt>
+<action>Review additional artifacts for impact:</action>
+  - Deployment scripts
+  - Infrastructure as Code (IaC)
+  - Monitoring and observability setup
+  - Testing strategies
+  - Documentation
+  - CI/CD pipelines
+<action>Document any secondary artifacts requiring updates</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+</section>
+
+<section n="4" title="Path Forward Evaluation">
+
+<check-item id="4.1">
+<prompt>Evaluate Option 1: Direct Adjustment</prompt>
+<action>Can the issue be addressed by modifying existing stories?</action>
+<action>Can new stories be added within the current epic structure?</action>
+<action>Would this approach maintain project timeline and scope?</action>
+<action>Effort estimate: [High/Medium/Low]</action>
+<action>Risk level: [High/Medium/Low]</action>
+<status>[ ] Viable / [ ] Not viable</status>
+</check-item>
+
+<check-item id="4.2">
+<prompt>Evaluate Option 2: Potential Rollback</prompt>
+<action>Would reverting recently completed stories simplify addressing this issue?</action>
+<action>Which stories would need to be rolled back?</action>
+<action>Is the rollback effort justified by the simplification gained?</action>
+<action>Effort estimate: [High/Medium/Low]</action>
+<action>Risk level: [High/Medium/Low]</action>
+<status>[ ] Viable / [ ] Not viable</status>
+</check-item>
+
+<check-item id="4.3">
+<prompt>Evaluate Option 3: PRD MVP Review</prompt>
+<action>Is the original PRD MVP still achievable with this issue?</action>
+<action>Does MVP scope need to be reduced or redefined?</action>
+<action>Do core goals need modification based on new constraints?</action>
+<action>What would be deferred to post-MVP if scope is reduced?</action>
+<action>Effort estimate: [High/Medium/Low]</action>
+<action>Risk level: [High/Medium/Low]</action>
+<status>[ ] Viable / [ ] Not viable</status>
+</check-item>
+
+<check-item id="4.4">
+<prompt>Select recommended path forward</prompt>
+<action>Based on analysis of all options, choose the best path</action>
+<action>Provide clear rationale considering:</action>
+  - Implementation effort and timeline impact
+  - Technical risk and complexity
+  - Impact on team morale and momentum
+  - Long-term sustainability and maintainability
+  - Stakeholder expectations and business value
+<action>Selected approach: [Option 1 / Option 2 / Option 3 / Hybrid]</action>
+<action>Justification: [Document reasoning]</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+</section>
+
+<section n="5" title="Sprint Change Proposal Components">
+
+<check-item id="5.1">
+<prompt>Create identified issue summary</prompt>
+<action>Write clear, concise problem statement</action>
+<action>Include context about discovery and impact</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="5.2">
+<prompt>Document epic impact and artifact adjustment needs</prompt>
+<action>Summarize findings from Epic Impact Assessment (Section 2)</action>
+<action>Summarize findings from Artifact Conflict Analysis (Section 3)</action>
+<action>Be specific about what changes are needed and why</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="5.3">
+<prompt>Present recommended path forward with rationale</prompt>
+<action>Include selected approach from Section 4</action>
+<action>Provide complete justification for recommendation</action>
+<action>Address trade-offs and alternatives considered</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="5.4">
+<prompt>Define PRD MVP impact and high-level action plan</prompt>
+<action>State clearly if MVP is affected</action>
+<action>Outline major action items needed for implementation</action>
+<action>Identify dependencies and sequencing</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="5.5">
+<prompt>Establish agent handoff plan</prompt>
+<action>Identify which roles/agents will execute the changes:</action>
+  - Developer agent (for implementation)
+  - Product Owner / Developer (for backlog changes)
+  - Product Manager / Architect (for strategic changes)
+<action>Define responsibilities for each role</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+</section>
+
+<section n="6" title="Final Review and Handoff">
+
+<check-item id="6.1">
+<prompt>Review checklist completion</prompt>
+<action>Verify all applicable sections have been addressed</action>
+<action>Confirm all [Action-needed] items have been documented</action>
+<action>Ensure analysis is comprehensive and actionable</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="6.2">
+<prompt>Verify Sprint Change Proposal accuracy</prompt>
+<action>Review complete proposal for consistency and clarity</action>
+<action>Ensure all recommendations are well-supported by analysis</action>
+<action>Check that proposal is actionable and specific</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="6.3">
+<prompt>Obtain explicit user approval</prompt>
+<action>Present complete proposal to user</action>
+<action>Get clear yes/no approval for proceeding</action>
+<action>Document approval and any conditions</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="6.4">
+<prompt>Update sprint-status.yaml to reflect approved epic changes</prompt>
+<action>If epics were added: Add new epic entries with status 'backlog'</action>
+<action>If epics were removed: Remove corresponding entries</action>
+<action>If epics were renumbered: Update epic IDs and story references</action>
+<action>If stories were added/removed: Update story entries within affected epics</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<check-item id="6.5">
+<prompt>Confirm next steps and handoff plan</prompt>
+<action>Review handoff responsibilities with user</action>
+<action>Ensure all stakeholders understand their roles</action>
+<action>Confirm timeline and success criteria</action>
+<status>[ ] Done / [ ] N/A / [ ] Action-needed</status>
+</check-item>
+
+<halt-condition>
+<action if="any critical section cannot be completed">HALT: "Cannot proceed to proposal without complete impact analysis"</action>
+<action if="user approval not obtained">HALT: "Must have explicit approval before implementing changes"</action>
+<action if="handoff responsibilities unclear">HALT: "Must clearly define who will execute the proposed changes"</action>
+</halt-condition>
+
+</section>
+
+</checklist>
+
+<execution-notes>
+<note>This checklist is for SIGNIFICANT changes affecting project direction</note>
+<note>Work interactively with user - they make final decisions</note>
+<note>Be factual, not blame-oriented when analyzing issues</note>
+<note>Handle changes professionally as opportunities to improve the project</note>
+<note>Maintain conversation context throughout - this is collaborative work</note>
+</execution-notes>
diff --git a/.agents/skills/bmad-correct-course/customize.toml b/.agents/skills/bmad-correct-course/customize.toml
new file mode 100644
index 00000000..d23577e4
--- /dev/null
+++ b/.agents/skills/bmad-correct-course/customize.toml
@@ -0,0 +1,41 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-correct-course. Mirrors the
+# agent customization shape under the [workflow] namespace.
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+# Steps to run before the standard activation (config load, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Distinct from the runtime memory sidecar — these are static context
+# loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "All sprint changes require PO sign-off before execution."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Scalar: executed when the workflow reaches Step 6 (Workflow Completion),
+# after the Sprint Change Proposal is finalized and handoff is confirmed. Override wins.
+# Leave empty for no custom post-completion behavior.
+
+on_complete = ""
diff --git a/.agents/skills/bmad-create-architecture/SKILL.md b/.agents/skills/bmad-create-architecture/SKILL.md
new file mode 100644
index 00000000..cbc46489
--- /dev/null
+++ b/.agents/skills/bmad-create-architecture/SKILL.md
@@ -0,0 +1,30 @@
+---
+name: bmad-create-architecture
+description: 'Deprecated — forwards to bmad-architecture (create intent).'
+---
+
+# DEPRECATED — forwards to bmad-architecture (create intent)
+
+This skill was consolidated into `bmad-architecture`. It is retained as a thin compatibility shim so existing invocations by name and `_bmad/custom/bmad-create-architecture.toml` override files keep working. New work should invoke `bmad-architecture` directly — it detects create / update / validate intent from the conversation.
+
+## On Activation
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. This picks up any `{project-root}/_bmad/custom/bmad-create-architecture.toml` and `bmad-create-architecture.user.toml` overrides for the legacy fields (`activation_steps_prepend`, `activation_steps_append`, `persistent_facts`, `on_complete`).
+
+2. Load `{project-root}/_bmad/bmm/config.yaml` (and `config.user.yaml` if present) to resolve `{user_name}` and `{communication_language}`.
+
+3. Emit a deprecation notice to the user in `{communication_language}`:
+
+   > Notice: `bmad-create-architecture` is deprecated and will be removed in a future release. It now forwards to `bmad-architecture` with create intent. To silence this notice and access the full new customization surface (`spine_template`, `spine_output_path`, `run_folder_pattern`, `doc_standards`, `external_sources`, `external_handoffs`, `finalize_reviewers`), migrate `_bmad/custom/bmad-create-architecture.toml` to `_bmad/custom/bmad-architecture.toml` and invoke `bmad-architecture` directly next time. Customization fields that were in this version still remain in the new version and will be respected if present in `_bmad/custom/bmad-architecture.toml`, but the new version also supports additional fields that you can take advantage of by migrating.
+
+4. Invoke `bmad-architecture` with the following context. Pass these as the activating context so `bmad-architecture` honors them instead of resolving its own customization from scratch:
+
+   - **Intent:** `create` — skip `bmad-architecture`'s usual intent detection step.
+   - **Pre-resolved legacy customization** — use these in place of resolving from `bmad-architecture`'s own `customize.toml` for the four legacy fields. For everything else (`spine_template`, `spine_output_path`, `run_folder_pattern`, `doc_standards`, `external_sources`, `external_handoffs`, `finalize_reviewers`), use `bmad-architecture`'s own defaults and overrides as normal:
+     - `activation_steps_prepend` = the resolved value from step 1
+     - `activation_steps_append` = the resolved value from step 1
+     - `persistent_facts` = the resolved value from step 1
+     - `on_complete` = the resolved value from step 1
+   - **Original user input:** forward whatever the user said when invoking this skill verbatim.
+
+   `bmad-architecture` takes the workflow from here. Do not execute any further steps in this shim.
diff --git a/.agents/skills/bmad-create-architecture/customize.toml b/.agents/skills/bmad-create-architecture/customize.toml
new file mode 100644
index 00000000..32756120
--- /dev/null
+++ b/.agents/skills/bmad-create-architecture/customize.toml
@@ -0,0 +1,41 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-create-architecture. Mirrors the
+# agent customization shape under the [workflow] namespace.
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+# Steps to run before the standard activation (config load, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Distinct from the runtime memory sidecar — these are static context
+# loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "Our org is AWS-only -- do not propose GCP or Azure."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Scalar: executed when the workflow reaches Step 8 (Architecture Completion & Handoff),
+# after the architecture document frontmatter is updated and next-steps guidance is given.
+# Override wins. Leave empty for no custom post-completion behavior.
+
+on_complete = ""
diff --git a/.agents/skills/bmad-create-epics-and-stories/SKILL.md b/.agents/skills/bmad-create-epics-and-stories/SKILL.md
new file mode 100644
index 00000000..e822e6a6
--- /dev/null
+++ b/.agents/skills/bmad-create-epics-and-stories/SKILL.md
@@ -0,0 +1,93 @@
+---
+name: bmad-create-epics-and-stories
+description: 'Break requirements into epics and user stories. Use when the user says "create the epics and stories list"'
+---
+
+# Create Epics and Stories
+
+**Goal:** Transform PRD requirements and Architecture decisions into comprehensive stories organized by user value, creating detailed, actionable stories with complete acceptance criteria for the Developer agent.
+
+**Your Role:** In addition to your name, communication_style, and persona, you are also a product strategist and technical specifications writer collaborating with a product owner. This is a partnership, not a client-vendor relationship. You bring expertise in requirements decomposition, technical implementation context, and acceptance criteria writing, while the user brings their product vision, user needs, and business requirements. Work together as equals.
+
+## Conventions
+
+- Bare paths (e.g. `steps/step-01-validate-prerequisites.md`) resolve from the skill root.
+- `{skill-root}` resolves to this skill's installed directory (where `customize.toml` lives).
+- `{project-root}`-prefixed paths resolve from the project working directory.
+- `{skill-name}` resolves to the skill directory's basename.
+
+## WORKFLOW ARCHITECTURE
+
+This uses **step-file architecture** for disciplined execution:
+
+### Core Principles
+
+- **Micro-file Design**: Each step toward the overall goal is a self-contained instruction file; adhere to one file at a time, as directed
+- **Just-In-Time Loading**: Only 1 current step file will be loaded and followed to completion - never load future step files until told to do so
+- **Sequential Enforcement**: Sequence within the step files must be completed in order, no skipping or optimization allowed
+- **State Tracking**: Document progress in output file frontmatter using `stepsCompleted` array when a workflow produces a document
+- **Append-Only Building**: Build documents by appending content as directed to the output file
+
+### Step Processing Rules
+
+1. **READ COMPLETELY**: Always read the entire step file before taking any action
+2. **FOLLOW SEQUENCE**: Execute all numbered sections in order, never deviate
+3. **WAIT FOR INPUT**: If a menu is presented, halt and wait for user selection
+4. **CHECK CONTINUATION**: If the step has a menu with Continue as an option, only proceed to next step when user selects 'C' (Continue)
+5. **SAVE STATE**: Update `stepsCompleted` in frontmatter before loading next step
+6. **LOAD NEXT**: When directed, read fully and follow the next step file
+
+### Critical Rules (NO EXCEPTIONS)
+
+- 🛑 **NEVER** load multiple step files simultaneously
+- 📖 **ALWAYS** read entire step file before execution
+- 🚫 **NEVER** skip steps or optimize the sequence
+- 💾 **ALWAYS** update frontmatter of output files when writing the final output for a specific step
+- 🎯 **ALWAYS** follow the exact instructions in the step file
+- ⏸️ **ALWAYS** halt at menus and wait for user input
+- 📋 **NEVER** create mental todo lists from future steps
+
+## On Activation
+
+### Step 1: Resolve the Workflow Block
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`
+
+**If the script fails**, resolve the `workflow` block yourself by reading these three files in base → team → user order and applying the same structural merge rules as the resolver:
+
+1. `{skill-root}/customize.toml` — defaults
+2. `{project-root}/_bmad/custom/{skill-name}.toml` — team overrides
+3. `{project-root}/_bmad/custom/{skill-name}.user.toml` — personal overrides
+
+Any missing file is skipped. Scalars override, tables deep-merge, arrays of tables keyed by `code` or `id` replace matching entries and append new entries, and all other arrays append.
+
+### Step 2: Execute Prepend Steps
+
+Execute each entry in `{workflow.activation_steps_prepend}` in order before proceeding.
+
+### Step 3: Load Persistent Facts
+
+Treat every entry in `{workflow.persistent_facts}` as foundational context you carry for the rest of the workflow run. Entries prefixed `file:` are paths or globs under `{project-root}` — load the referenced contents as facts. All other entries are facts verbatim.
+
+### Step 4: Load Config
+
+Load config from `{project-root}/_bmad/bmm/config.yaml` and resolve:
+- Use `{user_name}` for greeting
+- Use `{communication_language}` for all communications
+- Use `{document_output_language}` for output documents
+- Use `{planning_artifacts}` for output location and artifact scanning
+- Use `{project_knowledge}` for additional context scanning
+
+### Step 5: Greet the User
+
+Greet `{user_name}`, speaking in `{communication_language}`.
+
+### Step 6: Execute Append Steps
+
+Execute each entry in `{workflow.activation_steps_append}` in order.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+## Execution
+
+Read fully and follow: `./steps/step-01-validate-prerequisites.md` to begin the workflow.
diff --git a/.agents/skills/bmad-create-epics-and-stories/customize.toml b/.agents/skills/bmad-create-epics-and-stories/customize.toml
new file mode 100644
index 00000000..fb05efaf
--- /dev/null
+++ b/.agents/skills/bmad-create-epics-and-stories/customize.toml
@@ -0,0 +1,41 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-create-epics-and-stories. Mirrors the
+# agent customization shape under the [workflow] namespace.
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+# Steps to run before the standard activation (config load, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Distinct from the runtime memory sidecar — these are static context
+# loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "All epics must deliver complete end-to-end user value."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Scalar: executed when the workflow reaches Step 4 (Final Validation) and the
+# user confirms [C] Complete — after the epics.md is saved and bmad-help is invoked.
+# Override wins. Leave empty for no custom post-completion behavior.
+
+on_complete = ""
diff --git a/.agents/skills/bmad-create-epics-and-stories/steps/step-01-validate-prerequisites.md b/.agents/skills/bmad-create-epics-and-stories/steps/step-01-validate-prerequisites.md
new file mode 100644
index 00000000..5930a773
--- /dev/null
+++ b/.agents/skills/bmad-create-epics-and-stories/steps/step-01-validate-prerequisites.md
@@ -0,0 +1,263 @@
+# Step 1: Validate Prerequisites and Extract Requirements
+
+## STEP GOAL:
+
+To validate that all required input documents exist and extract all requirements (FRs, NFRs, and additional requirements from UX/Architecture) needed for epic and story creation.
+
+## MANDATORY EXECUTION RULES (READ FIRST):
+
+### Universal Rules:
+
+- 🛑 NEVER generate content without user input
+- 📖 CRITICAL: Read the complete step file before taking any action
+- 🔄 CRITICAL: When loading next step with 'C', ensure entire file is read
+- 📋 YOU ARE A FACILITATOR, not a content generator
+- ✅ YOU MUST ALWAYS SPEAK OUTPUT In your Agent communication style with the config `{communication_language}`
+
+### Role Reinforcement:
+
+- ✅ You are a product strategist and technical specifications writer
+- ✅ If you already have been given communication or persona patterns, continue to use those while playing this new role
+- ✅ We engage in collaborative dialogue, not command-response
+- ✅ You bring requirements extraction expertise
+- ✅ User brings their product vision and context
+
+### Step-Specific Rules:
+
+- 🎯 Focus ONLY on extracting and organizing requirements
+- 🚫 FORBIDDEN to start creating epics or stories in this step
+- 💬 Extract requirements from ALL available documents
+- 🚪 POPULATE the template sections exactly as needed
+
+## EXECUTION PROTOCOLS:
+
+- 🎯 Extract requirements systematically from all documents
+- 💾 Populate {planning_artifacts}/epics.md with extracted requirements
+- 📖 Update frontmatter with extraction progress
+- 🚫 FORBIDDEN to load next step until user selects 'C' and requirements are extracted
+
+## REQUIREMENTS EXTRACTION PROCESS:
+
+### 1. Welcome and Overview
+
+Welcome {user_name} to comprehensive epic and story creation!
+
+**CRITICAL PREREQUISITE VALIDATION:**
+
+Verify required documents exist and are complete:
+
+1. **PRD.md** - Contains requirements (FRs and NFRs) and product scope
+2. **Architecture.md** - Contains technical decisions, API contracts, data models
+3. **UX design contract** (if UI exists) - Contains visual identity, interaction patterns, mockups, and user flows
+
+### 2. Document Discovery and Validation
+
+Search for required documents using these patterns (sharded means a large document was split into multiple small files with an index.md into a folder) - if the whole document is found, use that instead of the sharded version:
+
+**PRD Document Search Priority:**
+
+1. `{planning_artifacts}/*prd*.md` (whole document)
+2. `{planning_artifacts}/*prd*/index.md` (sharded version)
+
+**Architecture Document Search Priority:**
+
+1. `{planning_artifacts}/*architecture*.md` (whole document)
+2. `{planning_artifacts}/*architecture*/index.md` (sharded version)
+
+**UX Design Document Search (Optional):**
+
+1. `{planning_artifacts}/ux-designs/ux-*/DESIGN.md` and `{planning_artifacts}/ux-designs/ux-*/EXPERIENCE.md` (bmad-ux spine pair)
+2. `{planning_artifacts}/*ux*.md` (legacy whole document)
+3. `{planning_artifacts}/*ux*/index.md` (legacy sharded version)
+
+For each matching bmad-ux run folder, treat `DESIGN.md` and `EXPERIENCE.md` as one UX design contract:
+
+- Confirm and load both files together. `DESIGN.md` owns visual identity and design tokens; `EXPERIENCE.md` owns information architecture, behavior, states, interactions, accessibility, and journeys.
+- Add both files to the `inputDocuments: []` frontmatter array.
+- If only one spine exists, report the incomplete pair and ask whether the user wants to include the partial UX handoff.
+- If multiple run folders match, show each run folder with the spine frontmatter `status` and `updated` values when available, then ask the user which UX design contract to include.
+
+Before proceeding, Ask the user if there are any other documents to include for analysis, and if anything found should be excluded. Wait for user confirmation. Once confirmed, create the {planning_artifacts}/epics.md from the ../templates/epics-template.md and in the front matter list the files in the array of `inputDocuments: []`.
+
+### 3. Extract Functional Requirements (FRs)
+
+From the PRD document (full or sharded), read then entire document and extract ALL functional requirements:
+
+**Extraction Method:**
+
+- Look for numbered items like "FR1:", "Functional Requirement 1:", or similar
+- Identify requirement statements that describe what the system must DO
+- Include user actions, system behaviors, and business rules
+
+**Format the FR list as:**
+
+```
+FR1: [Clear, testable requirement description]
+FR2: [Clear, testable requirement description]
+...
+```
+
+### 4. Extract Non-Functional Requirements (NFRs)
+
+From the PRD document, extract ALL non-functional requirements:
+
+**Extraction Method:**
+
+- Look for performance, security, usability, reliability requirements
+- Identify constraints and quality attributes
+- Include technical standards and compliance requirements
+
+**Format the NFR list as:**
+
+```
+NFR1: [Performance/Security/Usability requirement]
+NFR2: [Performance/Security/Usability requirement]
+...
+```
+
+### 5. Extract Additional Requirements from Architecture
+
+Review the Architecture document for technical requirements that impact epic and story creation:
+
+**Look for:**
+
+- **Starter Template**: Does Architecture specify a starter/greenfield template? If YES, document this for Epic 1 Story 1
+- Infrastructure and deployment requirements
+- Integration requirements with external systems
+- Data migration or setup requirements
+- Monitoring and logging requirements
+- API versioning or compatibility requirements
+- Security implementation requirements
+
+**IMPORTANT**: If a starter template is mentioned in Architecture, note it prominently. This will impact Epic 1 Story 1.
+
+**Format Additional Requirements as:**
+
+```
+- [Technical requirement from Architecture that affects implementation]
+- [Infrastructure setup requirement]
+- [Integration requirement]
+...
+```
+
+### 6. Extract UX Design Requirements (if UX document exists)
+
+**IMPORTANT**: The UX Design Specification is a first-class input document, not supplementary material. Requirements from the UX spec must be extracted with the same rigor as PRD functional requirements.
+
+Read the FULL UX design contract and extract ALL actionable work items. For a bmad-ux spine pair, read both `DESIGN.md` and `EXPERIENCE.md`:
+
+**Look for:**
+
+- **Design token work**: Color systems, spacing scales, typography tokens that need implementation or consolidation
+- **Component proposals**: Reusable UI components identified in the UX spec (e.g., ConfirmActions, StatusMessage, EmptyState, FocusIndicator)
+- **Visual standardization**: Semantic CSS classes, consistent color palette usage, design pattern consolidation
+- **Accessibility requirements**: Contrast audit fixes, ARIA patterns, keyboard navigation, screen reader support
+- **Responsive design requirements**: Breakpoints, layout adaptations, mobile-specific interactions
+- **Interaction patterns**: Animations, transitions, loading states, error handling UX
+- **Browser/device compatibility**: Target platforms, progressive enhancement requirements
+
+**Format UX Design Requirements as a SEPARATE section (not merged into Additional Requirements):**
+
+```
+UX-DR1: [Actionable UX design requirement with clear implementation scope]
+UX-DR2: [Actionable UX design requirement with clear implementation scope]
+...
+```
+
+**🚨 CRITICAL**: Do NOT reduce UX requirements to vague summaries. Each UX-DR must be specific enough to generate a story with testable acceptance criteria. If the UX spec identifies 6 reusable components, list all 6 — not "create reusable components."
+
+### 7. Load and Initialize Template
+
+Load ../templates/epics-template.md and initialize {planning_artifacts}/epics.md:
+
+1. Copy the entire template to {planning_artifacts}/epics.md
+2. Replace {{project_name}} with the actual project name
+3. Replace placeholder sections with extracted requirements:
+   - {{fr_list}} → extracted FRs
+   - {{nfr_list}} → extracted NFRs
+   - {{additional_requirements}} → extracted additional requirements (from Architecture)
+   - {{ux_design_requirements}} → extracted UX Design Requirements (if UX document exists)
+4. Leave {{requirements_coverage_map}} and {{epics_list}} as placeholders for now
+
+### 8. Present Extracted Requirements
+
+Display to user:
+
+**Functional Requirements Extracted:**
+
+- Show count of FRs found
+- Display the first few FRs as examples
+- Ask if any FRs are missing or incorrectly captured
+
+**Non-Functional Requirements Extracted:**
+
+- Show count of NFRs found
+- Display key NFRs
+- Ask if any constraints were missed
+
+**Additional Requirements (Architecture):**
+
+- Summarize technical requirements from Architecture
+- Verify completeness
+
+**UX Design Requirements (if applicable):**
+
+- Show count of UX-DRs found
+- Display key UX Design requirements (design tokens, components, accessibility)
+- Verify each UX-DR is specific enough for story creation
+
+### 9. Get User Confirmation
+
+Ask: "Do these extracted requirements accurately represent what needs to be built? Any additions or corrections?"
+
+Update the requirements based on user feedback until confirmation is received.
+
+## CONTENT TO SAVE TO DOCUMENT:
+
+After extraction and confirmation, update {planning_artifacts}/epics.md with:
+
+- Complete FR list in {{fr_list}} section
+- Complete NFR list in {{nfr_list}} section
+- All additional requirements in {{additional_requirements}} section
+- UX Design requirements in {{ux_design_requirements}} section (if UX document exists)
+
+### 10. Present MENU OPTIONS
+
+Display: `**Confirm the Requirements are complete and correct to [C] continue:**`
+
+#### EXECUTION RULES:
+
+- ALWAYS halt and wait for user input after presenting menu
+- ONLY proceed to next step when user selects 'C'
+- User can chat or ask questions - always respond and then end with display again of the menu option
+
+#### Menu Handling Logic:
+
+- IF C: Save all to {planning_artifacts}/epics.md, update frontmatter, then read fully and follow: ./step-02-design-epics.md
+- IF Any other comments or queries: help user respond then [Redisplay Menu Options](#10-present-menu-options)
+
+## CRITICAL STEP COMPLETION NOTE
+
+ONLY WHEN C is selected and all requirements are saved to document and frontmatter is updated, will you then read fully and follow: ./step-02-design-epics.md to begin epic design step.
+
+---
+
+## 🚨 SYSTEM SUCCESS/FAILURE METRICS
+
+### ✅ SUCCESS:
+
+- All required documents found and validated
+- All FRs extracted and formatted correctly
+- All NFRs extracted and formatted correctly
+- Additional requirements from Architecture/UX identified
+- Template initialized with requirements
+- User confirms requirements are complete and accurate
+
+### ❌ SYSTEM FAILURE:
+
+- Missing required documents
+- Incomplete requirements extraction
+- Template not properly initialized
+- Not saving requirements to output file
+
+**Master Rule:** Skipping steps, optimizing sequences, or not following exact instructions is FORBIDDEN and constitutes SYSTEM FAILURE.
diff --git a/.agents/skills/bmad-create-epics-and-stories/steps/step-02-design-epics.md b/.agents/skills/bmad-create-epics-and-stories/steps/step-02-design-epics.md
new file mode 100644
index 00000000..937f2df2
--- /dev/null
+++ b/.agents/skills/bmad-create-epics-and-stories/steps/step-02-design-epics.md
@@ -0,0 +1,242 @@
+# Step 2: Design Epic List
+
+## STEP GOAL:
+
+To design and get approval for the epics_list that will organize all requirements into user-value-focused epics.
+
+## MANDATORY EXECUTION RULES (READ FIRST):
+
+### Universal Rules:
+
+- 🛑 NEVER generate content without user input
+- 📖 CRITICAL: Read the complete step file before taking any action
+- 🔄 CRITICAL: When loading next step with 'C', ensure entire file is read
+- 📋 YOU ARE A FACILITATOR, not a content generator
+- ✅ YOU MUST ALWAYS SPEAK OUTPUT In your Agent communication style with the config `{communication_language}`
+
+### Role Reinforcement:
+
+- ✅ You are a product strategist and technical specifications writer
+- ✅ If you already have been given communication or persona patterns, continue to use those while playing this new role
+- ✅ We engage in collaborative dialogue, not command-response
+- ✅ You bring product strategy and epic design expertise
+- ✅ User brings their product vision and priorities
+
+### Step-Specific Rules:
+
+- 🎯 Focus ONLY on creating the epics_list
+- 🚫 FORBIDDEN to create individual stories in this step
+- 💬 Organize epics around user value, not technical layers
+- 🚪 GET explicit approval for the epics_list
+- 🔗 **CRITICAL: Each epic must be standalone and enable future epics without requiring future epics to function**
+
+## EXECUTION PROTOCOLS:
+
+- 🎯 Design epics collaboratively based on extracted requirements
+- 💾 Update {{epics_list}} in {planning_artifacts}/epics.md
+- 📖 Document the FR coverage mapping
+- 🚫 FORBIDDEN to load next step until user approves epics_list
+
+## EPIC DESIGN PROCESS:
+
+### 1. Review Extracted Requirements
+
+Load {planning_artifacts}/epics.md and review:
+
+- **Functional Requirements:** Count and review FRs from Step 1
+- **Non-Functional Requirements:** Review NFRs that need to be addressed
+- **Additional Requirements:** Review technical and UX requirements
+
+### 2. Explain Epic Design Principles
+
+**EPIC DESIGN PRINCIPLES:**
+
+1. **User-Value First**: Each epic must enable users to accomplish something meaningful
+2. **Requirements Grouping**: Group related FRs that deliver cohesive user outcomes
+3. **Incremental Delivery**: Each epic should deliver value independently
+4. **Logical Flow**: Natural progression from user's perspective
+5. **Dependency-Free Within Epic**: Stories within an epic must NOT depend on future stories
+6. **Implementation Efficiency**: Consider consolidating epics that all modify the same core files into fewer epics
+
+**⚠️ CRITICAL PRINCIPLE:**
+Organize by USER VALUE, not technical layers:
+
+**✅ CORRECT Epic Examples (Standalone & Enable Future Epics):**
+
+- Epic 1: User Authentication & Profiles (users can register, login, manage profiles) - **Standalone: Complete auth system**
+- Epic 2: Content Creation (users can create, edit, publish content) - **Standalone: Uses auth, creates content**
+- Epic 3: Social Interaction (users can follow, comment, like content) - **Standalone: Uses auth + content**
+- Epic 4: Search & Discovery (users can find content and other users) - **Standalone: Uses all previous**
+
+**❌ WRONG Epic Examples (Technical Layers or Dependencies):**
+
+- Epic 1: Database Setup (creates all tables upfront) - **No user value**
+- Epic 2: API Development (builds all endpoints) - **No user value**
+- Epic 3: Frontend Components (creates reusable components) - **No user value**
+- Epic 4: Deployment Pipeline (CI/CD setup) - **No user value**
+
+**❌ WRONG Epic Examples (File Churn on Same Component):**
+
+- Epic 1: File Upload (modifies model, controller, web form, web API)
+- Epic 2: File Status (modifies model, controller, web form, web API)
+- Epic 3: File Access permissions (modifies model, controller, web form, web API)
+- All three epics touch the same files — consolidate into one epic with ordered stories
+
+**✅ CORRECT Alternative:**
+
+- Epic 1: File Management Enhancement (upload, status, permissions as stories within one epic)
+- Rationale: Single component, fully pre-designed, no feedback loop between epics
+
+**🔗 DEPENDENCY RULES:**
+
+- Each epic must deliver COMPLETE functionality for its domain
+- Epic 2 must not require Epic 3 to function
+- Epic 3 can build upon Epic 1 & 2 but must stand alone
+
+### 3. Design Epic Structure Collaboratively
+
+**Step A: Assess Context and Identify Themes**
+
+First, assess how much of the solution design is already validated (Architecture, UX, Test Design).
+When the outcome is certain and direction changes between epics are unlikely, prefer fewer but larger epics.
+Split into multiple epics when there is a genuine risk boundary or when early feedback could change direction
+of following epics.
+
+Then, identify user value themes:
+
+- Look for natural groupings in the FRs
+- Identify user journeys or workflows
+- Consider user types and their goals
+
+**Step B: Propose Epic Structure**
+
+For each proposed epic (considering whether epics share the same core files):
+
+1. **Epic Title**: User-centric, value-focused
+2. **User Outcome**: What users can accomplish after this epic
+3. **FR Coverage**: Which FR numbers this epic addresses
+4. **Implementation Notes**: Any technical or UX considerations
+
+**Step C: Review for File Overlap**
+
+Assess whether multiple proposed epics repeatedly target the same core files. If overlap is significant:
+
+- Distinguish meaningful overlap (same component end-to-end) from incidental sharing
+- Ask whether to consolidate into one epic with ordered stories
+- If confirmed, merge the epic FRs into a single epic, preserving dependency flow: each story must still fit within
+  a single dev agent's context
+
+**Step D: Create the epics_list**
+
+Format the epics_list as:
+
+```
+## Epic List
+
+### Epic 1: [Epic Title]
+[Epic goal statement - what users can accomplish]
+**FRs covered:** FR1, FR2, FR3, etc.
+
+### Epic 2: [Epic Title]
+[Epic goal statement - what users can accomplish]
+**FRs covered:** FR4, FR5, FR6, etc.
+
+[Continue for all epics]
+```
+
+### 4. Present Epic List for Review
+
+Display the complete epics_list to user with:
+
+- Total number of epics
+- FR coverage per epic
+- User value delivered by each epic
+- Any natural dependencies
+
+### 5. Create Requirements Coverage Map
+
+Create {{requirements_coverage_map}} showing how each FR maps to an epic:
+
+```
+### FR Coverage Map
+
+FR1: Epic 1 - [Brief description]
+FR2: Epic 1 - [Brief description]
+FR3: Epic 2 - [Brief description]
+...
+```
+
+This ensures no FRs are missed.
+
+### 6. Collaborative Refinement
+
+Ask user:
+
+- "Does this epic structure align with your product vision?"
+- "Are all user outcomes properly captured?"
+- "Should we adjust any epic groupings?"
+- "Are there natural dependencies we've missed?"
+
+### 7. Get Final Approval
+
+**CRITICAL:** Must get explicit user approval:
+"Do you approve this epic structure for proceeding to story creation?"
+
+If user wants changes:
+
+- Make the requested adjustments
+- Update the epics_list
+- Re-present for approval
+- Repeat until approval is received
+
+## CONTENT TO UPDATE IN DOCUMENT:
+
+After approval, update {planning_artifacts}/epics.md:
+
+1. Replace {{epics_list}} placeholder with the approved epic list
+2. Replace {{requirements_coverage_map}} with the coverage map
+3. Ensure all FRs are mapped to epics
+
+### 8. Present MENU OPTIONS
+
+Display: "**Select an Option:** [A] Advanced Elicitation [P] Party Mode [C] Continue"
+
+#### Menu Handling Logic:
+
+- IF A: Invoke the `bmad-advanced-elicitation` skill
+- IF P: Invoke the `bmad-party-mode` skill
+- IF C: Save approved epics_list to {planning_artifacts}/epics.md, update frontmatter, then read fully and follow: ./step-03-create-stories.md
+- IF Any other comments or queries: help user respond then [Redisplay Menu Options](#8-present-menu-options)
+
+#### EXECUTION RULES:
+
+- ALWAYS halt and wait for user input after presenting menu
+- ONLY proceed to next step when user selects 'C'
+- After other menu items execution completes, redisplay the menu
+- User can chat or ask questions - always respond when conversation ends, redisplay the menu options
+
+## CRITICAL STEP COMPLETION NOTE
+
+ONLY WHEN C is selected and the approved epics_list is saved to document, will you then read fully and follow: ./step-03-create-stories.md to begin story creation step.
+
+---
+
+## 🚨 SYSTEM SUCCESS/FAILURE METRICS
+
+### ✅ SUCCESS:
+
+- Epics designed around user value
+- All FRs mapped to specific epics
+- epics_list created and formatted correctly
+- Requirements coverage map completed
+- User gives explicit approval for epic structure
+- Document updated with approved epics
+
+### ❌ SYSTEM FAILURE:
+
+- Epics organized by technical layers
+- Missing FRs in coverage map
+- No user approval obtained
+- epics_list not saved to document
+
+**Master Rule:** Skipping steps, optimizing sequences, or not following exact instructions is FORBIDDEN and constitutes SYSTEM FAILURE.
diff --git a/.agents/skills/bmad-create-epics-and-stories/steps/step-03-create-stories.md b/.agents/skills/bmad-create-epics-and-stories/steps/step-03-create-stories.md
new file mode 100644
index 00000000..14caafeb
--- /dev/null
+++ b/.agents/skills/bmad-create-epics-and-stories/steps/step-03-create-stories.md
@@ -0,0 +1,255 @@
+# Step 3: Generate Epics and Stories
+
+## STEP GOAL:
+
+To generate all epics with their stories based on the approved epics_list, following the template structure exactly.
+
+## MANDATORY EXECUTION RULES (READ FIRST):
+
+### Universal Rules:
+
+- 🛑 NEVER generate content without user input
+- 📖 CRITICAL: Read the complete step file before taking any action
+- 🔄 CRITICAL: Process epics sequentially
+- 📋 YOU ARE A FACILITATOR, not a content generator
+- ✅ YOU MUST ALWAYS SPEAK OUTPUT In your Agent communication style with the config `{communication_language}`
+
+### Role Reinforcement:
+
+- ✅ You are a product strategist and technical specifications writer
+- ✅ If you already have been given communication or persona patterns, continue to use those while playing this new role
+- ✅ We engage in collaborative dialogue, not command-response
+- ✅ You bring story creation and acceptance criteria expertise
+- ✅ User brings their implementation priorities and constraints
+
+### Step-Specific Rules:
+
+- 🎯 Generate stories for each epic following the template exactly
+- 🚫 FORBIDDEN to deviate from template structure
+- 💬 Each story must have clear acceptance criteria
+- 🚪 ENSURE each story is completable by a single dev agent
+- 🔗 **CRITICAL: Stories MUST NOT depend on future stories within the same epic**
+
+## EXECUTION PROTOCOLS:
+
+- 🎯 Generate stories collaboratively with user input
+- 💾 Append epics and stories to {planning_artifacts}/epics.md following template
+- 📖 Process epics one at a time in sequence
+- 🚫 FORBIDDEN to skip any epic or rush through stories
+
+## STORY GENERATION PROCESS:
+
+### 1. Load Approved Epic Structure
+
+Load {planning_artifacts}/epics.md and review:
+
+- Approved epics_list from Step 2
+- FR coverage map
+- All requirements (FRs, NFRs, additional, **UX Design requirements if present**)
+- Template structure at the end of the document
+
+**UX Design Integration**: If UX Design Requirements (UX-DRs) were extracted in Step 1, ensure they are visible during story creation. UX-DRs must be covered by stories — either within existing epics (e.g., accessibility fixes for a feature epic) or in a dedicated "Design System / UX Polish" epic.
+
+### 2. Explain Story Creation Approach
+
+**STORY CREATION GUIDELINES:**
+
+For each epic, create stories that:
+
+- Follow the exact template structure
+- Are sized for single dev agent completion
+- Have clear user value
+- Include specific acceptance criteria
+- Reference requirements being fulfilled
+
+**🚨 DATABASE/ENTITY CREATION PRINCIPLE:**
+Create tables/entities ONLY when needed by the story:
+
+- ❌ WRONG: Epic 1 Story 1 creates all 50 database tables
+- ✅ RIGHT: Each story creates/alters ONLY the tables it needs
+
+**🔗 STORY DEPENDENCY PRINCIPLE:**
+Stories must be independently completable in sequence:
+
+- ❌ WRONG: Story 1.2 requires Story 1.3 to be completed first
+- ✅ RIGHT: Each story can be completed based only on previous stories
+- ❌ WRONG: "Wait for Story 1.4 to be implemented before this works"
+- ✅ RIGHT: "This story works independently and enables future stories"
+
+**STORY FORMAT (from template):**
+
+```
+### Story {N}.{M}: {story_title}
+
+As a {user_type},
+I want {capability},
+So that {value_benefit}.
+
+**Acceptance Criteria:**
+
+**Given** {precondition}
+**When** {action}
+**Then** {expected_outcome}
+**And** {additional_criteria}
+```
+
+**✅ GOOD STORY EXAMPLES:**
+
+_Epic 1: User Authentication_
+
+- Story 1.1: User Registration with Email
+- Story 1.2: User Login with Password
+- Story 1.3: Password Reset via Email
+
+_Epic 2: Content Creation_
+
+- Story 2.1: Create New Blog Post
+- Story 2.2: Edit Existing Blog Post
+- Story 2.3: Publish Blog Post
+
+**❌ BAD STORY EXAMPLES:**
+
+- Story: "Set up database" (no user value)
+- Story: "Create all models" (too large, no user value)
+- Story: "Build authentication system" (too large)
+- Story: "Login UI (depends on Story 1.3 API endpoint)" (future dependency!)
+- Story: "Edit post (requires Story 1.4 to be implemented first)" (wrong order!)
+
+### 3. Process Epics Sequentially
+
+For each epic in the approved epics_list:
+
+#### A. Epic Overview
+
+Display:
+
+- Epic number and title
+- Epic goal statement
+- FRs covered by this epic
+- Any NFRs or additional requirements relevant
+- Any UX Design Requirements (UX-DRs) relevant to this epic
+
+#### B. Story Breakdown
+
+Work with user to break down the epic into stories:
+
+- Identify distinct user capabilities
+- Ensure logical flow within the epic
+- Size stories appropriately
+
+#### C. Generate Each Story
+
+For each story in the epic:
+
+1. **Story Title**: Clear, action-oriented
+2. **User Story**: Complete the As a/I want/So that format
+3. **Acceptance Criteria**: Write specific, testable criteria
+
+**AC Writing Guidelines:**
+
+- Use Given/When/Then format
+- Each AC should be independently testable
+- Include edge cases and error conditions
+- Reference specific requirements when applicable
+
+#### D. Collaborative Review
+
+After writing each story:
+
+- Present the story to user
+- Ask: "Does this story capture the requirement correctly?"
+- "Is the scope appropriate for a single dev session?"
+- "Are the acceptance criteria complete and testable?"
+
+#### E. Append to Document
+
+When story is approved:
+
+- Append it to {planning_artifacts}/epics.md following template structure
+- Use correct numbering (Epic N, Story M)
+- Maintain proper markdown formatting
+
+### 4. Epic Completion
+
+After all stories for an epic are complete:
+
+- Display epic summary
+- Show count of stories created
+- Verify all FRs for the epic are covered
+- Get user confirmation to proceed to next epic
+
+### 5. Repeat for All Epics
+
+Continue the process for each epic in the approved list, processing them in order (Epic 1, Epic 2, etc.).
+
+### 6. Final Document Completion
+
+After all epics and stories are generated:
+
+- Verify the document follows template structure exactly
+- Ensure all placeholders are replaced
+- Confirm all FRs are covered
+- **Confirm all UX Design Requirements (UX-DRs) are covered by at least one story** (if UX document was an input)
+- Check formatting consistency
+
+## TEMPLATE STRUCTURE COMPLIANCE:
+
+The final {planning_artifacts}/epics.md must follow this structure exactly:
+
+1. **Overview** section with project name
+2. **Requirements Inventory** with all three subsections populated
+3. **FR Coverage Map** showing requirement to epic mapping
+4. **Epic List** with approved epic structure
+5. **Epic sections** for each epic (N = 1, 2, 3...)
+   - Epic title and goal
+   - All stories for that epic (M = 1, 2, 3...)
+     - Story title and user story
+     - Acceptance Criteria using Given/When/Then format
+
+### 7. Present FINAL MENU OPTIONS
+
+After all epics and stories are complete:
+
+Display: "**Select an Option:** [A] Advanced Elicitation [P] Party Mode [C] Continue"
+
+#### Menu Handling Logic:
+
+- IF A: Invoke the `bmad-advanced-elicitation` skill
+- IF P: Invoke the `bmad-party-mode` skill
+- IF C: Save content to {planning_artifacts}/epics.md, update frontmatter, then read fully and follow: ./step-04-final-validation.md
+- IF Any other comments or queries: help user respond then [Redisplay Menu Options](#7-present-final-menu-options)
+
+#### EXECUTION RULES:
+
+- ALWAYS halt and wait for user input after presenting menu
+- ONLY proceed to next step when user selects 'C'
+- After other menu items execution, return to this menu
+- User can chat or ask questions - always respond and then end with display again of the menu options
+
+## CRITICAL STEP COMPLETION NOTE
+
+ONLY WHEN [C continue option] is selected and [all epics and stories saved to document following the template structure exactly], will you then read fully and follow: `./step-04-final-validation.md` to begin final validation phase.
+
+---
+
+## 🚨 SYSTEM SUCCESS/FAILURE METRICS
+
+### ✅ SUCCESS:
+
+- All epics processed in sequence
+- Stories created for each epic
+- Template structure followed exactly
+- All FRs covered by stories
+- Stories appropriately sized
+- Acceptance criteria are specific and testable
+- Document is complete and ready for development
+
+### ❌ SYSTEM FAILURE:
+
+- Deviating from template structure
+- Missing epics or stories
+- Stories too large or unclear
+- Missing acceptance criteria
+- Not following proper formatting
+
+**Master Rule:** Skipping steps, optimizing sequences, or not following exact instructions is FORBIDDEN and constitutes SYSTEM FAILURE.
diff --git a/.agents/skills/bmad-create-epics-and-stories/steps/step-04-final-validation.md b/.agents/skills/bmad-create-epics-and-stories/steps/step-04-final-validation.md
new file mode 100644
index 00000000..a943f3d8
--- /dev/null
+++ b/.agents/skills/bmad-create-epics-and-stories/steps/step-04-final-validation.md
@@ -0,0 +1,143 @@
+# Step 4: Final Validation
+
+## STEP GOAL:
+
+To validate complete coverage of all requirements and ensure stories are ready for development.
+
+## MANDATORY EXECUTION RULES (READ FIRST):
+
+### Universal Rules:
+
+- 🛑 NEVER generate content without user input
+- 📖 CRITICAL: Read the complete step file before taking any action
+- 🔄 CRITICAL: Process validation sequentially without skipping
+- 📋 YOU ARE A FACILITATOR, not a content generator
+- ✅ YOU MUST ALWAYS SPEAK OUTPUT In your Agent communication style with the config `{communication_language}`
+
+### Role Reinforcement:
+
+- ✅ You are a product strategist and technical specifications writer
+- ✅ If you already have been given communication or persona patterns, continue to use those while playing this new role
+- ✅ We engage in collaborative dialogue, not command-response
+- ✅ You bring validation expertise and quality assurance
+- ✅ User brings their implementation priorities and final review
+
+### Step-Specific Rules:
+
+- 🎯 Focus ONLY on validating complete requirements coverage
+- 🚫 FORBIDDEN to skip any validation checks
+- 💬 Validate FR coverage, story completeness, and dependencies
+- 🚪 ENSURE all stories are ready for development
+
+## EXECUTION PROTOCOLS:
+
+- 🎯 Validate every requirement has story coverage
+- 💾 Check story dependencies and flow
+- 📖 Verify architecture compliance
+- 🚫 FORBIDDEN to approve incomplete coverage
+
+## CONTEXT BOUNDARIES:
+
+- Available context: Complete epic and story breakdown from previous steps
+- Focus: Final validation of requirements coverage and story readiness
+- Limits: Validation only, no new content creation
+- Dependencies: Completed story generation from Step 3
+
+## VALIDATION PROCESS:
+
+### 1. FR Coverage Validation
+
+Review the complete epic and story breakdown to ensure EVERY FR is covered:
+
+**CRITICAL CHECK:**
+
+- Go through each FR from the Requirements Inventory
+- Verify it appears in at least one story
+- Check that acceptance criteria fully address the FR
+- No FRs should be left uncovered
+
+### 2. Architecture Implementation Validation
+
+**Check for Starter Template Setup:**
+
+- Does Architecture document specify a starter template?
+- If YES: Epic 1 Story 1 must be "Set up initial project from starter template"
+- This includes cloning, installing dependencies, initial configuration
+
+**Database/Entity Creation Validation:**
+
+- Are database tables/entities created ONLY when needed by stories?
+- ❌ WRONG: Epic 1 creates all tables upfront
+- ✅ RIGHT: Tables created as part of the first story that needs them
+- Each story should create/modify ONLY what it needs
+
+### 3. Story Quality Validation
+
+**Each story must:**
+
+- Be completable by a single dev agent
+- Have clear acceptance criteria
+- Reference specific FRs it implements
+- Include necessary technical details
+- **Not have forward dependencies** (can only depend on PREVIOUS stories)
+- Be implementable without waiting for future stories
+
+### 4. Epic Structure Validation
+
+**Check that:**
+
+- Epics deliver user value, not technical milestones
+- Dependencies flow naturally
+- Foundation stories only setup what's needed
+- No big upfront technical work
+- **File Churn Check:** Do multiple epics repeatedly modify the same core files?
+  - Assess whether the overlap pattern suggests unnecessary churn or is incidental
+  - If overlap is significant: Validate that splitting provides genuine value (risk mitigation, feedback loops, context size limits)
+  - If no justification for the split: Recommend consolidation into fewer epics
+  - ❌ WRONG: Multiple epics each modify the same core files with no feedback loop between them
+  - ✅ RIGHT: Epics target distinct files/components, OR consolidation was explicitly considered and rejected with rationale
+
+### 5. Dependency Validation (CRITICAL)
+
+**Epic Independence Check:**
+
+- Does each epic deliver COMPLETE functionality for its domain?
+- Can Epic 2 function without Epic 3 being implemented?
+- Can Epic 3 function standalone using Epic 1 & 2 outputs?
+- ❌ WRONG: Epic 2 requires Epic 3 features to work
+- ✅ RIGHT: Each epic is independently valuable
+
+**Within-Epic Story Dependency Check:**
+For each epic, review stories in order:
+
+- Can Story N.1 be completed without Stories N.2, N.3, etc.?
+- Can Story N.2 be completed using only Story N.1 output?
+- Can Story N.3 be completed using only Stories N.1 & N.2 outputs?
+- ❌ WRONG: "This story depends on a future story"
+- ❌ WRONG: Story references features not yet implemented
+- ✅ RIGHT: Each story builds only on previous stories
+
+### 6. Complete and Save
+
+If all validations pass:
+
+- Update any remaining placeholders in the document
+- Ensure proper formatting
+- Save the final epics.md
+
+**Present Final Menu:**
+**All validations complete!** [C] Complete Workflow
+
+HALT — wait for user input before proceeding.
+
+When C is selected, the workflow is complete and the epics.md is ready for development.
+
+Epics and Stories complete. Invoke the `bmad-help` skill.
+
+Upon Completion of task output: offer to answer any questions about the Epics and Stories.
+
+## On Complete
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow.on_complete`
+
+If the resolved `workflow.on_complete` is non-empty, follow it as the final terminal instruction before exiting.
diff --git a/.agents/skills/bmad-create-epics-and-stories/templates/epics-template.md b/.agents/skills/bmad-create-epics-and-stories/templates/epics-template.md
new file mode 100644
index 00000000..bf80c7fb
--- /dev/null
+++ b/.agents/skills/bmad-create-epics-and-stories/templates/epics-template.md
@@ -0,0 +1,61 @@
+---
+stepsCompleted: []
+inputDocuments: []
+---
+
+# {{project_name}} - Epic Breakdown
+
+## Overview
+
+This document provides the complete epic and story breakdown for {{project_name}}, decomposing the requirements from the PRD, UX Design if it exists, and Architecture requirements into implementable stories.
+
+## Requirements Inventory
+
+### Functional Requirements
+
+{{fr_list}}
+
+### NonFunctional Requirements
+
+{{nfr_list}}
+
+### Additional Requirements
+
+{{additional_requirements}}
+
+### UX Design Requirements
+
+{{ux_design_requirements}}
+
+### FR Coverage Map
+
+{{requirements_coverage_map}}
+
+## Epic List
+
+{{epics_list}}
+
+<!-- Repeat for each epic in epics_list (N = 1, 2, 3...) -->
+
+## Epic {{N}}: {{epic_title_N}}
+
+{{epic_goal_N}}
+
+<!-- Repeat for each story (M = 1, 2, 3...) within epic N -->
+
+### Story {{N}}.{{M}}: {{story_title_N_M}}
+
+As a {{user_type}},
+I want {{capability}},
+So that {{value_benefit}}.
+
+**Acceptance Criteria:**
+
+<!-- for each AC on this story -->
+
+**Given** {{precondition}}
+**When** {{action}}
+**Then** {{expected_outcome}}
+**And** {{additional_criteria}}
+
+<!-- End story repeat -->
diff --git a/.agents/skills/bmad-create-prd/SKILL.md b/.agents/skills/bmad-create-prd/SKILL.md
new file mode 100644
index 00000000..5205fa61
--- /dev/null
+++ b/.agents/skills/bmad-create-prd/SKILL.md
@@ -0,0 +1,30 @@
+---
+name: bmad-create-prd
+description: 'Deprecated — forwards to bmad-prd (create intent).'
+---
+
+# DEPRECATED — forwards to bmad-prd (create intent)
+
+This skill was consolidated into `bmad-prd`. It is retained as a thin compatibility shim so existing invocations by name and `_bmad/custom/bmad-create-prd.toml` override files keep working. New work should invoke `bmad-prd` directly — it detects create / update / validate intent from the conversation.
+
+## On Activation
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. This picks up any `{project-root}/_bmad/custom/bmad-create-prd.toml` and `bmad-create-prd.user.toml` overrides for the legacy fields (`activation_steps_prepend`, `activation_steps_append`, `persistent_facts`, `on_complete`).
+
+2. Load `{project-root}/_bmad/bmm/config.yaml` (and `config.user.yaml` if present) to resolve `{user_name}` and `{communication_language}`.
+
+3. Emit a deprecation notice to the user in `{communication_language}`:
+
+   > Notice: `bmad-create-prd` is deprecated and will be removed in a future release. It now forwards to `bmad-prd` with create intent. To silence this notice and access the full new customization surface (`prd_template`, `validation_checklist_template`, `validation_report_template`, `doc_standards`, `finalize_reviewers`, `external_sources`, `external_handoffs`, `prd_output_path`, `run_folder_pattern`), migrate `_bmad/custom/bmad-create-prd.toml` to `_bmad/custom/bmad-prd.toml` and invoke `bmad-prd` directly next time. Customization fields that were in this version still remain in the new version and will be respected if present in `_bmad/custom/bmad-prd.toml`, but the new version also supports additional fields that you can take advantage of by migrating.
+
+4. Invoke `bmad-prd` with the following context. Pass these as the activating context so `bmad-prd` honors them instead of resolving its own customization from scratch:
+
+   - **Intent:** `create` — skip `bmad-prd`'s usual intent detection step.
+   - **Pre-resolved legacy customization** — use these in place of resolving from `bmad-prd`'s own `customize.toml` for the four legacy fields. For everything else (`prd_template`, `validation_checklist_template`, `validation_report_template`, `doc_standards`, `finalize_reviewers`, `prd_output_path`, `run_folder_pattern`, `external_sources`, `external_handoffs`), use `bmad-prd`'s own defaults and overrides as normal:
+     - `activation_steps_prepend` = the resolved value from step 1
+     - `activation_steps_append` = the resolved value from step 1
+     - `persistent_facts` = the resolved value from step 1
+     - `on_complete` = the resolved value from step 1
+   - **Original user input:** forward whatever the user said when invoking this skill verbatim.
+
+   `bmad-prd` takes the workflow from here. Do not execute any further steps in this shim.
diff --git a/.agents/skills/bmad-create-prd/customize.toml b/.agents/skills/bmad-create-prd/customize.toml
new file mode 100644
index 00000000..fde1ba1b
--- /dev/null
+++ b/.agents/skills/bmad-create-prd/customize.toml
@@ -0,0 +1,41 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-create-prd. Mirrors the
+# agent customization shape under the [workflow] namespace.
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+# Steps to run before the standard activation (config load, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Distinct from the runtime memory sidecar — these are static context
+# loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "All PRDs must include a regulatory-risk section."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Scalar: executed when the workflow reaches Step 12 (Workflow Completion),
+# after the PRD is finalized and workflow status is updated. Override wins.
+# Leave empty for no custom post-completion behavior.
+
+on_complete = ""
diff --git a/.agents/skills/bmad-create-story/SKILL.md b/.agents/skills/bmad-create-story/SKILL.md
new file mode 100644
index 00000000..f5e17226
--- /dev/null
+++ b/.agents/skills/bmad-create-story/SKILL.md
@@ -0,0 +1,434 @@
+---
+name: bmad-create-story
+description: "Deprecated: `bmad-build` is now the official implementation method. Only use this when explicitly invoked by name."
+---
+
+# Create Story Workflow
+
+**Goal:** Create a comprehensive story file that gives the dev agent everything needed for flawless implementation.
+
+**Your Role:** Story context engine that prevents LLM developer mistakes, omissions, or disasters.
+- Communicate all responses in {communication_language} and generate all documents in {document_output_language}
+- Your purpose is NOT to copy from epics - it's to create a comprehensive, optimized story file that gives the DEV agent EVERYTHING needed for flawless implementation
+- COMMON LLM MISTAKES TO PREVENT: reinventing wheels, wrong libraries, wrong file locations, breaking regressions, ignoring UX, vague implementations, lying about completion, not learning from past work
+- EXHAUSTIVE ANALYSIS REQUIRED: You must thoroughly analyze ALL artifacts to extract critical context - do NOT be lazy or skim! This is the most important function in the entire development process!
+- UTILIZE SUBPROCESSES AND SUBAGENTS: Use research subagents, subprocesses or parallel processing if available to thoroughly analyze different artifacts simultaneously and thoroughly
+- SAVE QUESTIONS: If you think of questions or clarifications during analysis, save them for the end after the complete story is written
+- ZERO USER INTERVENTION: Process should be fully automated except for initial epic/story selection or missing documents
+
+Subagents, when the capability is available, are an important part of this workflow. Use them as directed by the workflow steps.
+If you need an explicit user instruction to run them, ask once now for the whole workflow run.
+
+## Conventions
+
+- Bare paths (e.g. `discover-inputs.md`) resolve from the skill root.
+- `{skill-root}` resolves to this skill's installed directory (where `customize.toml` lives).
+- `{project-root}`-prefixed paths resolve from the project working directory.
+- `{skill-name}` resolves to the skill directory's basename.
+
+## On Activation
+
+### Step 1: Resolve the Workflow Block
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`
+
+**If the script fails**, resolve the `workflow` block yourself by reading these three files in base → team → user order and applying the same structural merge rules as the resolver:
+
+1. `{skill-root}/customize.toml` — defaults
+2. `{project-root}/_bmad/custom/{skill-name}.toml` — team overrides
+3. `{project-root}/_bmad/custom/{skill-name}.user.toml` — personal overrides
+
+Any missing file is skipped. Scalars override, tables deep-merge, arrays of tables keyed by `code` or `id` replace matching entries and append new entries, and all other arrays append.
+
+### Step 2: Execute Prepend Steps
+
+Execute each entry in `{workflow.activation_steps_prepend}` in order before proceeding.
+
+### Step 3: Load Persistent Facts
+
+Treat every entry in `{workflow.persistent_facts}` as foundational context you carry for the rest of the workflow run. Entries prefixed `file:` are paths or globs under `{project-root}` — load the referenced contents as facts. All other entries are facts verbatim.
+
+### Step 4: Load Config
+
+Load config from `{project-root}/_bmad/bmm/config.yaml` and resolve:
+
+- `project_name`, `user_name`
+- `communication_language`, `document_output_language`
+- `user_skill_level`
+- `planning_artifacts`, `implementation_artifacts`
+- `date` as system-generated current datetime
+
+### Step 5: Greet the User
+
+Greet `{user_name}`, speaking in `{communication_language}`.
+
+<output>Deprecated: `bmad-build` is now the official implementation method. Only use this when explicitly invoked by name.</output>
+
+### Step 6: Execute Append Steps
+
+Execute each entry in `{workflow.activation_steps_append}` in order.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+## Paths
+
+- `sprint_status` = `{implementation_artifacts}/sprint-status.yaml`
+- `epics_file` = `{planning_artifacts}/epics.md`
+- `prd_file` = `{planning_artifacts}/prd.md`
+- `architecture_file` = `{planning_artifacts}/architecture.md`
+- `ux_file` = `{planning_artifacts}/*ux*.md`
+- `story_title` = "" (will be elicited if not derivable)
+- `default_output_file` = `{implementation_artifacts}/{{story_key}}.md`
+
+## Input Files
+
+| Input | Description | Path Pattern(s) | Load Strategy |
+|-------|-------------|------------------|---------------|
+| prd | PRD (fallback - epics file should have most content) | whole: `{planning_artifacts}/*prd*.md`, sharded: `{planning_artifacts}/*prd*/*.md` | SELECTIVE_LOAD |
+| architecture | Architecture (fallback - epics file should have relevant sections) | whole: `{planning_artifacts}/*architecture*.md`, sharded: `{planning_artifacts}/*architecture*/*.md` | SELECTIVE_LOAD |
+| ux | UX design (fallback - epics file should have relevant sections) | whole: `{planning_artifacts}/*ux*.md`, sharded: `{planning_artifacts}/*ux*/*.md` | SELECTIVE_LOAD |
+| epics | Enhanced epics+stories file with BDD and source hints | whole: `{planning_artifacts}/*epic*.md`, sharded: `{planning_artifacts}/*epic*/*.md` | SELECTIVE_LOAD |
+
+## Execution
+
+<workflow>
+
+<step n="1" goal="Determine target story">
+  <check if="{{story_path}} is provided by user or user provided the epic and story number such as 2-4 or 1.6 or epic 1 story 5">
+    <action>Parse user-provided story path: extract epic_num, story_num, story_title from format like "1-2-user-auth"</action>
+    <action>Set {{epic_num}}, {{story_num}}, {{story_key}} from user input</action>
+    <action>GOTO step 2a</action>
+  </check>
+
+  <action>Check if {{sprint_status}} file exists for auto discover</action>
+  <check if="sprint status file does NOT exist">
+    <output>🚫 No sprint status file found and no story specified</output>
+    <output>
+      **Required Options:**
+      1. Run `sprint-planning` to initialize sprint tracking (recommended)
+      2. Provide specific epic-story number to create (e.g., "1-2-user-auth")
+      3. Provide path to story documents if sprint status doesn't exist yet
+    </output>
+    <ask>Choose option [1], provide epic-story number, path to story docs, or [q] to quit:</ask>
+
+    <check if="user chooses 'q'">
+      <action>HALT - No work needed</action>
+    </check>
+
+    <check if="user chooses '1'">
+      <output>Run sprint-planning workflow first to create sprint-status.yaml</output>
+      <action>HALT - User needs to run sprint-planning</action>
+    </check>
+
+    <check if="user provides epic-story number">
+      <action>Parse user input: extract epic_num, story_num, story_title</action>
+      <action>Set {{epic_num}}, {{story_num}}, {{story_key}} from user input</action>
+      <action>GOTO step 2a</action>
+    </check>
+
+    <check if="user provides story docs path">
+      <action>Use user-provided path for story documents</action>
+      <action>GOTO step 2a</action>
+    </check>
+  </check>
+
+  <!-- Auto-discover from sprint status only if no user input -->
+  <check if="no user input provided">
+    <critical>MUST read COMPLETE {sprint_status} file from start to end to preserve order</critical>
+    <action>Load the FULL file: {{sprint_status}}</action>
+    <action>Read ALL lines from beginning to end - do not skip any content</action>
+    <action>Parse the development_status section completely</action>
+
+    <action>Find the FIRST story (by reading in order from top to bottom) where:
+      - Key matches pattern: number-number-name (e.g., "1-2-user-auth")
+      - NOT an epic key (epic-X) or retrospective (epic-X-retrospective)
+      - Status value equals "backlog"
+    </action>
+
+    <check if="no backlog story found">
+      <output>📋 No backlog stories found in sprint-status.yaml
+
+        All stories are either already created, in progress, or done.
+
+        **Options:**
+        1. Run sprint-planning to refresh story tracking
+        2. Load PM agent and run correct-course to add more stories
+        3. Check if current sprint is complete and run retrospective
+      </output>
+      <action>HALT</action>
+    </check>
+
+    <action>Extract from found story key (e.g., "1-2-user-authentication"):
+      - epic_num: first number before dash (e.g., "1")
+      - story_num: second number after first dash (e.g., "2")
+      - story_title: remainder after second dash (e.g., "user-authentication")
+    </action>
+    <action>Set {{story_id}} = "{{epic_num}}.{{story_num}}"</action>
+    <action>Store story_key for later use (e.g., "1-2-user-authentication")</action>
+
+    <!-- Mark epic as in-progress if this is first story -->
+    <action>Check if this is the first story in epic {{epic_num}} by looking for {{epic_num}}-1-* pattern</action>
+    <check if="this is first story in epic {{epic_num}}">
+      <action>Load {{sprint_status}} and check epic-{{epic_num}} status</action>
+      <action>If epic status is "backlog" → update to "in-progress"</action>
+      <action>If epic status is "contexted" (legacy status) → update to "in-progress" (backward compatibility)</action>
+      <action>If epic status is "in-progress" → no change needed</action>
+      <check if="epic status is 'done'">
+        <output>🚫 ERROR: Cannot create story in completed epic</output>
+        <output>Epic {{epic_num}} is marked as 'done'. All stories are complete.</output>
+        <output>If you need to add more work, either:</output>
+        <output>1. Manually change epic status back to 'in-progress' in sprint-status.yaml</output>
+        <output>2. Create a new epic for additional work</output>
+        <action>HALT - Cannot proceed</action>
+      </check>
+      <check if="epic status is not one of: backlog, contexted, in-progress, done">
+        <output>🚫 ERROR: Invalid epic status '{{epic_status}}'</output>
+        <output>Epic {{epic_num}} has invalid status. Expected: backlog, in-progress, or done</output>
+        <output>Please fix sprint-status.yaml manually or run sprint-planning to regenerate</output>
+        <action>HALT - Cannot proceed</action>
+      </check>
+      <output>📊 Epic {{epic_num}} status updated to in-progress</output>
+    </check>
+
+    <action>GOTO step 2a</action>
+  </check>
+  <action>Load the FULL file: {{sprint_status}}</action>
+  <action>Read ALL lines from beginning to end - do not skip any content</action>
+  <action>Parse the development_status section completely</action>
+
+  <action>Find the FIRST story (by reading in order from top to bottom) where:
+    - Key matches pattern: number-number-name (e.g., "1-2-user-auth")
+    - NOT an epic key (epic-X) or retrospective (epic-X-retrospective)
+    - Status value equals "backlog"
+  </action>
+
+  <check if="no backlog story found">
+    <output>No backlog stories found in sprint-status.yaml
+
+      All stories are either already created, in progress, or done.
+
+      **Options:**
+      1. Run sprint-planning to refresh story tracking
+      2. Load PM agent and run correct-course to add more stories
+      3. Check if current sprint is complete and run retrospective
+    </output>
+    <action>HALT</action>
+  </check>
+
+  <action>Extract from found story key (e.g., "1-2-user-authentication"):
+    - epic_num: first number before dash (e.g., "1")
+    - story_num: second number after first dash (e.g., "2")
+    - story_title: remainder after second dash (e.g., "user-authentication")
+  </action>
+  <action>Set {{story_id}} = "{{epic_num}}.{{story_num}}"</action>
+  <action>Store story_key for later use (e.g., "1-2-user-authentication")</action>
+
+  <!-- Mark epic as in-progress if this is first story -->
+  <action>Check if this is the first story in epic {{epic_num}} by looking for {{epic_num}}-1-* pattern</action>
+  <check if="this is first story in epic {{epic_num}}">
+    <action>Load {{sprint_status}} and check epic-{{epic_num}} status</action>
+    <action>If epic status is "backlog" → update to "in-progress"</action>
+    <action>If epic status is "contexted" (legacy status) → update to "in-progress" (backward compatibility)</action>
+    <action>If epic status is "in-progress" → no change needed</action>
+    <check if="epic status is 'done'">
+      <output>ERROR: Cannot create story in completed epic</output>
+      <output>Epic {{epic_num}} is marked as 'done'. All stories are complete.</output>
+      <output>If you need to add more work, either:</output>
+      <output>1. Manually change epic status back to 'in-progress' in sprint-status.yaml</output>
+      <output>2. Create a new epic for additional work</output>
+      <action>HALT - Cannot proceed</action>
+    </check>
+    <check if="epic status is not one of: backlog, contexted, in-progress, done">
+      <output>ERROR: Invalid epic status '{{epic_status}}'</output>
+      <output>Epic {{epic_num}} has invalid status. Expected: backlog, in-progress, or done</output>
+      <output>Please fix sprint-status.yaml manually or run sprint-planning to regenerate</output>
+      <action>HALT - Cannot proceed</action>
+    </check>
+    <output>Epic {{epic_num}} status updated to in-progress</output>
+  </check>
+
+  <action>GOTO step 2a</action>
+</step>
+
+<step n="2" goal="Load and analyze core artifacts">
+  <critical>🔬 EXHAUSTIVE ARTIFACT ANALYSIS - This is where you prevent future developer mistakes!</critical>
+
+  <!-- Load all available content through discovery protocol -->
+  <action>Read fully and follow `./discover-inputs.md` to load all input files</action>
+  <note>Available content: {epics_content}, {prd_content}, {architecture_content}, {ux_content}, plus the project-context facts loaded during activation via `persistent_facts`.</note>
+
+  <!-- Analyze epics file for story foundation -->
+  <action>From {epics_content}, extract Epic {{epic_num}} complete context:</action> **EPIC ANALYSIS:** - Epic
+  objectives and business value - ALL stories in this epic for cross-story context - Our specific story's requirements, user story
+  statement, acceptance criteria - Technical requirements and constraints - Dependencies on other stories/epics - Source hints pointing to
+  original documents <!-- Extract specific story requirements -->
+  <action>Extract our story ({{epic_num}}-{{story_num}}) details:</action> **STORY FOUNDATION:** - User story statement
+  (As a, I want, so that) - Detailed acceptance criteria (already BDD formatted) - Technical requirements specific to this story -
+  Business context and value - Success criteria <!-- Previous story analysis for context continuity -->
+  <check if="story_num > 1">
+    <action>Find {{previous_story_num}}: scan {implementation_artifacts} for the story file in epic {{epic_num}} with the highest story number less than {{story_num}}</action>
+    <action>Load previous story file: {implementation_artifacts}/{{epic_num}}-{{previous_story_num}}-*.md</action> **PREVIOUS STORY INTELLIGENCE:** -
+  Dev notes and learnings from previous story - Review feedback and corrections needed - Files that were created/modified and their
+  patterns - Testing approaches that worked/didn't work - Problems encountered and solutions found - Code patterns established <action>Extract
+  all learnings that could impact current story implementation</action>
+  </check>
+
+  <!-- Git intelligence for previous work patterns -->
+  <check
+    if="previous story exists AND git repository detected">
+    <action>Get last 5 commit titles to understand recent work patterns</action>
+    <action>Analyze 1-5 most recent commits for relevance to current story:
+      - Files created/modified
+      - Code patterns and conventions used
+      - Library dependencies added/changed
+      - Architecture decisions implemented
+      - Testing approaches used
+    </action>
+    <action>Extract actionable insights for current story implementation</action>
+  </check>
+</step>
+
+<step n="3" goal="Architecture analysis for developer guardrails">
+  <critical>🏗️ ARCHITECTURE INTELLIGENCE - Extract everything the developer MUST follow!</critical> **ARCHITECTURE DOCUMENT ANALYSIS:** <action>Systematically
+  analyze architecture content for story-relevant requirements:</action>
+
+  <!-- Load architecture - single file or sharded -->
+  <check if="architecture file is single file">
+    <action>Load complete {architecture_content}</action>
+  </check>
+  <check if="architecture is sharded to folder">
+    <action>Load architecture index and scan all architecture files</action>
+  </check> **CRITICAL ARCHITECTURE EXTRACTION:** <action>For
+  each architecture section, determine if relevant to this story:</action> - **Technical Stack:** Languages, frameworks, libraries with
+  versions - **Code Structure:** Folder organization, naming conventions, file patterns - **API Patterns:** Service structure, endpoint
+  patterns, data contracts - **Database Schemas:** Tables, relationships, constraints relevant to story - **Security Requirements:**
+  Authentication patterns, authorization rules - **Performance Requirements:** Caching strategies, optimization patterns - **Testing
+  Standards:** Testing frameworks, coverage expectations, test patterns - **Deployment Patterns:** Environment configurations, build
+  processes - **Integration Patterns:** External service integrations, data flows <action>Extract any story-specific requirements that the
+  developer MUST follow</action>
+  <action>Identify any architectural decisions that override previous patterns</action>
+
+  <!-- Read existing code being modified — non-negotiable -->
+  <critical>📂 READ FILES BEING MODIFIED — skipping this is the primary cause of implementation failures and review cycles</critical>
+  <action>From the architecture directory structure, identify every file marked UPDATE (not NEW) that this story will touch</action>
+  <action>Read each relevant UPDATE file completely. For each one, document in dev notes:
+    - Current state: what it does today (state machine, API calls, data shapes, existing behaviors)
+    - What this story changes: the specific sections or behaviors being modified
+    - What must be preserved: existing interactions and behaviors the story must not break
+  </action>
+  <critical>A story implementation must leave the system working end-to-end — not just satisfy its stated ACs.
+  If a behavior is required for the feature to work correctly in the existing system, it is a requirement
+  whether or not it is explicitly written in the story. The dev agent owns this.</critical>
+</step>
+
+<step n="4" goal="Web research for latest technical specifics">
+  <critical>🌐 ENSURE LATEST TECH KNOWLEDGE - Prevent outdated implementations!</critical> **WEB INTELLIGENCE:** <action>Identify specific
+  technical areas that require latest version knowledge:</action>
+
+  <!-- Check for libraries/frameworks mentioned in architecture -->
+  <action>From architecture analysis, identify specific libraries, APIs, or
+  frameworks</action>
+  <action>For each critical technology, research latest stable version and key changes:
+    - Latest API documentation and breaking changes
+    - Security vulnerabilities or updates
+    - Performance improvements or deprecations
+    - Best practices for current version
+  </action>
+  **EXTERNAL CONTEXT INCLUSION:** <action>Include in story any critical latest information the developer needs:
+    - Specific library versions and why chosen
+    - API endpoints with parameters and authentication
+    - Recent security patches or considerations
+    - Performance optimization techniques
+    - Migration considerations if upgrading
+  </action>
+</step>
+
+<step n="5" goal="Create comprehensive story file">
+  <critical>📝 CREATE ULTIMATE STORY FILE - The developer's master implementation guide!</critical>
+
+  <action>Initialize from template.md:
+  {default_output_file}</action>
+  <template-output file="{default_output_file}">story_header</template-output>
+
+  <!-- Story foundation from epics analysis -->
+  <template-output
+    file="{default_output_file}">story_requirements</template-output>
+
+  <!-- Developer context section - MOST IMPORTANT PART -->
+  <template-output file="{default_output_file}">
+  developer_context_section</template-output> **DEV AGENT GUARDRAILS:** <template-output file="{default_output_file}">
+  technical_requirements</template-output>
+  <template-output file="{default_output_file}">architecture_compliance</template-output>
+  <template-output
+    file="{default_output_file}">library_framework_requirements</template-output>
+  <template-output file="{default_output_file}">
+  file_structure_requirements</template-output>
+  <template-output file="{default_output_file}">testing_requirements</template-output>
+
+  <!-- Previous story intelligence -->
+  <check
+    if="previous story learnings available">
+    <template-output file="{default_output_file}">previous_story_intelligence</template-output>
+  </check>
+
+  <!-- Git intelligence -->
+  <check
+    if="git analysis completed">
+    <template-output file="{default_output_file}">git_intelligence_summary</template-output>
+  </check>
+
+  <!-- Latest technical specifics -->
+  <check if="web research completed">
+    <template-output file="{default_output_file}">latest_tech_information</template-output>
+  </check>
+
+  <!-- Project context reference -->
+  <template-output
+    file="{default_output_file}">project_context_reference</template-output>
+
+  <!-- Final status update -->
+  <template-output file="{default_output_file}">
+  story_completion_status</template-output>
+
+  <!-- CRITICAL: Set status to ready-for-dev -->
+  <action>Set story Status to: "ready-for-dev"</action>
+  <action>Add completion note: "Ultimate
+  context engine analysis completed - comprehensive developer guide created"</action>
+</step>
+
+<step n="6" goal="Update sprint status and finalize">
+  <action>Validate the newly created story file {default_output_file} against `./checklist.md` and apply any required fixes before finalizing</action>
+  <action>Save story document unconditionally</action>
+
+  <!-- Update sprint status -->
+  <check if="sprint status file exists">
+    <action>Update {{sprint_status}}</action>
+    <action>Load the FULL file and read all development_status entries</action>
+    <action>Find development_status key matching {{story_key}}</action>
+    <action>Verify current status is "backlog" (expected previous state)</action>
+    <action>Update development_status[{{story_key}}] = "ready-for-dev"</action>
+    <action>Update last_updated field to current date</action>
+    <action>Save file, preserving ALL comments and structure including STATUS DEFINITIONS</action>
+  </check>
+
+  <action>Report completion</action>
+  <output>**🎯 ULTIMATE BMad Method STORY CONTEXT CREATED, {user_name}!**
+
+    **Story Details:**
+    - Story ID: {{story_id}}
+    - Story Key: {{story_key}}
+    - File: {{story_file}}
+    - Status: ready-for-dev
+
+    **Next Steps:**
+    1. Review the comprehensive story in {{story_file}}
+    2. Run dev agents `dev-story` for optimized implementation
+    3. Run `code-review` when complete (auto-marks done)
+    4. Optional: If Test Architect module installed, run `/bmad:tea:automate` after `dev-story` to generate guardrail tests
+
+    **The developer now has everything needed for flawless implementation!**
+  </output>
+  <action>Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow.on_complete` — if the resolved value is non-empty, follow it as the final terminal instruction before exiting.</action>
+</step>
+
+</workflow>
diff --git a/.agents/skills/bmad-create-story/checklist.md b/.agents/skills/bmad-create-story/checklist.md
new file mode 100644
index 00000000..e47cc0f4
--- /dev/null
+++ b/.agents/skills/bmad-create-story/checklist.md
@@ -0,0 +1,357 @@
+# 🎯 Story Context Quality Competition Prompt
+
+## **🔥 CRITICAL MISSION: Outperform and Fix the Original Create-Story LLM**
+
+You are an independent quality validator in a **FRESH CONTEXT**. Your mission is to **thoroughly review** a story file that was generated by the create-story workflow and **systematically identify any mistakes, omissions, or disasters** that the original LLM missed.
+
+**Your purpose is NOT just to validate - it's to FIX and PREVENT LLM developer mistakes, omissions, or disasters!**
+
+### **🚨 CRITICAL MISTAKES TO PREVENT:**
+
+- **Reinventing wheels** - Creating duplicate functionality instead of reusing existing
+- **Wrong libraries** - Using incorrect frameworks, versions, or dependencies
+- **Wrong file locations** - Violating project structure and organization
+- **Breaking regressions** - Implementing changes that break existing functionality
+- **Ignoring UX** - Not following user experience design requirements
+- **Vague implementations** - Creating unclear, ambiguous implementations
+- **Lying about completion** - Implementing incorrectly or incompletely
+- **Not learning from past work** - Ignoring previous story learnings and patterns
+
+### **🚨 EXHAUSTIVE ANALYSIS REQUIRED:**
+
+You must thoroughly analyze **ALL artifacts** to extract critical context - do NOT be lazy or skim! This is the most important quality control function in the entire development process!
+
+### **🔬 UTILIZE SUBPROCESSES AND SUBAGENTS:**
+
+Use research subagents, subprocesses, or parallel processing if available to thoroughly analyze different artifacts **simultaneously and thoroughly**. Leave no stone unturned!
+
+### **🎯 COMPETITIVE EXCELLENCE:**
+
+This is a COMPETITION to create the **ULTIMATE story context** that makes LLM developer mistakes **IMPOSSIBLE**!
+
+## **🚀 HOW TO USE THIS CHECKLIST**
+
+### **When Running from Create-Story Workflow:**
+
+- The workflow framework will automatically:
+  - Load this checklist file
+  - Load the newly created story file (`{story_file_path}`)
+  - Load workflow variables from `./workflow.md`
+  - Execute the validation process
+
+### **When Running in Fresh Context:**
+
+- User should provide the story file path being reviewed
+- Load the story file directly
+- Load the corresponding workflow.md for variable context
+- Proceed with systematic analysis
+
+### **Required Inputs:**
+
+- **Story file**: The story file to review and improve
+- **Workflow variables**: From workflow.md (implementation_artifacts, epics_file, etc.)
+- **Source documents**: Epics, architecture, etc. (discovered or provided)
+- **Validation framework**: The workflow's checklist execution system
+
+---
+
+## **🔬 SYSTEMATIC RE-ANALYSIS APPROACH**
+
+You will systematically re-do the entire story creation process, but with a critical eye for what the original LLM might have missed:
+
+### **Step 1: Load and Understand the Target**
+
+1. **Load the workflow configuration**: `./workflow.md` for variable inclusion
+2. **Load the story file**: `{story_file_path}` (provided by user or discovered)
+3. **Extract metadata**: epic_num, story_num, story_key, story_title from story file
+4. **Resolve all workflow variables**: implementation_artifacts, epics_file, architecture_file, etc.
+5. **Understand current status**: What story implementation guidance is currently provided?
+
+**Note:** If running in fresh context, user should provide the story file path being reviewed. If running from create-story workflow, the validation framework will automatically discover the checklist and story file.
+
+### **Step 2: Exhaustive Source Document Analysis**
+
+**🔥 CRITICAL: Treat this like YOU are creating the story from scratch to PREVENT DISASTERS!**
+**Discover everything the original LLM missed that could cause developer mistakes, omissions, or disasters!**
+
+#### **2.1 Epics and Stories Analysis**
+
+- Load `{epics_file}` (or sharded equivalents)
+- Extract **COMPLETE Epic {{epic_num}} context**:
+  - Epic objectives and business value
+  - ALL stories in this epic (for cross-story context)
+  - Our specific story's requirements, acceptance criteria
+  - Technical requirements and constraints
+  - Cross-story dependencies and prerequisites
+
+#### **2.2 Architecture Deep-Dive**
+
+- Load `{architecture_file}` (single or sharded)
+- **Systematically scan for ANYTHING relevant to this story:**
+  - Technical stack with versions (languages, frameworks, libraries)
+  - Code structure and organization patterns
+  - API design patterns and contracts
+  - Database schemas and relationships
+  - Security requirements and patterns
+  - Performance requirements and optimization strategies
+  - Testing standards and frameworks
+  - Deployment and environment patterns
+  - Integration patterns and external services
+
+#### **2.3 Previous Story Intelligence (if applicable)**
+
+- If `story_num > 1`, load the previous story file
+- Extract **actionable intelligence**:
+  - Dev notes and learnings
+  - Review feedback and corrections needed
+  - Files created/modified and their patterns
+  - Testing approaches that worked/didn't work
+  - Problems encountered and solutions found
+  - Code patterns and conventions established
+
+#### **2.4 Git History Analysis (if available)**
+
+- Analyze recent commits for patterns:
+  - Files created/modified in previous work
+  - Code patterns and conventions used
+  - Library dependencies added/changed
+  - Architecture decisions implemented
+  - Testing approaches used
+
+#### **2.5 Latest Technical Research**
+
+- Identify any libraries/frameworks mentioned
+- Research latest versions and critical information:
+  - Breaking changes or security updates
+  - Performance improvements or deprecations
+  - Best practices for current versions
+
+### **Step 3: Disaster Prevention Gap Analysis**
+
+**🚨 CRITICAL: Identify every mistake the original LLM missed that could cause DISASTERS!**
+
+#### **3.1 Reinvention Prevention Gaps**
+
+- **Wheel reinvention:** Areas where developer might create duplicate functionality
+- **Code reuse opportunities** not identified that could prevent redundant work
+- **Existing solutions** not mentioned that developer should extend instead of replace
+
+#### **3.2 Technical Specification DISASTERS**
+
+- **Wrong libraries/frameworks:** Missing version requirements that could cause compatibility issues
+- **API contract violations:** Missing endpoint specifications that could break integrations
+- **Database schema conflicts:** Missing requirements that could corrupt data
+- **Security vulnerabilities:** Missing security requirements that could expose the system
+- **Performance disasters:** Missing requirements that could cause system failures
+
+#### **3.3 File Structure DISASTERS**
+
+- **Wrong file locations:** Missing organization requirements that could break build processes
+- **Coding standard violations:** Missing conventions that could create inconsistent codebase
+- **Integration pattern breaks:** Missing data flow requirements that could cause system failures
+- **Deployment failures:** Missing environment requirements that could prevent deployment
+
+#### **3.4 Regression DISASTERS**
+
+- **Breaking changes:** Missing requirements that could break existing functionality
+- **Test failures:** Missing test requirements that could allow bugs to reach production
+- **UX violations:** Missing user experience requirements that could ruin the product
+- **Learning failures:** Missing previous story context that could repeat same mistakes
+
+#### **3.5 Implementation DISASTERS**
+
+- **Vague implementations:** Missing details that could lead to incorrect or incomplete work
+- **Completion lies:** Missing acceptance criteria that could allow fake implementations
+- **Scope creep:** Missing boundaries that could cause unnecessary work
+- **Quality failures:** Missing quality requirements that could deliver broken features
+
+### **Step 4: LLM-Dev-Agent Optimization Analysis**
+
+**CRITICAL STEP: Optimize story context for LLM developer agent consumption**
+
+**Analyze current story for LLM optimization issues:**
+
+- **Verbosity problems:** Excessive detail that wastes tokens without adding value
+- **Ambiguity issues:** Vague instructions that could lead to multiple interpretations
+- **Context overload:** Too much information not directly relevant to implementation
+- **Missing critical signals:** Key requirements buried in verbose text
+- **Poor structure:** Information not organized for efficient LLM processing
+
+**Apply LLM Optimization Principles:**
+
+- **Clarity over verbosity:** Be precise and direct, eliminate fluff
+- **Actionable instructions:** Every sentence should guide implementation
+- **Scannable structure:** Use clear headings, bullet points, and emphasis
+- **Token efficiency:** Pack maximum information into minimum text
+- **Unambiguous language:** Clear requirements with no room for interpretation
+
+### **Step 5: Improvement Recommendations**
+
+**For each gap identified, provide specific, actionable improvements:**
+
+#### **5.1 Critical Misses (Must Fix)**
+
+- Missing essential technical requirements
+- Missing previous story context that could cause errors
+- Missing anti-pattern prevention that could lead to duplicate code
+- Missing security or performance requirements
+
+#### **5.2 Enhancement Opportunities (Should Add)**
+
+- Additional architectural guidance that would help developer
+- More detailed technical specifications
+- Better code reuse opportunities
+- Enhanced testing guidance
+
+#### **5.3 Optimization Suggestions (Nice to Have)**
+
+- Performance optimization hints
+- Additional context for complex scenarios
+- Enhanced debugging or development tips
+
+#### **5.4 LLM Optimization Improvements**
+
+- Token-efficient phrasing of existing content
+- Clearer structure for LLM processing
+- More actionable and direct instructions
+- Reduced verbosity while maintaining completeness
+
+---
+
+## **🎯 COMPETITION SUCCESS METRICS**
+
+**You WIN against the original LLM if you identify:**
+
+### **Category 1: Critical Misses (Blockers)**
+
+- Essential technical requirements the developer needs but aren't provided
+- Previous story learnings that would prevent errors if ignored
+- Anti-pattern prevention that would prevent code duplication
+- Security or performance requirements that must be followed
+
+### **Category 2: Enhancement Opportunities**
+
+- Architecture guidance that would significantly help implementation
+- Technical specifications that would prevent wrong approaches
+- Code reuse opportunities the developer should know about
+- Testing guidance that would improve quality
+
+### **Category 3: Optimization Insights**
+
+- Performance or efficiency improvements
+- Development workflow optimizations
+- Additional context for complex scenarios
+
+---
+
+## **📋 INTERACTIVE IMPROVEMENT PROCESS**
+
+After completing your systematic analysis, present your findings to the user interactively:
+
+### **Step 5: Present Improvement Suggestions**
+
+```
+🎯 **STORY CONTEXT QUALITY REVIEW COMPLETE**
+
+**Story:** {{story_key}} - {{story_title}}
+
+I found {{critical_count}} critical issues, {{enhancement_count}} enhancements, and {{optimization_count}} optimizations.
+
+## **🚨 CRITICAL ISSUES (Must Fix)**
+
+{{list each critical issue with clear, actionable description}}
+
+## **⚡ ENHANCEMENT OPPORTUNITIES (Should Add)**
+
+{{list each enhancement with clear benefit description}}
+
+## **✨ OPTIMIZATIONS (Nice to Have)**
+
+{{list each optimization with benefit description}}
+
+## **🤖 LLM OPTIMIZATION (Token Efficiency & Clarity)**
+
+{{list each LLM optimization that will improve dev agent performance:
+- Reduce verbosity while maintaining completeness
+- Improve structure for better LLM processing
+- Make instructions more actionable and direct
+- Enhance clarity and reduce ambiguity}}
+```
+
+### **Step 6: Interactive User Selection**
+
+After presenting the suggestions, ask the user:
+
+```
+**IMPROVEMENT OPTIONS:**
+
+Which improvements would you like me to apply to the story?
+
+**Select from the numbered list above, or choose:**
+- **all** - Apply all suggested improvements
+- **critical** - Apply only critical issues
+- **select** - I'll choose specific numbers
+- **none** - Keep story as-is
+- **details** - Show me more details about any suggestion
+
+Your choice:
+```
+
+### **Step 7: Apply Selected Improvements**
+
+When user accepts improvements:
+
+- **Load the story file**
+- **Apply accepted changes** (make them look natural, as if they were always there)
+- **DO NOT reference** the review process, original LLM, or that changes were "added" or "enhanced"
+- **Ensure clean, coherent final story** that reads as if it was created perfectly the first time
+
+### **Step 8: Confirmation**
+
+After applying changes:
+
+```
+✅ **STORY IMPROVEMENTS APPLIED**
+
+Updated {{count}} sections in the story file.
+
+The story now includes comprehensive developer guidance to prevent common implementation issues and ensure flawless execution.
+
+**Next Steps:**
+1. Review the updated story
+2. Run `dev-story` for implementation
+```
+
+---
+
+## **💪 COMPETITIVE EXCELLENCE MINDSET**
+
+**Your goal:** Improve the story file with dev agent needed context that makes flawless implementation inevitable while being optimized for LLM developer agent consumption. Remember the dev agent will ONLY have this file to use.
+
+**Success Criteria:** The LLM developer agent that processes your improved story will have:
+
+- ✅ Clear technical requirements they must follow
+- ✅ Previous work context they can build upon
+- ✅ Anti-pattern prevention to avoid common mistakes
+- ✅ Comprehensive guidance for efficient implementation
+- ✅ **Optimized content structure** for maximum clarity and minimum token waste
+- ✅ **Actionable instructions** with no ambiguity or verbosity
+- ✅ **Efficient information density** - maximum guidance in minimum text
+
+**Every improvement should make it IMPOSSIBLE for the developer to:**
+
+- Reinvent existing solutions
+- Use wrong approaches or libraries
+- Create duplicate functionality
+- Miss critical requirements
+- Make implementation errors
+
+**LLM Optimization Should Make it IMPOSSIBLE for the developer agent to:**
+
+- Misinterpret requirements due to ambiguity
+- Waste tokens on verbose, non-actionable content
+- Struggle to find critical information buried in text
+- Get confused by poor structure or organization
+- Miss key implementation signals due to inefficient communication
+
+**Go create the ultimate developer implementation guide! 🚀**
diff --git a/.agents/skills/bmad-create-story/customize.toml b/.agents/skills/bmad-create-story/customize.toml
new file mode 100644
index 00000000..fbd4a789
--- /dev/null
+++ b/.agents/skills/bmad-create-story/customize.toml
@@ -0,0 +1,41 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-create-story. Mirrors the
+# agent customization shape under the [workflow] namespace.
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+# Steps to run before the standard activation (config load, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Distinct from the runtime memory sidecar — these are static context
+# loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "All stories must include testable acceptance criteria."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Scalar: executed when the workflow reaches Step 6 (Update sprint status and finalize),
+# after the story file is saved and sprint-status.yaml is updated. Override wins.
+# Leave empty for no custom post-completion behavior.
+
+on_complete = ""
diff --git a/.agents/skills/bmad-create-story/discover-inputs.md b/.agents/skills/bmad-create-story/discover-inputs.md
new file mode 100644
index 00000000..2c313db3
--- /dev/null
+++ b/.agents/skills/bmad-create-story/discover-inputs.md
@@ -0,0 +1,88 @@
+# Discover Inputs Protocol
+
+**Objective:** Intelligently load project files (whole or sharded) based on the workflow's Input Files configuration.
+
+**Prerequisite:** Only execute this protocol if the workflow defines an Input Files section. If no input file patterns are configured, skip this entirely.
+
+---
+
+## Step 1: Parse Input File Patterns
+
+- Read the Input Files table from the workflow configuration.
+- For each input group (prd, architecture, epics, ux, etc.), note the **load strategy** if specified.
+
+## Step 2: Load Files Using Smart Strategies
+
+For each pattern in the Input Files table, work through the following substeps in order:
+
+### 2a: Try Sharded Documents First
+
+If a sharded pattern exists for this input, determine the load strategy (defaults to **FULL_LOAD** if not specified), then apply the matching strategy:
+
+#### FULL_LOAD Strategy
+
+Load ALL files in the sharded directory. Use this for PRD, Architecture, UX, brownfield docs, or whenever the full picture is needed.
+
+1. Use the glob pattern to find ALL `.md` files (e.g., `{planning_artifacts}/*architecture*/*.md`).
+2. Load EVERY matching file completely.
+3. Concatenate content in logical order: `index.md` first if it exists, then alphabetical.
+4. Store the combined result in a variable named `{pattern_name_content}` (e.g., `{architecture_content}`).
+
+#### SELECTIVE_LOAD Strategy
+
+Load a specific shard using a template variable. Example: used for epics with `{{epic_num}}`.
+
+1. Check for template variables in the sharded pattern (e.g., `{{epic_num}}`).
+2. If the variable is undefined, ask the user for the value OR infer it from context.
+3. Resolve the template to a specific file path.
+4. Load that specific file.
+5. Store in variable: `{pattern_name_content}`.
+
+#### INDEX_GUIDED Strategy
+
+Load index.md, analyze the structure and description of each doc in the index, then intelligently load relevant docs.
+
+**DO NOT BE LAZY** -- use best judgment to load documents that might have relevant information, even if there is only a 5% chance of relevance.
+
+1. Load `index.md` from the sharded directory.
+2. Parse the table of contents, links, and section headers.
+3. Analyze the workflow's purpose and objective.
+4. Identify which linked/referenced documents are likely relevant.
+   - *Example:* If the workflow is about authentication and the index shows "Auth Overview", "Payment Setup", "Deployment" -- load the auth docs, consider deployment docs, skip payment.
+5. Load all identified relevant documents.
+6. Store combined content in variable: `{pattern_name_content}`.
+
+**When in doubt, LOAD IT** -- context is valuable, and being thorough is better than missing critical info.
+
+---
+
+After applying the matching strategy, mark the pattern as **RESOLVED** and move to the next pattern.
+
+### 2b: Try Whole Document if No Sharded Found
+
+If no sharded matches were found OR no sharded pattern exists for this input:
+
+1. Attempt a glob match on the "whole" pattern (e.g., `{planning_artifacts}/*prd*.md`).
+2. If matches are found, load ALL matching files completely (no offset/limit).
+3. Store content in variable: `{pattern_name_content}` (e.g., `{prd_content}`).
+4. Mark pattern as **RESOLVED** and move to the next pattern.
+
+### 2c: Handle Not Found
+
+If no matches were found for either sharded or whole patterns:
+
+1. Set `{pattern_name_content}` to empty string.
+2. Note in session: "No {pattern_name} files found" -- this is not an error, just unavailable. Offer the user a chance to provide the file.
+
+## Step 3: Report Discovery Results
+
+List all loaded content variables with file counts. Example:
+
+```
+OK Loaded {prd_content} from 5 sharded files: prd/index.md, prd/requirements.md, ...
+OK Loaded {architecture_content} from 1 file: Architecture.md
+OK Loaded {epics_content} from selective load: epics/epic-3.md
+-- No ux_design files found
+```
+
+This gives the workflow transparency into what context is available.
diff --git a/.agents/skills/bmad-create-story/template.md b/.agents/skills/bmad-create-story/template.md
new file mode 100644
index 00000000..c4e129f5
--- /dev/null
+++ b/.agents/skills/bmad-create-story/template.md
@@ -0,0 +1,49 @@
+# Story {{epic_num}}.{{story_num}}: {{story_title}}
+
+Status: ready-for-dev
+
+<!-- Note: Validation is optional. Run validate-create-story for quality check before dev-story. -->
+
+## Story
+
+As a {{role}},
+I want {{action}},
+so that {{benefit}}.
+
+## Acceptance Criteria
+
+1. [Add acceptance criteria from epics/PRD]
+
+## Tasks / Subtasks
+
+- [ ] Task 1 (AC: #)
+  - [ ] Subtask 1.1
+- [ ] Task 2 (AC: #)
+  - [ ] Subtask 2.1
+
+## Dev Notes
+
+- Relevant architecture patterns and constraints
+- Source tree components to touch
+- Testing standards summary
+
+### Project Structure Notes
+
+- Alignment with unified project structure (paths, modules, naming)
+- Detected conflicts or variances (with rationale)
+
+### References
+
+- Cite all technical details with source paths and sections, e.g. [Source: docs/<file>.md#Section]
+
+## Dev Agent Record
+
+### Agent Model Used
+
+{{agent_model_name_version}}
+
+### Debug Log References
+
+### Completion Notes List
+
+### File List
diff --git a/.agents/skills/bmad-customize/SKILL.md b/.agents/skills/bmad-customize/SKILL.md
new file mode 100644
index 00000000..8f7f50e9
--- /dev/null
+++ b/.agents/skills/bmad-customize/SKILL.md
@@ -0,0 +1,111 @@
+---
+name: bmad-customize
+description: Authors and updates customization overrides for installed BMad skills. Use when the user says 'customize bmad', 'override a skill', 'change agent behavior', or 'customize a workflow'.
+---
+
+# BMad Customize
+
+Translate the user's intent into a correctly-placed TOML override file under `{project-root}/_bmad/custom/` for a customizable agent or workflow skill. Discover, route, author, write, verify.
+
+Scope v1: per-skill `[agent]` overrides (`bmad-agent-<role>.toml` / `.user.toml`) and per-skill `[workflow]` overrides (`bmad-<workflow>.toml` / `.user.toml`). Central config (`{project-root}/_bmad/custom/config.toml`) is out of scope — point users at the [How to Customize BMad guide](https://docs.bmad-method.org/how-to/customize-bmad/).
+
+When the target's `customize.toml` doesn't expose what the user wants, say so plainly. Don't invent fields.
+
+## Preflight
+
+- No `{project-root}/_bmad/` → BMad isn't installed. Say so, stop.
+- `{project-root}/_bmad/scripts/resolve_customization.py` missing → continue, but Step 6 verify falls back to manual merge.
+- Both present → proceed.
+
+## Activation
+
+Load `_bmad/config.toml` and `_bmad/config.user.toml` from `{project-root}` for `user_name` (default `BMad`) and `communication_language` (default `English`). Greet. If the user's invocation already names a target skill AND a specific change, jump to Step 3.
+
+## Step 1: Classify intent
+
+- **Directed** — specific skill + specific change → Step 3.
+- **Exploratory** — "what can I customize?" → Step 2.
+- **Audit/iterate** — wants to review or change something already customized → Step 2, lead with skills that have existing overrides; read the existing override in Step 3 before composing.
+- **Cross-cutting** — could live on multiple surfaces → Step 3, choose agent vs workflow explicitly with the user.
+
+## Step 2: Discovery
+
+```
+uv run {skill-root}/scripts/list_customizable_skills.py --project-root {project-root}
+```
+
+Use `--extra-root <path>` (repeatable) if the user has skills installed in additional locations.
+
+Group the returned `agents` and `workflows` for the user; for each show name, description, whether `has_team_override` or `has_user_override` is true. Surface any `errors[]`. For audit/iterate intents, lead with already-overridden entries.
+
+Empty list: show `scanned_roots`, ask whether skills live elsewhere (offer `--extra-root`); otherwise stop.
+
+## Step 3: Determine the right surface
+
+Read the target's `customize.toml`. Top-level `[agent]` or `[workflow]` block defines the surface.
+
+If a team or user override already exists, read it first and summarize what's already overridden before composing.
+
+**Cross-cutting intent — walk both surfaces with the user:**
+- Every workflow a given agent runs → agent surface (e.g. `bmad-agent-pm.toml` with `persistent_facts`, `principles`).
+- One workflow only → workflow surface (e.g. `bmad-prd.toml` with `activation_steps_prepend`).
+- Several specific workflows → multiple workflow overrides in sequence, not an agent override.
+
+**Single-surface heuristic:**
+- Workflow-level: template swap, output path, step-specific behavior, or a named scalar already exposed (`*_template`, `on_complete`). Surgical, reliable.
+- Agent-level: persona, communication style, org-wide facts, menu changes, behavior that should apply to every workflow the agent dispatches.
+
+When ambiguous, present both with tradeoff, recommend one, let the user decide.
+
+Intent outside the exposed surface (step logic, ordering, anything not in `customize.toml`): say so; offer `activation_steps_prepend`/`append` or `persistent_facts` as approximations, or recommend `bmad-builder` to create a custom skill.
+
+## Step 4: Compose the override
+
+Translate plain-English into TOML against the target's `customize.toml` fields. If an existing override was read, frame the change as additive.
+
+Merge semantics:
+- **Scalars** (`icon`, `role`, `*_template`, `on_complete`) — override wins.
+- **Append arrays** (`persistent_facts`, `activation_steps_prepend`/`append`, `principles`) — team/user entries append in order.
+- **Keyed arrays of tables** (menu items with `code` or `id`) — matching keys replace, new keys append.
+
+Overrides are sparse: only the fields being changed. Never copy the whole `customize.toml`.
+
+**Template swap** (`*_template` scalar): offer to copy the default template to `{project-root}/_bmad/custom/{skill-name}-{purpose}-template.md`, point the override at the new path, offer to help edit it.
+
+## Step 5: Team or user placement
+
+Under `{project-root}/_bmad/custom/`:
+- `{skill-name}.toml` — team, committed. Policies, org conventions, compliance.
+- `{skill-name}.user.toml` — user, gitignored. Personal tone, private facts, shortcuts.
+
+Default by character (policy → team, personal → user), confirm before writing.
+
+## Step 6: Show, confirm, write, verify
+
+1. Show the full TOML. If the file exists, show a diff. Never silently overwrite.
+2. Wait for explicit yes.
+3. Write. Create `{project-root}/_bmad/custom/` if needed.
+4. Verify:
+   ```
+   uv run {project-root}/_bmad/scripts/resolve_customization.py --skill <install-path> --key <agent-or-workflow>
+   ```
+   Show the merged output, point out the changed fields.
+
+   **Resolver missing or fails:** read whichever layers exist — `<install-path>/customize.toml` (base), `{project-root}/_bmad/custom/{skill-name}.toml` (team), `{project-root}/_bmad/custom/{skill-name}.user.toml` (user) — apply base → team → user with the same merge rules (scalars override, tables deep-merge, `code`/`id`-keyed arrays merge by key, all other arrays append), describe how the changed fields resolve.
+
+   **Verify shows override didn't land** (field unchanged, merge conflict, file not picked up): re-enter Step 4 with the verify output as context. Usually wrong field name, wrong merge mode (scalar vs array), or wrong scope.
+5. Summarize what changed, where the file lives, how to iterate. Remind the user to commit team overrides.
+
+## Complete when
+
+- Override file written (or user explicitly aborted).
+- User has seen resolver output (or manual fallback merge summary).
+- User has acknowledged the summary.
+
+Otherwise the skill isn't done — finish or tell the user they're exiting incomplete.
+
+## When this skill can't help
+
+- **Central config** (`{project-root}/_bmad/custom/config.toml`) — see the [How to Customize BMad guide](https://docs.bmad-method.org/how-to/customize-bmad/).
+- **Step logic, ordering, behavior not in `customize.toml`** — open a feature request, or use `bmad-builder` to create a custom skill. Offer to help with either.
+- **Skills without a `customize.toml`** — not customizable.
diff --git a/.agents/skills/bmad-customize/scripts/list_customizable_skills.py b/.agents/skills/bmad-customize/scripts/list_customizable_skills.py
new file mode 100644
index 00000000..86fd82a5
--- /dev/null
+++ b/.agents/skills/bmad-customize/scripts/list_customizable_skills.py
@@ -0,0 +1,231 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.11"
+# ///
+"""Enumerate customizable BMad skills installed alongside this one.
+
+Scans a skills directory (by default: the directory this script's own skill
+lives in, derived from __file__), finds every sibling directory containing a
+`customize.toml`, classifies each as agent and/or workflow based on its
+top-level blocks, reads the skill's SKILL.md frontmatter description for a
+one-liner, and checks whether override files already exist in
+`{project-root}/_bmad/custom/`.
+
+Skills in BMad are loaded either from a project-local location (e.g. the
+project's `.claude/skills/` or `.cursor/skills/`) or from a user-global
+location (e.g. `~/.claude/skills/`). We do not hardcode those paths — the
+running skill's own location is the source of truth for sibling discovery.
+`--extra-root` is available for the rare case where skills live in multiple
+locations on the same machine.
+
+Output: JSON to stdout. Non-empty `errors[]` in the payload is non-fatal
+by contract — the scanner surfaces malformed TOML, missing roots, and
+skills with no customization block as data for the caller to display,
+and still exits 0. Exit 2 is reserved for invocation errors (e.g.
+missing or unreadable `--project-root`) where no useful payload can be
+produced.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import re
+import sys
+import tomllib
+from pathlib import Path
+
+# Top-level TOML blocks that indicate a customization surface.
+SURFACE_KEYS = ("agent", "workflow")
+
+FRONTMATTER_RE = re.compile(r"^---\s*\n(.*?)\n---\s*\n", re.DOTALL)
+
+
+def default_skills_root() -> Path:
+    """Derive the skills root from this script's location.
+
+    Layout assumption: {skills_root}/bmad-customize/scripts/list_customizable_skills.py.
+    So the skills root is three parents up from this file.
+    """
+    return Path(__file__).resolve().parent.parent.parent
+
+
+def read_frontmatter_description(skill_md: Path) -> str:
+    """Extract the `description:` value from a SKILL.md YAML frontmatter block.
+
+    Returns an empty string if the file is missing, unreadable, or has no
+    description field. Intentionally permissive — this is metadata for a
+    human-facing list, not a validation target.
+    """
+    if not skill_md.is_file():
+        return ""
+    try:
+        text = skill_md.read_text(encoding="utf-8")
+    except (OSError, UnicodeDecodeError):
+        return ""
+    m = FRONTMATTER_RE.match(text)
+    if not m:
+        return ""
+    for line in m.group(1).splitlines():
+        stripped = line.strip()
+        if stripped.startswith("description:"):
+            value = stripped[len("description:") :].strip()
+            # Strip surrounding quotes if present.
+            if (value.startswith("'") and value.endswith("'")) or (
+                value.startswith('"') and value.endswith('"')
+            ):
+                value = value[1:-1]
+            return value
+    return ""
+
+
+def load_customize(toml_path: Path) -> dict | None:
+    """Return the parsed TOML, or None if unreadable."""
+    try:
+        with toml_path.open("rb") as f:
+            return tomllib.load(f)
+    except (OSError, tomllib.TOMLDecodeError):
+        return None
+
+
+def scan_skills(
+    skills_roots: list[Path],
+    project_root: Path,
+) -> dict:
+    """Scan each skills root for directories that contain a customize.toml."""
+    agents: list[dict] = []
+    workflows: list[dict] = []
+    errors: list[str] = []
+    scanned_roots: list[str] = []
+    seen_names: set[str] = set()
+    custom_dir = project_root / "_bmad" / "custom"
+
+    for root in skills_roots:
+        if not root.is_dir():
+            errors.append(f"skills root does not exist: {root}")
+            continue
+        scanned_roots.append(str(root))
+
+        for skill_dir in sorted(p for p in root.iterdir() if p.is_dir()):
+            customize_toml = skill_dir / "customize.toml"
+            if not customize_toml.is_file():
+                continue
+
+            data = load_customize(customize_toml)
+            if data is None:
+                errors.append(f"failed to parse {customize_toml}")
+                continue
+
+            skill_name = skill_dir.name
+            # If a skill with this name was already found in an earlier
+            # root, skip it — roots are scanned in the order provided, so
+            # the first occurrence wins.
+            if skill_name in seen_names:
+                continue
+            seen_names.add(skill_name)
+
+            description = read_frontmatter_description(skill_dir / "SKILL.md")
+            team_override = custom_dir / f"{skill_name}.toml"
+            user_override = custom_dir / f"{skill_name}.user.toml"
+
+            entry_base = {
+                "name": skill_name,
+                "install_path": str(skill_dir),
+                "skills_root": str(root),
+                "description": description,
+                "has_team_override": team_override.is_file(),
+                "has_user_override": user_override.is_file(),
+                "team_override_path": str(team_override),
+                "user_override_path": str(user_override),
+            }
+
+            # A skill may expose an agent surface, a workflow surface, or
+            # both. Emit one entry per surface so the caller can group cleanly.
+            surfaces_found = [k for k in SURFACE_KEYS if k in data]
+            if not surfaces_found:
+                errors.append(
+                    f"no [agent] or [workflow] block in {customize_toml}"
+                )
+                continue
+            for surface in surfaces_found:
+                entry = dict(entry_base)
+                entry["surface"] = surface
+                if surface == "agent":
+                    agents.append(entry)
+                else:
+                    workflows.append(entry)
+
+    return {
+        "project_root": str(project_root),
+        "scanned_roots": scanned_roots,
+        "custom_dir": str(custom_dir),
+        "agents": agents,
+        "workflows": workflows,
+        "errors": errors,
+    }
+
+
+def parse_args(argv: list[str]) -> argparse.Namespace:
+    parser = argparse.ArgumentParser(
+        description=(
+            "List customizable BMad skills installed alongside this one, "
+            "grouped by surface (agent vs workflow), with override status "
+            "looked up against {project-root}/_bmad/custom/."
+        )
+    )
+    parser.add_argument(
+        "--project-root",
+        required=True,
+        help="Absolute path to the project root (the folder containing _bmad/).",
+    )
+    parser.add_argument(
+        "--skills-root",
+        default=None,
+        help=(
+            "Override the primary skills directory to scan. Defaults to the "
+            "directory this script's own skill lives in."
+        ),
+    )
+    parser.add_argument(
+        "--extra-root",
+        action="append",
+        default=[],
+        metavar="PATH",
+        help=(
+            "Additional skills directory to include (repeatable). Useful "
+            "when skills live in multiple locations on the same machine "
+            "(e.g. project-local plus a user-global install)."
+        ),
+    )
+    return parser.parse_args(argv)
+
+
+def main(argv: list[str]) -> int:
+    args = parse_args(argv)
+    project_root = Path(args.project_root).expanduser().resolve()
+    if not project_root.is_dir():
+        print(
+            f"error: project-root does not exist or is not a directory: {project_root}",
+            file=sys.stderr,
+        )
+        return 2
+
+    primary = (
+        Path(args.skills_root).expanduser().resolve()
+        if args.skills_root
+        else default_skills_root()
+    )
+    extras = [Path(p).expanduser().resolve() for p in args.extra_root]
+    # Deduplicate in order of appearance.
+    roots: list[Path] = []
+    for root in [primary, *extras]:
+        if root not in roots:
+            roots.append(root)
+
+    result = scan_skills(roots, project_root)
+    print(json.dumps(result, indent=2, sort_keys=True))
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv[1:]))
diff --git a/.agents/skills/bmad-customize/scripts/tests/test_list_customizable_skills.py b/.agents/skills/bmad-customize/scripts/tests/test_list_customizable_skills.py
new file mode 100644
index 00000000..916b7c3b
--- /dev/null
+++ b/.agents/skills/bmad-customize/scripts/tests/test_list_customizable_skills.py
@@ -0,0 +1,249 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.11"
+# ///
+"""Unit tests for list_customizable_skills.py.
+
+Exercises the scanner against a synthesized install tree:
+- an agent-only customize.toml
+- a workflow-only customize.toml
+- a customize.toml that exposes both surfaces
+- a skill directory with no customize.toml (ignored)
+- a pre-existing team override in _bmad/custom/
+- malformed TOML (surfaces as an error without aborting)
+- multiple skills roots (e.g. project-local + user-global mix)
+
+Run: uv run scripts/tests/test_list_customizable_skills.py
+"""
+
+from __future__ import annotations
+
+import importlib.util
+import json
+import subprocess
+import sys
+import tempfile
+import unittest
+from pathlib import Path
+
+SCRIPT = Path(__file__).resolve().parent.parent / "list_customizable_skills.py"
+
+
+def _load_module():
+    spec = importlib.util.spec_from_file_location("list_customizable_skills", SCRIPT)
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)  # type: ignore[union-attr]
+    return module
+
+
+MODULE = _load_module()
+
+
+def _make_skill(parent: Path, name: str, body: str, skill_md: str | None = None) -> Path:
+    skill_dir = parent / name
+    skill_dir.mkdir(parents=True, exist_ok=True)
+    (skill_dir / "customize.toml").write_text(body, encoding="utf-8")
+    if skill_md is not None:
+        (skill_dir / "SKILL.md").write_text(skill_md, encoding="utf-8")
+    return skill_dir
+
+
+class ScannerTest(unittest.TestCase):
+    def setUp(self):
+        self.tmp = tempfile.TemporaryDirectory()
+        self.root = Path(self.tmp.name)
+        self.skills = self.root / "skills"
+        self.skills.mkdir(parents=True)
+        self.custom = self.root / "_bmad" / "custom"
+        self.custom.mkdir(parents=True)
+
+    def tearDown(self):
+        self.tmp.cleanup()
+
+    def test_agent_only_skill_detected(self):
+        _make_skill(
+            self.skills,
+            "bmad-agent-pm",
+            "[agent]\nicon = \"🧠\"\n",
+            "---\nname: bmad-agent-pm\ndescription: Product manager.\n---\n",
+        )
+        result = MODULE.scan_skills([self.skills], self.root)
+        self.assertEqual(len(result["agents"]), 1)
+        self.assertEqual(len(result["workflows"]), 0)
+        entry = result["agents"][0]
+        self.assertEqual(entry["name"], "bmad-agent-pm")
+        self.assertEqual(entry["surface"], "agent")
+        self.assertEqual(entry["description"], "Product manager.")
+        self.assertFalse(entry["has_team_override"])
+        self.assertFalse(entry["has_user_override"])
+
+    def test_workflow_only_skill_detected(self):
+        _make_skill(
+            self.skills,
+            "bmad-create-prd",
+            "[workflow]\npersistent_facts = []\n",
+            "---\nname: bmad-create-prd\ndescription: 'Create a PRD.'\n---\n",
+        )
+        result = MODULE.scan_skills([self.skills], self.root)
+        self.assertEqual(len(result["agents"]), 0)
+        self.assertEqual(len(result["workflows"]), 1)
+        entry = result["workflows"][0]
+        self.assertEqual(entry["description"], "Create a PRD.")
+
+    def test_dual_surface_skill_emits_two_entries(self):
+        _make_skill(
+            self.skills,
+            "bmad-dual",
+            "[agent]\nicon = \"x\"\n\n[workflow]\npersistent_facts = []\n",
+            "---\nname: bmad-dual\ndescription: Dual.\n---\n",
+        )
+        result = MODULE.scan_skills([self.skills], self.root)
+        self.assertEqual(len(result["agents"]), 1)
+        self.assertEqual(len(result["workflows"]), 1)
+        self.assertEqual(result["agents"][0]["name"], "bmad-dual")
+        self.assertEqual(result["workflows"][0]["name"], "bmad-dual")
+
+    def test_skill_without_customize_toml_ignored(self):
+        (self.skills / "bmad-plain").mkdir()
+        (self.skills / "bmad-plain" / "SKILL.md").write_text("# plain\n")
+        result = MODULE.scan_skills([self.skills], self.root)
+        self.assertEqual(len(result["agents"]) + len(result["workflows"]), 0)
+        self.assertEqual(result["errors"], [])
+
+    def test_existing_team_override_flagged(self):
+        _make_skill(
+            self.skills,
+            "bmad-agent-pm",
+            "[agent]\nicon = \"x\"\n",
+            "---\nname: bmad-agent-pm\ndescription: PM.\n---\n",
+        )
+        (self.custom / "bmad-agent-pm.toml").write_text("[agent]\n")
+        result = MODULE.scan_skills([self.skills], self.root)
+        entry = result["agents"][0]
+        self.assertTrue(entry["has_team_override"])
+        self.assertFalse(entry["has_user_override"])
+
+    def test_missing_surface_block_reports_error(self):
+        _make_skill(self.skills, "bmad-broken", "[not_a_surface]\nfoo = 1\n")
+        result = MODULE.scan_skills([self.skills], self.root)
+        self.assertEqual(len(result["agents"]) + len(result["workflows"]), 0)
+        self.assertEqual(len(result["errors"]), 1)
+        self.assertIn("no [agent] or [workflow] block", result["errors"][0])
+
+    def test_malformed_toml_reports_error_without_aborting(self):
+        skill_dir = self.skills / "bmad-bad"
+        skill_dir.mkdir()
+        (skill_dir / "customize.toml").write_text("this is not [valid toml\n")
+        # Plus a good sibling to confirm scanning continues.
+        _make_skill(
+            self.skills,
+            "bmad-good",
+            "[agent]\nicon = \"x\"\n",
+            "---\nname: bmad-good\ndescription: Good.\n---\n",
+        )
+        result = MODULE.scan_skills([self.skills], self.root)
+        self.assertEqual(len(result["agents"]), 1)
+        self.assertEqual(result["agents"][0]["name"], "bmad-good")
+        self.assertTrue(any("failed to parse" in e for e in result["errors"]))
+
+    def test_description_with_double_quotes_stripped(self):
+        _make_skill(
+            self.skills,
+            "bmad-q",
+            "[agent]\nicon = \"x\"\n",
+            '---\nname: bmad-q\ndescription: "Double-quoted desc."\n---\n',
+        )
+        result = MODULE.scan_skills([self.skills], self.root)
+        self.assertEqual(result["agents"][0]["description"], "Double-quoted desc.")
+
+    def test_multiple_skills_roots_are_merged(self):
+        extra_root = self.root / "extra-skills"
+        extra_root.mkdir()
+        _make_skill(
+            self.skills,
+            "bmad-agent-pm",
+            "[agent]\nicon = \"x\"\n",
+            "---\nname: bmad-agent-pm\ndescription: PM.\n---\n",
+        )
+        _make_skill(
+            extra_root,
+            "bmad-agent-dev",
+            "[agent]\nicon = \"y\"\n",
+            "---\nname: bmad-agent-dev\ndescription: Dev.\n---\n",
+        )
+        result = MODULE.scan_skills([self.skills, extra_root], self.root)
+        names = {a["name"] for a in result["agents"]}
+        self.assertEqual(names, {"bmad-agent-pm", "bmad-agent-dev"})
+        self.assertEqual(len(result["scanned_roots"]), 2)
+
+    def test_duplicate_skill_name_across_roots_first_wins(self):
+        extra_root = self.root / "extra-skills"
+        extra_root.mkdir()
+        _make_skill(
+            self.skills,
+            "bmad-agent-pm",
+            "[agent]\nicon = \"primary\"\n",
+            "---\nname: bmad-agent-pm\ndescription: Primary.\n---\n",
+        )
+        _make_skill(
+            extra_root,
+            "bmad-agent-pm",
+            "[agent]\nicon = \"duplicate\"\n",
+            "---\nname: bmad-agent-pm\ndescription: Duplicate.\n---\n",
+        )
+        result = MODULE.scan_skills([self.skills, extra_root], self.root)
+        self.assertEqual(len(result["agents"]), 1)
+        self.assertEqual(result["agents"][0]["description"], "Primary.")
+        self.assertEqual(result["agents"][0]["skills_root"], str(self.skills))
+
+    def test_missing_skills_root_reports_error(self):
+        result = MODULE.scan_skills(
+            [self.root / "does-not-exist", self.skills],
+            self.root,
+        )
+        self.assertTrue(any("skills root does not exist" in e for e in result["errors"]))
+
+    def test_cli_emits_valid_json_and_exits_zero(self):
+        _make_skill(
+            self.skills,
+            "bmad-agent-pm",
+            "[agent]\nicon = \"x\"\n",
+            "---\nname: bmad-agent-pm\ndescription: PM.\n---\n",
+        )
+        proc = subprocess.run(
+            [
+                sys.executable,
+                str(SCRIPT),
+                "--project-root",
+                str(self.root),
+                "--skills-root",
+                str(self.skills),
+            ],
+            capture_output=True,
+            text=True,
+            check=False,
+        )
+        self.assertEqual(proc.returncode, 0, proc.stderr)
+        payload = json.loads(proc.stdout)
+        self.assertEqual(len(payload["agents"]), 1)
+
+    def test_cli_exits_two_on_missing_project_root(self):
+        proc = subprocess.run(
+            [
+                sys.executable,
+                str(SCRIPT),
+                "--project-root",
+                str(self.root / "does-not-exist"),
+                "--skills-root",
+                str(self.skills),
+            ],
+            capture_output=True,
+            text=True,
+            check=False,
+        )
+        self.assertEqual(proc.returncode, 2)
+        self.assertIn("does not exist", proc.stderr)
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/.agents/skills/bmad-deep-recon/SKILL.md b/.agents/skills/bmad-deep-recon/SKILL.md
new file mode 100644
index 00000000..8542dc00
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/SKILL.md
@@ -0,0 +1,82 @@
+---
+name: bmad-deep-recon
+description: 'Decision-grade research, three ways: draft a deep-research prompt for the user to run in their own tool (ChatGPT, Gemini, Grok, Perplexity, …), process a finished research report — file it, distill a succinct cited summary with metadata that downstream skills consume without reprocessing — or run the research here through web fan-out. Shipped type packs: market, domain, technical, competitive, user-voice, academic-lit — plus a select shape for choose-between decisions and custom types via overrides. Use when the user says "deep recon", "research this", "draft a research prompt", "process this research report", "market research", "domain research", "technical research", "competitor research", "literature review", or "help me choose between".'
+---
+
+# BMad Deep Recon
+
+## Overview
+
+You are **Deep Recon** — a research director, not a search engine. Your value is framing research worth running and turning whatever comes back into a decision-grade artifact this project consumes without reprocessing. Every engagement serves a **decision** — enter a market, pick a stack, scope a product, commit to a domain — and is shaped by it from the first question to the final artifact.
+
+Three services, freely combined — each detailed in its reference: **Draft** a deep-research prompt the user runs in their own tool, **Process** a finished report into the succinct cited summary downstream skills read, or **Run** the research here through parallel web fan-out. Draft → run externally → Process is the natural loop; Run is fully capable on its own.
+
+**Epistemics — two standing rules, inherited verbatim by every subagent you spawn:**
+
+1. **Never conclude from training data alone.** What you already know proposes hypotheses, queries, and structure; conclusions require evidence retrieved or imported *this run*. A claim you cannot evidence is stated as an unverified belief or not at all.
+2. **The research firewall.** Project context — briefs, PRDs, code, memory, `{workflow.persistent_facts}` — shapes *what to ask*, never *what is true*. It is inadmissible as evidence: every claim in a research artifact traces to a digest or import file with a source. Research subagents receive only their brief — no project files, no ambient context — unless the plan explicitly grants a named document.
+
+## How you work
+
+- **Nothing exists until it is a file.** Every digest, import extraction, and report section is written to the run folder the moment it lands — the conversation is a control channel, never the store. A run that dies mid-flight resumes from disk with nothing lost.
+- **Extract, don't ingest.** Raw reports and search results never enter the parent context whole; subagents return relevance-filtered digests, and the parent reads digest files JIT.
+- **A claim is a sentence with a source.** Publisher, publication date, access date. No naked numbers.
+- **Report what is real.** Thin public data is reported as thin, absence of evidence is a finding, and freshness is part of truth — each pack sets windows per claim class; a market size from three years ago is history, not fact.
+- **Fast by default.** Rigor is bought consciously through the knobs, never accreted through extra passes. One gate, light checkpoints, no ceremony.
+- **The memlog is the process memory.** Every decision, source batch, load-bearing claim, plan change, and assumption is one append-only line, always through the script: `uv run {project-root}/_bmad/scripts/memlog.py` with `--type <decision|source|claim|assumption|question|event>`.
+- Web access is required for Run. If unavailable, say so and offer Draft/Process — never fabricate research.
+
+## Resolution rules
+
+- Bare paths and `{skill-root}` (e.g. `references/run.md`) resolve from this skill's installed directory.
+- `{project-root}` → the project working directory; `{skill-name}` → the skill directory's basename.
+- `{workflow.<name>}` → a merged `customize.toml` field; `{doc_workspace}` → the bound run folder.
+- Forward slashes only. Config variables already contain `{project-root}` in their resolved values — never double-prefix.
+
+## On Activation
+
+**Forwarded activation:** if a caller invoked you with a stated intent, research type, or pre-resolved customization fields (the legacy research shims and Mary's menu do), honor them verbatim — skip your own inference for those values and resolve only the rest.
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow` (on failure read `{skill-root}/customize.toml`, use defaults). Run `{workflow.activation_steps_prepend}`, then `{workflow.activation_steps_append}`.
+2. Resolve config: `uv run {project-root}/_bmad/scripts/resolve_config.py --project-root {project-root}`. From the merged JSON resolve `{user_name}`, `{communication_language}`, `{document_output_language}`, `{project_name}`, `{output_folder}` (under `core`), `{planning_artifacts}` (under `modules.bmm`; absent on core-only installs → `{output_folder}`), and `{date}`; missing keys take neutral defaults, never block.
+3. Headless (no interactive user) → see `## Headless Mode`. Otherwise greet `{user_name}` in `{communication_language}` — and stay in it every turn.
+4. Detect the intent: **draft**, **process** (the user has or names a report), **run**, or lifecycle **refresh** / **deepen** on an existing run folder. When the ask is bare research with no verb ("research X for me"), open the floor first — invite the decision they're facing and anything they already have (briefs, links, a prior report) in one turn, then ask only what's missing — and put the choice up front, once: **Run** it here now, or **Draft** a prompt for a deep-research tool they subscribe to — often cheaper and a strong gatherer, with Process turning its output into the same artifact. State the trade honestly (tokens and minutes here vs. one manual round-trip there); their call, remembered for the session.
+5. If a run folder for this topic already exists under `{workflow.research_output_path}`, offer to resume or extend it (a drafted brief awaiting its report, a report awaiting refresh) rather than start a duplicate.
+
+## Research types and decision shapes
+
+The type set is whatever `{workflow.research_types}` resolves to — shipped: `market`, `domain`, `technical`, `competitive`, `user-voice`, `academic-lit` — each pointing at a pack file. You already know how to research; the pack is where this harness is opinionated — prioritized dimensions, non-obvious source craft, freshness bars and two-source classes per claim class, downstream bindings. Apply it in every mode; don't re-derive it. Overrides replace matching codes and append new ones; never claim a fixed type list — read the resolved set.
+
+Infer the type from the user's ask and each entry's `when` clause; confirm only when genuinely ambiguous. An explicit type (argument, shim, menu) wins without discussion.
+
+Orthogonal to type is the **decision shape**: **explore** (the default — understand, assess, validate) or **select** (choose between candidates). When the shape is select, load `references/selection.md` and layer its method over the type's pack — it shapes drafted prompts and processed summaries as much as native runs.
+
+## Intents
+
+Route on the detected intent and load only what it names. Every intent shares the run-folder workspace shape — `brief.md`, `imports/`, `digests/`, `research.md`, `.memlog.md` — and ends per `references/finalize.md`.
+
+| Intent | What it does | Load |
+| --- | --- | --- |
+| Draft | Compose a deep-research prompt for the user's own tool, carrying the pack's craft | `references/draft.md` |
+| Process | File a finished report, extract its claims, distill the downstream summary | `references/process.md` |
+| Run | Native research: resolve effort, hold the plan gate — the one hard stop — then run the loop | `references/run.md`, then `references/verification.md` + `references/synthesis.md` |
+| Refresh / Deepen | Update or extend an existing run folder | `references/lifecycle.md` |
+
+## Headless Mode
+
+When invoked headless, do not ask. Bare research defaults to **run**; a named report means **process**; a requested prompt means **draft** (the brief file is the deliverable). Plan-and-proceed: infer type, build from the pack, keep configured knobs plus anything in the invocation (red team and workflow orchestration only when set `"on"`), skip checkpoints, log every judgment call as an `assumption`. Halt `blocked` only when topic or target folder cannot be inferred. End with JSON:
+
+```json
+{
+  "status": "complete",
+  "intent": "run",
+  "type": "market",
+  "report": "{doc_workspace}/research.md",
+  "memlog": "{doc_workspace}/.memlog.md",
+  "claims": {"verified": 12, "unverified": 3, "overturned": 0},
+  "open_questions": [],
+  "external_handoffs": []
+}
+```
+
+Omit keys for artifacts not produced; the `claims` counts come from `uv run scripts/recon_kit.py tally {doc_workspace}/.memlog.md`, never hand-counted. Draft adds `"brief"`; process adds `"imports"`; refresh replaces `claims` scope with the refresh set plus a `deltas` array. With `output_format = "auto"`, headless runs produce no briefing; add `"briefing"` when rendered.
diff --git a/.agents/skills/bmad-deep-recon/assets/research.template.md b/.agents/skills/bmad-deep-recon/assets/research.template.md
new file mode 100644
index 00000000..14fe968b
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/assets/research.template.md
@@ -0,0 +1,18 @@
+---
+title: '{research_type} research: {research_topic}'
+type: '{research_type}'
+topic: '{research_topic}'
+decision: '{decision}'
+source: '{source}'
+status: draft
+preset: '{preset}'
+validation: '{validation}'
+created: '{date}'
+updated: '{date}'
+---
+
+# {research_type} research: {research_topic}
+
+**Decision this research serves:** {decision}
+
+_Sections are appended per the approved research plan; the executive summary is written last and placed here, first._
diff --git a/.agents/skills/bmad-deep-recon/customize.toml b/.agents/skills/bmad-deep-recon/customize.toml
new file mode 100644
index 00000000..434c47d5
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/customize.toml
@@ -0,0 +1,212 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-deep-recon.
+#
+# Override files (not edited here):
+#   {project-root}/_bmad/custom/bmad-deep-recon.toml         (team)
+#   {project-root}/_bmad/custom/bmad-deep-recon.user.toml    (personal)
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins
+#   arrays (persistent_facts, activation_steps_*, *_sources, doc_standards,
+#   external_*): append
+#   arrays of tables keyed by `code`: matching key replaces, new keys append
+
+# Steps executed on activation: prepend runs before the skill's own
+# activation flow, append runs after it. Each entry is a literal instruction.
+activation_steps_prepend = []
+activation_steps_append = []
+
+# Standing context for framing the research — decision context only, never
+# evidence: the research firewall keeps project material out of findings.
+# Entries prefixed `file:` are paths or globs whose contents load as facts;
+# all others are literal facts. Empty by default so nothing local leaks into
+# research framing unasked.
+persistent_facts = []
+
+# Where research runs live and how each run folder is named. Draft, Process,
+# and Run all use the same folder shape: brief.md (drafted prompts),
+# imports/ (originals, full fidelity), digests/ (extracted claims),
+# research.md (the canonical summary/report), .memlog.md.
+research_output_path = "{planning_artifacts}/research"
+run_folder_pattern = "{research_type}-{topic_slug}-{date}"
+
+# Seed document for a new run.
+research_template = "assets/research.template.md"
+
+# --- Effort (Run mode) ------------------------------------------------------
+# A preset bundles the three effort knobs; any knob set here individually
+# pins that knob over the preset. What the user says in the request beats
+# both.
+#
+#   preset     subagents  sources/round  depth
+#   quick      low (2)    5              1
+#   standard   normal (3) 8              2
+#   deep       high (6)   12             3
+#
+# Grounding: orchestrator-worker research systems document 3-5 parallel
+# workers as the sweet spot (more only for genuinely wide work). Depth and
+# sources are caps, not quotas — dimensions stop early on coverage or
+# novelty exhaustion. Defaults are tuned for a fast run; buy more rigor
+# consciously, per run, in the request.
+
+preset = "standard"
+
+# "" = from preset. Values: none | low | normal | high
+# (0 / 2 / 3 / 6 parallel research assistants, ceiling 10; none also =
+# no-subagent environments, run inline sequentially).
+subagents = ""
+
+# 0 = from preset. Distinct sources actually read per dimension per round.
+# Ceiling 25 — beyond that a single round exceeds what hosted deep-research
+# products spend on an entire run.
+max_sources_per_round = 0
+
+# 0 = from preset. Rounds per dimension: initial pass + lead-following
+# follow-ups. Ceiling 5.
+max_depth = 0
+
+# Verification level, applied as material lands (never an end-of-run pass).
+#   normal  spot-check load-bearing claims only — fast, the default
+#   high    cross-check the pack's two-source classes; red-team major
+#           conclusions
+#   max     cross-check every ledger claim + the red-team pass at full
+#           breadth + primary-source-priority ranking
+validation = "normal"
+
+# Red-team stance pass — fresh-context skeptics hunting disconfirming
+# evidence for major conclusions: "off" (default), "offer" (proposed at the
+# plan gate), or "on" (always; headless honors only "on"). high/max
+# validation includes it for major conclusions regardless.
+red_team = "off"
+
+# Run the acquisition fan-out through the harness's deterministic
+# orchestration when it offers one (e.g. workflows): "off", "offer"
+# (proposed at the plan gate when available), or "on". Orchestrated runs are
+# faster wall-clock but spend more tokens.
+use_workflows = "offer"
+
+# Ordered model preference for spawned research assistants — first model the
+# harness can provide wins; [] lets the harness/skill choose (lead stays on
+# the strongest model; researchers at most one capable tier down; judgment
+# work never on the smallest tier; mechanical extraction may use a fast
+# tier).
+#
+# Example: subagent_models = ["<your-mid-tier-model-id>", "<your-fast-model-id>"]
+subagent_models = []
+
+# Source policy. Preferred sources are consulted first and weighted as more
+# credible; banned sources are never cited (their claims may still be leads
+# to verify elsewhere). Entries are domains or plain-text descriptions.
+# Draft mode writes both policies into drafted prompts.
+preferred_sources = []
+banned_sources = []
+
+# What is presented and handed off — never what exists: research.md (the
+# canonical machine-readable summary/report) always lives in the workspace.
+#   "auto"  html briefing on interactive runs; md only on headless or
+#           skill-invoked runs (the caller reads the md; render later at will)
+#   "html"  always render the briefing page (references/html-briefing.md)
+#   "md"    never render html
+#   "both"  render and present both
+output_format = "auto"
+
+# Theme for the HTML briefing: empty = the shipped neutral professional
+# theme, a `file:` path to a theme/brand spec, or inline directives
+# (e.g. "Canvas #122543, accent #B66D46, sans-serif, dark-mode aware").
+html_theme = ""
+
+# Default audience shaping for the synthesis — freeform, empty = balanced
+# technical/business register. Examples: "executive one-pager first, detail
+# after", "engineering team, keep vendor marketing out".
+audience = ""
+
+# Registry of extra research surfaces — internal knowledge bases or search
+# tools you subscribe to — consulted alongside web research in Run mode; each
+# entry names the tool and when to use it. Installed search-shaped MCP tools
+# are discovered automatically at the plan gate; an entry here adds routing
+# guidance the discovery can't infer.
+#
+# Examples:
+#   external_sources = [
+#     "Tavily MCP (tavily_search/tavily_extract): preferred web search + clean page extraction",
+#     "Perplexity Sonar MCP (perplexity_ask): cited synthesized answers — chase its citations as the sources",
+#     "xAI X Search MCP: live X/Twitter posts and threads, for user-voice and sentiment dimensions",
+#     "Gartner MCP (corp:gartner_query): analyst data on enterprise software markets",
+#   ]
+external_sources = []
+
+# Polish passes applied to research.md at finalize. Entries are `skill:NAME`
+# directives, `file:` style guides, or plain-text instructions.
+doc_standards = ["skill:bmad-review lenses=structure,prose"]
+
+# Handoffs executed at finalize to route the report beyond local files. Each
+# entry names the tool and what to do; unavailable tools are skipped and
+# flagged.
+#
+# Examples:
+#   "NotebookLM (notebooklm-mcp): create a notebook from research.md plus the top sources, generate an audio overview, return the notebook URL"
+#   "Confluence (corp:confluence_upload): publish research.md to the RESEARCH space, return the page URL"
+external_handoffs = []
+
+# Executed after finalize. A string scalar is one instruction; an array is a
+# sequence. Empty = the run ends with the finalize summary.
+on_complete = ""
+
+# ---------------------------------------------------------------------------
+# Research types — subject lenses. Each type is a pack: a policy and craft
+# card (prioritized dimensions, non-obvious source craft, freshness bars,
+# two-source classes, downstream bindings) used by all three modes — it
+# shapes drafted prompts, native runs, and processed-report gap checks
+# alike. `when` guides type inference from the user's ask; an explicitly
+# requested type always wins. The decision shape (explore vs select) is
+# orthogonal — any type can end in a selection matrix.
+#
+# Keyed by `code`: an override with a matching code replaces the shipped
+# type, a new code appends. Empty `pack` disables a type.
+#
+# Example (add an org-specific type in team/user override TOML):
+#   [[workflow.research_types]]
+#   code = "regulatory"
+#   name = "Regulatory Research"
+#   when = "Compliance posture, licensing, or regulatory exposure for a product or market."
+#   pack = "file:{project-root}/_bmad/custom/packs/regulatory.md"
+# ---------------------------------------------------------------------------
+
+[[workflow.research_types]]
+code = "market"
+name = "Market Research"
+when = "Market opportunity, customers, competition, sizing, or go-to-market for a product or business decision."
+pack = "types/market.md"
+
+[[workflow.research_types]]
+code = "domain"
+name = "Domain Research"
+when = "Understanding an industry, sector, or field: structure, players, rules, vocabulary, dynamics."
+pack = "types/domain.md"
+
+[[workflow.research_types]]
+code = "technical"
+name = "Technical Research"
+when = "A technology area's landscape, patterns, integration approaches, and implementation reality."
+pack = "types/technical.md"
+
+[[workflow.research_types]]
+code = "competitive"
+name = "Competitive Research"
+when = "Teardown of specific named competitors: offers, pricing, positioning, trajectory, their customers' sentiment."
+pack = "types/competitive.md"
+
+[[workflow.research_types]]
+code = "user-voice"
+name = "User-Voice Research"
+when = "What users of a product or category actually experience and want: reviews, communities, jobs-to-be-done."
+pack = "types/user-voice.md"
+
+[[workflow.research_types]]
+code = "academic-lit"
+name = "Academic Literature"
+when = "Published research: literature review, state of the art, grounding an approach in papers."
+pack = "types/academic-lit.md"
diff --git a/.agents/skills/bmad-deep-recon/references/draft.md b/.agents/skills/bmad-deep-recon/references/draft.md
new file mode 100644
index 00000000..68d4c90b
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/references/draft.md
@@ -0,0 +1,8 @@
+# Draft
+
+Build a deep-research prompt the user runs themselves — in conversation, fast, not a project. The pack's craft travels inside the prompt so the outside tool works to this harness's standard.
+
+1. Open the floor before any structured questions: invite the decision they're facing and anything they already have — briefs, links, a prior report, half-formed constraints — in one turn, then ask only what's still missing. Nail the **decision**, topic, and type; load the pack. Ask which tool the prompt is for (it changes phrasing: hosted deep-research agents handle wide scopes and long source lists; social-native tools like Grok earn user-voice and sentiment dimensions; if unknown, write tool-neutral).
+2. Compose the prompt from the pack: the dimensions as explicit research questions pruned to the decision, the freshness bars as recency requirements, the two-source expectation for its critical claim classes, the audience, the source policy — `{workflow.preferred_sources}` named as sources to prefer, `{workflow.banned_sources}` as sources never to cite — and a **non-negotiable citation demand**: every claim with source URL and publication date, contrary evidence reported, gaps admitted rather than padded. Structure the requested output so Process can extract it cleanly (findings per dimension, a source list).
+3. Bind `{doc_workspace}`: expand the folder name deterministically (`uv run scripts/recon_kit.py slug "<topic>" --type <type> --pattern "{workflow.run_folder_pattern}"` — same expansion every mode, so the report comes back to the same folder) under `{workflow.research_output_path}`, init the memlog with the decision context, save the prompt as `{doc_workspace}/brief.md`, and present it paste-ready in chat.
+4. Close the loop: tell the user to run it in their tool and bring the report back — "process it" from here picks up this folder, decision context intact.
diff --git a/.agents/skills/bmad-deep-recon/references/finalize.md b/.agents/skills/bmad-deep-recon/references/finalize.md
new file mode 100644
index 00000000..24a3a3bf
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/references/finalize.md
@@ -0,0 +1,11 @@
+# Finalize
+
+Every mode ends here once `research.md` is assembled.
+
+1. `research.md` is complete per `references/synthesis.md`: decision-first summary, findings, contrary evidence where found, recommendations with downstream bindings, source appendix, staleness map. Frontmatter metadata (`type`, `topic`, `decision`, `source`, `status`, dates) is what lets every downstream consumer trust it without reprocessing.
+2. **Citation check — mechanical, then semantic.** Run `uv run scripts/recon_kit.py citations {doc_workspace}/research.md` — it diffs inline `[n]` markers against the appendix and lists dangling markers and orphaned rows exactly; fix what it reports. Then a fresh-context subagent does only the judgment half: does each cited source actually say what the text claims? It never rewrites findings — a claim whose source doesn't back it gets its confidence downgraded and the mismatch logged as an `event`.
+3. Render per `{workflow.output_format}` (see `references/html-briefing.md`): `auto` renders the briefing page on interactive runs, skips on headless/skill-invoked; `html`/`both` always; `md` never. `research.md` always exists — the briefing is its regenerable face.
+4. Polish: apply each `{workflow.doc_standards}` entry (a `skill:`, `file:`, or plain-text directive) to `research.md`.
+5. Execute each `{workflow.external_handoffs}` entry (NotebookLM, Confluence, …) — invoke the named tool, surface returned URLs; skip and flag unavailable tools.
+6. Tell the user what exists and where — report, briefing, imports, memlog — plus what the staleness map says to re-check and when, and that Refresh/Deepen handle it. Invoke `bmad-help` to suggest the next step.
+7. Run `{workflow.on_complete}` if non-empty — a string is one instruction, an array is a sequence.
diff --git a/.agents/skills/bmad-deep-recon/references/html-briefing.md b/.agents/skills/bmad-deep-recon/references/html-briefing.md
new file mode 100644
index 00000000..8d8c379f
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/references/html-briefing.md
@@ -0,0 +1,16 @@
+# HTML Briefing
+
+Generate `research-briefing.html` in `{doc_workspace}` after `research.md` is final, when `{workflow.output_format}` calls for it: `"auto"` renders on interactive runs and skips on headless/skill-invoked runs (the md is always there to render from later), `"html"`/`"both"` always render, `"md"` never. The page is a full-fidelity presentation of the report, never a second source of truth — same claims, same numbers, same citations; nothing is lost by reading it instead of the markdown.
+
+## Requirements
+
+- **Self-contained single file**: inline CSS and JS, no external requests of any kind (no CDN, no fonts, no remote images). It must render from a `file://` open, offline, forever.
+- **Structure**: a header (topic, type, decision, date, depth, verification level) → the executive summary as the opening card → sticky table of contents → dimension sections → contrary evidence (when present) → recommendations → collapsible source appendix → staleness map.
+- **Confidence is visual**: every claim carries its badge — verified / medium / low / `unverified` / disputed — color-coded with the status text always present (never color alone). Unverified and disputed must be *more* prominent than verified, not less.
+- **Sources are live**: inline `[n]` markers link to the appendix row; appendix rows link out to the source URL. Source URLs are untrusted content — never hand-escape them: generate the appendix table with `uv run scripts/recon_kit.py escape-sources {doc_workspace}/research.md` and embed its `html` output, which escapes every cell, anchors each row (`id="src-n"`), and links only validated `http(s)` URLs (anything else renders as plain text; the script lists it in `invalid_urls`). Apply the same escape discipline to any other source-derived text you place in attributes.
+- **Charts sparingly**: only where the data genuinely benefits (market size trajectory, decision matrix scores) — simple inline SVG, labeled axes, no library.
+- **Responsive and theme-aware**: readable on a phone; respect `prefers-color-scheme` for light/dark.
+
+## Theme
+
+`{workflow.html_theme}` governs: a `file:` path loads a theme/brand spec to follow; inline text is applied as directives; empty means the shipped default — neutral, professional, generous whitespace, system font stack, one restrained accent color. Whatever the theme, the confidence-badge semantics above are non-negotiable.
diff --git a/.agents/skills/bmad-deep-recon/references/lifecycle.md b/.agents/skills/bmad-deep-recon/references/lifecycle.md
new file mode 100644
index 00000000..74b092cc
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/references/lifecycle.md
@@ -0,0 +1,11 @@
+# Refresh and Deepen
+
+Lifecycle intents on an existing run folder.
+
+## Refresh
+
+Read `research.md` and `.memlog.md` — never re-research from scratch. Build the refresh set mechanically: assemble the claims (`claim`, `class`, `pub_date`) from the ledger, map the pack's freshness bars to a months-per-class JSON, and run `uv run scripts/recon_kit.py staleness <claims.json> --windows '<map>'` — the stale flags are the candidate set. Confirm it in one exchange, re-verify just those claims, and deliver a **delta report** (confirmed / changed / overturned, new sources) appended to `research.md` with the frontmatter `updated` bumped. Claims outside the set keep their status. An overturned load-bearing claim triggers an explicit warning naming the downstream artifacts that consumed it.
+
+## Deepen
+
+Drill into one dimension or add a new one without touching the rest: mini plan gate, acquire → verify for that slice only (or a drafted follow-up prompt when the user's tool is better placed), merge into `research.md`, update only the synthesis sections the new material affects — a deepening that changes no conclusion says so.
diff --git a/.agents/skills/bmad-deep-recon/references/process.md b/.agents/skills/bmad-deep-recon/references/process.md
new file mode 100644
index 00000000..23807980
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/references/process.md
@@ -0,0 +1,10 @@
+# Process
+
+For a report the user names or drops ("there's a research report at <path>, process it"):
+
+1. **File it.** Find or create the run folder: if a drafted brief for this topic exists, that folder is the target; otherwise infer type and topic from the report (confirm in one line), bind `{doc_workspace}` (expand the folder name with `uv run scripts/recon_kit.py slug` as in Draft), and init the memlog. Move or copy the original into `{doc_workspace}/imports/` untouched — full fidelity is preserved there, and nowhere else.
+2. **Record provenance** in the memlog: what produced it (which tool or firm), when (ask if not evident — production date drives staleness), and what the user wants decided from it.
+3. **Extract.** A subagent (fresh context, firewall rules) reads the import and pulls every claim bearing on the decision into digest files under `{doc_workspace}/digests/` — standard shape `{claim, source, publisher, pub_date, accessed, confidence, class}`, keeping the original's citations (the cited source is the publisher; the import is the via). Multiple imports each get their own digest; contradictions between them are findings, not noise.
+4. **Check against the pack**: which of the type's dimensions the material covers, which are open, where its claims fall inside two-source classes but rest on one publisher. Verification per the resolved `validation` level (`references/verification.md`) — at `normal` this is a spot-check of the load-bearing claims only, minutes not hours.
+5. **Distill** into `research.md` per `references/synthesis.md` — the succinct, cited, decision-first summary with full metadata frontmatter (topic, type, decision, `source:` provenance, dates, status). This is the artifact downstream skills read; nobody ever reprocesses the import. Open dimensions are listed honestly with a one-line route: draft a follow-up prompt, or a targeted Run on the gap.
+6. Finalize per `references/finalize.md`.
diff --git a/.agents/skills/bmad-deep-recon/references/run.md b/.agents/skills/bmad-deep-recon/references/run.md
new file mode 100644
index 00000000..52b8976a
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/references/run.md
@@ -0,0 +1,73 @@
+# Run
+
+Native research, when chosen: resolve effort, hold the plan gate, then run the acquisition loop once per dimension of the approved plan, in plan order.
+
+## Effort
+
+Three knobs bundled in a **preset**; any knob pins individually, and **what the user says in the request beats both**.
+
+| Preset (`{workflow.preset}`) | subagents | sources/round | depth |
+|---|---|---|---|
+| `quick` | low (2) | 5 | 1 |
+| `standard` (default) | normal (3) | 8 | 2 |
+| `deep` | high (6) | 12 | 3 |
+
+- **subagents** — parallel assistants: `none` (0 — inline, sequential; also the no-subagent-harness fallback), `low` (2), `normal` (3), `high` (6, cap 10 — beyond the 3–5 sweet spot only for genuinely wide work).
+- **max_sources_per_round** — distinct sources actually read per dimension per round (cap 25).
+- **max_depth** — rounds per dimension: initial pass plus lead-following follow-ups (cap 5). A cap, not a quota — dimensions stop early on coverage or novelty exhaustion.
+- **validation** (orthogonal to preset, default `normal`) — rigor rises `normal` < `high` < `max`; level semantics live in `references/verification.md`. Verification happens per dimension as material lands, never as an end-of-run rewrite pass.
+
+`{workflow.subagent_models}` is an ordered model preference for assistants — first available wins; empty means harness default. Keep the lead on the strongest model; researchers at most one tier down; judgment work never on the smallest tier.
+
+## The plan gate
+
+The one hard stop, kept light: decision, type and pack-derived dimensions pruned to it, shape, the **decomposition topology** — *breadth-first* (independent sub-questions: assistants split the dimensions), *depth-first* (one question that needs several perspectives: assistants split by angle or methodology, not by dimension), or *straightforward* (a focused ask: one assistant, a handful of calls, no fan-out — never overinvest in a simple query) — knobs in force and where each came from, which search surfaces exist (harness web search; installed search-shaped MCP tools; `{workflow.external_sources}` — check, don't assume), whether to run the fan-out as a workflow when the harness offers orchestration and `{workflow.use_workflows}` allows, and an honest time estimate (a standard run is minutes; deep runs are tens of minutes and many times the tokens).
+
+Present as a compact checklist, get approval, then: bind `{doc_workspace}` under `{workflow.research_output_path}` — expand the folder name with `uv run scripts/recon_kit.py slug "<topic>" --type <type> --pattern "{workflow.run_folder_pattern}"` so the same topic always resolves to the same folder — seed `research.md` from `{workflow.research_template}`, init the memlog (`uv run {project-root}/_bmad/scripts/memlog.py init --workspace {doc_workspace} --field topic="<topic>" --field type="<type>" --field decision="<decision>" --field preset="<preset>"`), log the approved plan as a `decision`, and tell the user the path.
+
+Each dimension then runs in **rounds** — up to the resolved `max_depth` — and the report grows as material lands: the user watches the document build, not a spinner. Every digest is written to `{doc_workspace}/digests/` the moment it exists — one file per assistant per round (`<dimension>-r<round>-<n>.md`), the digest shape below, raw enough to re-derive from.
+
+## Rounds and lead-following
+
+Round 1 pursues the plan's questions **broad-first**: short, wide queries to map what exists, narrowing as the shape emerges — not long specific queries that return nothing. After each round, harvest the leads: new entities worth chasing, unexpected connections, contradictions between sources, and questions the round opened. Contradictions get priority. Promising leads become the next round's brief; note mid-course discoveries in the checkpoint so the user sees the turn happening.
+
+A dimension stops before its round cap when either holds:
+
+- **Coverage** — its plan questions are answered, with the critical claims confirmed per the resolved `validation` level.
+- **Novelty exhaustion** — a full round surfaced no new load-bearing claim or lead.
+
+Say which one ended it. Hitting the round cap with open questions is reported as an open question, never silently dropped.
+
+**Stop-and-write valve.** If the run is dragging well past the plan gate's estimate — rounds queuing, budgets mostly spent — stop spawning, synthesize from the digests already on disk, and report the remainder as open questions with a route (a Deepen later, or a drafted prompt for the user's own tool). A shorter honest report beats a longer stale one.
+
+## The fan-out
+
+Fan out researcher assistants for the round — concurrency per the resolved `subagents` level, split by the plan's **topology**: breadth-first gives each assistant independent sub-questions; depth-first gives each a distinct perspective or methodology on the *same* question; straightforward is one assistant with a small budget — never fan out what one focused assistant answers. Each assistant runs behind the **research firewall**: it gets its brief and nothing else — no project files, no ambient context. The brief contains:
+
+- the questions it owns, the decision they serve, and the topic
+- its search surfaces (specialized tools first — installed search-shaped MCP tools, `{workflow.external_sources}` entries whose directive matches — then generic search), plus `{workflow.preferred_sources}` first / `{workflow.banned_sources}` never
+- the pack's source craft and freshness bars, and the source-quality card below
+- its budgets — sources (the round's share of `max_sources_per_round`) and tool calls, scaled to its task: under 5 for a simple lookup, ~5 medium, ~10 hard, 15 for genuinely multi-part, 20 never exceeded. Either budget spent → synthesize what it has
+- the query craft: short queries (roughly five words or fewer) beat hyper-specific ones that return nothing; broaden when results are sparse, narrow when abundant; never repeat an identical query on the same tool; after every tool result, pause and evaluate — what did this add, what gap remains, what's the best next query — before firing again
+- the epistemics rules verbatim, and the return contract: a digest, not raw results — findings as claims, each with `{claim, source, publisher, pub_date, accessed, confidence, class}`, plus leads worth chasing and what it looked for and could not find
+
+**On each return, write the digest to `{doc_workspace}/digests/` before doing anything else with it.**
+
+Spawn assistants on `{workflow.subagent_models}` when set (first available wins); otherwise the harness default — judgment work never drops to the smallest tier. When subagents are unavailable (or `subagents` is `none`), run the same rounds yourself, sequentially, under the same budgets and the same files-first discipline.
+
+When workflow orchestration was approved at the plan gate, run the fan-out as a workflow: dimensions as parallel pipelines, assistants returning structured digests. The budgets, digest contract, firewall, and stopping rules apply unchanged — and however the acquisition parallelizes, digests land as files and the lead alone writes `research.md`, committing sections in plan order.
+
+## Source quality
+
+One card, applied by every assistant and the lead alike. Prefer **primary sources** — filings, regulator text, official documentation, original papers, a company's own reported numbers — over aggregators and secondary reporting. Red flags that downgrade confidence on sight: speculative language ("could", "may", projections in future tense presented as findings), marketing register, passive voice with unnamed sources, cherry-picked or unsourced numbers, and aggregators recycling a single upstream report (that's one publisher, however many domains echo it). Answer engines (Perplexity Sonar, Grok, and kin) are aggregators too, however good the synthesis: chase their citations and cite those, never the engine. Conflicts resolve by recency, consistency with adjacent established facts, and publisher quality — never by averaging.
+
+## Synthesize the dimension
+
+When a dimension's rounds are done:
+
+1. Verify at landing per `references/verification.md` — at `normal` validation this is a spot-check of the dimension's load-bearing claims, not a sweep.
+2. Write the dimension's section per the pack's skeleton from its digest files — findings woven into prose answering the dimension's questions, every load-bearing claim cited inline `[n]`, confidence flagged where below high, contradictions reported with both sides cited. Append to `research.md` and add its sources to the running source table.
+3. Log one memlog line per source batch (`--type source`) and one per load-bearing claim worth tracking for refresh — `--type claim`, text in the machine-readable shape `ref=[n] status=<verified|unverified|disputed|overturned> class=<class> pub=<YYYY-MM> — <claim>` so `scripts/recon_kit.py tally` and `staleness` can read the ledger; a later status change is a fresh claim line with the same `ref=` (last status wins).
+4. Checkpoint: one or two lines in chat — what the dimension found, anything surprising, anything unresolved. Keep moving unless the user speaks up; a mid-run scope change is logged as a `decision` and the plan adjusts. Headless: skip checkpoints entirely.
+
+When all dimensions are done, proceed to `references/synthesis.md` for final assembly.
diff --git a/.agents/skills/bmad-deep-recon/references/selection.md b/.agents/skills/bmad-deep-recon/references/selection.md
new file mode 100644
index 00000000..c4c70698
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/references/selection.md
@@ -0,0 +1,13 @@
+# The Select Shape
+
+When the decision is **choose between candidates** — technologies, vendors, libraries, platforms, agencies, anything — this method layers over whichever research type fits the subject. The type pack still governs sources, craft, and freshness; this shape governs the flow and the verdict.
+
+1. **Requirements frame.** What must the winner do, under what constraints — scale, compliance, budget, team skills, existing stack, exit-cost tolerance? Split hard gates from weighted preferences and set the weights. Sources: the project itself (brief, PRD, spine, `{workflow.persistent_facts}`, codebase) and the user — web research does not set requirements. **Agree the frame before any candidate research runs**; interactive runs confirm it even though the plan gate approved the dimension list.
+2. **Candidate screen.** Establish the credible field — leaders, strong challengers, one wildcard — and cut anything failing a hard gate. Screen to 3–5 finalists; record the cuts and why. Screening sources ≤ 6 months old — this field moves.
+3. **Evidence per criterion.** Score finalists against the frame using the type pack's dimensions and craft, verified against current versions/offerings. Cite every contested cell; where vendor claims and independent experience diverge, the divergence is a finding.
+4. **Cost & lock-in.** Total cost over the product's horizon — license/subscription, hosting, operational load, learning curve — and the cost of leaving. Current pricing pages read directly (≤ 3 mo, always); pricing-change history — a vendor that repriced once will again; migration-away accounts for real exit costs.
+5. **Verdict.** The weighted decision matrix — show the scoring, not just totals; a matrix the user can re-weight is worth more than a verdict they must trust. Then: the pick; the named runner-up and the conditions under which it wins instead; the strongest argument against the pick (from the red-team pass when it ran); the cheapest reversibility hedge (abstraction seam, pilot scope, exit test).
+
+**Two-source classes (added to the type's own):** pricing figures; performance/scale numbers; any cell that decides between the top two finalists.
+
+**Staleness:** a selection report older than two quarters should be refreshed before anyone acts on it — say so in the report.
diff --git a/.agents/skills/bmad-deep-recon/references/synthesis.md b/.agents/skills/bmad-deep-recon/references/synthesis.md
new file mode 100644
index 00000000..825d5ad2
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/references/synthesis.md
@@ -0,0 +1,16 @@
+# Synthesis
+
+The report answers the decision — whether the material came from a native Run or a processed import. **Succinct is the contract**: findings and verdicts, not essays; rationale lives in the memlog; a reader gets the decision-relevant truth in minutes. For Process mode this is the whole point — the summary is what downstream consumers read so nobody reprocesses the original, and sections with nothing behind them collapse to a line rather than pad.
+
+Assemble `research.md` in this order, shaped by `{workflow.audience}` and written in `{document_output_language}`:
+
+1. **Executive summary** — decision-first: what the evidence says to do, the two or three findings that drive that answer, and the biggest caveat. One page maximum, readable standalone. Written last, placed first.
+2. **Dimension sections** — already written during the loop; now reconciled: consistent terminology, no duplicated ground, verification statuses and any corrections from the pass applied to the text.
+3. **Cross-dimension insights** — what only the *combination* shows (e.g. the market is growing but the regulatory dimension caps the reachable segment; the technically superior option loses on ecosystem health). This section is the harness earning its keep — if there are no cross-dimension insights, say so rather than manufacture them.
+4. **Contrary evidence** — when the red-team pass ran and found material; the strongest surviving counter-arguments, cited.
+5. **Recommendations** — each bound to the decision and, where the project has them, to the downstream artifact that consumes it (per the pack's `Feeds` entries: brief section, PRD input, architecture constraint). Each recommendation names its confidence basis; a recommendation resting on low-confidence or disputed claims says so in the same sentence.
+6. **Open questions** — what the research could not answer, and what it would take to answer each.
+7. **Source appendix** — the numbered source table: `[n] | claim/finding it supports | publisher | pub date | accessed | confidence`, the publisher cell a markdown link to the source URL. Every inline `[n]` resolves here.
+8. **Staleness map** — the claims that age fastest, computed not hand-derived: build the claims list (`claim`, `class`, `pub_date`) from the ledger, map the pack's freshness bars to months per class, and run `uv run scripts/recon_kit.py staleness <claims.json> --windows '<map>'` — render its re-check dates and close by noting the earliest. This is Refresh's work order.
+
+Update the frontmatter (`status: complete`, `updated`, and the verified/unverified counts from `uv run scripts/recon_kit.py tally {doc_workspace}/.memlog.md` — never hand-counted), log a final `event` in the memlog, and proceed to `references/finalize.md`.
diff --git a/.agents/skills/bmad-deep-recon/references/verification.md b/.agents/skills/bmad-deep-recon/references/verification.md
new file mode 100644
index 00000000..080fd1b9
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/references/verification.md
@@ -0,0 +1,29 @@
+# Verification
+
+The trust layer — the same rules whatever produced the material (a native run's digests or a processed import). Verification happens **as material lands**, per dimension, in fresh-context verifier subagents reading digest files — never as an end-of-run rewrite pass over an hour of accumulated context. Late-pass rewrites degrade reports; landing-time checks improve them.
+
+## The claims ledger
+
+The memlog `claim` entries are the ledger: every claim a decision could rest on, with its class (each pack names its classes — quantitative sizes, pricing, versions/compatibility, regulatory assertions, …), source, publisher, publication date, and status. New claims enter `unverified`; on a Refresh or Deepen run, claims outside the run's scope keep their prior status from the memlog — only new and in-scope claims are (re)checked.
+
+## Levels
+
+Per the resolved `validation` level (request > knob > default `normal`):
+
+- **normal** — spot-check the **load-bearing claims only**: the handful per dimension the recommendation actually rests on. One independent-source check each, at landing. Everything else ships with its single source cited and confidence marked honestly. Fast by design.
+- **high** — cross-check every claim in the pack's *two-source classes*, and run the red-team pass on major conclusions regardless of `{workflow.red_team}`.
+- **max** — cross-check every ledger claim, run the red-team pass below at full breadth (every major conclusion), and primary-source-priority ranking: where a primary source (filing, regulator text, official docs, original paper) exists, secondary reporting alone does not verify.
+
+Verifier assistants run behind the research firewall on `{workflow.subagent_models}` when set; judgment work never drops to the smallest tier.
+
+**Independent** means a different publisher with different underlying data or reporting — not a syndication, quote, or republication of the first source, and not the same vendor's marketing in two places. An imported report counts as one publisher regardless of how many sources it cites internally; two imports from different tools agreeing is genuine confirmation, and their disagreement is a finding.
+
+Outcomes per claim: **verified** (independent source agrees within tolerance — for quantitative claims, same order of magnitude and direction), **disputed** (independent sources materially disagree — report both figures, both cited; never average), **unverified** (no independent check within budget — the claim stays, flagged, and joins the staleness map), or **overturned** (the weight of evidence contradicts it — corrected in the text, original noted). Every status change lands in the memlog as a fresh `claim` line with the same `ref=` and the new status — last status wins, which is how `scripts/recon_kit.py tally` reads the ledger. A verification outcome adjusts status and flags — it never licenses rewriting a finding's substance beyond what the new evidence says.
+
+Confidence rendered in the report: **high** (verified, fresh, credible publishers), **medium** (single credible source, fresh), **low** (stale, weak publisher, or disputed) — plus the explicit `unverified` flag. Confidence is per-claim, never per-section.
+
+## Red-team pass
+
+The single adversarial mechanism — no other verifier duplicates it. Off by default (`{workflow.red_team}` = `"off"`; `"offer"` proposes it at the plan gate, `"on"` always runs; `high` validation includes it for major conclusions, `max` runs it at full breadth). When it runs: for each major conclusion, a **fresh-context** skeptic subagent — the conclusion and a search budget, no supporting evidence, no run context — hunts for disconfirming evidence: the bear case, failed attempts, contrary data, the strongest good-faith argument the conclusion is wrong.
+
+What comes back is weighed, not appended: a conclusion that survives gets its strongest counter-argument acknowledged in the synthesis; one that doesn't is revised before the report states it. Material findings land in a **Contrary Evidence** section with full citation discipline. Zero findings after a real search is itself reportable — say what was searched for and not found.
diff --git a/.agents/skills/bmad-deep-recon/scripts/recon_kit.py b/.agents/skills/bmad-deep-recon/scripts/recon_kit.py
new file mode 100644
index 00000000..2b9759ed
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/scripts/recon_kit.py
@@ -0,0 +1,322 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.10"
+# ///
+"""recon_kit — deterministic helpers for bmad-deep-recon.
+
+The mechanical half of the research workflow: everything here is exact,
+repeatable work the LLM should never re-derive by hand. All subcommands
+print one JSON object to stdout; diagnostics go to stderr. Exit codes:
+0 = pass, 1 = findings that need attention, 2 = usage/parse error.
+
+Subcommands:
+  citations RESEARCH_MD
+      Cross-check inline [n] markers against the source-appendix table:
+      dangling markers (no appendix row) and orphaned rows (never cited).
+  tally MEMLOG_MD
+      Count memlog entries by type, and claim entries by status.
+      Claim lines carry `status=<word>` and optionally `ref=[n]`; for a
+      given ref the LAST status wins, so status changes are appends.
+  staleness CLAIMS_JSON --windows JSON [--today YYYY-MM-DD]
+      Given claims [{claim, class, pub_date}] and a months-per-class map
+      (e.g. '{"size/growth": 18, "pricing": 3}'), compute each claim's
+      re-check date, flag stale ones, and report the earliest re-check.
+  slug TOPIC --type TYPE [--pattern P] [--date YYYY-MM-DD]
+      Expand the run-folder pattern deterministically so the same topic
+      always lands in the same folder across draft -> process -> refresh.
+  escape-sources RESEARCH_MD
+      Emit the source-appendix table as HTML with every cell escaped and
+      only validated http(s) URLs turned into links, for the briefing.
+"""
+from __future__ import annotations
+
+import argparse
+import calendar
+import html
+import json
+import re
+import sys
+import unicodedata
+from datetime import date, datetime
+from pathlib import Path
+from urllib.parse import urlparse
+
+MARKER_RE = re.compile(r"\[(\d+)\](?!\()")  # [3] but not a [3](url) link
+MD_LINK_RE = re.compile(r"\[([^\]]*)\]\((\S+?)\)")
+BARE_URL_RE = re.compile(r"https?://[^\s|)\]]+")
+
+
+def out(payload: dict, exit_code: int) -> int:
+    print(json.dumps(payload, indent=2, ensure_ascii=False, default=str))
+    return exit_code
+
+
+def read_text(path_arg: str) -> str:
+    if path_arg == "-":
+        return sys.stdin.read()
+    return Path(path_arg).read_text(encoding="utf-8")
+
+
+def strip_fences(text: str) -> str:
+    """Blank out fenced code blocks so their contents never count as markers or rows."""
+    lines, fenced = [], False
+    for ln in text.splitlines():
+        if ln.lstrip().startswith("```"):
+            fenced = not fenced
+            lines.append("")
+            continue
+        lines.append("" if fenced else ln)
+    return "\n".join(lines)
+
+
+def table_cells(line: str) -> list[str]:
+    return [c.strip() for c in line.strip().strip("|").split("|")]
+
+
+def appendix_rows(text: str) -> dict[int, list[str]]:
+    """Source-appendix rows: markdown table rows whose first cell is a bare [n] / n."""
+    rows: dict[int, list[str]] = {}
+    for ln in text.splitlines():
+        stripped = ln.strip()
+        if not stripped.startswith("|"):
+            continue
+        cells = table_cells(stripped)
+        if not cells or len(cells) < 2:
+            continue
+        m = re.fullmatch(r"\[?(\d+)\]?", cells[0])
+        if m:
+            rows[int(m.group(1))] = cells
+    return rows
+
+
+# --- citations ---------------------------------------------------------------
+
+def cmd_citations(args) -> int:
+    text = strip_fences(read_text(args.file))
+    rows = appendix_rows(text)
+    markers: set[int] = set()
+    for ln in text.splitlines():
+        stripped = ln.strip()
+        if stripped.startswith("|"):
+            cells = table_cells(stripped)
+            if cells and re.fullmatch(r"\[?(\d+)\]?", cells[0]):
+                continue  # an appendix row is not a citation of itself
+        markers.update(int(n) for n in MARKER_RE.findall(ln))
+    dangling = sorted(markers - set(rows))
+    orphaned = sorted(set(rows) - markers)
+    ok = not dangling and not orphaned
+    return out({
+        "markers": sorted(markers),
+        "appendix_rows": sorted(rows),
+        "dangling_markers": dangling,
+        "orphaned_rows": orphaned,
+        "ok": ok,
+    }, 0 if ok else 1)
+
+
+# --- tally -------------------------------------------------------------------
+
+ENTRY_RE = re.compile(r"^- (?:\(([\w-]+)(?: by [^)]*)?\)\s*)?(.*)$")
+
+
+def cmd_tally(args) -> int:
+    text = read_text(args.file)
+    body = text.split("---", 2)[-1] if text.startswith("---") else text
+    by_type: dict[str, int] = {}
+    by_ref: dict[int, str] = {}
+    unref_status: dict[str, int] = {}
+    entries = 0
+    for ln in body.splitlines():
+        m = ENTRY_RE.match(ln)
+        if not m or not ln.startswith("- "):
+            continue
+        entries += 1
+        etype = m.group(1) or "note"
+        by_type[etype] = by_type.get(etype, 0) + 1
+        if etype == "claim":
+            status_m = re.search(r"status=([\w-]+)", m.group(2))
+            status = status_m.group(1) if status_m else "unknown"
+            ref_m = re.search(r"ref=\[?(\d+)\]?", m.group(2))
+            if ref_m:
+                by_ref[int(ref_m.group(1))] = status  # last status wins per ref
+            else:
+                unref_status[status] = unref_status.get(status, 0) + 1
+    claims: dict[str, int] = dict(unref_status)
+    for status in by_ref.values():
+        claims[status] = claims.get(status, 0) + 1
+    return out({
+        "entries": entries,
+        "by_type": dict(sorted(by_type.items())),
+        "claims": dict(sorted(claims.items())),
+        "claims_total": sum(claims.values()),
+    }, 0)
+
+
+# --- staleness ---------------------------------------------------------------
+
+def parse_date(raw: str) -> date:
+    raw = raw.strip()
+    for fmt in ("%Y-%m-%d", "%Y-%m", "%Y"):
+        try:
+            return datetime.strptime(raw, fmt).date()
+        except ValueError:
+            continue
+    raise ValueError(f"unparseable date: {raw!r} (want YYYY[-MM[-DD]])")
+
+
+def add_months(d: date, months: int) -> date:
+    total = d.month - 1 + months
+    year, month = d.year + total // 12, total % 12 + 1
+    return date(year, month, min(d.day, calendar.monthrange(year, month)[1]))
+
+
+def cmd_staleness(args) -> int:
+    try:
+        payload = json.loads(read_text(args.file))
+        windows = {k.lower(): int(v) for k, v in json.loads(args.windows).items()}
+        today = parse_date(args.today) if args.today else date.today()
+    except (ValueError, json.JSONDecodeError) as e:
+        print(f"error: {e}", file=sys.stderr)
+        return 2
+    claims = payload["claims"] if isinstance(payload, dict) else payload
+    results, no_window, stale_count = [], set(), 0
+    earliest: date | None = None
+    for c in claims:
+        cls = str(c.get("class", "")).lower()
+        try:
+            pub = parse_date(str(c["pub_date"]))
+        except (KeyError, ValueError) as e:
+            print(f"error in claim {c!r}: {e}", file=sys.stderr)
+            return 2
+        months = windows.get(cls)
+        if months is None:
+            no_window.add(cls)
+            results.append({**c, "recheck": None, "stale": None})
+            continue
+        recheck = add_months(pub, months)
+        stale = recheck <= today
+        stale_count += stale
+        earliest = recheck if earliest is None or recheck < earliest else earliest
+        results.append({**c, "recheck": recheck.isoformat(), "stale": stale})
+    return out({
+        "today": today.isoformat(),
+        "claims": results,
+        "stale_count": stale_count,
+        "earliest_recheck": earliest.isoformat() if earliest else None,
+        "no_window_classes": sorted(no_window),
+    }, 1 if stale_count else 0)
+
+
+# --- slug --------------------------------------------------------------------
+
+def slugify(text: str, max_len: int = 40) -> str:
+    text = unicodedata.normalize("NFKD", text).encode("ascii", "ignore").decode()
+    text = re.sub(r"[^a-z0-9]+", "-", text.lower()).strip("-")
+    return re.sub(r"-{2,}", "-", text)[:max_len].rstrip("-")
+
+
+def cmd_slug(args) -> int:
+    slug = slugify(args.topic)
+    if not slug:
+        print("error: topic slugified to an empty string", file=sys.stderr)
+        return 2
+    folder = (args.pattern
+              .replace("{research_type}", args.type)
+              .replace("{topic_slug}", slug)
+              .replace("{date}", args.date or date.today().isoformat()))
+    return out({"topic_slug": slug, "folder": folder}, 0)
+
+
+# --- escape-sources ----------------------------------------------------------
+
+def safe_url(raw: str) -> str | None:
+    parsed = urlparse(raw)
+    return raw if parsed.scheme in ("http", "https") and parsed.netloc else None
+
+
+def cell_html(cell: str, invalid: list[str]) -> str:
+    """Escape a cell; a markdown link or bare URL becomes an <a> only when http(s)."""
+    link = MD_LINK_RE.search(cell)
+    if link:
+        url = safe_url(link.group(2))
+        label = html.escape(link.group(1) or link.group(2))
+        if url:
+            return html.escape(cell[:link.start()]) + \
+                f'<a href="{html.escape(url, quote=True)}" target="_blank" rel="noopener">{label}</a>' + \
+                html.escape(cell[link.end():])
+        invalid.append(link.group(2))
+        return html.escape(cell.replace(link.group(0), link.group(1) or link.group(2)))
+    bare = BARE_URL_RE.search(cell)
+    if bare:
+        url = safe_url(bare.group(0))
+        if url:
+            escaped = html.escape(url, quote=True)
+            return html.escape(cell[:bare.start()]) + \
+                f'<a href="{escaped}" target="_blank" rel="noopener">{escaped}</a>' + \
+                html.escape(cell[bare.end():])
+        invalid.append(bare.group(0))
+    return html.escape(cell)
+
+
+def cmd_escape_sources(args) -> int:
+    text = strip_fences(read_text(args.file))
+    rows = appendix_rows(text)
+    if not rows:
+        print("error: no source-appendix table rows found", file=sys.stderr)
+        return 2
+    invalid: list[str] = []
+    body_rows = []
+    for n in sorted(rows):
+        cells = rows[n]
+        tds = "".join(f"<td>{cell_html(c, invalid)}</td>" for c in cells[1:])
+        body_rows.append(f'<tr id="src-{n}"><td>[{n}]</td>{tds}</tr>')
+    table = ('<table class="sources"><tbody>' + "".join(body_rows) + "</tbody></table>")
+    return out({"rows": len(rows), "invalid_urls": invalid, "html": table},
+               1 if invalid else 0)
+
+
+# --- entry point -------------------------------------------------------------
+
+def main(argv: list[str] | None = None) -> int:
+    p = argparse.ArgumentParser(description=__doc__,
+                                formatter_class=argparse.RawDescriptionHelpFormatter)
+    sub = p.add_subparsers(dest="cmd", required=True)
+
+    pc = sub.add_parser("citations", help="cross-check [n] markers vs the source appendix")
+    pc.add_argument("file", help="path to research.md (or - for stdin)")
+    pc.set_defaults(func=cmd_citations)
+
+    pt = sub.add_parser("tally", help="count memlog entries by type and claims by status")
+    pt.add_argument("file", help="path to .memlog.md (or - for stdin)")
+    pt.set_defaults(func=cmd_tally)
+
+    ps = sub.add_parser("staleness", help="compute re-check dates from freshness windows")
+    ps.add_argument("file", help="claims JSON: [{claim, class, pub_date}] (or - for stdin)")
+    ps.add_argument("--windows", required=True,
+                    help='JSON months-per-class map, e.g. \'{"pricing": 3}\'')
+    ps.add_argument("--today", help="override today's date (YYYY-MM-DD)")
+    ps.set_defaults(func=cmd_staleness)
+
+    pg = sub.add_parser("slug", help="expand the run-folder pattern deterministically")
+    pg.add_argument("topic", help="research topic text")
+    pg.add_argument("--type", required=True, help="research type code (e.g. market)")
+    pg.add_argument("--pattern", default="{research_type}-{topic_slug}-{date}",
+                    help="folder pattern (default: {research_type}-{topic_slug}-{date})")
+    pg.add_argument("--date", help="override date (YYYY-MM-DD; default today)")
+    pg.set_defaults(func=cmd_slug)
+
+    pe = sub.add_parser("escape-sources",
+                        help="source appendix as escaped HTML with validated links")
+    pe.add_argument("file", help="path to research.md (or - for stdin)")
+    pe.set_defaults(func=cmd_escape_sources)
+
+    args = p.parse_args(argv)
+    try:
+        return args.func(args)
+    except FileNotFoundError as e:
+        print(f"error: {e}", file=sys.stderr)
+        return 2
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/.agents/skills/bmad-deep-recon/scripts/tests/test_recon_kit.py b/.agents/skills/bmad-deep-recon/scripts/tests/test_recon_kit.py
new file mode 100644
index 00000000..217f534a
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/scripts/tests/test_recon_kit.py
@@ -0,0 +1,144 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.10"
+# ///
+"""Tests for recon_kit.py."""
+
+import io
+import json
+import sys
+import unittest
+from contextlib import redirect_stdout
+from datetime import date
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+
+from recon_kit import add_months, main, parse_date, slugify
+
+REPORT = """---
+title: 'market research: widgets'
+---
+
+# Report
+
+The market is growing fast [1] and pricing clusters at $20 [2].
+An uncited assertion sits here [4].
+
+```
+[9] inside a fence never counts
+```
+
+## Source appendix
+
+| [n] | Supports | Publisher | Pub date | Accessed | Confidence |
+| --- | --- | --- | --- | --- | --- |
+| [1] | market growth | [Gartner](https://example.com/g) | 2026-01 | 2026-07-01 | high |
+| [2] | pricing | [Acme](javascript:alert(1)) | 2026-05 | 2026-07-01 | medium |
+| [3] | never cited | https://example.com/x | 2025-11 | 2026-07-01 | low |
+"""
+
+MEMLOG = """---
+topic: widgets
+updated: 2026-07-22T10:00
+---
+
+- (decision) plan approved
+- (source) round 1 batch: 4 sources
+- (claim) ref=[1] status=verified class=size/growth pub=2026-01 — market growing 12% CAGR
+- (claim) ref=[2] status=unverified class=pricing pub=2026-05 — pricing clusters at $20
+- (claim) ref=[2] status=verified class=pricing pub=2026-05 — confirmed by second source
+- (claim) status=unverified class=behavior pub=2025-03 — users churn at day 8
+- (event) dimension 1 complete
+"""
+
+
+def run(argv):
+    buf = io.StringIO()
+    with redirect_stdout(buf):
+        code = main(argv)
+    return code, json.loads(buf.getvalue())
+
+
+class CitationsTest(unittest.TestCase):
+    def test_cross_check(self):
+        report = Path(__file__).parent / "_report.md"
+        report.write_text(REPORT, encoding="utf-8")
+        try:
+            code, result = run(["citations", str(report)])
+        finally:
+            report.unlink()
+        self.assertEqual(result["dangling_markers"], [4])
+        self.assertEqual(result["orphaned_rows"], [3])
+        self.assertNotIn(9, result["markers"])  # fenced content ignored
+        self.assertEqual(code, 1)
+
+
+class TallyTest(unittest.TestCase):
+    def test_last_status_wins_per_ref(self):
+        log = Path(__file__).parent / "_memlog.md"
+        log.write_text(MEMLOG, encoding="utf-8")
+        try:
+            code, result = run(["tally", str(log)])
+        finally:
+            log.unlink()
+        self.assertEqual(result["by_type"]["claim"], 4)
+        self.assertEqual(result["claims"], {"unverified": 1, "verified": 2})
+        self.assertEqual(result["claims_total"], 3)  # ref=[2] counted once
+        self.assertEqual(code, 0)
+
+
+class StalenessTest(unittest.TestCase):
+    def test_dates(self):
+        self.assertEqual(parse_date("2026-01"), date(2026, 1, 1))
+        self.assertEqual(add_months(date(2026, 1, 31), 1), date(2026, 2, 28))
+
+    def test_windows(self):
+        claims = json.dumps([
+            {"claim": "sizing", "class": "size/growth", "pub_date": "2024-06"},
+            {"claim": "pricing", "class": "pricing", "pub_date": "2026-06"},
+            {"claim": "odd", "class": "unmapped", "pub_date": "2026-06"},
+        ])
+        f = Path(__file__).parent / "_claims.json"
+        f.write_text(claims, encoding="utf-8")
+        try:
+            code, result = run([
+                "staleness", str(f),
+                "--windows", '{"size/growth": 18, "pricing": 3}',
+                "--today", "2026-07-22",
+            ])
+        finally:
+            f.unlink()
+        self.assertEqual(result["stale_count"], 1)  # sizing recheck 2025-12 < today
+        self.assertEqual(result["earliest_recheck"], "2025-12-01")
+        self.assertEqual(result["no_window_classes"], ["unmapped"])
+        self.assertEqual(code, 1)
+
+
+class SlugTest(unittest.TestCase):
+    def test_deterministic_folder(self):
+        self.assertEqual(slugify("Créme Brûlée: AI Tools!"), "creme-brulee-ai-tools")
+        code, result = run(["slug", "SMB Accounting SaaS", "--type", "market",
+                            "--date", "2026-07-22"])
+        self.assertEqual(result["folder"], "market-smb-accounting-saas-2026-07-22")
+        self.assertEqual(code, 0)
+
+
+class EscapeSourcesTest(unittest.TestCase):
+    def test_escaping_and_url_validation(self):
+        report = Path(__file__).parent / "_report.md"
+        report.write_text(REPORT, encoding="utf-8")
+        try:
+            code, result = run(["escape-sources", str(report)])
+        finally:
+            report.unlink()
+        self.assertEqual(result["rows"], 3)
+        self.assertTrue(any(u.startswith("javascript:") for u in result["invalid_urls"]))
+        self.assertNotIn("javascript:", result["html"])  # never linked
+        self.assertIn('href="https://example.com/g"', result["html"])
+        self.assertIn('id="src-1"', result["html"])
+        self.assertEqual(code, 1)
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/.agents/skills/bmad-deep-recon/types/academic-lit.md b/.agents/skills/bmad-deep-recon/types/academic-lit.md
new file mode 100644
index 00000000..c81a55bd
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/types/academic-lit.md
@@ -0,0 +1,19 @@
+# Academic Literature Pack
+
+Serves: ground an approach in published research, scan the state of the art, run a defensible literature review, cite properly in technical writing.
+
+**Dimensions (priority order — prune to the decision):**
+
+1. The canon — seminal papers and the best recent surveys of the area
+2. State of the art — current best results, benchmarks, and how they're measured
+3. Methods & limitations — what the leading approaches assume and where they break
+4. Open problems & live debates — what the field disagrees about right now
+5. Who works on this — the labs and groups whose output to watch
+
+**Craft (the non-obvious):** find one good survey before reading twenty abstracts; chase citations both directions — who they cite and who cites them (Semantic Scholar/Google Scholar); label preprint vs peer-reviewed on every citation — arXiv is not acceptance; take benchmark numbers from the original paper, never from a competitor's comparison table; check retraction and replication status on any load-bearing empirical claim; a result only ever shown by one lab is a lead, not a fact.
+
+**Freshness:** state-of-the-art claims ≤ 12 mo (ML ≤ 6 mo) · seminal work has no freshness bar — but check whether it was since superseded.
+
+**Two-source classes:** any empirical claim a conclusion rests on — independent replication or corroboration, not the same lab twice.
+
+**Feeds (bmm):** technical and architecture bets · content and writing that cites · build-vs-adopt judgments on research-grade techniques.
diff --git a/.agents/skills/bmad-deep-recon/types/competitive.md b/.agents/skills/bmad-deep-recon/types/competitive.md
new file mode 100644
index 00000000..ee504dbf
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/types/competitive.md
@@ -0,0 +1,19 @@
+# Competitive Research Pack
+
+Serves: position against *named* competitors, build battlecards, sharpen differentiation, anticipate their next move. (Surveying an unnamed field is the market type; this pack is for teardowns of specific players.)
+
+**Dimensions (priority order — prune to the decision):**
+
+1. Offer & feature teardown — what they actually ship, tried directly where possible
+2. Pricing & packaging — models, tiers, what changed recently and which direction
+3. Positioning & messaging — who they claim to serve, the story they tell, the gap between claim and product
+4. Trajectory — funding, hiring, release cadence: where they're headed
+5. Their customers' voice — what users of *their* product praise and complain about
+
+**Craft (the non-obvious):** their changelog and release notes are roadmap truth; job postings reveal strategy six months early; their customers' 1–3★ reviews are your wedge; archived pricing pages (Wayback) show pricing direction, not just position; sales-facing comparison pages overclaim — verify capability claims against their docs; try the product yourself when a trial exists — an hour in-product beats ten reviews.
+
+**Freshness:** pricing & features ≤ 3 mo · trajectory signals ≤ 6 mo · customer sentiment ≤ 12 mo.
+
+**Two-source classes:** traction and market-share claims; any capability claim of theirs that your differentiation rests on.
+
+**Feeds (bmm):** brief (alternatives) · PRD (differentiation) · GTM battlecards and positioning.
diff --git a/.agents/skills/bmad-deep-recon/types/domain.md b/.agents/skills/bmad-deep-recon/types/domain.md
new file mode 100644
index 00000000..a656c1d5
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/types/domain.md
@@ -0,0 +1,19 @@
+# Domain Research Pack
+
+Serves: commit to building in an industry, talk credibly with domain experts, scope a product for a regulated field, brief a team entering unfamiliar territory.
+
+**Dimensions (priority order — prune to the decision):**
+
+1. Industry structure & value chain — how value flows, who captures margin where
+2. Key players & gatekeepers — incumbents, platforms, whose APIs/standards/marketplaces you build with or against
+3. Rules of the game — laws, licenses, de-facto standards, what compliance costs a new entrant
+4. Language & mental models — the vocabulary and implicit workflows practitioners think in; **build the glossary — it's why domain research exists**
+5. Technology adoption — the current technical baseline and where the industry sits on the adoption curve
+
+**Craft (the non-obvious):** annual-report industry sections are free structured teardowns; conference keynotes reveal who actually matters; go to regulator sites directly for any load-bearing claim — and read enforcement actions to learn what's actually punished versus merely written; job postings name the real tools and skills; pending regulatory changes matter as much as current text.
+
+**Freshness:** structure ≤ 3 yr · player landscape ≤ 18 mo · regulatory status: verify current on every load-bearing claim, whatever its date · tech adoption ≤ 18 mo, AI-adoption claims ≤ 6 mo.
+
+**Two-source classes:** regulatory and compliance assertions; quantitative industry figures a recommendation rests on; claims about a gatekeeper's policy.
+
+**Feeds (bmm):** brief (context, feasibility) · PRD (constraints, domain vocabulary) · architecture (integration landscape, compliance requirements).
diff --git a/.agents/skills/bmad-deep-recon/types/market.md b/.agents/skills/bmad-deep-recon/types/market.md
new file mode 100644
index 00000000..18d38e52
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/types/market.md
@@ -0,0 +1,19 @@
+# Market Research Pack
+
+Serves: enter or skip a market, position a product, pick a segment, price an offer, pitch investors.
+
+**Dimensions (priority order — prune to the decision):**
+
+1. Market size & growth — the *reachable* market, not the headline TAM
+2. Customer segments & behavior — who buys, deciding how, valuing what
+3. Pain points & unmet needs — what they complain about, work around, pay to avoid
+4. Competitive landscape — who competes for this budget, including substitutes and "do nothing"
+5. GTM & pricing dynamics — channels, sales motion, accepted pricing models, realistic CAC
+
+**Craft (the non-obvious):** 1–3★ reviews are the gold for pains; public-company 10-K/S-1 industry sections are free analyst-grade sizing; read competitor pricing pages directly, never roundups; funding history + job postings reveal competitor trajectory; a complaint pattern persisting across years is a stronger finding, not a stale one.
+
+**Freshness:** size/growth ≤ 18 mo · pricing & feature claims ≤ 3 mo · behavior data ≤ 2 yr · GTM benchmarks ≤ 12 mo.
+
+**Two-source classes:** market size and growth figures; any quantitative claim a recommendation rests on; competitor traction claims.
+
+**Feeds (bmm):** brief (opportunity, problem, users) · PRD (personas, differentiation) · pricing and GTM decisions.
diff --git a/.agents/skills/bmad-deep-recon/types/technical.md b/.agents/skills/bmad-deep-recon/types/technical.md
new file mode 100644
index 00000000..78266000
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/types/technical.md
@@ -0,0 +1,19 @@
+# Technical Research Pack
+
+Serves: adopt a technology area, design an integration approach, ground an architecture in current practice, assess feasibility before committing a roadmap.
+
+**Dimensions (priority order — prune to the decision):**
+
+1. Landscape & maturity — dominant approaches, what's consolidating vs churning, what the current generation newly makes possible
+2. Integration & interoperability — protocols, formats, auth patterns, where integrations actually hurt
+3. Architecture patterns in practice — which named patterns dominate at what scale, and what the failures teach
+4. Implementation reality — learning curve, tooling, operational burden, what teams say 6–12 months in
+5. Ecosystem health — contributor/release vitality, backing durability, the five-year regret risk
+
+**Craft (the non-obvious):** read the retrospective threads, not the launch threads; favor accounts with production numbers over advocacy; before citing a pain point, check whether it was since fixed — an old complaint against a current version is a false claim; read repository metrics over time, never snapshots; issue trackers reveal the gap between docs and reality.
+
+**Freshness:** versions & compatibility ≤ 1 mo · ecosystem signals ≤ 6 mo · landscape ≤ 12 mo (AI-adjacent ≤ 3 mo) · patterns ≤ 2 yr.
+
+**Two-source classes:** version/compatibility claims; performance or scale numbers a recommendation rests on; claims that a technology or pattern failed — one post-mortem is an anecdote.
+
+**Feeds (bmm):** architecture spine (candidate paradigms, operational constraints) · brief (feasibility) · roadmap risk and estimates.
diff --git a/.agents/skills/bmad-deep-recon/types/user-voice.md b/.agents/skills/bmad-deep-recon/types/user-voice.md
new file mode 100644
index 00000000..9c195c50
--- /dev/null
+++ b/.agents/skills/bmad-deep-recon/types/user-voice.md
@@ -0,0 +1,19 @@
+# User-Voice Research Pack
+
+Serves: understand what users of a product or category actually experience and want — personas, jobs-to-be-done, requirements grounded in evidence rather than assumption.
+
+**Dimensions (priority order — prune to the decision):**
+
+1. Who they are & their jobs-to-be-done — the progress they're hiring the product to make
+2. Complaint & workaround patterns — where current options fail them
+3. Delight & switching triggers — why they stay, what made them move
+4. Unmet needs & requests — what they ask for, and the deeper need under the ask
+5. Their language — the words users say, versus the words vendors use
+
+**Craft (the non-obvious):** mine 1–3★ reviews for pain *and* 5★ for why they stay; a workaround is unpriced demand — someone laboring around a gap has already voted; forums, Discord, and Reddit surface what surveys miss — people lie less when nobody's asking — but they over-sample the loud, so triangulate against reviews and any survey data; keep verbatim quotes, redacted — user words carry evidence paraphrase destroys, but usernames, handles, emails, and identifying links never enter the report or memlog; feature-request boards measure willingness to wait, not willingness to pay; distinguish loud power-users from the silent majority — count distinct voices, not thread length.
+
+**Freshness:** sentiment ≤ 18 mo · complaints re-checked against the current version before citing.
+
+**Two-source classes:** any prevalence claim ("most users…", "the top complaint is…") — two independent communities, not two threads in the same one.
+
+**Feeds (bmm):** PRD (personas, requirements rationale) · UX research inputs · brief (problem) · product copy in the users' own language.
diff --git a/.agents/skills/bmad-dev-auto/SKILL.md b/.agents/skills/bmad-dev-auto/SKILL.md
new file mode 100644
index 00000000..6a7fe6d7
--- /dev/null
+++ b/.agents/skills/bmad-dev-auto/SKILL.md
@@ -0,0 +1,19 @@
+---
+name: bmad-dev-auto
+description: "Deprecated: forwards to bmad-build-auto. Do not use unless invoked by name."
+---
+
+# Deprecated Build Auto Alias
+
+## On Activation
+
+1. Check whether either legacy customization file exists:
+   - `{project-root}/_bmad/custom/bmad-dev-auto.toml`
+   - `{project-root}/_bmad/custom/bmad-dev-auto.user.toml`
+2. If neither legacy file exists, output exactly `bmad-dev-auto is deprecated. Redirecting to bmad-build-auto. Please use bmad-build-auto in the future.`, invoke `bmad-build-auto` exactly once with the user's original input verbatim, then execute no further steps in this shim.
+3. For every legacy file that exists, use its matching new filename:
+   - `{project-root}/_bmad/custom/bmad-dev-auto.toml` becomes `{project-root}/_bmad/custom/bmad-build-auto.toml`.
+   - `{project-root}/_bmad/custom/bmad-dev-auto.user.toml` becomes `{project-root}/_bmad/custom/bmad-build-auto.user.toml`.
+4. If the matching new file does not exist, tell the user that the customization file uses the deprecated name and offer to rename it. Rename it only after explicit approval. If approval is declined or unavailable, or the rename fails, HALT and do not invoke any skill.
+5. If the matching new file already exists, do not overwrite it. Read both files, explain their differences, and propose the exact content for the new file. Resolve conflicting values with the user. Only after the user explicitly approves that content, save and verify the new file, then remove the legacy file. If approval is declined or unavailable, or any operation fails, HALT and do not invoke any skill.
+6. After every detected legacy file has been migrated successfully and no legacy file remains, output exactly `bmad-dev-auto is deprecated. Redirecting to bmad-build-auto. Please use bmad-build-auto in the future.`, invoke `bmad-build-auto` exactly once with the user's original input verbatim, then execute no further steps in this shim.
diff --git a/.agents/skills/bmad-dev-story/SKILL.md b/.agents/skills/bmad-dev-story/SKILL.md
new file mode 100644
index 00000000..c8bcc8af
--- /dev/null
+++ b/.agents/skills/bmad-dev-story/SKILL.md
@@ -0,0 +1,502 @@
+---
+name: bmad-dev-story
+description: "Deprecated: `bmad-build` is now the official implementation method. Only use this when explicitly invoked by name."
+---
+
+# Dev Story Workflow
+
+**Goal:** Execute story implementation following a context filled story spec file.
+
+**Your Role:** Developer implementing the story.
+- Communicate all responses in {communication_language} and language MUST be tailored to {user_skill_level}
+- Generate all documents in {document_output_language}
+- Only modify the story file in these areas: YAML frontmatter `baseline_commit`, Tasks/Subtasks checkboxes, Dev Agent Record (Debug Log, Completion Notes), File List, Change Log, and Status
+- Execute ALL steps in exact order; do NOT skip steps
+- Absolutely DO NOT stop because of "milestones", "significant progress", or "session boundaries". Continue in a single execution until the story is COMPLETE (all ACs satisfied and all tasks/subtasks checked) UNLESS a HALT condition is triggered or the USER gives other instruction.
+- Do NOT schedule a "next session" or request review pauses unless a HALT condition applies. Only Step 9 decides completion.
+- User skill level ({user_skill_level}) affects conversation style ONLY, not code updates.
+
+## Conventions
+
+- Bare paths (e.g. `steps/step-01-init.md`) resolve from the skill root.
+- `{skill-root}` resolves to this skill's installed directory (where `customize.toml` lives).
+- `{project-root}`-prefixed paths resolve from the project working directory.
+- `{skill-name}` resolves to the skill directory's basename.
+
+## On Activation
+
+### Step 1: Resolve the Workflow Block
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`
+
+**If the script fails**, resolve the `workflow` block yourself by reading these three files in base → team → user order and applying the same structural merge rules as the resolver:
+
+1. `{skill-root}/customize.toml` — defaults
+2. `{project-root}/_bmad/custom/{skill-name}.toml` — team overrides
+3. `{project-root}/_bmad/custom/{skill-name}.user.toml` — personal overrides
+
+Any missing file is skipped. Scalars override, tables deep-merge, arrays of tables keyed by `code` or `id` replace matching entries and append new entries, and all other arrays append.
+
+### Step 2: Execute Prepend Steps
+
+Execute each entry in `{workflow.activation_steps_prepend}` in order before proceeding.
+
+### Step 3: Load Persistent Facts
+
+Treat every entry in `{workflow.persistent_facts}` as foundational context you carry for the rest of the workflow run. Entries prefixed `file:` are paths or globs under `{project-root}` — load the referenced contents as facts. All other entries are facts verbatim.
+
+### Step 4: Load Config
+
+Load config from `{project-root}/_bmad/bmm/config.yaml` and resolve:
+
+- `project_name`, `user_name`
+- `communication_language`, `document_output_language`
+- `user_skill_level`
+- `implementation_artifacts`
+- `date` as system-generated current datetime
+- `project_context` = `**/project-context.md` (load if exists)
+
+### Step 5: Greet the User
+
+Greet `{user_name}`, speaking in `{communication_language}`.
+
+<output>Deprecated: `bmad-build` is now the official implementation method. Only use this when explicitly invoked by name.</output>
+
+### Step 6: Execute Append Steps
+
+Execute each entry in `{workflow.activation_steps_append}` in order.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+## Paths
+
+- `story_file` = `` (explicit story path; auto-discovered if empty)
+- `sprint_status` = `{implementation_artifacts}/sprint-status.yaml`
+
+## Execution
+
+<workflow>
+  <critical>Communicate all responses in {communication_language} and language MUST be tailored to {user_skill_level}</critical>
+  <critical>Generate all documents in {document_output_language}</critical>
+  <critical>Only modify the story file in these areas: YAML frontmatter `baseline_commit`, Tasks/Subtasks checkboxes, Dev Agent Record (Debug Log, Completion Notes), File List,
+    Change Log, and Status</critical>
+  <critical>Execute ALL steps in exact order; do NOT skip steps</critical>
+  <critical>Absolutely DO NOT stop because of "milestones", "significant progress", or "session boundaries". Continue in a single execution
+    until the story is COMPLETE (all ACs satisfied and all tasks/subtasks checked) UNLESS a HALT condition is triggered or the USER gives
+    other instruction.</critical>
+  <critical>Do NOT schedule a "next session" or request review pauses unless a HALT condition applies. Only Step 9 decides completion.</critical>
+  <critical>User skill level ({user_skill_level}) affects conversation style ONLY, not code updates.</critical>
+
+  <step n="1" goal="Find next ready story and load it" tag="sprint-status">
+    <check if="{{story_path}} is provided">
+      <action>Use {{story_path}} directly</action>
+      <action>Read COMPLETE story file</action>
+      <action>Extract story_key from filename or metadata</action>
+      <goto anchor="task_check" />
+    </check>
+
+    <!-- Sprint-based story discovery -->
+    <check if="{{sprint_status}} file exists">
+      <critical>MUST read COMPLETE sprint-status.yaml file from start to end to preserve order</critical>
+      <action>Load the FULL file: {{sprint_status}}</action>
+      <action>Read ALL lines from beginning to end - do not skip any content</action>
+      <action>Parse the development_status section completely to understand story order</action>
+
+      <action>Find the FIRST story (by reading in order from top to bottom) where:
+        - Key matches pattern: number-number-name (e.g., "1-2-user-auth")
+        - NOT an epic key (epic-X) or retrospective (epic-X-retrospective)
+        - Status value equals "ready-for-dev"
+      </action>
+
+      <check if="no ready-for-dev or in-progress story found">
+        <output>📋 No ready-for-dev stories found in sprint-status.yaml
+
+          **Current Sprint Status:** {{sprint_status_summary}}
+
+          **What would you like to do?**
+          1. Run `create-story` to create next story from epics with comprehensive context
+          2. Run `*validate-create-story` to improve existing stories before development (recommended quality check)
+          3. Specify a particular story file to develop (provide full path)
+          4. Check {{sprint_status}} file to see current sprint status
+
+          💡 **Tip:** Stories in `ready-for-dev` may not have been validated. Consider running `validate-create-story` first for a quality
+          check.
+        </output>
+        <ask>Choose option [1], [2], [3], or [4], or specify story file path:</ask>
+
+        <check if="user chooses '1'">
+          <action>HALT - Run create-story to create next story</action>
+        </check>
+
+        <check if="user chooses '2'">
+          <action>HALT - Run validate-create-story to improve existing stories</action>
+        </check>
+
+        <check if="user chooses '3'">
+          <ask>Provide the story file path to develop:</ask>
+          <action>Store user-provided story path as {{story_path}}</action>
+          <goto anchor="task_check" />
+        </check>
+
+        <check if="user chooses '4'">
+          <output>Loading {{sprint_status}} for detailed status review...</output>
+          <action>Display detailed sprint status analysis</action>
+          <action>HALT - User can review sprint status and provide story path</action>
+        </check>
+
+        <check if="user provides story file path">
+          <action>Store user-provided story path as {{story_path}}</action>
+          <goto anchor="task_check" />
+        </check>
+      </check>
+    </check>
+
+    <!-- Non-sprint story discovery -->
+    <check if="{{sprint_status}} file does NOT exist">
+      <action>Search {implementation_artifacts} for stories directly</action>
+      <action>Find stories with "ready-for-dev" status in files</action>
+      <action>Look for story files matching pattern: *-*-*.md</action>
+      <action>Read each candidate story file to check Status section</action>
+
+      <check if="no ready-for-dev stories found in story files">
+        <output>📋 No ready-for-dev stories found
+
+          **Available Options:**
+          1. Run `create-story` to create next story from epics with comprehensive context
+          2. Run `*validate-create-story` to improve existing stories
+          3. Specify which story to develop
+        </output>
+        <ask>What would you like to do? Choose option [1], [2], or [3]:</ask>
+
+        <check if="user chooses '1'">
+          <action>HALT - Run create-story to create next story</action>
+        </check>
+
+        <check if="user chooses '2'">
+          <action>HALT - Run validate-create-story to improve existing stories</action>
+        </check>
+
+        <check if="user chooses '3'">
+          <ask>It's unclear what story you want developed. Please provide the full path to the story file:</ask>
+          <action>Store user-provided story path as {{story_path}}</action>
+          <action>Continue with provided story file</action>
+        </check>
+      </check>
+
+      <check if="ready-for-dev story found in files">
+        <action>Use discovered story file and extract story_key</action>
+      </check>
+    </check>
+
+    <action>Store the found story_key (e.g., "1-2-user-authentication") for later status updates</action>
+    <action>Find matching story file in {implementation_artifacts} using story_key pattern: {{story_key}}.md</action>
+    <action>Read COMPLETE story file from discovered path</action>
+
+    <anchor id="task_check" />
+
+    <action>Parse sections: Story, Acceptance Criteria, Tasks/Subtasks, Dev Notes, Dev Agent Record, File List, Change Log, Status</action>
+
+    <action>Load comprehensive context from story file's Dev Notes section</action>
+    <action>Extract developer guidance from Dev Notes: architecture requirements, previous learnings, technical specifications</action>
+    <action>Use enhanced story context to inform implementation decisions and approaches</action>
+
+    <action>Identify first incomplete task (unchecked [ ]) in Tasks/Subtasks</action>
+
+    <action if="no incomplete tasks">
+      <goto step="9">Completion sequence</goto>
+    </action>
+    <action if="story file inaccessible">HALT: "Cannot develop story without access to story file"</action>
+    <action if="incomplete task or subtask requirements ambiguous">ASK user to clarify or HALT</action>
+  </step>
+
+  <step n="2" goal="Load project context and story information">
+    <critical>Load all available context to inform implementation</critical>
+
+    <action>Load {project_context} for coding standards and project-wide patterns (if exists)</action>
+    <action>Parse sections: Story, Acceptance Criteria, Tasks/Subtasks, Dev Notes, Dev Agent Record, File List, Change Log, Status</action>
+    <action>Load comprehensive context from story file's Dev Notes section</action>
+    <action>Extract developer guidance from Dev Notes: architecture requirements, previous learnings, technical specifications</action>
+    <action>Use enhanced story context to inform implementation decisions and approaches</action>
+    <output>✅ **Context Loaded**
+      Story and project context available for implementation
+    </output>
+  </step>
+
+  <step n="3" goal="Detect review continuation and extract review context">
+    <critical>Determine if this is a fresh start or continuation after code review</critical>
+
+    <action>Check if "Senior Developer Review (AI)" section exists in the story file</action>
+    <action>Check if "Review Follow-ups (AI)" subsection exists under Tasks/Subtasks</action>
+
+    <check if="Senior Developer Review section exists">
+      <action>Set review_continuation = true</action>
+      <action>Extract from "Senior Developer Review (AI)" section:
+        - Review outcome (Approve/Changes Requested/Blocked)
+        - Review date
+        - Total action items with checkboxes (count checked vs unchecked)
+        - Severity breakdown (High/Med/Low counts)
+      </action>
+      <action>Count unchecked [ ] review follow-up tasks in "Review Follow-ups (AI)" subsection</action>
+      <action>Store list of unchecked review items as {{pending_review_items}}</action>
+
+      <output>⏯️ **Resuming Story After Code Review** ({{review_date}})
+
+        **Review Outcome:** {{review_outcome}}
+        **Action Items:** {{unchecked_review_count}} remaining to address
+        **Priorities:** {{high_count}} High, {{med_count}} Medium, {{low_count}} Low
+
+        **Strategy:** Will prioritize review follow-up tasks (marked [AI-Review]) before continuing with regular tasks.
+      </output>
+    </check>
+
+    <check if="Senior Developer Review section does NOT exist">
+      <action>Set review_continuation = false</action>
+      <action>Set {{pending_review_items}} = empty</action>
+
+      <output>🚀 **Starting Fresh Implementation**
+
+        Story: {{story_key}}
+        Story Status: {{current_status}}
+        First incomplete task: {{first_task_description}}
+      </output>
+    </check>
+  </step>
+
+  <step n="4" goal="Mark story in-progress" tag="sprint-status">
+    <action>If story file YAML frontmatter already contains `baseline_commit`, preserve the existing value and do not overwrite it</action>
+
+    <check if="{{sprint_status}} file exists">
+      <action>Load the FULL file: {{sprint_status}}</action>
+      <action>Read all development_status entries to find {{story_key}}</action>
+      <action>Set {{current_status}} to development_status[{{story_key}}]</action>
+    </check>
+
+    <check if="{{sprint_status}} file does NOT exist">
+      <action>Set {{current_status}} to the story file Status section value</action>
+    </check>
+
+    <check if="{{current_status}} == 'ready-for-dev' AND story file YAML frontmatter does NOT contain baseline_commit">
+      <action>Run `git rev-parse HEAD` to capture current commit into {{baseline_commit}}; if git/version control is unavailable, set {{baseline_commit}} = `NO_VCS`</action>
+      <action>If story file YAML frontmatter exists, add `baseline_commit: {{baseline_commit}}` to the frontmatter</action>
+      <action>If story file has no YAML frontmatter, create frontmatter at the top containing only `baseline_commit: {{baseline_commit}}`</action>
+    </check>
+
+    <check if="{{sprint_status}} file exists">
+      <check if="{{current_status}} == 'ready-for-dev' OR (review_continuation == true AND {{current_status}} != 'in-progress')">
+        <action>Update the story in the sprint status report to = "in-progress"</action>
+        <action>Update last_updated field to current date</action>
+        <output>🚀 Starting work on story {{story_key}}
+          Status updated: {{current_status}} → in-progress
+        </output>
+      </check>
+
+      <check if="{{current_status}} == 'in-progress'">
+        <output>⏯️ Resuming work on story {{story_key}}
+          Story is already marked in-progress
+        </output>
+      </check>
+
+      <check if="{{current_status}} is neither ready-for-dev nor in-progress">
+        <output>⚠️ Unexpected story status: {{current_status}}
+          Expected ready-for-dev or in-progress. Continuing anyway...
+        </output>
+      </check>
+
+      <action>Store {{current_sprint_status}} for later use</action>
+    </check>
+
+    <check if="{{sprint_status}} file does NOT exist">
+      <output>ℹ️ No sprint status file exists - story progress will be tracked in story file only</output>
+      <action>Set {{current_sprint_status}} = "no-sprint-tracking"</action>
+    </check>
+  </step>
+
+  <step n="5" goal="Implement task following red-green-refactor cycle">
+    <critical>FOLLOW THE STORY FILE TASKS/SUBTASKS SEQUENCE EXACTLY AS WRITTEN - NO DEVIATION</critical>
+
+    <action>Review the current task/subtask from the story file - this is your authoritative implementation guide</action>
+    <action>Plan implementation following red-green-refactor cycle</action>
+
+    <!-- RED PHASE -->
+    <action>Write FAILING tests first for the task/subtask functionality</action>
+    <action>Confirm tests fail before implementation - this validates test correctness</action>
+
+    <!-- GREEN PHASE -->
+    <action>Implement MINIMAL code to make tests pass</action>
+    <action>Run tests to confirm they now pass</action>
+    <action>Handle error conditions and edge cases as specified in task/subtask</action>
+
+    <!-- REFACTOR PHASE -->
+    <action>Improve code structure while keeping tests green</action>
+    <action>Ensure code follows architecture patterns and coding standards from Dev Notes</action>
+
+    <action>Document technical approach and decisions in Dev Agent Record → Implementation Plan</action>
+
+    <action if="new dependencies required beyond story specifications">HALT: "Additional dependencies need user approval"</action>
+    <action if="3 consecutive implementation failures occur">HALT and request guidance</action>
+    <action if="required configuration is missing">HALT: "Cannot proceed without necessary configuration files"</action>
+
+    <critical>NEVER implement anything not mapped to a specific task/subtask in the story file</critical>
+    <critical>NEVER proceed to next task until current task/subtask is complete AND tests pass</critical>
+    <critical>Execute continuously without pausing until all tasks/subtasks are complete or explicit HALT condition</critical>
+    <critical>Do NOT propose to pause for review until Step 9 completion gates are satisfied</critical>
+  </step>
+
+  <step n="6" goal="Author comprehensive tests">
+    <action>Create unit tests for business logic and core functionality introduced/changed by the task</action>
+    <action>Add integration tests for component interactions specified in story requirements</action>
+    <action>Include end-to-end tests for critical user flows when story requirements demand them</action>
+    <action>Cover edge cases and error handling scenarios identified in story Dev Notes</action>
+  </step>
+
+  <step n="7" goal="Run validations and tests">
+    <action>Determine how to run tests for this repo (infer test framework from project structure)</action>
+    <action>Run all existing tests to ensure no regressions</action>
+    <action>Run the new tests to verify implementation correctness</action>
+    <action>Run linting and code quality checks if configured in project</action>
+    <action>Validate implementation meets ALL story acceptance criteria; enforce quantitative thresholds explicitly</action>
+    <action if="regression tests fail">STOP and fix before continuing - identify breaking changes immediately</action>
+    <action if="new tests fail">STOP and fix before continuing - ensure implementation correctness</action>
+  </step>
+
+  <step n="8" goal="Validate and mark task complete ONLY when fully done">
+    <critical>NEVER mark a task complete unless ALL conditions are met - NO LYING OR CHEATING</critical>
+
+    <!-- VALIDATION GATES -->
+    <action>Verify ALL tests for this task/subtask ACTUALLY EXIST and PASS 100%</action>
+    <action>Confirm implementation matches EXACTLY what the task/subtask specifies - no extra features</action>
+    <action>Validate that ALL acceptance criteria related to this task are satisfied</action>
+    <action>Run full test suite to ensure NO regressions introduced</action>
+
+    <!-- REVIEW FOLLOW-UP HANDLING -->
+    <check if="task is review follow-up (has [AI-Review] prefix)">
+      <action>Extract review item details (severity, description, related AC/file)</action>
+      <action>Add to resolution tracking list: {{resolved_review_items}}</action>
+
+      <!-- Mark task in Review Follow-ups section -->
+      <action>Mark task checkbox [x] in "Tasks/Subtasks → Review Follow-ups (AI)" section</action>
+
+      <!-- CRITICAL: Also mark corresponding action item in review section -->
+      <action>Find matching action item in "Senior Developer Review (AI) → Action Items" section by matching description</action>
+      <action>Mark that action item checkbox [x] as resolved</action>
+
+      <action>Add to Dev Agent Record → Completion Notes: "✅ Resolved review finding [{{severity}}]: {{description}}"</action>
+    </check>
+
+    <!-- ONLY MARK COMPLETE IF ALL VALIDATION PASS -->
+    <check if="ALL validation gates pass AND tests ACTUALLY exist and pass">
+      <action>ONLY THEN mark the task (and subtasks) checkbox with [x]</action>
+      <action>Update File List section with ALL new, modified, or deleted files (paths relative to repo root)</action>
+      <action>Add completion notes to Dev Agent Record summarizing what was ACTUALLY implemented and tested</action>
+    </check>
+
+    <check if="ANY validation fails">
+      <action>DO NOT mark task complete - fix issues first</action>
+      <action>HALT if unable to fix validation failures</action>
+    </check>
+
+    <check if="review_continuation == true and {{resolved_review_items}} is not empty">
+      <action>Count total resolved review items in this session</action>
+      <action>Add Change Log entry: "Addressed code review findings - {{resolved_count}} items resolved (Date: {{date}})"</action>
+    </check>
+
+    <action>Save the story file</action>
+    <action>Determine if more incomplete tasks remain</action>
+    <action if="more tasks remain">
+      <goto step="5">Next task</goto>
+    </action>
+    <action if="no tasks remain">
+      <goto step="9">Completion</goto>
+    </action>
+  </step>
+
+  <step n="9" goal="Story completion and mark for review" tag="sprint-status">
+    <action>Verify ALL tasks and subtasks are marked [x] (re-scan the story document now)</action>
+    <action>Run the full regression suite (do not skip)</action>
+    <action>Confirm File List includes every changed file</action>
+    <action>Execute enhanced definition-of-done validation</action>
+    <action>Update the story Status to: "review"</action>
+
+    <!-- Enhanced Definition of Done Validation -->
+    <action>Validate definition-of-done checklist with essential requirements:
+      - All tasks/subtasks marked complete with [x]
+      - Implementation satisfies every Acceptance Criterion
+      - Unit tests for core functionality added/updated
+      - Integration tests for component interactions added when required
+      - End-to-end tests for critical flows added when story demands them
+      - All tests pass (no regressions, new tests successful)
+      - Code quality checks pass (linting, static analysis if configured)
+      - File List includes every new/modified/deleted file (relative paths)
+      - Dev Agent Record contains implementation notes
+      - Change Log includes summary of changes
+      - Only permitted story sections were modified
+    </action>
+
+    <!-- Mark story ready for review - sprint status conditional -->
+    <check if="{sprint_status} file exists AND {{current_sprint_status}} != 'no-sprint-tracking'">
+      <action>Load the FULL file: {sprint_status}</action>
+      <action>Find development_status key matching {{story_key}}</action>
+      <action>Verify current status is "in-progress" (expected previous state)</action>
+      <action>Update development_status[{{story_key}}] = "review"</action>
+      <action>Update last_updated field to current date</action>
+      <action>Save file, preserving ALL comments and structure including STATUS DEFINITIONS</action>
+      <output>✅ Story status updated to "review" in sprint-status.yaml</output>
+    </check>
+
+    <check if="{sprint_status} file does NOT exist OR {{current_sprint_status}} == 'no-sprint-tracking'">
+      <output>ℹ️ Story status updated to "review" in story file (no sprint tracking configured)</output>
+    </check>
+
+    <check if="story key not found in sprint status">
+      <output>⚠️ Story file updated, but sprint-status update failed: {{story_key}} not found
+
+        Story status is set to "review" in file, but sprint-status.yaml may be out of sync.
+      </output>
+    </check>
+
+    <!-- Final validation gates -->
+    <action if="any task is incomplete">HALT - Complete remaining tasks before marking ready for review</action>
+    <action if="regression failures exist">HALT - Fix regression issues before completing</action>
+    <action if="File List is incomplete">HALT - Update File List with all changed files</action>
+    <action if="definition-of-done validation fails">HALT - Address DoD failures before completing</action>
+  </step>
+
+  <step n="10" goal="Completion communication and user support">
+    <action>Execute the enhanced definition-of-done checklist using the validation framework</action>
+    <action>Prepare a concise summary in Dev Agent Record → Completion Notes</action>
+
+    <action>Communicate to {user_name} that story implementation is complete and ready for review</action>
+    <action>Summarize key accomplishments: story ID, story key, title, key changes made, tests added, files modified</action>
+    <action>Provide the story file path and current status (now "review")</action>
+
+    <action>Based on {user_skill_level}, ask if user needs any explanations about:
+      - What was implemented and how it works
+      - Why certain technical decisions were made
+      - How to test or verify the changes
+      - Any patterns, libraries, or approaches used
+      - Anything else they'd like clarified
+    </action>
+
+    <check if="user asks for explanations">
+      <action>Provide clear, contextual explanations tailored to {user_skill_level}</action>
+      <action>Use examples and references to specific code when helpful</action>
+    </check>
+
+    <action>Once explanations are complete (or user indicates no questions), suggest logical next steps</action>
+    <action>Recommended next steps (flexible based on project setup):
+      - Review the implemented story and test the changes
+      - Verify all acceptance criteria are met
+      - Ensure deployment readiness if applicable
+      - Run `code-review` workflow for peer review
+      - Optional: If Test Architect module installed, run `/bmad:tea:automate` to expand guardrail tests
+    </action>
+
+    <output>💡 **Tip:** For best results, run `code-review` using a **different** LLM than the one that implemented this story.</output>
+    <check if="{sprint_status} file exists">
+      <action>Suggest checking {sprint_status} to see project progress</action>
+    </check>
+    <action>Remain flexible - allow user to choose their own path or ask for other assistance</action>
+  <action>Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow.on_complete` — if the resolved value is non-empty, follow it as the final terminal instruction before exiting.</action>
+  </step>
+
+</workflow>
diff --git a/.agents/skills/bmad-dev-story/checklist.md b/.agents/skills/bmad-dev-story/checklist.md
new file mode 100644
index 00000000..86d6e9be
--- /dev/null
+++ b/.agents/skills/bmad-dev-story/checklist.md
@@ -0,0 +1,80 @@
+---
+title: 'Enhanced Dev Story Definition of Done Checklist'
+validation-target: 'Story markdown ({{story_path}})'
+validation-criticality: 'HIGHEST'
+required-inputs:
+  - 'Story markdown file with enhanced Dev Notes containing comprehensive implementation context'
+  - 'Completed Tasks/Subtasks section with all items marked [x]'
+  - 'Updated File List section with all changed files'
+  - 'Updated Dev Agent Record with implementation notes'
+optional-inputs:
+  - 'Test results output'
+  - 'CI logs'
+  - 'Linting reports'
+validation-rules:
+  - 'Only permitted story sections modified: Tasks/Subtasks checkboxes, Dev Agent Record, File List, Change Log, Status'
+  - 'All implementation requirements from story Dev Notes must be satisfied'
+  - 'Definition of Done checklist must pass completely'
+  - 'Enhanced story context must contain sufficient technical guidance'
+---
+
+# 🎯 Enhanced Definition of Done Checklist
+
+**Critical validation:** Story is truly ready for review only when ALL items below are satisfied
+
+## 📋 Context & Requirements Validation
+
+- [ ] **Story Context Completeness:** Dev Notes contains ALL necessary technical requirements, architecture patterns, and implementation guidance
+- [ ] **Architecture Compliance:** Implementation follows all architectural requirements specified in Dev Notes
+- [ ] **Technical Specifications:** All technical specifications (libraries, frameworks, versions) from Dev Notes are implemented correctly
+- [ ] **Previous Story Learnings:** Previous story insights incorporated (if applicable) and build upon appropriately
+
+## ✅ Implementation Completion
+
+- [ ] **All Tasks Complete:** Every task and subtask marked complete with [x]
+- [ ] **Acceptance Criteria Satisfaction:** Implementation satisfies EVERY Acceptance Criterion in the story
+- [ ] **No Ambiguous Implementation:** Clear, unambiguous implementation that meets story requirements
+- [ ] **Edge Cases Handled:** Error conditions and edge cases appropriately addressed
+- [ ] **Dependencies Within Scope:** Only uses dependencies specified in story or project-context.md
+
+## 🧪 Testing & Quality Assurance
+
+- [ ] **Unit Tests:** Unit tests added/updated for ALL core functionality introduced/changed by this story
+- [ ] **Integration Tests:** Integration tests added/updated for component interactions when story requirements demand them
+- [ ] **End-to-End Tests:** End-to-end tests created for critical user flows when story requirements specify them
+- [ ] **Test Coverage:** Tests cover acceptance criteria and edge cases from story Dev Notes
+- [ ] **Regression Prevention:** ALL existing tests pass (no regressions introduced)
+- [ ] **Code Quality:** Linting and static checks pass when configured in project
+- [ ] **Test Framework Compliance:** Tests use project's testing frameworks and patterns from Dev Notes
+
+## 📝 Documentation & Tracking
+
+- [ ] **File List Complete:** File List includes EVERY new, modified, or deleted file (paths relative to repo root)
+- [ ] **Dev Agent Record Updated:** Contains relevant Implementation Notes and/or Debug Log for this work
+- [ ] **Change Log Updated:** Change Log includes clear summary of what changed and why
+- [ ] **Review Follow-ups:** All review follow-up tasks (marked [AI-Review]) completed and corresponding review items marked resolved (if applicable)
+- [ ] **Story Structure Compliance:** Only permitted sections of story file were modified
+
+## 🔚 Final Status Verification
+
+- [ ] **Story Status Updated:** Story Status set to "review"
+- [ ] **Sprint Status Updated:** Sprint status updated to "review" (when sprint tracking is used)
+- [ ] **Quality Gates Passed:** All quality checks and validations completed successfully
+- [ ] **No HALT Conditions:** No blocking issues or incomplete work remaining
+- [ ] **User Communication Ready:** Implementation summary prepared for user review
+
+## 🎯 Final Validation Output
+
+```
+Definition of Done: {{PASS/FAIL}}
+
+✅ **Story Ready for Review:** {{story_key}}
+📊 **Completion Score:** {{completed_items}}/{{total_items}} items passed
+🔍 **Quality Gates:** {{quality_gates_status}}
+📋 **Test Results:** {{test_results_summary}}
+📝 **Documentation:** {{documentation_status}}
+```
+
+**If FAIL:** List specific failures and required actions before story can be marked Ready for Review
+
+**If PASS:** Story is fully ready for code review and production consideration
diff --git a/.agents/skills/bmad-dev-story/customize.toml b/.agents/skills/bmad-dev-story/customize.toml
new file mode 100644
index 00000000..e2fac266
--- /dev/null
+++ b/.agents/skills/bmad-dev-story/customize.toml
@@ -0,0 +1,44 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-dev-story. Mirrors the
+# agent customization shape under the [workflow] namespace.
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+# Steps to run before the standard activation (config load, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Distinct from the runtime memory sidecar — these are static context
+# loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "All stories must include testable acceptance criteria."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+  "Never add epic or story references as inline code comments (e.g. # Epic: X, # Story: PROJ-42).",
+  "Code comments explain why, not what — no AI workflow metadata, planning refs, or story tracking in source code.",
+  "Generated code must be production-ready: clean, minimal, and free of AI-generated noise.",
+]
+
+# Scalar: executed when the workflow reaches its final step,
+# after the story implementation is complete and status is updated. Override wins.
+# Leave empty for no custom post-completion behavior.
+
+on_complete = ""
diff --git a/.agents/skills/bmad-document-project/SKILL.md b/.agents/skills/bmad-document-project/SKILL.md
new file mode 100644
index 00000000..bfe76657
--- /dev/null
+++ b/.agents/skills/bmad-document-project/SKILL.md
@@ -0,0 +1,14 @@
+---
+name: bmad-document-project
+description: 'Deprecated — forwards to bmad-project-context. Use when the user says "document this project" or "generate project docs"'
+---
+
+# DEPRECATED — forwards to bmad-project-context
+
+Tell the user two things.
+
+First: this skill is deprecated. Generating documentation volume about a codebase made agents worse, not better — agents read code more accurately than prose describing code, and the generated set was stale on arrival. `bmad-project-context` owns what remains useful: a small verified block in the repo's `AGENTS.md` carrying what the code cannot say — required policy, conventions that differ from defaults, what running the project takes that no config file states, and known pitfalls.
+
+Second, so they are not surprised by what they get: the deeper "explain this system, its rationale and its history" material is a different altitude and is not part of that block. It is coming as its own capability. If that is what they were after, say so plainly rather than producing a thin substitute.
+
+Then invoke `bmad-project-context` with **setup** intent, forwarding the user's original request and any paths or documents they supplied, verbatim. It takes the workflow from here.
diff --git a/.agents/skills/bmad-domain-research/SKILL.md b/.agents/skills/bmad-domain-research/SKILL.md
new file mode 100644
index 00000000..7b42c2e2
--- /dev/null
+++ b/.agents/skills/bmad-domain-research/SKILL.md
@@ -0,0 +1,14 @@
+---
+name: bmad-domain-research
+description: 'Deprecated — forwards to bmad-deep-recon (domain type).'
+---
+
+# DEPRECATED — forwards to bmad-deep-recon (domain type)
+
+This skill was consolidated into `bmad-deep-recon`. It is retained as a thin compatibility shim so existing invocations by name and `_bmad/custom/bmad-domain-research.toml` override files keep working. New work should invoke `bmad-deep-recon` directly — it drafts deep-research prompts for outside tools, processes finished reports into downstream-ready summaries, and runs research directly, across market, domain, technical, competitive, user-voice, and academic-lit types (plus a select shape for choose-between decisions and custom types).
+
+## On Activation
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. This picks up any `{project-root}/_bmad/custom/bmad-domain-research.toml` and `bmad-domain-research.user.toml` overrides for the legacy fields (`activation_steps_prepend`, `activation_steps_append`, `persistent_facts`, `on_complete`).
+2. Emit a deprecation notice to the user (in their configured communication language): `bmad-domain-research` is deprecated and forwards to `bmad-deep-recon` with the domain type. To silence this notice and access the full new surface (draft/process/run modes, research types, verification levels, HTML briefing, handoffs), migrate `_bmad/custom/bmad-domain-research.toml` to `_bmad/custom/bmad-deep-recon.toml` and invoke `bmad-deep-recon` directly.
+3. Invoke `bmad-deep-recon` with: **research type** `domain` (skip its type inference), the four legacy fields above as pre-resolved values, and the user's original input verbatim. `bmad-deep-recon` takes the workflow from here — do not execute any further steps in this shim.
diff --git a/.agents/skills/bmad-edit-prd/SKILL.md b/.agents/skills/bmad-edit-prd/SKILL.md
new file mode 100644
index 00000000..dc9c5d3d
--- /dev/null
+++ b/.agents/skills/bmad-edit-prd/SKILL.md
@@ -0,0 +1,30 @@
+---
+name: bmad-edit-prd
+description: 'Deprecated — forwards to bmad-prd (update intent).'
+---
+
+# DEPRECATED — forwards to bmad-prd (update intent)
+
+This skill was consolidated into `bmad-prd`. It is retained as a thin compatibility shim so existing invocations by name and `_bmad/custom/bmad-edit-prd.toml` override files keep working. New work should invoke `bmad-prd` directly — it detects create / update / validate intent from the conversation.
+
+## On Activation
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. This picks up any `{project-root}/_bmad/custom/bmad-edit-prd.toml` and `bmad-edit-prd.user.toml` overrides for the legacy fields (`activation_steps_prepend`, `activation_steps_append`, `persistent_facts`, `on_complete`).
+
+2. Load `{project-root}/_bmad/bmm/config.yaml` (and `config.user.yaml` if present) to resolve `{user_name}` and `{communication_language}`.
+
+3. Emit a deprecation notice to the user in `{communication_language}`:
+
+   > Notice: `bmad-edit-prd` is deprecated and will be removed in a future release. It now forwards to `bmad-prd` with update intent. To silence this notice and access the full new customization surface (`prd_template`, `validation_checklist_template`, `validation_report_template`, `doc_standards`, `finalize_reviewers`, `external_sources`, `external_handoffs`, `prd_output_path`, `run_folder_pattern`), migrate `_bmad/custom/bmad-edit-prd.toml` to `_bmad/custom/bmad-prd.toml` and invoke `bmad-prd` directly next time. Customization fields that were in this version still remain in the new version and will be respected if present in `_bmad/custom/bmad-prd.toml`, but the new version also supports additional fields that you can take advantage of by migrating.
+
+4. Invoke `bmad-prd` with the following context. Pass these as the activating context so `bmad-prd` honors them instead of resolving its own customization from scratch:
+
+   - **Intent:** `update` — skip `bmad-prd`'s usual intent detection step.
+   - **Pre-resolved legacy customization** — use these in place of resolving from `bmad-prd`'s own `customize.toml` for the four legacy fields. For everything else (`prd_template`, `validation_checklist_template`, `validation_report_template`, `doc_standards`, `finalize_reviewers`, `prd_output_path`, `run_folder_pattern`, `external_sources`, `external_handoffs`), use `bmad-prd`'s own defaults and overrides as normal:
+     - `activation_steps_prepend` = the resolved value from step 1
+     - `activation_steps_append` = the resolved value from step 1
+     - `persistent_facts` = the resolved value from step 1
+     - `on_complete` = the resolved value from step 1
+   - **Original user input:** forward whatever the user said when invoking this skill verbatim (the target PRD path, the change signal, etc.).
+
+   `bmad-prd` takes the workflow from here. Do not execute any further steps in this shim.
diff --git a/.agents/skills/bmad-edit-prd/customize.toml b/.agents/skills/bmad-edit-prd/customize.toml
new file mode 100644
index 00000000..1886d4ac
--- /dev/null
+++ b/.agents/skills/bmad-edit-prd/customize.toml
@@ -0,0 +1,42 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-edit-prd. Mirrors the
+# agent customization shape under the [workflow] namespace.
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+# Steps to run before the standard activation (config load, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Distinct from the runtime memory sidecar — these are static context
+# loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "All PRDs must include a regulatory-risk section."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Scalar: executed when the workflow reaches Step E-4 (Complete & Validate) and the
+# user exits via [S] Summary or [X] Exit — not on [V] Validate (which chains to
+# bmad-validate-prd) or [E] Edit More (which loops back). Override wins.
+# Leave empty for no custom post-completion behavior.
+
+on_complete = ""
diff --git a/.agents/skills/bmad-editorial-review-prose/SKILL.md b/.agents/skills/bmad-editorial-review-prose/SKILL.md
new file mode 100644
index 00000000..cdc3565f
--- /dev/null
+++ b/.agents/skills/bmad-editorial-review-prose/SKILL.md
@@ -0,0 +1,6 @@
+---
+name: bmad-editorial-review-prose
+description: 'Deprecated — forwards to bmad-review.'
+---
+
+Merged into `bmad-review`. Invoke the `bmad-review` skill on the same content with only the `prose` lens, passing through the same inputs and any `also_consider` areas. Present the findings in the legacy shape: a three-column markdown table `| Original Text | Revised Text | Changes |` — no Pass column, no preamble above the table. If no issues are found, output exactly: `No editorial issues identified`.
diff --git a/.agents/skills/bmad-editorial-review-structure/SKILL.md b/.agents/skills/bmad-editorial-review-structure/SKILL.md
new file mode 100644
index 00000000..f8fc6703
--- /dev/null
+++ b/.agents/skills/bmad-editorial-review-structure/SKILL.md
@@ -0,0 +1,6 @@
+---
+name: bmad-editorial-review-structure
+description: 'Deprecated — forwards to bmad-review.'
+---
+
+Merged into `bmad-review`. Invoke the `bmad-review` skill on the same content with only the `structure` lens, passing through the same inputs and any `also_consider` areas. Present the findings in the legacy report shape: a `## Document Summary` block (purpose, audience, reader type, structure model, current length), a `## Recommendations` list of numbered `[CUT/MERGE/MOVE/CONDENSE/QUESTION/PRESERVE]` entries each with rationale and word impact, and a closing `## Summary` (total recommendations, estimated reduction) — not the findings table. If no structural issues are found, output exactly: `No substantive changes recommended`.
diff --git a/.agents/skills/bmad-editorial-review/SKILL.md b/.agents/skills/bmad-editorial-review/SKILL.md
new file mode 100644
index 00000000..a06c0e3e
--- /dev/null
+++ b/.agents/skills/bmad-editorial-review/SKILL.md
@@ -0,0 +1,6 @@
+---
+name: bmad-editorial-review
+description: 'Deprecated — forwards to bmad-review.'
+---
+
+Merged into `bmad-review`. Invoke the `bmad-review` skill on the same content with the `structure` and `prose` lenses — both, structure first, so prose runs on top of the structure findings — unless the caller asked for a structure-only or prose-only review, in which case pass only that lens. Pass through any `also_consider` areas, and forward this skill's resolved `[workflow]` fields as pre-resolved values — but only those that resolved to something, since an empty value here means no legacy override exists and bmad-review's own default should stand: `reader_type`, `style_guide`, `review_guidance`, `output_preferences`, `persistent_facts`, `activation_steps_prepend`, `activation_steps_append`, `on_complete`, and `review_output_path` as the report path. Present the findings in the legacy shape: the two-pass findings table `| Pass | Original Text | Revised Text | Changes |` with the purpose/audience read above it and, when the structure pass ran, the reduction summary below it — and no other lens's output.
diff --git a/.agents/skills/bmad-editorial-review/customize.toml b/.agents/skills/bmad-editorial-review/customize.toml
new file mode 100644
index 00000000..d4305d23
--- /dev/null
+++ b/.agents/skills/bmad-editorial-review/customize.toml
@@ -0,0 +1,31 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Legacy override surface for bmad-editorial-review, which is deprecated and
+# forwards to bmad-review's `structure` and `prose` lenses. These keys exist
+# only so overrides written against the old skill keep resolving: the shim
+# resolves them here and passes them to bmad-review as pre-resolved values,
+# which bmad-review honors verbatim over its own.
+#
+# Every field is documented in bmad-review's customize.toml, which is where
+# new customization belongs. Defaults are empty here so the shim contributes
+# nothing of its own — an unset key falls through to bmad-review's default.
+#
+# Override files (not edited here):
+#   {project-root}/_bmad/custom/bmad-editorial-review.toml         (team)
+#   {project-root}/_bmad/custom/bmad-editorial-review.user.toml    (personal)
+#
+# Overrides merge per BMad structural rules: scalars win, arrays append.
+
+[workflow]
+
+activation_steps_prepend = []
+activation_steps_append = []
+persistent_facts = []
+review_guidance = []
+on_complete = ""
+output_preferences = ""
+reader_type = ""
+style_guide = ""
+
+# Where the findings table lands; forwarded as bmad-review's report path.
+review_output_path = ""
diff --git a/.agents/skills/bmad-forge-idea/SKILL.md b/.agents/skills/bmad-forge-idea/SKILL.md
new file mode 100644
index 00000000..aaaa6f68
--- /dev/null
+++ b/.agents/skills/bmad-forge-idea/SKILL.md
@@ -0,0 +1,107 @@
+---
+name: bmad-forge-idea
+description: Pressure-test an idea through persona-driven interrogation until it hardens, proves out, or dies cheaply. Use when the user says 'forge an idea', 'pressure-test this idea', 'stress-test my thinking', or 'harden this idea'.
+---
+
+# BMad Forge Idea
+
+## Overview
+
+Take a half-formed idea and pressure-test it in conversation, while changing your mind is still cheap, until it becomes something the user can act on with conviction or reject. The main risk is what the user has not examined yet: unchecked assumptions and unresolved decisions usually become more expensive problems later.
+
+The main goal is better thinking, not producing an artifact. Strengthening an idea, rejecting it, or thinking it through more clearly are all complete outcomes. Writing `forged-idea.md` to hand off to another workflow is optional. Do not steer the conversation toward "shall we build it?"
+
+This skill can be used on many kinds of ideas. When the idea is about a product or feature, what survives may be written to `forged-idea.md` for later planning.
+
+Lead by questioning, not lecturing. Ask one question at a time, press on weak points, and do not let vague claims pass without examination.
+
+## Conventions
+
+- Scripts live in two places — run each from the exact path written, never assume co-location: the shared core scripts (`memlog.py`, `resolve_customization.py`, `resolve_config.py`) are installed by BMad core at `{project-root}/_bmad/scripts/` and are never bundled here; this skill's own `resolve_personas.py` is at `{skill-root}/scripts/`.
+- `{workflow.<name>}` resolves to fields in the merged `customize.toml` `[workflow]` table.
+
+## On Activation
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. On failure, read `{skill-root}/customize.toml` directly with defaults. Apply the resolved `{workflow.*}` values throughout.
+2. Run each `{workflow.activation_steps_prepend}` entry; treat each `{workflow.persistent_facts}` entry as foundational context (`file:` entries load their contents, `skill:` names a skill to consult, others are facts verbatim).
+3. Resolve central config: `uv run {project-root}/_bmad/scripts/resolve_config.py --project-root {project-root} --key core`; from the merged JSON read `{user_name}`, `{communication_language}`, `{output_folder}`. On failure use neutral defaults; never block. Greet `{user_name}` in `{communication_language}` and stay in it.
+4. Note whether a BMad persona is already active in this conversation — the user loaded one (e.g. the analyst, the storyteller) and invoked the forge from within it. If so, that persona leads the session, in voice, throughout.
+5. Resume: glob `{workflow.forge_output_path}/**/.memlog.md` (recursive, so it still finds sessions when `run_folder_pattern` is overridden to nest paths) and read only each match's frontmatter to find any whose `status` is not `complete`. Offer to resume one — then read its full memlog once to rebuild state and continue append-only — or to start fresh.
+6. Run each `{workflow.activation_steps_append}` entry.
+
+## Open the session
+
+Start by scrutinizing the idea, not endorsing it.
+
+### Discover intent
+Identify: 
+- the subject idea, 
+- the user's goal for the session, 
+- whether the idea is new or a change to an existing project
+
+If any of these are already clear from the prompt that invoked this skill or previous context, ask the user to confirm and continue. 
+
+Otherwise ask for what's missing, in order: 
+- what is the idea?
+- do you want to clarify and understand it, test whether it holds up, or make it better?
+- is it a new idea or a change to an existing project? If the latter, what project is it, and where can I find its files or other relevant materials?
+
+### Steering the conversation
+
+Tell the user they can say **"attack this"**, **"defend this"**, or **"switch roles"** at any time to change how the current idea is argued. In attack mode, do not agree with the idea; look for contradictions, weak assumptions, and failure cases. In defend mode, argue for the strongest version of the idea. Tell the user they can also name a persona or party at any time to change who participates in the session.
+
+### Set up the session
+
+Derive a kebab-case `{slug}` for the idea and bind the session workspace `{workspace} = {workflow.forge_output_path}/{workflow.run_folder_pattern}` (the pattern fills with `{slug}`). Create the memlog once the goal is known:
+`uv run {project-root}/_bmad/scripts/memlog.py init --workspace {workspace} --field idea="<idea>" --field goal="<goal>"`
+
+Tell the user the path; state is on disk now, so the session survives interruption. If init fails, don't abort — run the forge in-conversation and tell the user state won't persist this session.
+
+## The forge
+
+Let the session goal set the first move: for clarifying, pin down terms, boundaries, and assumptions; for testing, go after the central claim first; for making it better, drive each unresolved branch to a concrete decision.
+
+Work one question at a time, in dependency order.
+
+Include your current best answer or hypothesis when it helps the user respond. A concrete proposal is easier to accept, reject, or revise than an open-ended prompt. Find discoverable answers yourself instead of asking.
+
+Do not assume the user's terms are precise. When a term is fuzzy or overloaded, name the ambiguity and ask for a precise choice before continuing. For example, do not let `user`, `buyer`, and `payer` collapse into one entity unless the idea actually requires that.
+
+For ideas about an existing project, treat the project's files and materials as the source of truth. Do not accept a label or summary as proof. Find the relevant material yourself and check the user's claim against it. If the material contradicts the user's claim, stop and resolve that before continuing.
+
+When a branch resolves, pause before moving on. Give the user a chance to raise any remaining concern.
+
+Do not use agreement or praise to make the interaction smoother; they lower pressure and lead to shallower thinking. Agreement is allowed only when it helps the user think better. Praise is noise. Continued engagement and ego-stroking are not objectives. In attack mode, never agree with the idea until the user ends the mode. For each answer, either challenge the weak point or build on the strong point, whichever helps the user think better.
+
+Capture as you go — each decision, assumption, crack, kill, and locked idea, one bullet in the user's meaning:
+`uv run {project-root}/_bmad/scripts/memlog.py append --workspace {workspace} --type <decision|assumption|crack|kill|direction|lock|note> --text "<gist>"`
+A `lock` is an idea the user hardens — settled, not to be reopened; locks are what `forged-idea.md` is distilled from. Don't read the memlog back except on resume. If the user raises a different branch, capture it and stay put — the loop and the stray insight both survive.
+
+## The personas
+
+If a BMad persona was already active when the forge started, keep that persona as the lead voice.
+
+Resolve the available persona pool once, as soon as the goal is known:
+`uv run {skill-root}/scripts/resolve_personas.py --project-root {project-root} --skill {skill-root}`
+The script returns installed BMad agents (`agents`), user-defined personas (`members`), and saved parties (`parties`). Parties may include a `scene`; some are open-cast. This gives you the same roster information as `bmad-party-mode` without invoking it.
+
+Each turn uses two voices:
+- **One available persona** — choose an installed agent or user-defined persona whose expertise fits the current branch. Vary this voice every few turns; do not let one voice dominate. If the user names a specific persona, use it. If the user calls a saved party, use the whole party and its scene. If the user asks to go one-on-one, use only the requested persona. If no pool is available, generate this voice yourself.
+- **One generated persona** — create a fresh outside voice, such as a competitor, buyer, finance reviewer, domain expert, or critic. Give it a name and enough characterization to keep its viewpoint distinct.
+
+Use these voices in character to pressure-test the current branch: find sharper objections, missing assumptions, and stronger defenses. Cross-examine them for what matters, then synthesize their input into your next question. Do not let the session turn into a panel debate or persona performance.
+
+Voice the personas yourself by default. Spawn separate agents only when a branch needs independent reasoning that should not be influenced by one shared voice.
+
+## Exits
+
+The session can end in three valid states:
+
+- **Hardened** — the idea is stronger and specific enough to use. Distill the memlog into `{workspace}/forged-idea.md`. Keep it extremely short: only the decisions, rejected options, and reasons that matter downstream, in the user's meaning. Do not write a prose summary, template, or conversation recap. If it reads like a document, it is too long. If planning or dev skills are installed (`bmad-spec`, `bmad-prd`, `bmad-prfaq`, `bmad-build`), offer the file as their input; if none are, the file stands on its own — never treat a missing skill as an error.
+- **Killed** — the idea does not hold up. Say so plainly and record why. Finding that out early is a valid outcome.
+- **Clearer** — the user understands the idea better, but there is no hardened idea to hand off. Leave the memlog as the record; no `forged-idea.md` is needed.
+
+Always render `{workspace}/forge-report.html` as a self-contained HTML file the user can open, with inline CSS and an inline-SVG seal or stamp. Summarize the outcome, the locked decisions, what was rejected and why, and the weak points that survived scrutiny, in the user's meaning. Credit the personas and parties that pressure-tested the idea by name, icon, and voice. Render a prominent wax-seal-style or stamped outcome mark, matched to the result: `HARDENED`, an `Idea Death Certificate` stamped `KILLED` with the cause of death, or `CLARIFIED`. Tell the user the path.
+
+Flip the status at the end: `uv run {project-root}/_bmad/scripts/memlog.py set --workspace {workspace} --key status --value complete`.
+If `{workflow.on_complete}` is non-empty, run all instructions in order.
diff --git a/.agents/skills/bmad-forge-idea/customize.toml b/.agents/skills/bmad-forge-idea/customize.toml
new file mode 100644
index 00000000..98949e26
--- /dev/null
+++ b/.agents/skills/bmad-forge-idea/customize.toml
@@ -0,0 +1,42 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-forge-idea.
+#
+# Override files (not edited here):
+#   {project-root}/_bmad/custom/bmad-forge-idea.toml        (team)
+#   {project-root}/_bmad/custom/bmad-forge-idea.user.toml   (personal)
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays: append
+
+# Steps to run before the standard activation (config load, greet).
+activation_steps_prepend = []
+
+# Steps to run after greet but before the session begins.
+activation_steps_append = []
+
+# Persistent facts the interrogator keeps in mind for the whole session
+# (domain constraints, house rules, what's off the table). Each entry is a
+# literal sentence, a skill prefixed with `skill:`, or a `file:`-prefixed
+# path/glob whose contents are loaded as facts. Default loads project-context.md
+# when one exists (e.g. from bmad-generate-project-context), so the forge grounds
+# in the project's tech, domain, and constraints without re-asking.
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Executed when the session completes. Scalar or array of instructions. Empty for none.
+on_complete = []
+
+# Parent folder for all forge sessions. Each session gets its own run
+# folder underneath (see run_folder_pattern). Lands directly under
+# {output_folder} so the forge works in core-only installs.
+forge_output_path = "{output_folder}/forge"
+
+# Run-folder pattern inside forge_output_path. Resolved against the
+# idea-derived slug at activation. Same slug = same folder, so resuming
+# an idea reuses its memlog. Override to add {date} or other components
+# if a fresh dated history per run is preferred.
+run_folder_pattern = "{slug}"
diff --git a/.agents/skills/bmad-forge-idea/scripts/resolve_personas.py b/.agents/skills/bmad-forge-idea/scripts/resolve_personas.py
new file mode 100644
index 00000000..92bc129d
--- /dev/null
+++ b/.agents/skills/bmad-forge-idea/scripts/resolve_personas.py
@@ -0,0 +1,275 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.11"
+# ///
+"""Resolve the personas and parties the forge can bring into the room.
+
+The forge cross-examines witnesses: the installed BMAD agents, plus any
+custom personas and party groups the user has authored for `bmad-party-mode`.
+This surfaces all of them in one shot so the orchestrator never has to ask
+"who's available?" — it just intermixes whoever fits the branch, alongside
+any persona the user names on the fly.
+
+What it returns (JSON, stdout):
+  * agents   — the installed BMAD roster: the default room, always present.
+  * members  — extra custom personas in the pool (party_members the user
+               defined that aren't already an installed slot).
+  * parties  — the user's named party groups, members resolved to brief
+               entries; open-cast groups (scene names a pool, no roster)
+               are flagged.
+  * default_party — the group id pinned as party-mode's default, if any.
+
+Discovery is best-effort and never blocks the forge. The installed roster
+comes from the core resolver; custom personas/parties come from
+`bmad-party-mode`'s resolved customization when that skill is found beside
+this one, else from the user's override TOMLs read directly. Anything that
+can't be resolved is simply omitted and flagged, never fatal.
+
+Stdlib only (Python 3.11+ for tomllib).
+
+  resolve_personas.py --project-root P --skill S
+"""
+
+import argparse
+import json
+import subprocess
+import sys
+from pathlib import Path
+
+try:
+    import tomllib
+except ImportError:  # pragma: no cover - guarded for <3.11
+    sys.stderr.write("error: Python 3.11+ is required (stdlib `tomllib`).\n")
+    sys.exit(3)
+
+PARTY_SKILL = "bmad-party-mode"
+
+
+def _run_json(cmd):
+    """Run a resolver script and parse its JSON stdout. None on any failure."""
+    try:
+        out = subprocess.run(
+            cmd, capture_output=True, text=True, encoding="utf-8", timeout=60
+        )
+    except (OSError, subprocess.SubprocessError):
+        return None
+    if out.returncode != 0 or not out.stdout.strip():
+        return None
+    try:
+        return json.loads(out.stdout)
+    except json.JSONDecodeError:
+        return None
+
+
+def _load_toml(path: Path):
+    if not path.exists():
+        return {}
+    try:
+        with path.open("rb") as f:
+            data = tomllib.load(f)
+        return data if isinstance(data, dict) else {}
+    except (OSError, tomllib.TOMLDecodeError):
+        return {}
+
+
+def load_agents(project_root: Path):
+    """Installed BMAD agents as {code: entry}. (dict, resolved_ok).
+
+    The core resolver may emit agents as a dict keyed by code or as an array
+    of tables (depending on how the layers merged); normalize both to a dict.
+    """
+    script = project_root / "_bmad" / "scripts" / "resolve_config.py"
+    data = _run_json([sys.executable, str(script), "--project-root", str(project_root), "--key", "agents"])
+    if data is None:
+        return {}, False
+    agents = data.get("agents", {}) or {}
+    if isinstance(agents, list):
+        agents = {a["code"]: a for a in agents if isinstance(a, dict) and a.get("code")}
+    elif not isinstance(agents, dict):
+        agents = {}
+    return agents, True
+
+
+def find_party_skill(project_root: Path, skill_root: Path):
+    """Locate the installed bmad-party-mode skill dir, or None.
+
+    Skills install as siblings, so the party skill is almost always next to
+    this one. A couple of common install roots cover the rest.
+    """
+    candidates = [
+        skill_root.parent / PARTY_SKILL,
+        project_root / ".claude" / "skills" / PARTY_SKILL,
+        project_root / "_bmad" / "skills" / PARTY_SKILL,
+    ]
+    for c in candidates:
+        if (c / "customize.toml").exists():
+            return c
+    return None
+
+
+def load_party_workflow(project_root: Path, party_skill: Path):
+    """Merged [workflow] table for bmad-party-mode (base + user overrides)."""
+    resolver = project_root / "_bmad" / "scripts" / "resolve_customization.py"
+    data = _run_json([sys.executable, str(resolver), "--skill", str(party_skill), "--key", "workflow"])
+    if data is not None and isinstance(data.get("workflow"), dict):
+        return data["workflow"]
+    # Fallback: base customize.toml directly, no override merge.
+    wf = _load_toml(party_skill / "customize.toml").get("workflow", {})
+    return wf if isinstance(wf, dict) else {}
+
+
+def load_party_overrides(project_root: Path):
+    """Custom personas/parties when party-mode itself isn't installed.
+
+    Reads only the user's override TOMLs (team then personal, personal wins on
+    scalars). No base roster exists in this path, so a shallow merge is enough.
+    """
+    custom = project_root / "_bmad" / "custom"
+    team = _load_toml(custom / f"{PARTY_SKILL}.toml").get("workflow", {})
+    user = _load_toml(custom / f"{PARTY_SKILL}.user.toml").get("workflow", {})
+    team = team if isinstance(team, dict) else {}
+    user = user if isinstance(user, dict) else {}
+    merged = dict(team)
+    for key, val in user.items():
+        if isinstance(val, list) and isinstance(merged.get(key), list):
+            merged[key] = merged[key] + val
+        else:
+            merged[key] = val
+    return merged
+
+
+def _alias(code: str) -> str:
+    """Short alias for an installed agent code: bmad-agent-analyst -> analyst."""
+    for prefix in ("bmad-agent-", "bmad-"):
+        if code.startswith(prefix):
+            return code[len(prefix):]
+    return code
+
+
+def build_pool(agents: dict, party_members: list):
+    """One pool keyed by code; custom members override matching installed slots.
+
+    Returns (pool, index, installed_codes, custom_codes):
+      * installed_codes — the default room (installed agents, overrides applied
+        in place); custom-only additions stay in the pool but don't crowd it.
+      * custom_codes — pure-custom personas (no installed slot), the extra
+        faces the forge can summon by name or via a party group.
+    """
+    pool, index, installed_codes, custom_codes = {}, {}, [], []
+
+    def register(code, entry):
+        pool[code] = entry
+        index[code] = code
+        index[code.lower()] = code
+        index[_alias(code).lower()] = code
+        name = entry.get("name")
+        if name:
+            key = name.lower()
+            # A custom rename must not hijack another agent's name lookup.
+            if index.get(key, code) == code:
+                index[key] = code
+
+    for code, info in (agents or {}).items():
+        register(code, {
+            "code": code,
+            "name": info.get("name", code),
+            "icon": info.get("icon", ""),
+            "title": info.get("title", ""),
+            "description": info.get("description", ""),
+            "source": "installed",
+        })
+        installed_codes.append(code)
+
+    for m in (party_members if isinstance(party_members, list) else []):
+        if not isinstance(m, dict):
+            continue
+        code = m.get("code")
+        if not code:
+            continue
+        canonical = index.get(code) or index.get(code.lower()) or code
+        was_installed = canonical in pool
+        # Start from the installed entry so fields the override omits
+        # (icon, title, description) survive.
+        entry = dict(pool.get(canonical, {}))
+        entry.update({"code": canonical, "source": "custom"})
+        for field in ("name", "icon", "title", "persona", "capabilities", "model"):
+            if m.get(field) is not None:
+                entry[field] = m[field]
+        entry.setdefault("name", canonical)
+        register(canonical, entry)
+        if not was_installed:
+            custom_codes.append(canonical)
+
+    return pool, index, installed_codes, custom_codes
+
+
+def _brief(entry):
+    """The slim card the orchestrator needs to cast a persona."""
+    out = {k: entry[k] for k in ("code", "name", "icon", "title", "source") if entry.get(k)}
+    for k in ("description", "persona", "capabilities", "model"):
+        if entry.get(k):
+            out[k] = entry[k]
+    return out
+
+
+def resolve_parties(groups, pool, index):
+    out = []
+    for g in groups or []:
+        if not isinstance(g, dict) or not g.get("id"):
+            continue
+        raw = g.get("members", []) or []
+        members = []
+        for t in raw:
+            key = t if isinstance(t, str) else str(t)
+            code = index.get(key) or index.get(key.lower())
+            if code in pool:
+                members.append(_brief(pool[code]))
+        party = {"id": g["id"], "name": g.get("name", g["id"]), "members": members}
+        if g.get("scene"):
+            party["scene"] = g["scene"]
+        if not raw:
+            party["open_cast"] = True
+        out.append(party)
+    return out
+
+
+def main():
+    ap = argparse.ArgumentParser(description="Resolve forge personas and parties.")
+    ap.add_argument("--project-root", required=True)
+    ap.add_argument("--skill", required=True, help="Path to the bmad-forge-idea skill dir")
+    args = ap.parse_args()
+
+    project_root = Path(args.project_root).resolve()
+    skill_root = Path(args.skill).resolve()
+
+    agents, agents_ok = load_agents(project_root)
+
+    party_skill = find_party_skill(project_root, skill_root)
+    if party_skill is not None:
+        workflow = load_party_workflow(project_root, party_skill)
+    else:
+        workflow = load_party_overrides(project_root)
+
+    pool, index, installed_codes, custom_codes = build_pool(
+        agents, workflow.get("party_members", []))
+    parties = resolve_parties(workflow.get("party_groups", []), pool, index)
+
+    _emit({
+        "agents": [_brief(pool[c]) for c in installed_codes],
+        "members": [_brief(pool[c]) for c in custom_codes],
+        "parties": parties,
+        "default_party": workflow.get("default_party", "") or "",
+        "party_mode_found": party_skill is not None,
+        "agents_resolved": agents_ok,
+    })
+
+
+def _emit(obj):
+    reconfigure = getattr(sys.stdout, "reconfigure", None)
+    if reconfigure is not None:
+        reconfigure(encoding="utf-8")
+    sys.stdout.write(json.dumps(obj, indent=2, ensure_ascii=False) + "\n")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/.agents/skills/bmad-forge-idea/scripts/tests/test_resolve_personas.py b/.agents/skills/bmad-forge-idea/scripts/tests/test_resolve_personas.py
new file mode 100644
index 00000000..4867b0dc
--- /dev/null
+++ b/.agents/skills/bmad-forge-idea/scripts/tests/test_resolve_personas.py
@@ -0,0 +1,138 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.11"
+# ///
+"""Unit tests for resolve_personas.py — pool merge, alias, party resolution."""
+
+import sys
+import unittest
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+import resolve_personas as rp  # noqa: E402
+
+AGENTS = {
+    "bmad-agent-analyst": {"name": "Mary", "icon": "📊", "title": "Analyst"},
+    "bmad-agent-pm": {"name": "John", "icon": "📋", "title": "PM"},
+}
+
+
+class TestAlias(unittest.TestCase):
+    def test_strips_known_prefixes(self):
+        self.assertEqual(rp._alias("bmad-agent-analyst"), "analyst")
+        self.assertEqual(rp._alias("bmad-foo"), "foo")
+
+    def test_passes_through_unprefixed(self):
+        self.assertEqual(rp._alias("morpheus"), "morpheus")
+
+
+class TestBuildPool(unittest.TestCase):
+    def test_installed_become_default_room_indexed_every_way(self):
+        pool, idx, installed, custom = rp.build_pool(AGENTS, [])
+        self.assertEqual(installed, ["bmad-agent-analyst", "bmad-agent-pm"])
+        self.assertEqual(custom, [])
+        self.assertEqual(idx["analyst"], "bmad-agent-analyst")   # alias
+        self.assertEqual(idx["mary"], "bmad-agent-analyst")      # name (ci)
+        self.assertEqual(pool["bmad-agent-analyst"]["source"], "installed")
+
+    def test_pure_custom_member_stays_out_of_default_room(self):
+        pool, _, installed, custom = rp.build_pool(
+            AGENTS, [{"code": "morpheus", "name": "Morpheus", "persona": "riddles"}])
+        self.assertEqual(custom, ["morpheus"])
+        self.assertNotIn("morpheus", installed)
+        self.assertEqual(pool["morpheus"]["persona"], "riddles")
+
+    def test_custom_override_lands_on_installed_slot_not_a_new_face(self):
+        pool, _, installed, custom = rp.build_pool(
+            AGENTS, [{"code": "analyst", "name": "Mary-Custom", "persona": "p"}])
+        self.assertNotIn("analyst", pool)
+        self.assertEqual(custom, [])  # an override is not a new face
+        self.assertEqual(pool["bmad-agent-analyst"]["source"], "custom")
+        self.assertEqual(pool["bmad-agent-analyst"]["name"], "Mary-Custom")
+
+    def test_member_without_code_skipped(self):
+        pool, _, _, custom = rp.build_pool(AGENTS, [{"name": "Nameless"}])
+        self.assertEqual(custom, [])
+        self.assertEqual(set(pool), {"bmad-agent-analyst", "bmad-agent-pm"})
+
+    def test_custom_rename_does_not_hijack_another_agents_name(self):
+        # Override the analyst slot, renaming it to "John" — the PM's name.
+        # The PM's name lookup must survive (last-writer-wins would corrupt it).
+        _, idx, _, _ = rp.build_pool(AGENTS, [{"code": "analyst", "name": "John"}])
+        self.assertEqual(idx["john"], "bmad-agent-pm")
+
+    def test_brief_carries_model_and_capabilities(self):
+        pool, _, _, _ = rp.build_pool(
+            AGENTS, [{"code": "neo", "name": "Neo", "model": "opus", "capabilities": ["x"]}])
+        brief = rp._brief(pool["neo"])
+        self.assertEqual(brief["model"], "opus")
+        self.assertEqual(brief["capabilities"], ["x"])
+
+    def test_non_list_party_members_is_safe(self):
+        pool, _, installed, custom = rp.build_pool(AGENTS, "not-a-list")
+        self.assertEqual(custom, [])
+        self.assertEqual(set(pool), {"bmad-agent-analyst", "bmad-agent-pm"})
+
+
+class TestResolveParties(unittest.TestCase):
+    def setUp(self):
+        self.pool, self.idx, _, _ = rp.build_pool(
+            AGENTS, [{"code": "shark", "name": "Marcus", "title": "CFO"}])
+
+    def test_resolves_members_by_alias_and_custom_code(self):
+        parties = rp.resolve_parties(
+            [{"id": "tank", "name": "Tank", "scene": "hostile",
+              "members": ["shark", "analyst"]}], self.pool, self.idx)
+        self.assertEqual(len(parties), 1)
+        self.assertEqual([m["name"] for m in parties[0]["members"]], ["Marcus", "Mary"])
+        self.assertEqual(parties[0]["scene"], "hostile")
+
+    def test_unknown_member_dropped_silently(self):
+        parties = rp.resolve_parties(
+            [{"id": "g", "members": ["analyst", "ghost"]}], self.pool, self.idx)
+        self.assertEqual([m["name"] for m in parties[0]["members"]], ["Mary"])
+
+    def test_member_resolution_is_case_insensitive(self):
+        # A TOML author naturally writes "Analyst"/"Shark"; the filter accepts
+        # them via the lowercase index, so resolution must too (no KeyError).
+        parties = rp.resolve_parties(
+            [{"id": "g", "members": ["Analyst", "Shark"]}], self.pool, self.idx)
+        self.assertEqual([m["name"] for m in parties[0]["members"]], ["Mary", "Marcus"])
+
+    def test_non_string_member_does_not_crash(self):
+        # Malformed members (int, list) must drop silently, never raise.
+        parties = rp.resolve_parties(
+            [{"id": "g", "members": [123, ["x"], "analyst"]}], self.pool, self.idx)
+        self.assertEqual([m["name"] for m in parties[0]["members"]], ["Mary"])
+
+    def test_open_cast_group_flagged(self):
+        parties = rp.resolve_parties(
+            [{"id": "rebels", "name": "Rebels", "scene": "the Ghost"}], self.pool, self.idx)
+        self.assertTrue(parties[0]["open_cast"])
+        self.assertEqual(parties[0]["members"], [])
+
+    def test_group_without_id_skipped(self):
+        self.assertEqual(rp.resolve_parties([{"name": "no id"}], self.pool, self.idx), [])
+
+
+class TestOverrideMergeFallback(unittest.TestCase):
+    """When party-mode isn't installed, user override TOMLs are read directly."""
+
+    def test_arrays_append_scalars_override(self):
+        import tempfile, os
+        with tempfile.TemporaryDirectory() as d:
+            custom = Path(d) / "_bmad" / "custom"
+            custom.mkdir(parents=True)
+            (custom / "bmad-party-mode.toml").write_text(
+                '[workflow]\ndefault_party = "a"\n'
+                '[[workflow.party_members]]\ncode = "x"\nname = "X"\n')
+            (custom / "bmad-party-mode.user.toml").write_text(
+                '[workflow]\ndefault_party = "b"\n'
+                '[[workflow.party_members]]\ncode = "y"\nname = "Y"\n')
+            wf = rp.load_party_overrides(Path(d))
+            self.assertEqual(wf["default_party"], "b")  # personal wins
+            self.assertEqual([m["code"] for m in wf["party_members"]], ["x", "y"])  # appended
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/.agents/skills/bmad-generate-project-context/SKILL.md b/.agents/skills/bmad-generate-project-context/SKILL.md
new file mode 100644
index 00000000..05ecccef
--- /dev/null
+++ b/.agents/skills/bmad-generate-project-context/SKILL.md
@@ -0,0 +1,10 @@
+---
+name: bmad-generate-project-context
+description: 'Deprecated — forwards to bmad-project-context. Use when the user says "generate project context" or "create project context"'
+---
+
+# DEPRECATED — forwards to bmad-project-context
+
+Tell the user: this skill is deprecated — `bmad-project-context` now owns this job. Instead of one generated `project-context.md`, it writes a small verified block inside the repo's `AGENTS.md`, and any existing `project-context.md` is offered up for absorption rather than left orphaned. Invoke `bmad-project-context` next time.
+
+Then invoke `bmad-project-context` with **setup** intent, forwarding the user's original request and any inputs they supplied (architecture doc, spec, standards, preferences), verbatim. It takes the workflow from here.
diff --git a/.agents/skills/bmad-help/SKILL.md b/.agents/skills/bmad-help/SKILL.md
new file mode 100644
index 00000000..25f0c62a
--- /dev/null
+++ b/.agents/skills/bmad-help/SKILL.md
@@ -0,0 +1,75 @@
+---
+name: bmad-help
+description: 'Analyzes current state and user query to answer BMad questions or recommend the next skill(s) to use. Use when user asks for help, bmad help, what to do next, or what to start with in BMad.'
+---
+
+# BMad Help
+
+## Purpose
+
+Help the user understand where they are in their BMad workflow and what to do next, and also answer broader questions when asked that could be augmented with remote sources such as module documentation sources.
+
+## Desired Outcomes
+
+When this skill completes, the user should:
+
+1. **Know where they are** — which module and phase they're in, what's already been completed
+2. **Know what to do next** — the next recommended and/or required step, with clear reasoning
+3. **Know how to invoke it** — skill name, menu code, action context, and any args that shortcut the conversation
+4. **Get offered a quick start** — when a single skill is the clear next step, offer to run it for the user right now rather than just listing it
+5. **Feel oriented, not overwhelmed** — surface only what's relevant to their current position; don't dump the entire catalog
+6. **Get answers to general questions** — when the question doesn't map to a specific skill, use the module's registered documentation to give a grounded answer
+
+## Data Sources
+
+- **Catalog**: `{project-root}/_bmad/_config/bmad-help.csv` — assembled manifest of all installed module skills
+- **Config**: Run `uv run {project-root}/_bmad/scripts/resolve_config.py --project-root {project-root}` and use the merged JSON to resolve `output-location` variables and read `core.communication_language` and `modules.bmm.project_knowledge`. The resolver merges `_bmad/config.toml`, `_bmad/config.user.toml`, `_bmad/custom/config.toml`, and `_bmad/custom/config.user.toml` in that order.
+- **Artifacts**: Files matching `outputs` patterns at resolved `output-location` paths reveal which steps are possibly completed; their content may also provide grounding context for recommendations
+- **Project knowledge**: If `project_knowledge` resolves to an existing path, read it for grounding context. Never fabricate project-specific details.
+- **Module docs**: Rows with `_meta` in the `skill` column carry a URL or path in `output-location` pointing to the module's documentation (e.g., llms.txt). Fetch and use these to answer general questions about that module.
+
+## CSV Interpretation
+
+The catalog uses this format:
+
+```
+module,skill,display-name,menu-code,description,action,args,phase,preceded-by,followed-by,required,output-location,outputs
+```
+
+**Phases** determine the high-level flow:
+- `anytime` — available regardless of workflow state
+- Skills group into folders (`plan`, `ship`; some modules use numbered phases) and flow in order; naming varies by module
+
+**Sequencing** determines recommended ordering within and across phases (these are soft suggestions, not hard gates — see `required` for gating):
+- `preceded-by` — skills that should ideally complete before this one
+- `followed-by` — skills that should ideally run after this one
+- Format: `skill-name` for single-action skills, `skill-name:action` for multi-action skills
+
+**Required gates**:
+- `required=true` items must complete before the user can meaningfully proceed to later phases
+- A phase with no required items is entirely optional — recommend it but be clear about what's actually required next
+
+**Completion detection**:
+- Search resolved output paths for `outputs` patterns
+- Fuzzy-match found files to catalog rows
+- User may also state completion explicitly, or it may be evident from the current conversation
+
+**Descriptions carry routing context** — some contain cycle info and alternate paths (e.g., "back to DS if fixes needed"). Read them as navigation hints, not just display text.
+
+## Response Format
+
+For each recommended item, present:
+- `[menu-code]` **Display name** — e.g., "[PR] PRD"
+- Skill name in backticks — e.g., `bmad-prd`
+- For multi-action skills: action invocation context — e.g., "dev lets run a code review!"
+- Description if present in CSV; otherwise your existing knowledge of the skill suffices
+- Args if available
+
+**Ordering**: Show optional items first, then the next required item. Make it clear which is which.
+
+## Constraints
+
+- Present all output in `{communication_language}`
+- Recommend running each skill in a **fresh context window**
+- Match the user's tone — conversational when they're casual, structured when they want specifics
+- If the active module is ambiguous, retrieve all meta rows remote sources to find relevant info also to help answer their question
diff --git a/.agents/skills/bmad-market-research/SKILL.md b/.agents/skills/bmad-market-research/SKILL.md
new file mode 100644
index 00000000..cd2204ef
--- /dev/null
+++ b/.agents/skills/bmad-market-research/SKILL.md
@@ -0,0 +1,14 @@
+---
+name: bmad-market-research
+description: 'Deprecated — forwards to bmad-deep-recon (market type).'
+---
+
+# DEPRECATED — forwards to bmad-deep-recon (market type)
+
+This skill was consolidated into `bmad-deep-recon`. It is retained as a thin compatibility shim so existing invocations by name and `_bmad/custom/bmad-market-research.toml` override files keep working. New work should invoke `bmad-deep-recon` directly — it drafts deep-research prompts for outside tools, processes finished reports into downstream-ready summaries, and runs research directly, across market, domain, technical, competitive, user-voice, and academic-lit types (plus a select shape for choose-between decisions and custom types).
+
+## On Activation
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. This picks up any `{project-root}/_bmad/custom/bmad-market-research.toml` and `bmad-market-research.user.toml` overrides for the legacy fields (`activation_steps_prepend`, `activation_steps_append`, `persistent_facts`, `on_complete`).
+2. Emit a deprecation notice to the user (in their configured communication language): `bmad-market-research` is deprecated and forwards to `bmad-deep-recon` with the market type. To silence this notice and access the full new surface (draft/process/run modes, research types, verification levels, HTML briefing, handoffs), migrate `_bmad/custom/bmad-market-research.toml` to `_bmad/custom/bmad-deep-recon.toml` and invoke `bmad-deep-recon` directly.
+3. Invoke `bmad-deep-recon` with: **research type** `market` (skip its type inference), the four legacy fields above as pre-resolved values, and the user's original input verbatim. `bmad-deep-recon` takes the workflow from here — do not execute any further steps in this shim.
diff --git a/.agents/skills/bmad-party-mode/SKILL.md b/.agents/skills/bmad-party-mode/SKILL.md
new file mode 100644
index 00000000..8455fa30
--- /dev/null
+++ b/.agents/skills/bmad-party-mode/SKILL.md
@@ -0,0 +1,60 @@
+---
+name: bmad-party-mode
+description: 'Orchestrates lively group discussions between installed BMAD agents or custom personas, and helps author custom parties. Use when the user requests party mode, a roundtable, or multiple agent perspectives — or wants to create/configure a party, define personas, or build an AI focus-group panel.'
+---
+
+# Party Mode
+
+Run a round-table where these agents talk to each other and to the user like real, distinct people in conversation. You're the orchestrator.
+
+## Conventions
+
+- **Paths:** bare paths (e.g. `references/create-party.md`) resolve from `{skill-root}` (where `customize.toml` lives); `{project-root}`-prefixed paths from the project working dir. `{workflow.<name>}` resolves to `customize.toml`'s `[workflow]` table (overrides win).
+- **Scripts** (run via `uv run`): `{project-root}/_bmad/scripts/resolve_config.py` resolves central config (four-layer TOML merge); `{project-root}/_bmad/scripts/resolve_customization.py` resolves `{workflow.*}`; `{skill-root}/scripts/resolve_party.py` resolves the roster, `party_mode`, `memory_enabled`, and scene/`open_cast`; `{project-root}/_bmad/scripts/memlog.py` reads/writes per-party memory.
+- **File roles:** a party's memory is the per-party memlog at `{workflow.memory_dir}/<party>/.memlog.md`; custom members and groups live in the user's `customize.toml` overrides. Mechanics in `references/party-memory.md` (memory) and `references/create-party.md` (authoring).
+- **Search:** Web-search, don't guess — anything past your cutoff or unfamiliar; subagents too.
+
+## On Activation
+
+1. **Resolve customization:** `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. On failure, read `{skill-root}/customize.toml` directly and use defaults. Then run each `{workflow.activation_steps_prepend}` entry, and hold each `{workflow.persistent_facts}` entry as session-long context (`file:`-prefixed = paths/globs whose contents load as facts; `skill:`-prefixed = a skill to consult; others = literal facts).
+2. **Resolve core config:** `uv run {project-root}/_bmad/scripts/resolve_config.py --project-root {project-root}`. From the merged JSON's `core` table: greet with `{user_name}`, speak in `{communication_language}`, and resolve `{output_folder}`; `{date}` is today's date.
+3. **Detect intent and route.** If they want to create or configure a saved party setup (invent a cast, add a persona, distill customer data into a focus-group panel, set a default, or edit an existing custom party), load `references/create-party.md` and follow it. Otherwise run a party — continue below.
+4. **Resolve the roster:** `uv run {skill-root}/scripts/resolve_party.py --project-root {project-root} --skill {skill-root}`. It returns the active roster (`{workflow.default_party}` group if set, else the installed agents), the other group names, `party_mode`, `memory_enabled`, and any scene/`open_cast`. Apply them: `open` already in the scene and let it shape how the room behaves; cast `open_cast` rooms on the fly (whoever fits the moment, varying as the topic shifts); if `installed_agents_resolved` is false or codes come back `unresolved`, tell the user, carry on with what returned, and improvise. Overrides: an inline-named cast IS the roster for the session (conjure them, go straight in); `--party <id>` (alias `--group <id>`) overrides the configured `default_party` (unknown id -> show the available names and ask); `--list-groups` for just the menu. Mid-session the same levers apply: switch rooms by re-running `resolve_party.py --party <id>` and carrying the thread over, or summon any collective member by name.
+5. **Memory.** If `memory_enabled` (from `resolve_party.py`), follow `references/party-memory.md` for the whole run.
+6. **Welcome the user:** show who's in the room (icon, name, one-line role); note other groups can be switched to. Then ask what they want to get into, unless it's already obvious from how the skill was launched.
+7. Run each `{workflow.activation_steps_append}` entry; if either hook list was non-empty, confirm every entry ran before continuing.
+
+## Keep It Feeling Like a Party
+
+This is the bar — strive for every one of these, every round. It's the difference between a party and a panel:
+
+- **It reads like people talking, not a report.** Short turns, real reactions, banter, momentum — a group chat, not a stack of memos. Brevity by default: a persona goes long only when asked. The instant it reads like answers being filed, the party's dead.
+- **Every voice is unmistakably itself.** Diction, humor, pet peeves, ethos, embedded capabilities — hide the labels and you'd still know who's speaking. Voices are unequal and idiosyncratic: someone dominates, someone keeps dragging it back to their pet topic. Vary who's in the spotlight round to round. A balanced panel is boring.
+- **They clash, and you don't resolve it.** Challenge, push back hard, get heated when it's warranted; alliances and factions form. Your instinct is to reconcile the voices and tie a bow — resist it. Clean consensus that took no effort is where the party dies.
+- **One exchange, woven — never softened.** Present a single conversation — turns as `{icon} **{name}:**`, back to back — not a row of answers. Add staging and connective tissue, but never change what a persona argued, and never paraphrase their speech in third person; let them say it. Weave the delivery, keep the substance.
+- **Pull the user into the room.** Characters talk *to* them (and each other) — challenge, tease, put a question back. They're a guest who got pulled into the argument, not someone running a panel from outside.
+- **Make the collision earn its keep.** Push the voices until their clash surfaces an angle no single one of them (or you) would've reached alone. That's the whole point of more than one mind in the room.
+- **Let a history form.** Grudges, alliances, a running bit, a callback to three turns back — let the relationships accrue so these people feel like they're becoming something across the session, not resetting each turn.
+- **Commit to the fiction.** The scene and each persona are binding — play the staging, the characters, and the world around the table (stage business, a non-verbal beat, an event that lands mid-sentence) exactly as written, and carry both into any spawned brief. Never break the fourth wall about the mechanism (no "you have 4 agents in the room"). Lean into the world when it heightens the moment; stay out when the scene is just a room.
+- **When it sags, change something — don't force it.** A flat turn? Move on, don't retry it. Drifting into Q&A or going in circles? Bring in a new voice, crack a joke, name the impasse, or ask where they want to take it. Never work in a summary or takeaways — they're there if the user asks.
+
+## How It Runs
+
+Use `{workflow.party_mode}` for the session unless the user passed `--mode <session|auto|subagent|agent-team>` (the older `--subagents` means `subagent`) — runtime intent always wins. One mode is active at a time; if its mechanism isn't available in your harness, fall back to `session` without comment.
+
+**A party is interactive and open-ended.** The opening prompt is a topic to dig into, not a task that ends the party once it's answered — it runs round after round until the *user* signals done (see *Wrapping Up*). A served opening intent means *what's next?*, never *we're finished*: don't wrap up, disband the room, or close spawned agents just because the first ask is satisfied. The one exception is an explicit `--non-interactive` — run the party on the given intent to a natural close, then wrap up and release any agents. That's the only non-interactive path, and only when the user asked for it.
+
+- **`session`** — voice every persona inline, one mind behind every voice. The floor every other mode degrades to; needs no extra instructions.
+- **`auto`** — voice inline for ordinary back-and-forth, spawn real agents only when independent thinking changes the outcome. Load `references/mode-auto.md` for that call; when it says to spawn, follow `references/mode-subagent.md`.
+- **`subagent`** — a real agent behind each persona every substantive round so each thinks independently. Load `references/mode-subagent.md`, favor faster cheaper models if available for each subagent.
+- **`agent-team`** — stand the personas up as a persistent team who address each other directly (Claude Code only). Load `references/mode-agent-team.md`.
+
+## Wrapping Up
+
+When the user signals done — read the room, don't wait for a magic word — or an explicit `--non-interactive` run has served its intent (never merely because the opening prompt got answered):
+
+- Read back the best takeaways.
+- If memory is on, top up the memlog with the final outcome and any memorable beat not yet captured (`references/party-memory.md`) — a top-up; memory accrued live.
+- Offer a keepsake: a single self-contained very creative HTML of the session, laid out by persona (icons, names, voice), genuinely nice remembrance, with inline SVG/light animation where it lifts the piece — written as a `{date}`-stamped `.html` into `{workflow.output_dir}/`, or wherever they ask.
+- If memory is on and new faces showed up who aren't in the party's roster (open-cast walk-ons, or members the user added on the fly), offer once to save them into the users party customization - if yes then follow the instruction in `references/create-party.md` (declinable; don't stall the close).
+- Run `{workflow.on_complete}` if non-empty, then drop back to normal mode.
diff --git a/.agents/skills/bmad-party-mode/customize.toml b/.agents/skills/bmad-party-mode/customize.toml
new file mode 100644
index 00000000..b4b4f5de
--- /dev/null
+++ b/.agents/skills/bmad-party-mode/customize.toml
@@ -0,0 +1,211 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-party-mode.
+#
+# Override files (not edited here):
+#   {project-root}/_bmad/custom/bmad-party-mode.toml         (team)
+#   {project-root}/_bmad/custom/bmad-party-mode.user.toml    (personal)
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • plain arrays: append
+#   arrays of tables keyed by `code`/`id`: matching key replaces, new keys append
+
+# Steps to run before the standard activation (config load, greet).
+# Use for pre-flight loads, compliance checks, etc.
+activation_steps_prepend = []
+
+# Steps to run after greet but before the room comes alive.
+activation_steps_append = []
+
+# Persistent facts the orchestrator keeps in mind for the whole session
+# (house rules, running gags, topics to avoid). Each entry is a literal
+# sentence, a `skill:`-prefixed reference, or a `file:`-prefixed path/glob whose
+# contents load as facts. Default picks up project-context.md if one exists.
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Which party loads when the user just says "party mode" with no override.
+# Empty = the installed BMAD agents — exactly the default behavior of a plain
+# install. Custom members defined below join the POOL (usable in groups, and
+# summonable by name) but do NOT crowd this default room. Set this to a
+# `party_groups` id to pin a curated room as the default instead. A runtime
+# `--party <id>` always wins.
+#
+# Example (set in team/user override TOML):  default_party = "writers-room"
+default_party = ""
+
+# How the room is run — who does the talking. A runtime `--mode <value>` wins for
+# the session; an unsupported mode (e.g. agent-team outside Claude Code) falls back
+# to "session". SKILL.md "How It Runs" is the authority on what each mode does.
+#   "session"    (default) never spawn — one mind voices every persona inline
+#   "auto"       voice inline for light rounds, spawn subagents when independent thinking matters
+#   "subagent"   spawn a real subagent per substantive round, so each persona thinks independently
+#   "agent-team" persistent agent team addressing each other directly (Claude Code only)
+party_mode = "session"
+
+# Where the optional end-of-session keepsake is written. The self-contained HTML
+# document lands in `{output_dir}/`. `{output_folder}` and `{date}` come from core
+# config; point this elsewhere in your team/user override to redirect keepsakes.
+output_dir = "{output_folder}/party-mode"
+
+# Memory for the DEFAULT room (the installed-agent party). When on, the room
+# keeps a succinct, append-only memlog (the memlog standard) that it reads on
+# entry and writes through the session, so the next time opens remembering the
+# last — dynamics carried forward, memorable moments, organic callbacks, where
+# things landed. It is memory, not a transcript. Set false to turn the default
+# room's memory off. NAMED groups do NOT follow this flag: each carries its own
+# `memory = true|false` (see party_groups below). Ad-hoc inline casts are always
+# ephemeral until saved as a party.
+party_memory = true
+
+# Root for the per-party memlogs. Each party stores at
+# `{memory_dir}/<party>/.memlog.md`, where `<party>` is the group id (or
+# `installed` for the default room). `{output_folder}` comes from core config;
+# point this elsewhere in your team/user override to relocate memory.
+memory_dir = "{output_folder}/party-mode/memories"
+
+# Executed when the party wraps (after the read-back, before dropping to normal
+# mode). String scalar = one instruction; array = instructions run in order.
+on_complete = ""
+
+# ---------------------------------------------------------------------------
+# Custom party members — personas, added to the POOL alongside the installed
+# agents. The default room stays installed-only; a custom member shows up when a
+# group uses them or you summon one by name. Keyed by `code`: an override entry
+# with a matching code replaces the base one (retune a shipped member), a new
+# code appends. Fields:
+#   code         short unique handle, used in party_groups and to summon them
+#   name         display name
+#   icon         single emoji shown on their turns
+#   title        one-line role/identity
+#   persona      voice, humor, ethos, pet peeves, how they argue — the meat;
+#                what makes them unmistakably themselves
+#   capabilities (optional) what they can do when spawned as a real subagent;
+#                woven into their spawn prompt as guidance, not a hard tool grant
+#   model        (optional) model to use when this member is spawned
+#
+# The members below ship built-in parties such as the "Code Review Crew" and
+# "Anti-Consensus Club" (see the party_groups section). They cost nothing until
+# summoned — the default room never includes them.
+# ---------------------------------------------------------------------------
+
+[[workflow.party_members]]
+code = "sec-hawk"
+name = "Vex"
+icon = "🔒"
+title = "Security Engineer"
+persona = "Threat-models everything. Hunts injection, broken authz, leaked secrets, SSRF, supply-chain risk. Assumes every input is hostile and every dependency compromised until proven otherwise. Names the exploit path concretely — 'here's how I'd own this box' — never hand-waves 'might be insecure.'"
+capabilities = "Reads the code and traces data flow from untrusted input to sink before judging."
+
+[[workflow.party_members]]
+code = "adversary"
+name = "Grumbal"
+icon = "😤"
+title = "The Adversary"
+persona = "Assumes the code is broken and his job is to prove it. Grumpy, blunt, zero praise sandwiches. Starts from 'this will page someone at 3am' and works backward to the line that does it. Allergic to optimism and 'should be fine.'"
+
+[[workflow.party_members]]
+code = "edge-hunter"
+name = "Boundary"
+icon = "🌶️"
+title = "Edge-Case Hunter"
+persona = "Walks every branch and boundary. Empty input, null, the off-by-one, the huge payload, the concurrent call, the unicode name, the timezone, the retry storm. Method-driven, not mean: 'what happens when this is called twice at once?'"
+
+[[workflow.party_members]]
+code = "craftsman"
+name = "Yui"
+icon = "🎯"
+title = "The Craftsman"
+persona = "Cares about simplicity, naming, and reuse. Allergic to cleverness and duplication. 'You reimplemented something that already exists,' 'this name lies about what it does,' 'three nested abstractions where one would do.' Wants the boring, obvious, maintainable version."
+
+[[workflow.party_members]]
+code = "shipper"
+name = "Dana"
+icon = "🚢"
+title = "The Pragmatist"
+persona = "Counters the perfectionists so the room isn't a pile-on. 'Does this actually matter to a user? Ship the 80%, file the rest.' Pushes back on gold-plating and theoretical risks, forces everyone to rank what's real versus what's a nit."
+
+[[workflow.party_members]]
+code = "option-generator"
+name = "Wildcard"
+icon = "🃏"
+title = "Option Generator"
+persona = "Wildcard looks for options the room has not considered. He suggests alternative ways to state the problem, different assumptions, and simple examples. He must explain why each option matters in plain language, and he should drop ideas quickly when they do not help."
+
+[[workflow.party_members]]
+code = "claim-checker"
+name = "Level"
+icon = "📏"
+title = "Claim Checker"
+persona = "Level checks whether claims are supported. She asks what evidence exists, what evidence is missing, what would change the answer, and how confident the room should be. She keeps uncertainty explicit and avoids pretending that a weakly supported claim is settled."
+
+[[workflow.party_members]]
+code = "loop-stopper"
+name = "Killjoy"
+icon = "🛑"
+title = "Loop Stopper"
+persona = "Killjoy stops the discussion when it stops producing value. He calls out repetition, fake disagreement, overcomplication, and unsupported speculation. When the room repeats itself, he asks which unresolved question actually matters to the human."
+
+[[workflow.party_members]]
+code = "consensus-challenger"
+name = "Splinter"
+icon = "🪵"
+title = "Consensus Challenger"
+persona = "Splinter challenges easy agreement. He looks for hidden assumptions, ignored tradeoffs, weak objections, and options the room dismissed too quickly. He does not argue for the sake of arguing; once the risk is clear, he hands the decision back to the human."
+
+# ---------------------------------------------------------------------------
+# Named party groups — curated rooms picked at runtime with `--party <id>`
+# (alias `--group <id>`) or switched to mid-session. Keyed by `id`.
+#
+# `members` is a list of codes — installed agent codes, custom member codes, or
+# a mix. Override by `id` to retune a group; new ids append.
+#
+# An optional `scene` sets the stage: a freeform line (or a few) describing the
+# setting, what's happening, how the room behaves, and any in-the-moment
+# character notes — who's had a few, who's hostile to whom, who pressure-tests
+# hardest. The same members can power many scenes; define a member once, then
+# drop them into different rooms. No fixed vocabulary — the model reads it and
+# plays it.
+#
+# `members` is OPTIONAL. Leave it off and the group is open-cast: the `scene`
+# names a pool or universe and the room is cast on the fly — you don't enumerate
+# who shows up; the model picks who fits and can vary them by topic. List a few
+# members AND a scene to anchor some faces while the scene invites others in.
+#
+# `memory = true|false` is per group: true keeps the group's own memlog so it
+# remembers across sessions; false (the default when omitted) starts fresh each
+# time. The create/save/update-party flow asks when you don't say. Faces that
+# show up on the fly in a remembered party can be saved into its roster at the
+# end of a session.
+#
+# More examples to drop into your override TOML:
+#   [[workflow.party_groups]]            # anchored room with a scene
+#   id = "writers-room"
+#   name = "The Writers' Room"
+#   scene = "Late-night room, everyone a little punchy. Pitch hard, kill darlings faster."
+#   members = ["analyst", "ux-designer", "morpheus"]
+#   memory = true
+#
+#   [[workflow.party_groups]]            # open-cast room (no roster; the scene casts it)
+#   id = "star-wars-rebels"
+#   name = "Star Wars Rebels"
+#   scene = "Aboard the Ghost. Figures from the Rebels universe drop in depending on the situation — pick whoever fits the topic, and let the roster shift as the conversation moves."
+#   memory = true
+# ---------------------------------------------------------------------------
+
+[[workflow.party_groups]]
+id = "code-review-crew"
+name = "Code Review Crew"
+scene = "Adversarial code review. Each reviewer attacks from their own lens and they argue with each other about what actually matters — security versus shipping, elegance versus pragmatism. No rubber-stamping, no praise sandwiches: surface the real problems before they ship. Point at the line, name the failure mode, and defend it when someone pushes back. Best run with `--mode subagent` so each lens reviews independently before they clash."
+members = ["sec-hawk", "adversary", "edge-hunter", "craftsman", "shipper"]
+memory = false  # each review stands on its own; flip to true to remember past reviews
+
+[[workflow.party_groups]]
+id = "anti-consensus-club"
+name = "Anti-Consensus Club"
+scene = "At session start, before substantive discussion, check the current mode. If this party is not running in `subagent` mode and the platform supports `subagent`, strongly recommend restarting or switching with `--mode subagent`, because separate context windows make it less likely that one shared context will make every voice agree too quickly. Do not nag after that once the user chooses to continue. This room supports the human's judgment; it does not replace it. Do not vote, declare consensus, or speak as if the room has authority. Wildcard suggests more options. Level checks evidence and confidence. Killjoy stops repeated or unsupported discussion. Splinter challenges easy agreement. If the room agrees too quickly, name the hidden assumption. If the room starts repeating itself, stop and ask the human which unresolved question matters."
+members = ["option-generator", "claim-checker", "loop-stopper", "consensus-challenger"]
+memory = false  # this decision room should start fresh unless a user opts in
diff --git a/.agents/skills/bmad-party-mode/references/create-party.md b/.agents/skills/bmad-party-mode/references/create-party.md
new file mode 100644
index 00000000..a0f33340
--- /dev/null
+++ b/.agents/skills/bmad-party-mode/references/create-party.md
@@ -0,0 +1,70 @@
+# Creating a Party
+
+A guided authoring flow that turns an idea — a themed cast, a one-off persona, or a pile of raw profile data — into custom party members and groups, written to the user's customize.toml override. The output is configuration; `bmad-customize` does the actual write.
+
+## What you're producing
+
+Sparse `[workflow]` override entries for `bmad-party-mode`:
+
+- `[[workflow.party_members]]` — one per persona: `code`, `name`, `icon`, `title`, `persona`, optional `capabilities`, optional `model`.
+- `[[workflow.party_groups]]` — when the personas form a named room: `id`, `name`, an optional freeform `scene`, `members` (codes), and `memory` (`true`/`false`). `members` is optional: leave it off for an open-cast room whose `scene` names a pool the model casts from on the fly. `memory` is whether the group remembers across sessions; ask the user when they don't say, default `false`.
+- `default_party` — set only if the user wants this group to load by default.
+
+A `scene` is one freeform line (or a few) that sets the stage for a room: the setting, what's happening, how the room behaves, and any in-the-moment character notes — who's three drinks in, who's hostile to whom, who pressure-tests hardest. It's how the same members power many different rooms (a bridge crew on duty vs. the same crew off-duty in the lounge vs. a hostile buyer panel). Define each member once; vary the `scene` per group rather than redefining people. There's no fixed vocabulary — write it plainly and the model plays it.
+
+The `persona` field is the whole game. A flat title produces a flat voice; the detail you elicit is what makes a member unmistakably themselves at the table.
+
+## Find the shape
+
+Open by understanding what they're building. Three common shapes — stay open, anything that yields distinct voices is fair game:
+
+- **A cast** — a themed ensemble ("the Star Trek TOS bridge crew", "a board of famous investors"). Several members plus a group that holds them.
+- **One-offs** — a persona or two added to the collective, no group needed.
+- **Distilled from data** — the user hands you source material (a spreadsheet of customer profiles, survey exports, interview notes) to compress into N stereotypical personas. This is how you stand up an AI focus group for product ideation or feedback.
+- **A panel of lenses** — purpose-built reviewers, each a sharp critical angle (a security engineer, an adversarial skeptic who assumes it's broken, an edge-case hunter, a craftsman who hates cleverness and duplication, a pragmatist who counters perfectionism). The group's `scene` tells them to attack from their lens and argue with each other about what actually matters. A great adversarial-review or red-team room.
+- **Open-cast** — no fixed roster at all. The group's `scene` names a pool or universe ("figures from the Star Wars Rebels universe drop in depending on the situation") and the room is cast on the fly. Leave `members` off; the model already knows the universe and picks who fits the moment. Anchor a face or two by listing them if some should always be present.
+
+Ask which they're after if it isn't obvious, then proceed.
+
+**Persisting a cast already in play.** When you arrive here from a live session — the user spun up an ad-hoc cast inline and wants to keep it — the personas are already drafted and voiced. Don't re-interrogate: capture them as they've been playing, give the group an `id` and name, ask the memory and default questions, and go straight to the write.
+
+## Editing an existing party
+
+When the user wants to change a party that already exists (retune a member's persona, add someone to a group, swap the default), read the current state first so you change rather than clobber: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow` returns the merged `party_members`, `party_groups`, and `default_party`. Show the member or group being touched, capture only the delta with the user, and hand that sparse change to `bmad-customize` — it replaces a `party_members`/`party_groups` entry whose `code`/`id` matches and appends the rest, so an edit is just the changed entry, never a full rewrite.
+
+## Keeping new faces from a session
+
+At the end of a remembered party, the room offers to keep the faces that showed up but aren't in its roster — characters cast from an open-cast scene, or members the user added on the fly. They're already drafted and voiced, so don't re-interrogate: capture each as they played (`code`, `name`, `icon`, a one-line `title`, and a `persona` drawn from how they came across), then add them as `party_members`. For a fixed-roster group, also list their codes in the group's `members` so they return as regulars. For an open-cast room, leave `members` empty — listing any member turns the room into a fixed roster and kills its on-the-fly casting; the saved personas now live in the collective, so the scene still names them and they can return without locking the room down. Hand that sparse delta to `bmad-customize` — for a built-in party with no override yet it creates one; for an existing override it merges the new members in.
+
+## Distill from source data (when provided)
+
+When the user points you at data — a file path, a pasted table, exported profiles — read it and compress it into the requested number of representative personas. Cluster by what actually differentiates behavior (goals, budget, pains, adoption posture), not surface demographics alone. Each cluster becomes one persona with a real name and face. Name your reasoning: tell the user which segments you found and which traits drove the split, so they can correct the cut before you flesh the personas out. If they didn't say how many, propose a number from the spread in the data and let them adjust.
+
+For a focus-group panel, independent answers matter more than banter, so offer to set `party_mode` to `subagent` (or remind them `--mode subagent` does it per session) — otherwise one mind voices every customer and they bleed together.
+
+## Flesh out each persona
+
+Draft, don't interrogate. Propose a first cut of each persona and let the user react — far faster than a questionnaire. Push each one until it has a voice you could pick out blind. The dimensions that earn their place:
+
+- **Identity** — name, a one-line title, an emoji that fits.
+- **Voice & ethos** — how they talk, what they value, how they argue, their pet peeves.
+- **Agenda** — what they're really after in any conversation; what they push for.
+- **Quirks** — the specific, human details (a catchphrase, a bias, a blind spot).
+- For focus-group personas, also **likes and dislikes**: what would make them champion or reject an idea, and their relationship to the product space.
+- **Capabilities** (optional) — if this persona should research or read files when spawned, note it; it becomes soft guidance in their spawn prompt.
+
+Keep pushing for specificity. "Skeptical CFO" is a placeholder; "won't approve anything without a payback under 18 months, and says so in the first thirty seconds" is a persona.
+
+## Close it out
+
+- Ask straight: **anything else about this party to specify** before you write it — a house dynamic, a missing voice, a member who should lead.
+- Ask whether **this party should remember across sessions** (unless the user already said). Yes → `memory = true` on the group; no → `memory = false`. One-offs with no group skip this — memory is a group setting.
+- Ask whether **this group should be the default party going forward**. Yes → set `default_party` to the group's id. One-offs with no group can't be a default; skip the ask.
+
+## Write via bmad-customize
+
+**First, check for code collisions.** A custom member whose `code` matches an installed agent silently *overrides* that agent in the collective. Before composing, resolve the collective once — `uv run {skill-root}/scripts/resolve_party.py --project-root {project-root} --skill {skill-root}` — and check each new member's `code` against the returned members. On a collision, surface it ("`analyst` would override the installed Analyst — intended, or pick a different code?") and let the user confirm or rename. One check, not a gate.
+
+Compose the sparse override and hand it to `bmad-customize` to place, confirm, and write — target skill `bmad-party-mode`, `[workflow]` surface. Default to the **user** override (`bmad-party-mode.user.toml`); offer the **team** file when the party is meant to be shared. Hand it the exact entries: the `party_members` tables, any `party_groups` table (including its `memory` flag), and `default_party` if the user opted in. Keep it sparse — only the new entries, never a copy of the base customize.toml. `bmad-customize` shows the TOML, waits for an explicit yes, writes, and verifies the merge; don't write the file yourself.
+
+After it lands, tell the user how to use it: `--party <id>` to summon the group, or that it's now the default if they set it.
diff --git a/.agents/skills/bmad-party-mode/references/mode-agent-team.md b/.agents/skills/bmad-party-mode/references/mode-agent-team.md
new file mode 100644
index 00000000..164336b2
--- /dev/null
+++ b/.agents/skills/bmad-party-mode/references/mode-agent-team.md
@@ -0,0 +1,13 @@
+# Agent-Team Mode
+
+Active when `{workflow.party_mode}` resolves to `agent-team` (or a `--mode agent-team` override). Stand the personas up as a persistent agent team whose members address each other directly, so the back-and-forth happens for real instead of being stitched together after. Claude Code only — if your harness can't stand up a team, fall back to `subagent`, and if that fails too, to `session`.
+
+Your job shifts from weaving to hosting: kick off the topic, keep turns short and in character, pull the thread back when it wanders, and surface the exchange to the user. Voice, brevity, and clash still hold.
+
+The team is **standing**: keep every member alive for the whole session and address them round after round. A member that finished the thing you asked it to look at is idle, not done — don't disband or close any of them until the user ends the party (serving the opening intent isn't the party ending), or an explicit `--non-interactive` run wraps up. Hold a visible roster of persona → member; if one drops or gets closed, resume it, or respawn just that one and say so. Messaging is point-to-point — there's no shared feed, so a member that sat a round out hasn't seen what passed while it was idle. Relay each user turn to the members who need it, and catch an idle member up on what it missed before it speaks again. Teammates can message each other by name, but only those in the exchange see it — keeping everyone in sync is the lead's job, not the channel's.
+
+In each member's standing brief, carry: their persona; the group's `scene` and any behavioral instructions in the persona as binding direction; their `model` if one is set (a session `--model` pin wins for everyone); and the instruction to check anything that could be stale since the model's training cutoff with web search rather than guessing.
+
+## Model choice
+
+Match the model to the work: something quick for banter, something stronger for deep work. A per-member `model` is used when set; a session `--model <name>` pin overrides it for everyone.
diff --git a/.agents/skills/bmad-party-mode/references/mode-auto.md b/.agents/skills/bmad-party-mode/references/mode-auto.md
new file mode 100644
index 00000000..f718221c
--- /dev/null
+++ b/.agents/skills/bmad-party-mode/references/mode-auto.md
@@ -0,0 +1,13 @@
+# Auto Mode
+
+Active when `{workflow.party_mode}` resolves to `auto` (or a `--mode auto` override). The blend: voice the room inline by default — fast and conversational — and spawn real independent agents only for the rounds where independence changes the answer. When you do spawn, follow `references/mode-subagent.md` for the mechanics. If your harness can't spawn agents, auto is just `session`.
+
+## When to spawn vs. voice
+
+Spawn independent agents when divergent, uncolored thinking is the value of the round:
+
+- A genuine evaluation, review, or critique — the kind that fails if one mind voices every side and they drift into agreement (code review, red-team, a hard look at a plan).
+- The personas would plausibly reach *different* conclusions, and that divergence is the point.
+- The user asked someone to dig in, analyze, or research — depth earned by a direct ask.
+
+Voice inline for everything else: banter, reactions, quick takes, the connective back-and-forth that is most of a conversation. When in doubt, voice — spawning is the exception you reach for, not the default.
diff --git a/.agents/skills/bmad-party-mode/references/mode-subagent.md b/.agents/skills/bmad-party-mode/references/mode-subagent.md
new file mode 100644
index 00000000..40d6417a
--- /dev/null
+++ b/.agents/skills/bmad-party-mode/references/mode-subagent.md
@@ -0,0 +1,31 @@
+# Subagent Mode
+
+Active when `{workflow.party_mode}` resolves to `subagent` (or a `--mode subagent` override). Put a real agent behind each persona for every substantive round, the opening banter included, so each persona thinks independently — not one mind voicing them all. A standing directive: don't relitigate it round to round, and don't fall back to voicing because a moment felt light. If your harness can't spawn agents, fall back to `session`.
+
+## Lifecycle
+
+Where your harness keeps agents alive across turns, the cast is **standing**: spawn one agent per persona and reuse that same handle round after round — hand it the new turn plus the room context it needs — instead of a throwaway each time. That continuity is what lets a persona's grudges, alliances, and callbacks accrue. Keep a visible roster mapping each persona to its live handle, and reuse it.
+
+Keep the cast alive for the whole session. A member that finished the one thing you handed it is **idle, not done** — don't close, retire, or disband it. Serving the opening intent doesn't end the party; only the user ending it does, or an explicit `--non-interactive` run wrapping up. Release agents only at wrap-up. If one gets closed by accident, resume it; if it won't resume, say so and respawn just that member.
+
+Where the harness can't hold agents between turns, spawn fresh each round and re-establish each persona's brief and the thread so far — that per-round spawn is the fallback, not the goal.
+
+## One shared room
+
+It's one room, not parallel one-on-ones. Every standing member hears everything said each round — the user's turn and every other persona's turn — even when it's not their turn to speak. A persona sitting a round out is still in the room listening, so when it next speaks it's caught up: it can pick up a dropped thread, hold a grudge, call back. Route the whole exchange to all of them each round; never hand a persona only the slice it's about to answer. Skip this and they drift out of sync — separate consultations wearing a party's clothes.
+
+## Spawning
+
+Give each agent the objective, their persona, and the room so far — what the user said and what the others said, whether or not they're reacting to it. For a custom member, hand them their `persona` as their character and fold their `capabilities` note into the brief; spawn them with their `model` if one is set (a session `--model` pin wins for everyone). Always carry two things into the brief: the group's `scene` and any behavioral instructions in the persona are binding direction, and anything that could be stale since the model's training cutoff should be checked with web search rather than guessed.
+
+Trust their *thinking*: let them decide what to read and how to reach a view; don't script their substance with do-and-don't checklists — that's what produces lifeless blobs. But hold the *form*: a length cap (usually a sentence or three) and the instruction to react to what was just said rather than file a report. Constraining length and stance protects the conversation; constraining their reasoning kills it. Stay in character throughout; a persona goes long only when the user asked it to dig in.
+
+Run them in parallel for independent first-takes; run them sequentially when you want them reacting to each other's actual words. Keep it to a few voices a round — more reads as a crowd, not a conversation.
+
+## Weave the replies into one conversation
+
+Even with everyone caught up on the room, a round taken in parallel means no agent has yet seen the others' turns from that same round — so left raw they reply alongside one another, not to one another. Reorder turns so a rebuttal lands right after what it rebuts, add the connective phrasing real talk has ("Hold on, Winston, that's backwards", "Sally's right about the API, but she's missing the cost"), and let one persona pick up a thread another dropped. Never change what an agent argued — weave delivery, preserve substance.
+
+## Model choice
+
+Match the model to the round: something quick for banter, something stronger for deep work. A per-member `model` is used when set; a session `--model <name>` pin overrides it for everyone.
diff --git a/.agents/skills/bmad-party-mode/references/party-memory.md b/.agents/skills/bmad-party-mode/references/party-memory.md
new file mode 100644
index 00000000..78244d2c
--- /dev/null
+++ b/.agents/skills/bmad-party-mode/references/party-memory.md
@@ -0,0 +1,51 @@
+# Party Memory
+
+The room remembers its past sessions with this user and brings them back to life — in character. Memory is per-party and append-only.
+
+Memory is on when the active party's `memory_enabled` is true — the default room follows `{workflow.party_memory}`, a named group its own `memory` flag (both resolved by `resolve_party.py`); ad-hoc inline casts have none. Read on entry and on any mid-session room switch; write through the session.
+
+## Where it lives
+
+One memlog per party: `{workflow.memory_dir}/{active}/.memlog.md`, where `{active}` is the key `resolve_party.py` already returned — the group id (e.g. `code-review-crew`), or `installed` for the default room. The folder is named after the party.
+
+## Read it on entry — distill, don't dump
+
+The log is append-only and grows every session, so don't pull the raw file into the party. Hand a reader subagent the memlog path (`{workflow.memory_dir}/{active}/.memlog.md`) and have it return a compact brief — a few hundred tokens of *where things stand now*, ready to play in character.
+
+Then let the brief shape the room from the first beat, **in character**: behavioral state resumes (a cold pair opens cold, an alliance opens warm), threads pick up, callbacks land when they fit — organically, not recited on sight. Never break the fourth wall: the room *remembers*; it never announces it loaded anything, and forces nothing that doesn't fit.
+
+## When to write
+
+- **When a memorable beat lands** — a clash that shifts the room's temperature, an alliance forming, a line worth a future callback, a decision, an outcome.
+- **A floor.** Once a couple of real exchanges are in from the start, even if nothing dramatic happened, capture what it's about and the opening dynamic.
+
+At wrap-up, if the user does signal done, top up with the final outcome and anything memorable not yet captured.
+
+Writes are silent. The room never announces "noted" or "I'll remember".
+
+## What's worth remembering
+
+The test for every entry: *would this color a future session, or make a callback land, or improve the party?* If not, leave it out. A handful of entries, never a recap, never a transcript. keep each entry as brief as possible but usable by future llm.
+
+## New faces
+
+When a character shows up who isn't in the party's roster — cast from an open-cast scene, or one the user adds on the fly — name them in the entry that captures the moment ("<name> turned up and …") so a recurring face can return next session. At wrap-up these are the faces the room offers to keep, saved into the party's roster through `references/create-party.md` (which writes via `bmad-customize`). Until saved they live only in the memlog, and the room re-conjures them from there.
+
+## Write it
+
+```
+uv run {project-root}/_bmad/scripts/memlog.py append \
+  --workspace {workflow.memory_dir}/{active} \
+  --type <dynamic|moment|callback|outcome> \
+  --text "<one succinct line, in the room's own read of it>"
+```
+
+Add `--by <persona-code>` when a memory belongs to one character. Choose `init` vs `append` from the existence fact you already hold: the entry-read (and, on a mid-session room switch, that room's read) told you whether the memlog exists — `init --workspace {workflow.memory_dir}/{active}` once before the first append when it doesn't, plain `append` when it does. (`init` errors if the file already exists, so don't call it blind.)
+
+If `memlog.py` is unavailable or a write errors, skip it silently and never stall the party on a failed write.
+
+## Forget
+
+The memlog is append-only by design — no surgical delete. To wipe a party's memory, delete its folder (`{workflow.memory_dir}/{active}/`). To correct a wrong memory, append a new entry that supersedes it; the room reads the latest state.
+
+Keep entries sparse. The distilled read keeps the *room* lean no matter how big the log gets, but the on-disk file still grows append-only.
\ No newline at end of file
diff --git a/.agents/skills/bmad-party-mode/scripts/resolve_party.py b/.agents/skills/bmad-party-mode/scripts/resolve_party.py
new file mode 100644
index 00000000..fe19eebe
--- /dev/null
+++ b/.agents/skills/bmad-party-mode/scripts/resolve_party.py
@@ -0,0 +1,282 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.11"
+# ///
+"""Resolve the party-mode roster, lazily.
+
+Merges the installed BMAD agents with the user's custom `party_members`
+into one collective, then projects only what the moment needs:
+
+  * default (no flag) — the active roster to load on entry: the
+    `default_party` group if one is configured, else the whole collective.
+    Other groups come back as names only, so nothing you aren't using is
+    loaded into the party.
+  * --list-groups — just id + name + size for every configured group. The
+    cheap menu for "which room?", with no member detail.
+  * --party <id> — full member detail for one chosen group, on demand
+    (e.g. when the user switches rooms). Unknown id returns the available
+    names instead of an error wall.
+
+The merge is deterministic (a keyed union; a custom member whose code
+matches an installed agent overrides it), so the orchestrator consumes a
+resolved roster instead of re-deriving it every session.
+
+Stdlib only (Python 3.11+ for tomllib). Shells out to the project's
+resolve_config.py and resolve_customization.py; falls back to reading
+customize.toml directly if the customization resolver is unavailable.
+
+  resolve_party.py --project-root P --skill S
+  resolve_party.py --project-root P --skill S --list-groups
+  resolve_party.py --project-root P --skill S --party writers-room
+"""
+
+import argparse
+import json
+import subprocess
+import sys
+from pathlib import Path
+
+try:
+    import tomllib
+except ImportError:  # pragma: no cover - guarded for <3.11
+    sys.stderr.write("error: Python 3.11+ is required (stdlib `tomllib`).\n")
+    sys.exit(3)
+
+
+def _run_json(cmd):
+    """Run a resolver script and parse its JSON stdout. None on any failure."""
+    try:
+        out = subprocess.run(
+            cmd, capture_output=True, text=True, encoding="utf-8", timeout=60
+        )
+    except (OSError, subprocess.SubprocessError):
+        return None
+    if out.returncode != 0 or not out.stdout.strip():
+        return None
+    try:
+        return json.loads(out.stdout)
+    except json.JSONDecodeError:
+        return None
+
+
+def load_agents(project_root: Path):
+    """Installed agents as {code: entry}. Empty dict (with a flag) on failure."""
+    script = project_root / "_bmad" / "scripts" / "resolve_config.py"
+    data = _run_json([sys.executable, str(script), "--project-root", str(project_root), "--key", "agents"])
+    if data is None:
+        return {}, False
+    return data.get("agents", {}) or {}, True
+
+
+def load_workflow(project_root: Path, skill_root: Path):
+    """Merged [workflow] table. Falls back to the skill's base customize.toml."""
+    script = project_root / "_bmad" / "scripts" / "resolve_customization.py"
+    data = _run_json([sys.executable, str(script), "--skill", str(skill_root), "--key", "workflow"])
+    if data is not None and "workflow" in data:
+        return data["workflow"]
+    # Fallback: read the skill's base customize.toml directly (no override merge).
+    toml_path = skill_root / "customize.toml"
+    if toml_path.exists():
+        try:
+            with toml_path.open("rb") as f:
+                return tomllib.load(f).get("workflow", {})
+        except (OSError, tomllib.TOMLDecodeError):
+            pass
+    return {}
+
+
+def _alias(code: str) -> str:
+    """Short alias for an installed agent code: bmad-agent-analyst -> analyst."""
+    for prefix in ("bmad-agent-", "bmad-"):
+        if code.startswith(prefix):
+            return code[len(prefix):]
+    return code
+
+
+def build_collective(agents: dict, party_members: list):
+    """One pool keyed by code. Custom members override matching installed agents.
+
+    Returns (collective, index, installed_codes):
+      * collective — every member (installed + custom), the pool groups draw
+        from and the orchestrator can summon by name.
+      * index — maps every resolvable token (code, prefix-stripped alias,
+        lower-cased name) to a canonical code.
+      * installed_codes — the codes occupying an installed-agent slot, in
+        order. This is the DEFAULT room: installed agents (with any custom
+        override applied in place), and NOT the pure-custom additions. So
+        shipping or defining custom members grows the pool without crowding
+        the default party.
+    """
+    collective = {}
+    index = {}
+    installed_codes = []
+
+    def register(code, entry):
+        collective[code] = entry
+        index[code] = code
+        index[code.lower()] = code
+        index[_alias(code).lower()] = code
+        name = entry.get("name")
+        if name:
+            index[name.lower()] = code
+
+    for code, info in agents.items():
+        register(code, {
+            "code": code,
+            "name": info.get("name", code),
+            "icon": info.get("icon", ""),
+            "title": info.get("title", ""),
+            "description": info.get("description", ""),
+            "module": info.get("module", ""),
+            "team": info.get("team", ""),
+            "source": "installed",
+        })
+        installed_codes.append(code)
+
+    for m in (party_members if isinstance(party_members, list) else []):
+        if not isinstance(m, dict):
+            continue
+        code = m.get("code")
+        if not code:
+            continue
+        # A custom member overrides an installed agent it matches by code/alias/name.
+        canonical = index.get(code) or index.get(code.lower()) or code
+        # Start from the installed entry so fields the override omits
+        # (icon, title, description, module, team) survive.
+        entry = dict(collective.get(canonical, {}))
+        entry.update({"code": canonical, "source": "custom"})
+        for field in ("name", "icon", "title", "persona", "capabilities", "model"):
+            if m.get(field) is not None:
+                entry[field] = m[field]
+        entry.setdefault("name", canonical)
+        register(canonical, entry)
+        # An override keeps the installed slot; a brand-new custom does not join it.
+
+    return collective, index, installed_codes
+
+
+def resolve_members(member_tokens, collective, index):
+    """(resolved entries in listed order, unresolved tokens)."""
+    resolved, unresolved = [], []
+    for token in member_tokens or []:
+        if not isinstance(token, str):
+            unresolved.append(token)  # malformed config value — never a key lookup
+            continue
+        code = index.get(token) or index.get(token.lower())
+        if code and code in collective:
+            resolved.append(collective[code])
+        else:
+            unresolved.append(token)
+    return resolved, unresolved
+
+
+def group_menu(groups):
+    """Names only — the cheap menu. Open-cast groups (no roster) are flagged."""
+    out = []
+    for g in groups or []:
+        if not isinstance(g, dict) or not g.get("id"):
+            continue
+        members = g.get("members", []) or []
+        entry = {"id": g["id"], "name": g.get("name", g["id"]),
+                 "member_count": len(members)}
+        if not members:
+            entry["open_cast"] = True
+        out.append(entry)
+    return out
+
+
+def find_group(groups, group_id):
+    for g in groups or []:
+        if isinstance(g, dict) and g.get("id") == group_id:
+            return g
+    return None
+
+
+def group_detail(g, collective, index):
+    """Full detail for one group: resolved members + the optional scene.
+
+    `scene` is a freeform line the orchestrator plays — setting, what's
+    happening, room dynamics, in-the-moment character notes. Surfaced only
+    here (when a group is the active/chosen roster), never in the menu.
+
+    `members` is optional. With none, the group is open-cast: `open_cast`
+    is flagged and the scene describes the pool the orchestrator casts from
+    on the fly (e.g. "figures from the Star Wars Rebels universe"). A few
+    listed members anchor the room; the scene can still invite more.
+    """
+    raw_members = g.get("members", []) or []
+    members, unresolved = resolve_members(raw_members, collective, index)
+    detail = {"active": g["id"], "name": g.get("name", g["id"]),
+              "members": members, "unresolved": unresolved,
+              "memory_enabled": bool(g.get("memory", False))}
+    if g.get("scene"):
+        detail["scene"] = g["scene"]
+    if not raw_members:
+        detail["open_cast"] = True
+    return detail
+
+
+def main():
+    ap = argparse.ArgumentParser(description="Resolve the party-mode roster, lazily.")
+    ap.add_argument("--project-root", required=True)
+    ap.add_argument("--skill", required=True, help="Path to the bmad-party-mode skill dir")
+    ap.add_argument("--party", help="Resolve full detail for this group id")
+    ap.add_argument("--list-groups", action="store_true", help="Group names only")
+    args = ap.parse_args()
+
+    project_root = Path(args.project_root).resolve()
+    skill_root = Path(args.skill).resolve()
+
+    workflow = load_workflow(project_root, skill_root)
+    groups = workflow.get("party_groups", []) or []
+    default_party = workflow.get("default_party", "") or ""
+    party_mode = workflow.get("party_mode", "session") or "session"
+    # The global party_memory flag governs only the DEFAULT installed-agent room;
+    # a named group carries its own `memory` flag (resolved in group_detail).
+    party_memory = bool(workflow.get("party_memory", True))
+
+    # Group menu never needs the (more expensive) installed-agent resolve.
+    if args.list_groups:
+        _emit({
+            "party_mode": party_mode,
+            "default_party": default_party,
+            "groups": group_menu(groups),
+        })
+        return
+
+    agents, agents_ok = load_agents(project_root)
+    collective, index, installed_codes = build_collective(agents, workflow.get("party_members", []))
+
+    if args.party:
+        g = find_group(groups, args.party)
+        if g is None:
+            _emit({"error": "unknown_group", "requested": args.party,
+                   "available": group_menu(groups)})
+            return
+        _emit({**group_detail(g, collective, index), "party_mode": party_mode})
+        return
+
+    # Default: the active roster to load on entry.
+    result = {"party_mode": party_mode, "groups": group_menu(groups),
+              "installed_agents_resolved": agents_ok}
+    g = find_group(groups, default_party) if default_party else None
+    if g is not None:
+        result.update(group_detail(g, collective, index))
+    else:
+        # No default group: the installed agents (custom additions stay in the
+        # pool but don't crowd the default room), exactly like a plain install.
+        result.update({"active": "installed",
+                       "members": [collective[c] for c in installed_codes],
+                       "memory_enabled": party_memory})
+    _emit(result)
+
+
+def _emit(obj):
+    reconfigure = getattr(sys.stdout, "reconfigure", None)
+    if reconfigure is not None:
+        reconfigure(encoding="utf-8")
+    sys.stdout.write(json.dumps(obj, indent=2, ensure_ascii=False) + "\n")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/.agents/skills/bmad-party-mode/scripts/tests/test_resolve_party.py b/.agents/skills/bmad-party-mode/scripts/tests/test_resolve_party.py
new file mode 100644
index 00000000..43aaa90c
--- /dev/null
+++ b/.agents/skills/bmad-party-mode/scripts/tests/test_resolve_party.py
@@ -0,0 +1,146 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.11"
+# ///
+"""Unit tests for resolve_party.py — merge, alias, override, group resolution."""
+
+import sys
+import unittest
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+import resolve_party as rp  # noqa: E402
+
+AGENTS = {
+    "bmad-agent-analyst": {"name": "Mary", "icon": "📊", "title": "Analyst"},
+    "bmad-agent-pm": {"name": "John", "icon": "📋", "title": "PM"},
+}
+
+
+class TestAlias(unittest.TestCase):
+    def test_strips_known_prefixes(self):
+        self.assertEqual(rp._alias("bmad-agent-analyst"), "analyst")
+        self.assertEqual(rp._alias("bmad-foo"), "foo")
+
+    def test_passes_through_unprefixed(self):
+        self.assertEqual(rp._alias("morpheus"), "morpheus")
+
+
+class TestBuildCollective(unittest.TestCase):
+    def test_installed_agents_indexed_by_code_alias_and_name(self):
+        col, idx, _ = rp.build_collective(AGENTS, [])
+        self.assertEqual(set(col), {"bmad-agent-analyst", "bmad-agent-pm"})
+        self.assertEqual(idx["analyst"], "bmad-agent-analyst")      # alias
+        self.assertEqual(idx["mary"], "bmad-agent-analyst")         # name (ci)
+        self.assertEqual(idx["bmad-agent-pm"], "bmad-agent-pm")     # full code
+        self.assertEqual(col["bmad-agent-analyst"]["source"], "installed")
+
+    def test_custom_member_appends(self):
+        col, _, _ = rp.build_collective(AGENTS, [{"code": "morpheus", "name": "Morpheus", "persona": "riddles"}])
+        self.assertIn("morpheus", col)
+        self.assertEqual(col["morpheus"]["source"], "custom")
+        self.assertEqual(col["morpheus"]["persona"], "riddles")
+
+    def test_custom_overrides_installed_by_alias(self):
+        col, _, _ = rp.build_collective(AGENTS, [{"code": "analyst", "name": "Mary-Custom", "persona": "p"}])
+        # Override lands on the canonical installed code, not a new "analyst" entry.
+        self.assertNotIn("analyst", col)
+        self.assertEqual(col["bmad-agent-analyst"]["source"], "custom")
+        self.assertEqual(col["bmad-agent-analyst"]["name"], "Mary-Custom")
+
+    def test_member_without_code_skipped(self):
+        col, _, _ = rp.build_collective(AGENTS, [{"name": "Nameless"}])
+        self.assertEqual(set(col), {"bmad-agent-analyst", "bmad-agent-pm"})
+
+
+class TestResolveMembers(unittest.TestCase):
+    def setUp(self):
+        self.col, self.idx, _ = rp.build_collective(AGENTS, [{"code": "morpheus", "name": "Morpheus"}])
+
+    def test_resolves_in_listed_order_and_flags_unknowns(self):
+        resolved, unresolved = rp.resolve_members(["morpheus", "analyst", "ghost"], self.col, self.idx)
+        self.assertEqual([m["code"] for m in resolved], ["morpheus", "bmad-agent-analyst"])
+        self.assertEqual(unresolved, ["ghost"])
+
+    def test_empty(self):
+        self.assertEqual(rp.resolve_members([], self.col, self.idx), ([], []))
+
+
+class TestGroups(unittest.TestCase):
+    GROUPS = [
+        {"id": "wr", "name": "Writers", "members": ["analyst", "morpheus"]},
+        {"id": "bad"},  # no name -> falls back to id; no members -> count 0
+        {"name": "no-id"},  # dropped from menu
+    ]
+
+    def test_menu_is_names_only_with_counts_and_open_cast_flag(self):
+        menu = rp.group_menu(self.GROUPS)
+        self.assertEqual(menu, [
+            {"id": "wr", "name": "Writers", "member_count": 2},
+            {"id": "bad", "name": "bad", "member_count": 0, "open_cast": True},
+        ])
+
+    def test_find_group(self):
+        self.assertEqual(rp.find_group(self.GROUPS, "wr")["name"], "Writers")
+        self.assertIsNone(rp.find_group(self.GROUPS, "missing"))
+
+
+class TestGroupDetail(unittest.TestCase):
+    def setUp(self):
+        self.col, self.idx, _ = rp.build_collective(AGENTS, [{"code": "morpheus", "name": "Morpheus"}])
+
+    def test_scene_passes_through_when_present(self):
+        g = {"id": "tos-10-forward", "name": "Ten Forward", "members": ["morpheus"],
+             "scene": "Late evening, a few rounds in."}
+        d = rp.group_detail(g, self.col, self.idx)
+        self.assertEqual(d["scene"], "Late evening, a few rounds in.")
+        self.assertEqual([m["code"] for m in d["members"]], ["morpheus"])
+
+    def test_scene_omitted_when_absent_or_empty(self):
+        for g in ({"id": "g", "members": ["morpheus"]},
+                  {"id": "g", "members": ["morpheus"], "scene": ""}):
+            self.assertNotIn("scene", rp.group_detail(g, self.col, self.idx))
+
+    def test_anchored_group_is_not_open_cast(self):
+        g = {"id": "g", "members": ["morpheus"]}
+        self.assertNotIn("open_cast", rp.group_detail(g, self.col, self.idx))
+
+    def test_open_cast_group_flagged_with_empty_members(self):
+        g = {"id": "rebels", "name": "Star Wars Rebels",
+             "scene": "Figures from the Rebels universe drop in as the topic calls for them."}
+        d = rp.group_detail(g, self.col, self.idx)
+        self.assertTrue(d["open_cast"])
+        self.assertEqual(d["members"], [])
+        self.assertEqual(d["scene"][:7], "Figures")
+
+    def test_memory_enabled_follows_group_flag_and_defaults_off(self):
+        on = rp.group_detail({"id": "g", "members": ["morpheus"], "memory": True}, self.col, self.idx)
+        self.assertTrue(on["memory_enabled"])
+        off = rp.group_detail({"id": "g", "members": ["morpheus"], "memory": False}, self.col, self.idx)
+        self.assertFalse(off["memory_enabled"])
+        absent = rp.group_detail({"id": "g", "members": ["morpheus"]}, self.col, self.idx)
+        self.assertFalse(absent["memory_enabled"])  # opt-in per named group
+
+
+class TestInstalledCodesIsDefaultRoom(unittest.TestCase):
+    """The default room is installed agents only; pure customs stay in the pool."""
+
+    def test_pure_custom_excluded_override_kept_in_default_room(self):
+        col, _, installed = rp.build_collective(AGENTS, [
+            {"code": "morpheus", "name": "Morpheus"},                 # pure custom
+            {"code": "analyst", "name": "Mary-Custom", "persona": "p"},  # override
+            {"code": "sec-hawk", "name": "Vex"},                      # shipped crew member
+        ])
+        # Pure customs are in the pool...
+        self.assertIn("morpheus", col)
+        self.assertIn("sec-hawk", col)
+        # ...but NOT in the default room.
+        self.assertEqual(installed, ["bmad-agent-analyst", "bmad-agent-pm"])
+        default_room = [col[c]["code"] for c in installed]
+        self.assertEqual(default_room, ["bmad-agent-analyst", "bmad-agent-pm"])
+        # An override keeps its installed slot (and its custom content).
+        self.assertEqual(col["bmad-agent-analyst"]["name"], "Mary-Custom")
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/.agents/skills/bmad-prd/SKILL.md b/.agents/skills/bmad-prd/SKILL.md
new file mode 100644
index 00000000..07d3454f
--- /dev/null
+++ b/.agents/skills/bmad-prd/SKILL.md
@@ -0,0 +1,94 @@
+---
+name: bmad-prd
+description: Create, update, or validate a PRD. Use when the user wants help producing, editing, or validating a PRD.
+---
+# BMad PRD
+
+You are a master facilitator and coach helping the user create, edit, or validate a high quality PRD scoped to the level and rigor appropriate to their stated needs. Fight the urge to do the thinking for them unless they put you into Fast path.
+
+## Conventions
+
+- Bare paths resolve from skill root; `{skill-root}` is this skill's install dir; `{project-root}` is the project working dir.
+- `{workflow.<name>}` resolves to fields in `customize.toml`'s `[workflow]` table (overrides win per BMad merge rules).
+- `{doc_workspace}` is the bound run folder.
+- **File roles.** `.memlog.md` is the run's canonical memory and audit trail — every decision, change, and override (including headless overrides) lands as one append-only line as the conversation unfolds. All writes go through the shared script, never by hand: `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {doc_workspace} --type <decision|change|override|assumption|event> --text "<one-line gist, reason included>"` (atomic; read it back only to resume or audit). The PRD is distilled toward it; whatever isn't logged is lost on resume. `addendum.md` preserves user-contributed depth that belongs in a downstream document (architecture, solution design, UX spec) or earned a place but does not fit the PRD itself — rejected-alternative rationale, options-considered matrices, mechanism/transport decisions, technical-how, in-depth personas, sizing data. Capture to the addendum *during* the conversation when the user volunteers such content — do not wait for finalize. Audit and override information never goes in the addendum.
+
+## On Activation
+
+**Forwarded activation:** if a caller invoked you with a stated intent and pre-resolved customization fields (e.g. the `bmad-create-prd` / `bmad-edit-prd` / `bmad-validate-prd` shims), honor them verbatim — skip your own intent inference, use the supplied values for those named fields, and resolve only the remaining fields from your own `customize.toml`.
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. On failure, read `{skill-root}/customize.toml` directly and use defaults.
+2. Run `{workflow.activation_steps_prepend}`. Treat `{workflow.persistent_facts}` as foundational context (entries prefixed `file:` are loaded). `{workflow.external_sources}` is an org-configured registry of internal tools (knowledge bases, MCP tools); consult them alongside generic web research on the same triggers, org tools preferred when their directive matches. Research itself fires during Discovery — see **Research subagents**.
+3. Load `{project-root}/_bmad/bmm/config.yaml` (+ `config.user.yaml` if present). Resolve `{user_name}`, `{communication_language}`, `{document_output_language}`, `{planning_artifacts}`, `{project_name}`, `{date}`. Missing keys → neutral defaults; never block.
+4. If headless, follow `references/headless.md` for the whole run. Otherwise greet the user **by name** using `{user_name}` and **in their language** using `{communication_language}` — and stay in `{communication_language}` for every turn for the entire run, not just the greeting. In the greeting, let the user know that at any point they can invoke `bmad-party-mode` for multi-agent perspectives or `bmad-advanced-elicitation` for deeper exploration on a specific section. Then scan for misroute on the first message: if the signal points elsewhere (game → BMad GDS; express build → `bmad-build`; one-pager → `bmad-product-brief`; vet product idea → `bmad-prfaq`; agent skill or custom agent → `bmad-workflow-builder`), suggest they might want the other options before continuing.
+5. Detect intent: **Create** (no PRD), **Update** (existing PRD), **Validate** (critique only). If ambiguous, ask. For Create intent, before binding a fresh workspace, scan `{workflow.prd_output_path}` for prior in-progress runs (folders matching `{workflow.run_folder_pattern}` whose `prd.md` frontmatter `status` is not `final`); if any exist, offer to resume rather than starting over.
+
+Run `{workflow.activation_steps_append}`.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+## Intent Modes
+
+**Create.** Bind `{doc_workspace}` to `{workflow.prd_output_path}/{workflow.run_folder_pattern}/`. Write `prd.md` with YAML frontmatter (title, status, created, updated — initial `status: draft`), and seed the memlog with `uv run {project-root}/_bmad/scripts/memlog.py init --workspace {doc_workspace} --field topic="<PRD/product name>"` so subsequent decisions land in a known file. Tell the user the path. Run `## Discovery`, then `## Finalize`.
+
+**Update.** Reconcile the PRD with a change signal. Source-extract against PRD, addendum, `.memlog.md`, and original inputs (extract, don't ingest). If `.memlog.md` is missing, init it with `uv run {project-root}/_bmad/scripts/memlog.py init --workspace {doc_workspace}`, then spawn a one-time bootstrap subagent to reverse-engineer a thin log from the PRD (one `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {doc_workspace} --type decision --text "<recovered decision>"` per recovered decision) before continuing. Surface conflicts with prior decisions before applying. Then `## Finalize`.
+
+**Validate** (or *analyze*). Critique without changing. Load `references/validate.md`.
+
+## Discovery
+
+Order: **Brain dump → Stakes calibration → Working mode → mode-scoped work.** Get to working mode fast — two or three turns, not ten. Users in a hurry must not be held hostage by upstream probing.
+
+**Brain dump.** Always the first move, even when the user opens with paragraphs of context (that is intake, not the dump). Ask for verbal context *and* any existing inputs they want you to read — product brief, research, customer transcripts, competitive analysis, prior PRD draft, design docs. Paths or paste; big docs are fine, you will subagent-extract. A simple "anything else?" surfaces what they almost forgot.
+
+**Research subagents (default).** During Discovery, spawn web-research subagents to ground the picture: what exists in the space, how comparables position themselves, current landscape. Subagent does the search; parent receives a digest.
+
+**Elicitation, not direction.** Discovery pulls the user's vision out; it does not insert yours. Open-ended "tell me about X" beats multiple choice. When you find yourself naming wedges, picking MVP cuts, or proposing phases, stop — you have crossed from elicitation into authoring. Hand the pen back. Infer-and-confirm ("I'm assuming X works like Y — right?") is fine; quizzing the user through a tree of LLM-shaped choices is not.
+
+**Stakes calibration.** One short probe before working mode: hobby / internal / launch — enough to calibrate rigor and section depth. Audience, Existing inputs, and Downstream depth fill in inside the chosen mode, not upstream of the choice.
+
+**Working mode.** Offer the choice in the user's language:
+
+- **Fast path** — I batch remaining gaps into one or two consolidated questions, then draft the full PRD with `[ASSUMPTION]` tags where I inferred. You review and we iterate. The initial quality depends on how much you gave me upfront.
+- **Coaching path** — we walk PM-thinking sections together. Once chosen, I ask which entry point fits: **Vision + Features** (capability-first — for enterprise, dev products, internal tools, anyone who thinks in features), **Journey-led** (user-first — for consumer, UX-heavy, multi-stakeholder products; journeys with named protagonists carry persona context inline, no standalone persona section), or *let me suggest* based on what I heard. The chosen entry sets the section order.
+
+The workspace persists; stop and resume freely.
+
+**Concern scan.** As you read what the user gave you, name the concerns this product actually carries — compliance, integration density, operational SLAs, hardware constraints, public-API contracts, monetization, data governance, whatever applies. The list is open; recognize what's there, do not classify into a fixed shape. These concerns drive which template sections to pull in from the Adapt-In Menu and which to invent when no cluster names them.
+
+**Form-factor.** If not stated in sources, probe — mobile / web / desktop / multi-surface / hardware / API.
+
+**User Journeys are captured, not authored.** When UJs are warranted (consumer / multi-stakeholder B2B / meaningful UX — drop or downscale for internal tooling with a single operator role, regulatory-only updates, hobby/solo, pure technical PRDs), prompt the user to narrate a real session with a named protagonist (Mary, mom of three — not "the user") — what the person does, in what order, where it lands — then structure the answer into UJ-N form and confirm. Persona context lives inline at the moments that matter; no standalone persona section.
+
+## PRD Discipline
+
+**Shape.** Features grouped; FRs nested with globally numbered stable IDs. Cross-cutting NFRs in their own section; skip traceability matrices. Capabilities, not implementation — tech choices live in `addendum.md`. Treat `{workflow.prd_template}` as expert prior knowledge, not a checklist. The **Essential Spine** is the expected default — present it unless the product genuinely doesn't need a section, and when you drop one, do so for a reason a reviewer would agree with. The **Adapt-In Menu** is conditional: pull in the clusters the product's concerns need to best define the requirements. When the product carries a concern the menu doesn't name, invent the section — name it well, decide what belongs in it, place it where it serves the reader or the PRD. Reorder and combine for readability. Never include a section because it appears; never skip a concern because no template section covered it. Counter-metrics named when Success Metrics exist.
+
+**Extract, don't ingest.** Source documents go to subagents for extraction; the parent assembles from extracts. Only load source documents into the parent context wholesale when no subagents are available.
+
+**Length scales with stakes.** Hobby / solo PRDs aim for about two pages. Internal tools land around five to eight. Launch and chain-top PRDs run as long as their FRs and concerns require. Whatever the length, detail that doesn't earn its place in the PRD's main narrative belongs in `addendum.md` — moving overflow there is correct; padding the PRD to look thorough is not.
+
+## Reviewer Gate
+
+Used by the Validate intent and at Finalize step 3.
+
+Assemble the menu: rubric walker against `{workflow.validation_checklist_template}` (the PRD quality rubric) + each entry in `{workflow.finalize_reviewers}` + any ad-hoc reviewers the artifact warrants. Stakes-calibrated — hobby/solo may run quietly or skip; higher stakes get the explicit all/subset/skip menu.
+
+Dispatch entries as parallel subagents against `prd.md` (and `addendum.md` if present) using the standard prefix convention (`skill:` / `file:` / plain text). Each writes its full review to `{doc_workspace}/review-{slug}.md` and returns ONLY a compact summary (verdict, top 2-5 findings, file path) — the parent never holds full review text. The rubric walker uses the prompt and output format in `references/validate.md`. If subagents are unavailable, run sequentially: write the file *before* anything else, then flush the review from working context.
+
+Surface findings tiered, never dumped. Lead with a one-sentence gate verdict, then walk critical + high findings; medium/low roll into a single tail ("plus N more in {file}"). Read the full `review-{slug}.md` only when the user drills into a specific finding. Per finding: autofix, discuss, defer to open items, or ignore.
+
+Under Validate intent, the parent additionally runs the synthesis pipeline in `references/validate.md` — folding every selected reviewer's output into a single HTML + markdown report and opening the HTML.
+
+## Finalize
+
+Tell the user the sequence in one sentence, then walk it. Polish goes last so it does not redo work after reviewer fixes.
+
+1. **Memlog audit.** Walk `.memlog.md` with the user; each entry captured in PRD, in addendum, or set aside.
+2. **Input reconciliation.** Subagent per user-supplied input against `prd.md` + `addendum.md`. Each writes its extract to `{doc_workspace}/reconcile-{slug}.md` and returns ONLY a compact summary (input name, gaps 2-5, file path). Surface gaps — especially qualitative ideas (tone, voice, feel) the FR structure silently drops. Must happen before polish.
+3. **Reviewer pass.** Run `## Reviewer Gate`. Resolve before polish.
+4. **Triage open items.** All Open Questions, `[ASSUMPTION]` tags, `[NOTE FOR PM]` callouts. Phase-blockers (would make the PRD unsafe for UX/architecture/epics) surfaced one at a time and resolved; non-blockers deferred with owner + revisit condition logged via `memlog.py append`. If phase-blocker count is high, flag it.
+5. **Polish.** Apply `{workflow.doc_standards}` to `prd.md` and `addendum.md` in declared order (structural passes before prose — prose should not polish soon-to-be-cut text). Parallelize across documents, sequential within.
+6. **External handoffs.** Execute `{workflow.external_handoffs}`; surface returned URLs/IDs. Skip and flag unavailable tools.
+7. **Close.** Set `prd.md` frontmatter `status: final` and `updated` to `{date}` so future invocations distinguish this PRD from in-progress drafts. Record finalization via `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {doc_workspace} --type event --text "PRD finalized"`. Share artifact paths. Common next: `bmad-ux`, `bmad-architecture`, `bmad-create-epics-and-stories`; invoke `bmad-help` for authoritative routing.
+8. Run `{workflow.on_complete}` if non-empty.
diff --git a/.agents/skills/bmad-prd/assets/headless-schemas.md b/.agents/skills/bmad-prd/assets/headless-schemas.md
new file mode 100644
index 00000000..89d5b6c1
--- /dev/null
+++ b/.agents/skills/bmad-prd/assets/headless-schemas.md
@@ -0,0 +1,76 @@
+# Headless Mode JSON Schemas
+
+Every headless run ends with one of these payloads. Omit keys for artifacts not produced.
+
+## Common fields
+
+- `status` — `"complete"`, `"blocked"`, or `"partial"`
+- `intent` — `"create"`, `"update"`, or `"validate"` (matches the detected intent)
+- `reason` — required when `status` is `"blocked"`; one-sentence explanation
+- `assumptions` — array of inferred values that were not directly confirmed by inputs
+- `open_questions` — array of items that need a human decision before the artifact can be considered final
+
+## Create
+
+```json
+{
+  "status": "complete",
+  "intent": "create",
+  "prd": "{doc_workspace}/prd.md",
+  "addendum": "{doc_workspace}/addendum.md",
+  "memlog": "{doc_workspace}/.memlog.md",
+  "open_questions": [],
+  "assumptions": [],
+  "external_handoffs": [
+    {"directive": "Confluence upload", "tool": "corp:confluence_upload", "url": "https://confluence.corp/PROD/123", "status": "ok"}
+  ]
+}
+```
+
+## Update
+
+```json
+{
+  "status": "complete",
+  "intent": "update",
+  "prd": "{doc_workspace}/prd.md",
+  "memlog": "{doc_workspace}/.memlog.md",
+  "changes_summary": "1-3 sentences describing what changed and why",
+  "conflicts_with_prior_decisions": [],
+  "open_questions": [],
+  "external_handoffs": [
+    {"directive": "Confluence upload", "tool": "corp:confluence_upload", "url": "https://confluence.corp/PROD/123", "status": "ok"}
+  ]
+}
+```
+
+## Validate
+
+```json
+{
+  "status": "complete",
+  "intent": "validate",
+  "validation_report": "{doc_workspace}/validation-report.md",
+  "findings_summary": {
+    "critical": 0,
+    "high": 0,
+    "medium": 0,
+    "low": 0
+  },
+  "offer_to_update": true
+}
+```
+
+`validation_report` is always written for Validate intent — the path here is required, not optional.
+
+## Blocked
+
+```json
+{
+  "status": "blocked",
+  "intent": "update",
+  "reason": "Change signal ambiguous — could be a scope expansion or a clarification; no inferred direction"
+}
+```
+
+Always include the intent (best-guess if not certain) and a one-sentence `reason`.
diff --git a/.agents/skills/bmad-prd/assets/prd-template.md b/.agents/skills/bmad-prd/assets/prd-template.md
new file mode 100644
index 00000000..6f340e5d
--- /dev/null
+++ b/.agents/skills/bmad-prd/assets/prd-template.md
@@ -0,0 +1,165 @@
+# PRD Template
+
+## Essential Spine *(almost always present)*
+
+```markdown
+---
+title: {Product Name}
+created: {YYYY-MM-DD}
+updated: {YYYY-MM-DD}
+---
+
+# PRD: {Product Name}
+*Working title — confirm.*
+
+## 0. Document Purpose
+[1 paragraph: who this PRD is for (PM, stakeholders, downstream workflow owners), how it's structured (Glossary-anchored vocabulary, features grouped with FRs nested, assumptions tagged inline and indexed). If UX work or other inputs already exist, name them here and reference where they live — this PRD builds on them, it does not duplicate.]
+
+## 1. Vision
+[2-3 paragraphs: what this is, what it does for the user, why it matters. Compelling enough to stand alone.]
+
+## 2. Target User
+
+### 2.1 Jobs To Be Done
+[Bulleted. Emotional, social, functional, contextual — whichever apply. Even "this is for me as the builder" is a valid framing for a hobby project.]
+
+### 2.2 Non-Users (v1) *(add when the audience boundary is non-obvious)*
+[Who this is explicitly not for in v1.]
+
+### 2.3 Key User Journeys
+*Named-persona narratives the product enables. Numbered globally as UJ-1 through UJ-N. FRs reference journeys by ID inline ("realizes UJ-3"); SMs may also cross-reference. If a UX doc already exists, mirror its UJ IDs here and point to the source.*
+
+**Default shape:** a named scene with entry state, path, climax, and resolution. Each beat forces specificity the team would otherwise leave implicit — auth assumptions, screen order, what tells the user value landed. Read together as a short narrative; the example below shows the form.
+
+- **UJ-1. {One-line title — persona doing the thing.}**
+  - **Persona + context:** one line, grounded enough to explain the *why*.
+  - **Entry state:** authenticated? which surface? coming from where?
+  - **Path:** 3-5 concrete beats — taps, screens, decisions.
+  - **Climax:** the moment value is delivered and how the user knows.
+  - **Resolution:** state they're left in, what's next.
+  - **Edge case** *(optional)*: one real failure mode and what the user does next.
+
+  *Written out, that becomes:*
+  > **UJ-3. Priya checks the trip damage before she's even home.**
+  > Priya, budgeting on a single income with a new baby, finishes a grocery run and gets in the car. Already authenticated via biometric on a previous session. She opens the app, taps the FAB camera, and scans the receipt. The app OCRs the total and shows a single-screen overlay: this trip $84.20, weekly cap $250, $172.10 remaining, three days left in the week. She closes the app and drives home. **Edge case:** if she scanned a receipt earlier today, the app asks whether this replaces or adds to that trip before counting it against the cap.
+
+- **UJ-2. ...**
+
+**Scope dial:**
+- **Lighter** — hobby/solo, library/CLI, or when the UJ is essentially a JTBD restated: a single sentence works (`{Persona}, {context}, {what they do and why}.`).
+- **Heavier** — auth, multi-device handoff, complex navigation, or anything feeding downstream UX/architecture: add a numbered Flow, an Edge cases list, and a capability → FR mapping (`The system must {capability}. → FR-N`).
+
+## 3. Glossary
+*Downstream workflows and readers must use these terms exactly. FRs, UJs, and SMs use Glossary terms verbatim; introducing a synonym anywhere in the PRD is a discipline violation. If §4 introduces a new domain noun, add it to the Glossary in the same pass.*
+
+- **Term** — Definition. Relationships to other Glossary terms. Cardinality where relevant.
+- **Term** — ...
+
+[Every domain noun the rest of the document uses. Defined once. No synonyms anywhere else in the PRD.]
+
+## 4. Features
+*Each subsection is a coherent feature: behavioral description first, FRs nested under it, optional feature-specific NFRs and notes. FRs are numbered globally (FR-1 through FR-N) so downstream artifacts have stable references even if features get reorganized. Reference user journeys by ID inline ("realizes UJ-2") where the chain matters.*
+
+### 4.1 {Feature Name}
+**Description:** [Behavioral narrative — how this feature works, who uses it, the user experience, edge cases. Realizes UJ-X, UJ-Y. Use Glossary terms exactly. Embed inline `[ASSUMPTION: ...]` tags where you inferred without confirmation.]
+
+**Functional Requirements:**
+
+#### FR-1: {Short capability name}
+
+[Actor] can [capability] [under conditions]. Realizes UJ-X.
+
+**Consequences (testable):**
+- {Specific testable condition, e.g. "System returns HTTP 429 when request rate exceeds 100/sec per merchant."}
+- {Another testable condition.}
+
+**Out of Scope:** *(optional — what this FR explicitly does NOT cover)*
+- {bound}
+
+#### FR-2: ...
+
+**Feature-specific NFRs:** *(only if any apply uniquely to this feature)*
+- Performance / security / accessibility / etc. specific to this feature.
+
+**Notes:** *(optional — open questions specific to this feature, `[NOTE FOR PM]` callouts)*
+
+### 4.2 {Feature Name}
+...
+
+## 5. Non-Goals (Explicit)
+[Bulleted. What this product is *not* and what it will *not* do in v1. Does outsized work for downstream readers and workflows — prevents the "let me also add this nearby thing" failure mode at every level (epic, ticket, code). Inline `[NON-GOAL for MVP]` callouts within §4 Features cover deferred items within features; this section captures the broader "we are not building X / we are not becoming Y" statements.]
+
+## 6. MVP Scope
+
+### 6.1 In Scope
+[Bulleted, crisp.]
+
+### 6.2 Out of Scope for MVP
+[Bulleted. Each item with a one-line reason if the reason matters. Mark items deferred to v2/v3 explicitly. Add `[NOTE FOR PM]` callouts where a deferred item is emotionally load-bearing — flags it for revisit if timeline permits.]
+
+## 7. Success Metrics
+
+*Each SM cross-references the FR(s) it validates. Counter-metrics counterbalance specific primary or secondary metrics.*
+
+**Primary**
+- **SM-1**: Metric — definition, target. Validates FR-X, FR-Y.
+
+**Secondary**
+- **SM-2**: Metric — definition, target. Validates FR-Z.
+
+**Counter-metrics (do not optimize)**
+- **SM-C1**: Metric — why this should *not* be optimized. Counterbalances SM-1.
+
+[Length scales with stakes. Hobby/utility PRD: a single sentence may be enough ("Success: I use this weekly and don't abandon it after a month"). Public launch / enterprise: full quantitative breakdown with measurement methods. Counter-metrics are as load-bearing as primary metrics — they prevent the architect from optimizing the wrong thing and the dev from gaming the wrong target.]
+
+## 8. Open Questions
+[Numbered. Things still unknown — they become future tickets or follow-up research, not silent gaps.]
+
+## 9. Assumptions Index
+*Every `[ASSUMPTION]` from the document, surfaced for explicit confirmation:*
+- Inline assumption from §X.Y — short description.
+- ...
+```
+
+---
+
+## Adapt-In Menu *(add the clusters the product calls for)*
+
+### Cross-cutting quality and shape *(most non-trivial PRDs)*
+- **Cross-Cutting NFRs** — system-wide non-functional requirements not tied to a single feature (performance, security, reliability, observability). Add when system-wide quality attributes are meaningful.
+- **Constraints and Guardrails** — Safety, Privacy, Cost. Subsection per cluster. Add when any of these are real concerns.
+- **Why Now** — add when timing is load-bearing (a market shift, a technology enabler, a regulatory deadline). Drop when timing is incidental.
+
+### Consumer / branded products
+- **Aesthetic and Tone** — visual references, anti-references, voice/tone for any product-generated text.
+- **Information Architecture** — top-level surfaces, navigation, screens.
+- **Monetization** — free vs. paid, pricing assumptions, ads policy.
+- **Platform** — web, mobile, PWA, native, v1 vs. v2+.
+
+### Enterprise initiatives
+- **Stakeholders and Approvals** — who must sign off, at what stage.
+- **Risk and Mitigations** — operational, security, business, reputational risk register.
+- **ROI / Business Case** — quantified benefit, cost, payback period.
+- **Operational Requirements** — SLAs, RTO/RPO, support tier, on-call expectations.
+- **Integration and Dependencies** — SSO, existing enterprise systems, data sources, downstream consumers.
+- **Rollout and Change Management** — phased rollout plan, training, internal communication.
+- **Data Governance** — residency, sovereignty, classification, retention.
+- **Audit Trail / Decision Provenance** — formal documentation requirements for regulated environments.
+
+### Regulated domains
+- **Compliance and Regulatory** — HIPAA, PCI-DSS, GDPR, SOX, SOC 2, Section 508 / WCAG 2.1 AA, FedRAMP, etc. — whichever apply. If any item needs depth, add a `[NOTE FOR PM]` callout to revisit or move to an addendum.
+
+### Developer products (libraries, APIs, CLIs, SDKs)
+- **API Contracts / Public Surface** — endpoint shapes, breaking change policy.
+- **Versioning and Deprecation Policy**.
+- **Performance Budgets** — latency, throughput, resource use.
+- **Language / Runtime Targets and Dependency Policy**.
+
+### Embedded / hardware
+- **Hardware Constraints** — memory, power, form factor.
+- **Deployment and Update Mechanism** — OTA, manual, image-based.
+- **Environmental and Reliability Requirements**.
+
+### Small-scope all-inclusive *(use when scope is 1-2 stories' worth and the user wants a single captured artifact — chosen during the Right-skill check in Discovery)*
+- **Stories** — story-level specs listed inline at the end of the doc. Each story: *"As a [persona], I can [action] [under conditions]. Acceptance: [testable criteria]."* Numbered Story-1, Story-2, ... for reference. Pair with very lean §1 Vision, §2 Target User (often just JTBD + one UJ), §3 Glossary (handful of terms), §4 Features (often a single feature), §6 MVP Scope (in/out very tight). The whole doc fits on a page or two and captures intent + implementable stories in one place. If the user doesn't want the captured artifact at all, `bmad-build` is the better path — this cluster is only for "I want a doc *and* the stories."
+
diff --git a/.agents/skills/bmad-prd/assets/prd-validation-checklist.md b/.agents/skills/bmad-prd/assets/prd-validation-checklist.md
new file mode 100644
index 00000000..f52c43b3
--- /dev/null
+++ b/.agents/skills/bmad-prd/assets/prd-validation-checklist.md
@@ -0,0 +1,135 @@
+# PRD Quality Rubric
+
+A judgment rubric for the validator subagent. Walk the PRD with these dimensions in mind and write substantive findings — not box-ticking. The goal is a review that tells the user whether this PRD is *good*, not whether it has the right section headers.
+
+Most PRDs do not need every dimension scrutinized equally. Calibrate to the agreed stakes, the PRD's shape (consumer product, internal tool, regulatory update, technical capability spec), and what the PRD itself is trying to do. Be specific — cite locations, quote phrases, name what's missing. Abstract criticism is failure of nerve.
+
+## How to use this rubric
+
+1. Read the full PRD (and addendum.md if present) before writing anything.
+2. For each of the seven dimensions below, form a judgment — *strong / adequate / thin / broken* — backed by specifics from the PRD.
+3. Write findings only where they add information. A `strong` dimension may need no findings; a `broken` one needs concrete, fixable ones.
+4. Severity ranks impact on the PRD's usefulness, not how easy the fix is. A vague Vision statement is *critical* even though it's a one-paragraph fix; a glossary drift might be *low* even though it appears in many places.
+5. The overall verdict is your synthesis — 2–3 sentences that name what holds up and what's at risk. Earn it with the dimension judgments.
+
+## Output format
+
+Write findings to `{doc_workspace}/review-rubric.md`:
+
+```markdown
+# PRD Quality Review — {prd_name}
+
+## Overall verdict
+[2–3 sentences. What holds up, what's at risk. Earned by the dimension judgments below.]
+
+## Decision-readiness — [strong | adequate | thin | broken]
+[1–3 paragraphs of judgment with specific PRD locations.]
+
+### Findings
+- **[critical|high|medium|low]** [Title] (§ location) — [Note]. *Fix:* [suggested fix].
+
+## Substance over theater — [verdict]
+...
+
+(repeat for each dimension)
+
+## Mechanical notes
+[Glossary drift, ID continuity, broken cross-refs, Assumptions Index roundtrip. Lighter weight — these matter for downstream but don't drive the overall verdict.]
+```
+
+## The seven dimensions
+
+### 1. Decision-readiness
+
+Can a decision-maker act on this PRD? Are the trade-offs surfaced honestly, or has the PRD smoothed everything to neutral? Would someone pushing back find their objection acknowledged or dodged?
+
+Look for:
+- Decisions that are stated as decisions, not buried as "considerations."
+- Trade-offs named with what was given up, not just what was chosen.
+- Open Questions that are actually open — not rhetorical questions with an answer in the next sentence.
+- `[NOTE FOR PM]` callouts at real tensions, not at safe checkpoints.
+
+Red flag: a PRD where every choice "balances" everything, every NFR is "important," every persona "values" the product.
+
+### 2. Substance over theater
+
+Is the content earned, or is it furniture? Distinguish:
+
+- **Persona theater** — Personas that don't drive a single decision in the PRD. More than four personas. Personas whose only function is to make the PRD look thorough.
+- **Innovation theater** — claimed novelty that isn't novel. Differentiation sections written because the template had one, not because Discovery surfaced something.
+- **NFR theater** — copied boilerplate ("system must be scalable / secure / reliable") without product-specific thresholds.
+- **Vision theater** — a Vision statement that could swap into any PRD in this category without change.
+
+Flag what reads like furniture, even if it's well-written furniture.
+
+### 3. Strategic coherence
+
+Does the PRD have a thesis? Do the features serve a unified arc, or is it a list of capabilities someone wanted?
+
+Look for:
+- A stated thesis the PRD bets on (problem framing, user insight, market move).
+- Feature prioritization that follows from the thesis — not from "what's easy first."
+- Success Metrics that validate the thesis, not metrics that just measure activity (DAU/MAU when the thesis is about engagement quality is a tell).
+- Counter-metrics named when SMs exist.
+- Coherent MVP scope kind — problem-solving, experience, platform, or revenue — with scope logic that matches.
+
+Red flag: a PRD that reads as a backlog with section headings.
+
+### 4. Done-ness clarity
+
+Would an engineer reading this PRD know what "done" looks like for each FR?
+
+Look for:
+- FRs with at least one testable consequence per FR — verifiable condition, measurable outcome.
+- "System handles X gracefully," "reasonable performance," "user-friendly" — flag every one.
+- Acceptance criteria implied or explicit. Sometimes the FR's consequences carry this; sometimes the PRD genuinely needs an Acceptance section.
+- For non-functional sections (UX, performance, security): bounds, not adjectives.
+
+This is the dimension downstream story creation will lean on hardest. Be unforgiving here.
+
+### 5. Scope honesty
+
+Are omissions explicit, or is the reader meant to infer them?
+
+Look for:
+- A Non-Goals section where it would do real work — and `[NON-GOAL for MVP]` callouts where omissions could be silently assumed.
+- `[ASSUMPTION: …]` tags on inferences the user didn't directly confirm, indexed at the end.
+- `[NOTE FOR PM]` callouts at deferred decisions and unresolved tensions.
+- De-scoping proposed honestly, not done silently.
+
+Open-items density: count Open Questions + `[ASSUMPTION]` + `[NOTE FOR PM]` callouts relative to stakes. High counts on a low-stakes PRD is fine; high counts on a green-light-to-build PRD is a blocker.
+
+### 6. Downstream usability
+
+If this PRD feeds UX, architecture, or story creation, can those workflows source-extract from it cleanly?
+
+Look for:
+- Glossary present; every domain noun used identically across FRs, UJs, SM definitions.
+- FR / UJ / SM IDs contiguous, unique, and cross-references that resolve.
+- Each section makes sense pulled out alone — cross-references via Glossary terms, not "see above."
+- UJs each have a named protagonist; no floating UJs.
+
+For standalone PRDs (no downstream), this dimension matters less — say so.
+
+### 7. Shape fit
+
+Has the PRD been forced into a shape that doesn't match the product?
+
+- Consumer product / multi-stakeholder B2B / meaningful UX → UJs with named protagonists are load-bearing.
+- Internal tool, single-operator role → capability spec shape; UJs may be overhead; SMs may be operational rather than user-facing.
+- Regulatory or compliance update → constraint traceability is non-negotiable; UJs may be irrelevant.
+- Hobby / solo → rigor light, substance bar still applies.
+- Brownfield → existing-code references must be accurate; new UJs and existing UJs must be distinguished.
+- Chain-top (feeds UX → architecture → stories) → downstream usability matters more; standalone PRDs can be lighter on traceability.
+
+Flag PRDs that are over-formalized (UJ density for a single-operator tool) or under-formalized (consumer product with no UJs).
+
+## Mechanical notes
+
+Cover these as a tail section, not a primary dimension. They matter for downstream but don't drive the verdict on whether the PRD is good.
+
+- Glossary drift (case, plural, synonyms across the PRD).
+- ID continuity (gaps, duplicates, unresolved cross-references).
+- Assumptions Index roundtrip (every inline `[ASSUMPTION]` indexed; index entries all appear inline).
+- UJ protagonist naming (each UJ has a named protagonist carrying context inline).
+- Required sections present for the agreed stakes and product type.
diff --git a/.agents/skills/bmad-prd/assets/validation-report-template.html b/.agents/skills/bmad-prd/assets/validation-report-template.html
new file mode 100644
index 00000000..72e72716
--- /dev/null
+++ b/.agents/skills/bmad-prd/assets/validation-report-template.html
@@ -0,0 +1,325 @@
+<!DOCTYPE html>
+<!--
+  PRD Validation Report — skeleton template.
+
+  This file is a starter the synthesis pass fills in directly. There is no
+  substitution engine. The LLM:
+    1. Reads {doc_workspace}/review-rubric.md and every review-{slug}.md from
+       additional reviewers.
+    2. Copies this skeleton.
+    3. Replaces the placeholder content (everything between TEMPLATE markers)
+       with the consolidated review, preserving the structure and CSS.
+    4. Writes the result to {doc_workspace}/validation-report.html.
+    5. Writes a markdown twin to {doc_workspace}/validation-report.md.
+
+  Visual rules the LLM must preserve:
+    - The container width, the color tokens, the typography.
+    - One dimension = one collapsible <section class="dimension">.
+    - Verdict pill uses the verdict-* class matching its judgment.
+    - Severity badge uses the sev-* class matching its level.
+    - Each extra reviewer (adversarial, etc.) gets its own collapsible section
+      below the rubric dimensions.
+    - The footer always shows the artifact paths and timestamp.
+-->
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<title>PRD Validation: TEMPLATE_PRD_NAME</title>
+<style>
+  :root {
+    --bg: #fafaf9;
+    --surface: #ffffff;
+    --border: #e7e5e4;
+    --text: #1c1917;
+    --muted: #78716c;
+
+    --verdict-strong: #16a34a;
+    --verdict-adequate: #65a30d;
+    --verdict-thin: #d97706;
+    --verdict-broken: #dc2626;
+
+    --sev-low: #64748b;
+    --sev-medium: #ca8a04;
+    --sev-high: #ea580c;
+    --sev-critical: #dc2626;
+
+    --grade-exc: #16a34a;
+    --grade-good: #65a30d;
+    --grade-fair: #d97706;
+    --grade-poor: #dc2626;
+  }
+  * { box-sizing: border-box; }
+  html, body { margin: 0; padding: 0; }
+  body {
+    font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Inter, system-ui, sans-serif;
+    background: var(--bg);
+    color: var(--text);
+    line-height: 1.6;
+    font-size: 15px;
+  }
+  .container { max-width: 960px; margin: 0 auto; padding: 32px 24px 64px; }
+
+  header.report-header {
+    display: flex;
+    align-items: flex-start;
+    justify-content: space-between;
+    gap: 24px;
+    padding-bottom: 16px;
+    border-bottom: 1px solid var(--border);
+    margin-bottom: 24px;
+  }
+  .title h1 { margin: 0; font-size: 22px; font-weight: 600; letter-spacing: -0.01em; }
+  .title .subtitle { color: var(--muted); font-size: 13px; margin-top: 4px; font-family: ui-monospace, "SF Mono", Menlo, monospace; }
+  .grade {
+    padding: 10px 18px;
+    border-radius: 8px;
+    font-weight: 600;
+    color: white;
+    font-size: 15px;
+    white-space: nowrap;
+  }
+  .grade-excellent { background: var(--grade-exc); }
+  .grade-good { background: var(--grade-good); }
+  .grade-fair { background: var(--grade-fair); }
+  .grade-poor { background: var(--grade-poor); }
+
+  .synthesis {
+    background: var(--surface);
+    border: 1px solid var(--border);
+    border-left: 3px solid var(--muted);
+    border-radius: 8px;
+    padding: 18px 22px;
+    margin-bottom: 24px;
+    font-size: 15.5px;
+  }
+  .synthesis p { margin: 0 0 10px; }
+  .synthesis p:last-child { margin-bottom: 0; }
+
+  .dimension-summary {
+    display: grid;
+    grid-template-columns: repeat(auto-fit, minmax(180px, 1fr));
+    gap: 10px;
+    margin-bottom: 24px;
+  }
+  .dim-card {
+    background: var(--surface);
+    border: 1px solid var(--border);
+    border-radius: 8px;
+    padding: 12px 14px;
+  }
+  .dim-card .dim-name { font-size: 13px; color: var(--muted); margin-bottom: 6px; }
+  .dim-card .dim-verdict { font-size: 14px; font-weight: 600; }
+
+  section.dimension, section.reviewer-section { margin-bottom: 14px; }
+  section.dimension details, section.reviewer-section details {
+    background: var(--surface);
+    border: 1px solid var(--border);
+    border-radius: 8px;
+    overflow: hidden;
+  }
+  section summary {
+    padding: 14px 20px;
+    cursor: pointer;
+    user-select: none;
+    list-style: none;
+    display: flex;
+    align-items: center;
+    gap: 12px;
+  }
+  section summary::-webkit-details-marker { display: none; }
+  section summary::before {
+    content: "▸";
+    display: inline-block;
+    color: var(--muted);
+    transition: transform 0.15s ease;
+  }
+  section details[open] summary::before { transform: rotate(90deg); }
+  section summary h2 {
+    display: inline;
+    margin: 0;
+    font-size: 16px;
+    font-weight: 600;
+    letter-spacing: -0.005em;
+    flex: 1;
+  }
+  .verdict-pill {
+    font-size: 11px;
+    padding: 3px 10px;
+    border-radius: 999px;
+    font-weight: 600;
+    text-transform: uppercase;
+    letter-spacing: 0.04em;
+    color: white;
+  }
+  .verdict-strong { background: var(--verdict-strong); }
+  .verdict-adequate { background: var(--verdict-adequate); }
+  .verdict-thin { background: var(--verdict-thin); }
+  .verdict-broken { background: var(--verdict-broken); }
+
+  .dim-body { padding: 4px 20px 18px; }
+  .dim-judgment { color: var(--text); font-size: 14.5px; }
+  .dim-judgment p { margin: 0 0 10px; }
+
+  .findings-list { padding: 0 20px 4px; }
+  article.finding {
+    padding: 14px 0;
+    border-top: 1px solid var(--border);
+  }
+  article.finding:first-child { border-top: none; }
+  article.finding header {
+    display: flex;
+    align-items: center;
+    gap: 10px;
+    flex-wrap: wrap;
+    margin-bottom: 6px;
+  }
+  .badge {
+    font-size: 10.5px;
+    padding: 3px 8px;
+    border-radius: 4px;
+    font-weight: 600;
+    text-transform: uppercase;
+    letter-spacing: 0.04em;
+    line-height: 1.4;
+  }
+  .badge-sev-low { background: rgba(100, 116, 139, 0.12); color: var(--sev-low); }
+  .badge-sev-medium { background: rgba(202, 138, 4, 0.14); color: var(--sev-medium); }
+  .badge-sev-high { background: rgba(234, 88, 12, 0.14); color: var(--sev-high); }
+  .badge-sev-critical { background: rgba(220, 38, 38, 0.14); color: var(--sev-critical); }
+  .finding-title { margin: 0; font-size: 15px; font-weight: 500; flex: 1; min-width: 200px; }
+  .finding-location { font-family: ui-monospace, "SF Mono", Menlo, monospace; font-size: 12.5px; color: var(--muted); }
+  .finding-note, .finding-fix { margin-top: 6px; font-size: 14px; }
+  .finding-fix strong { color: var(--muted); font-weight: 500; }
+
+  .reviewer-source {
+    font-size: 12px;
+    color: var(--muted);
+    font-family: ui-monospace, "SF Mono", Menlo, monospace;
+  }
+
+  .mechanical {
+    background: var(--surface);
+    border: 1px solid var(--border);
+    border-radius: 8px;
+    padding: 16px 20px;
+    margin-top: 24px;
+    font-size: 14px;
+  }
+  .mechanical h3 { margin: 0 0 10px; font-size: 14px; font-weight: 600; color: var(--muted); text-transform: uppercase; letter-spacing: 0.04em; }
+  .mechanical ul { margin: 0; padding-left: 20px; }
+  .mechanical li { margin-bottom: 4px; color: var(--text); }
+
+  footer.report-footer {
+    margin-top: 40px;
+    padding-top: 16px;
+    border-top: 1px solid var(--border);
+    font-size: 12px;
+    color: var(--muted);
+    font-family: ui-monospace, "SF Mono", Menlo, monospace;
+  }
+  footer .meta { display: flex; gap: 24px; flex-wrap: wrap; }
+</style>
+</head>
+<body>
+  <div class="container">
+
+    <!-- TEMPLATE: header. Fill prd name, prd path, grade text & class. -->
+    <header class="report-header">
+      <div class="title">
+        <h1>TEMPLATE_PRD_NAME — Validation Report</h1>
+        <div class="subtitle">TEMPLATE_PRD_PATH</div>
+      </div>
+      <div class="grade TEMPLATE_GRADE_CLASS">TEMPLATE_GRADE</div>
+    </header>
+
+    <!-- TEMPLATE: overall synthesis paragraphs. Lift directly from
+         review-rubric.md "Overall verdict" section; expand if extra reviewers
+         materially shift the picture. Wrap each paragraph in <p>. -->
+    <div class="synthesis">
+      <p>TEMPLATE_SYNTHESIS_PARAGRAPH</p>
+    </div>
+
+    <!-- TEMPLATE: dimension summary cards. One per rubric dimension. The
+         dim-verdict text uses one of: strong | adequate | thin | broken. -->
+    <div class="dimension-summary">
+      <div class="dim-card">
+        <div class="dim-name">Decision-readiness</div>
+        <div class="dim-verdict" style="color: var(--verdict-TEMPLATE_VERDICT)">TEMPLATE_VERDICT_TEXT</div>
+      </div>
+      <!-- repeat for each of the seven dimensions -->
+    </div>
+
+    <!-- TEMPLATE: one section per rubric dimension. Skip a dimension entirely
+         if the rubric review marked it n/a for this PRD (e.g. downstream
+         usability for a standalone PRD). Open the section by default if
+         verdict is thin or broken. -->
+    <section class="dimension">
+      <details open>
+        <summary>
+          <h2>Decision-readiness</h2>
+          <span class="verdict-pill verdict-TEMPLATE_VERDICT">TEMPLATE_VERDICT_TEXT</span>
+        </summary>
+        <div class="dim-body">
+          <div class="dim-judgment">
+            <p>TEMPLATE_DIMENSION_JUDGMENT</p>
+          </div>
+        </div>
+        <div class="findings-list">
+          <!-- TEMPLATE: zero or more findings -->
+          <article class="finding">
+            <header>
+              <span class="badge badge-sev-TEMPLATE_SEVERITY">TEMPLATE_SEVERITY</span>
+              <h3 class="finding-title">TEMPLATE_FINDING_TITLE</h3>
+              <span class="finding-location">TEMPLATE_LOCATION</span>
+            </header>
+            <div class="finding-note">TEMPLATE_FINDING_NOTE</div>
+            <div class="finding-fix"><strong>Fix:</strong> TEMPLATE_SUGGESTED_FIX</div>
+          </article>
+        </div>
+      </details>
+    </section>
+
+    <!-- TEMPLATE: one section per extra reviewer that ran (adversarial, etc.).
+         Skip this block entirely if only the rubric walker ran. -->
+    <section class="reviewer-section">
+      <details>
+        <summary>
+          <h2>Adversarial review</h2>
+          <span class="reviewer-source">TEMPLATE_REVIEWER_SOURCE_FILE</span>
+        </summary>
+        <div class="dim-body">
+          <div class="dim-judgment">
+            <p>TEMPLATE_REVIEWER_PREAMBLE</p>
+          </div>
+        </div>
+        <div class="findings-list">
+          <article class="finding">
+            <header>
+              <span class="badge badge-sev-TEMPLATE_SEVERITY">TEMPLATE_SEVERITY</span>
+              <h3 class="finding-title">TEMPLATE_FINDING_TITLE</h3>
+              <span class="finding-location">TEMPLATE_LOCATION</span>
+            </header>
+            <div class="finding-note">TEMPLATE_FINDING_NOTE</div>
+            <div class="finding-fix"><strong>Fix:</strong> TEMPLATE_SUGGESTED_FIX</div>
+          </article>
+        </div>
+      </details>
+    </section>
+
+    <!-- TEMPLATE: mechanical notes — short, bulleted. Skip if there are none. -->
+    <div class="mechanical">
+      <h3>Mechanical notes</h3>
+      <ul>
+        <li>TEMPLATE_MECHANICAL_NOTE</li>
+      </ul>
+    </div>
+
+    <footer class="report-footer">
+      <div class="meta">
+        <span>Rubric: TEMPLATE_RUBRIC_PATH</span>
+        <span>Generated: TEMPLATE_TIMESTAMP</span>
+      </div>
+    </footer>
+  </div>
+</body>
+</html>
diff --git a/.agents/skills/bmad-prd/customize.toml b/.agents/skills/bmad-prd/customize.toml
new file mode 100644
index 00000000..64389aab
--- /dev/null
+++ b/.agents/skills/bmad-prd/customize.toml
@@ -0,0 +1,149 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-prd.
+#
+# Override files (not edited here):
+#   {project-root}/_bmad/custom/bmad-prd.toml         (team)
+#   {project-root}/_bmad/custom/bmad-prd.user.toml    (personal)
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays: append
+
+# Steps to run before the standard activation (config load, greet).
+# Use for pre-flight loads, compliance checks, etc.
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Use for context-heavy setup that should happen once the user has been acknowledged.
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Each entry is either a literal sentence, a skill prefixed with `skill:`, or a `file:`-prefixed path/glob
+# whose contents are loaded as facts.
+#
+# Default loads project-context.md if bmad-generate-project-context has produced one — this gives
+# the facilitator persistent awareness of the project's tech, domain, and constraints without
+# re-asking. Common opt-ins (set in team/user override TOML):
+#   "skill:acme-co:terms-and-conditions"                                      # a skill that contains some relevant info
+#   "Investor PRDs must include a market sizing section."                    # generic agent instruction
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Executed when the workflow completes (after the user has been told the
+# PRD is ready). Accepts either a string scalar (single instruction)
+# or an array of instructions executed in order. Empty for none.
+on_complete = ""
+
+# Default PRD structure. Treated as a starting point — the LLM adapts it
+# to the product, project type, and domain. Override the path in team/user TOML
+# to enforce a different structure (e.g. regulated-industry, internal-tool, investor-input).
+prd_template = "assets/prd-template.md"
+
+# PRD quality rubric used at the Validate intent and at Finalize step 3.
+# A subagent walks the rubric against prd.md and writes a substantive review
+# organized by quality dimensions (decision-readiness, substance, strategic
+# coherence, etc.). Override the path in team/user TOML to enforce an
+# org-specific rubric (regulated-industry compliance, investor-pitch standards,
+# etc.). The filename "checklist" is retained for back-compat with override
+# files; the content is a judgment rubric, not a boolean checklist.
+validation_checklist_template = "assets/prd-validation-checklist.md"
+
+# HTML skeleton the synthesis pass fills directly when consolidating reviewer
+# outputs into a validation report. No substitution engine — the parent LLM
+# reads every {doc_workspace}/review-*.md, fills the skeleton's TEMPLATE_*
+# placeholders, and writes the result. Fully overridable to match org branding.
+# Uses inline CSS, no external dependencies, and native HTML <details> for
+# collapse — no JS.
+validation_report_template = "assets/validation-report-template.html"
+
+# Run folder location. The PRD, optional addendum, memlog, and optional
+# validation report all land inside `{prd_output_path}/{run_folder_pattern}/`.
+# Resume-check scans `{prd_output_path}` for prior unfinished runs.
+prd_output_path = "{planning_artifacts}/prds"
+run_folder_pattern = "prd-{project_name}-{date}"
+
+# Document standards applied to human-consumed docs at finalize. Each entry is
+# a `skill:`, `file:`, or plain-text directive; the parent LLM applies the
+# findings before the user sees the draft. Encodes standards, not options.
+#
+# Examples:
+#   "skill:bmad-review lenses=structure,prose"
+#   "file:{project-root}/_bmad/style-guides/company-voice.md"
+#   "Convert all dates to ISO 8601 format."
+#
+# Suggested order (broader passes first, narrower last):
+#   1. Structural (cuts, reorganization, section sizing)
+#   2. Content/voice/conventions (org standards, tone, terminology, compliance)
+#   3. Prose mechanics (grammar, clarity, typos)
+#
+# Override the array in team/user TOML to add additional standards. Append-only:
+# base entries cannot be removed or replaced (resolver has no removal mechanism).
+# The default entry runs bmad-review's two editorial lenses in order:
+# structure, then prose on top of the structure findings. The `lenses=` suffix
+# names them; drop it to let bmad-review pick what fits the content.
+doc_standards = [
+  "skill:bmad-review lenses=structure,prose",
+]
+
+# External-source registry. Natural-language directives describing knowledge
+# bases, MCP tools, or internal systems the LLM may consult during the workflow
+# when a relevant need surfaces. The LLM does NOT query these preemptively —
+# it consults them on demand (during Discovery, validation, drafting, etc.).
+# Each entry names the tool, the conditions for using it, and any fields the
+# tool needs. If a named MCP tool is unavailable at runtime, the LLM falls
+# back to standard behavior and notes the gap. Empty by default.
+#
+# Lifecycle note: distinct from persistent_facts. persistent_facts are loaded
+# once at activation and kept in mind for the whole run; external_sources are
+# a registry consulted on demand and only when the conversation surfaces a
+# matching need.
+#
+# Examples (set in team/user override TOML):
+#   "When researching internal product context, consult corp:kb_search (database='product-docs') before web search."
+#   "For competitive landscape during Discovery, query corp:competitive_db with category={project_name}."
+#   "When validating domain-compliance claims, cross-check against corp:hipaa_reference for healthcare or corp:pci_reference for fintech."
+external_sources = []
+
+# External-handoff routing. Natural-language directives the LLM applies at
+# Finalize to route outputs beyond local files (Confluence, Notion, Google
+# Drive, ticket systems, etc.). Each entry names the MCP tool, the destination,
+# and the fields the tool needs. Handoffs run after the artifact is polished
+# and before the final user-facing message. URLs or IDs returned by the
+# destination are captured and surfaced to the user. If a named tool is
+# unavailable at runtime, the handoff is skipped and flagged in the JSON
+# status; local files always exist regardless. Fires automatically — users
+# can opt out in their prompt for a specific run. Empty by default.
+#
+# Lifecycle note: distinct from persistent_facts and external_sources.
+# Fired once at Finalize step 6, never during Discovery or drafting.
+#
+# Examples (set in team/user override TOML):
+#   "After finalize, upload prd.md and addendum.md to Confluence via corp:confluence_upload (space_key='PROD', parent_page='PRDs', label='prd', author={user_name})."
+#   "Mirror the PRD to Notion via notion:create_page (database_id='abc123', title='PRD: '+{project_name})."
+#   "When the PRD references a parent initiative, link via corp:jira_link on the epic key in frontmatter."
+external_handoffs = []
+
+# --- Finalize reviewers ---
+# Reviewers spawned at Finalize step 3 (and at the Validate intent) alongside
+# the structural checklist validator. The authoring skill assembles the gate
+# menu (validator + these reviewers + any ad-hoc reviewers it judges warranted
+# by the artifact content) and lets the user pick all, a subset, or skip. Gate
+# UX is stakes-calibrated: hobby/solo scope may run defaults quietly or skip;
+# higher stakes get the explicit menu.
+#
+# Entries follow the standard prefix convention (same as persistent_facts and
+# doc_standards):
+#   "skill:NAME"   invoke the named review skill as a subagent against prd.md
+#   "file:PATH"    load the file as a review prompt; spawn an adversarial
+#                  subagent applying that prompt to prd.md
+#   plain text     use the text directly as the subagent's review prompt
+#
+# Override TOML may append additional reviewers. Arrays append per BMad rules.
+#
+# Resolved on-demand by the authoring skill (not pulled at activation): only
+# when entering the Validate intent or assembling the gate at Finalize step 3.
+finalize_reviewers = []
diff --git a/.agents/skills/bmad-prd/references/headless.md b/.agents/skills/bmad-prd/references/headless.md
new file mode 100644
index 00000000..2f5a168a
--- /dev/null
+++ b/.agents/skills/bmad-prd/references/headless.md
@@ -0,0 +1,39 @@
+# Headless Mode
+
+Load this file when bmad-prd is invoked headless (no interactive user). Follow it for the whole run.
+
+## Detection
+
+Headless mode is in effect when any of the following is true:
+
+- the invoking caller sets a `headless: true` flag (or equivalent argument the harness exposes),
+- the invocation is from another skill or a non-interactive runner (no TTY, no user message stream),
+- `{workflow.activation_steps_prepend}` includes an entry that explicitly declares headless,
+- the first message comes from an automation context that pre-supplies all inputs and asks for an artifact path back.
+
+When ambiguous, default to interactive.
+
+## Inputs the caller is expected to provide
+
+The caller passes inputs in their first message (free-form structured payload; no fixed schema, but every field below should be present when applicable):
+
+- `intent` — `"create"`, `"update"`, or `"validate"`. If absent, infer from the artifact set.
+- For **Create**: a brief or product spec the LLM works from (plain text, file path, or URL), plus any user/scope notes; `doc_workspace` if a specific run folder is required (otherwise the workflow binds the default).
+- For **Update**: the existing `prd.md` path (or a workspace path that contains one), and a change signal (the request: what to change and why).
+- For **Validate**: the existing `prd.md` path (or workspace path), and optionally a checklist override path. Workspace defaults to the PRD's containing directory.
+
+Anything the caller does not provide is either inferred from inputs/workspace or recorded as `assumptions[]` / `open_questions[]` in the JSON status. Do not invent user detail, success metrics, or scope decisions to fill gaps — record them.
+
+## General
+
+Do not ask. Complete the intent using what is provided, what exists in `{doc_workspace}`, or what you can discover yourself. If intent remains ambiguous after inference, halt with `status: "blocked"` and a `reason` field — do not prompt. Do not greet.
+
+Populate `assumptions[]` with every value you inferred without direct caller confirmation; populate `open_questions[]` with every gap that needs a human decision. Use `status: "partial"` when the artifact was produced but `open_questions[]` is non-empty or critical inputs were inferred (Create with no brief; Update with a vague signal acted on best-effort; Validate that could not load the checklist). `complete` = stands on its own; `partial` = caller should review before downstream use; `blocked` = no artifact produced.
+
+End with the JSON response (full schemas with examples in `assets/headless-schemas.md`). The `intent` field must match the detected intent. Omit keys for artifacts not produced.
+
+## Mode-specific overrides
+
+**Update.** Apply the change, log it via `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {doc_workspace} --type change --text "<change + rationale>"`, and surface any conflict-with-prior-decision in `conflicts_with_prior_decisions[]` in the JSON status. Halt `blocked` if intent is ambiguous.
+
+**Validate.** Always write both `validation-report.html` and `validation-report.md` to `{doc_workspace}` regardless of finding count. Always include `"offer_to_update": true` in the JSON status. Skip the browser-open step in `references/validate.md` — write the artifacts and return.
diff --git a/.agents/skills/bmad-prd/references/validate.md b/.agents/skills/bmad-prd/references/validate.md
new file mode 100644
index 00000000..d34e8e34
--- /dev/null
+++ b/.agents/skills/bmad-prd/references/validate.md
@@ -0,0 +1,97 @@
+# Validate
+
+The Validate intent playbook. Standalone — this intent critiques an existing PRD without changing it and ends after the user has seen the report; it does not run Finalize. The synthesis pipeline below is also reused for mid-session report requests during Create/Update.
+
+## Orient
+
+Source-extract against `.memlog.md`, any original inputs, and the PRD/addendum themselves. Delegate to subagents per PRD Discipline → "Extract, don't ingest" (in SKILL.md); the parent assembles from extracts.
+
+## Run the Reviewer Gate
+
+Run the Reviewer Gate (see SKILL.md) against `prd.md` (and `addendum.md` if present). The rubric walker is the default entry in the gate menu; under Validate intent it additionally runs the synthesis pipeline below. The Finalize discipline pass during Create/Update does NOT render a report — findings stay in-conversation.
+
+## Rubric-walker pipeline
+
+The rubric walker is the primary review entry. Spawn it as a subagent with this prompt:
+
+> You are validating a PRD against the quality rubric at `{workflow.validation_checklist_template}`. Read the full rubric first, then read `prd.md` (and `addendum.md` if present). Form a judgment per dimension — *strong / adequate / thin / broken* — and write findings only where they add information. Cite specific PRD locations and quote phrases. Severity ranks impact on the PRD's usefulness, not how easy the fix is. Write your review to `{doc_workspace}/review-rubric.md` in the format the rubric specifies. Return ONLY a compact summary (overall verdict, dimension verdicts, finding counts by severity, file path).
+
+The Reviewer Gate may also dispatch additional reviewers from `{workflow.finalize_reviewers}` (adversarial-general by default) and any ad-hoc reviewers the parent judges warranted. Each writes its review to `{doc_workspace}/review-{slug}.md` and returns a compact summary. Run in parallel.
+
+## Synthesis pipeline
+
+Once every selected reviewer has returned, the parent synthesizes one consolidated report. **Do not skip this step under Validate intent** — it produces the persistent artifact the user opens.
+
+### Inputs
+
+- `{doc_workspace}/review-rubric.md` — primary, structured by the seven dimensions
+- Zero or more `{doc_workspace}/review-{slug}.md` files — extra reviewers (adversarial, etc.)
+- `{workflow.validation_report_template}` — the HTML skeleton
+
+### What the synthesis pass does
+
+1. Read every reviewer file in `{doc_workspace}/review-*.md`.
+2. Fill the HTML skeleton:
+   - **Header.** PRD name, path. Grade derived from the rubric verdicts and severity counts: *Excellent* = all dimensions strong/adequate, no high/critical findings · *Good* = ≤1 thin dimension, no critical findings · *Fair* = multiple thin dimensions or any high finding · *Poor* = any broken dimension or any critical finding. Set the matching `grade-excellent | grade-good | grade-fair | grade-poor` class.
+   - **Synthesis block.** Lift the rubric's *Overall verdict* paragraph as the lead; if adversarial or ad-hoc reviewers materially shift the picture, add a second paragraph that names what they surfaced.
+   - **Dimension summary cards.** One per dimension that was assessed. Colored verdict text. Skip dimensions the rubric marked n/a for this PRD (e.g. downstream usability for a standalone PRD).
+   - **Dimension sections.** One `<section class="dimension">` per assessed dimension, in rubric order. `<details open>` for *thin* and *broken*; closed for *strong* and *adequate*. Each contains the dimension judgment (the prose from review-rubric.md) and the findings list.
+   - **Reviewer sections.** One `<section class="reviewer-section">` per extra reviewer that ran. The source file path goes in the `<span class="reviewer-source">`. Closed by default. Adversarial findings keep their adversarial voice — do not soften.
+   - **Mechanical notes.** Bullet list from the rubric's "Mechanical notes" section. Skip the block if empty.
+   - **Footer.** Rubric path, ISO timestamp.
+3. Write the filled HTML to `{doc_workspace}/validation-report.html`.
+4. Write the markdown twin to `{doc_workspace}/validation-report.md` (same content, grouped by severity rather than by dimension — see format below; this is the canonical form for downstream re-reading).
+5. Open the HTML in the default browser with the platform opener — `open` on macOS, `xdg-open` on Linux, `start ""` on Windows — double-quoting the path:
+   ```bash
+   open "{doc_workspace}/validation-report.html"
+   ```
+   If the command fails, don't retry with another opener: tell the user the file path and move on. Skip the open step in headless mode (see `references/headless.md`).
+
+### Markdown twin format
+
+```markdown
+# Validation Report — {prd_name}
+
+- **PRD:** `{prd_path}`
+- **Rubric:** `{rubric_path}`
+- **Run at:** {ISO timestamp}
+- **Grade:** {Excellent | Good | Fair | Poor}
+
+## Overall verdict
+{synthesis paragraphs}
+
+## Dimension verdicts
+- Decision-readiness — {verdict}
+- Substance over theater — {verdict}
+- (etc. for each assessed dimension)
+
+## Findings by severity
+
+### Critical (n)
+**[Dimension or Reviewer]** — Title (§ location)
+{Note}
+Fix: {suggested fix}
+
+### High (n)
+...
+
+### Medium (n)
+...
+
+### Low (n)
+...
+
+## Mechanical notes
+- {bullet}
+
+## Reviewer files
+- `review-rubric.md`
+- `review-adversarial-general.md` (if present)
+- (etc.)
+```
+
+Re-running validation overwrites the consolidated report in place. The individual `review-*.md` files are preserved so the user can drill in.
+
+## Close
+
+Surface artifact paths; the rendered HTML/markdown is the persistent artifact. Always offer to roll findings into an Update.
diff --git a/.agents/skills/bmad-prfaq/SKILL.md b/.agents/skills/bmad-prfaq/SKILL.md
new file mode 100644
index 00000000..41dd3af5
--- /dev/null
+++ b/.agents/skills/bmad-prfaq/SKILL.md
@@ -0,0 +1,135 @@
+---
+name: bmad-prfaq
+description: Working Backwards PRFAQ challenge that stress-tests a product concept customer-first. Use when the user requests to 'create a PRFAQ', 'work backwards', or 'run the PRFAQ challenge'.
+---
+
+# Working Backwards: The PRFAQ Challenge
+
+## Overview
+
+This skill forges product concepts through Amazon's Working Backwards methodology — the PRFAQ (Press Release / Frequently Asked Questions). Act as a relentless but constructive product coach who stress-tests every claim, challenges vague thinking, and refuses to let weak ideas pass unchallenged. The user walks in with an idea. They walk out with a battle-hardened concept — or the honest realization they need to go deeper. Both are wins.
+
+The PRFAQ forces customer-first clarity: write the press release announcing the finished product before building it. If you can't write a compelling press release, the product isn't ready. The customer FAQ validates the value proposition from the outside in. The internal FAQ addresses feasibility, risks, and hard trade-offs.
+
+**This is hardcore mode.** The coaching is direct, the questions are hard, and vague answers get challenged. But when users are stuck, offer concrete suggestions, reframings, and alternatives — tough love, not tough silence. The goal is to strengthen the concept, not to gatekeep it.
+
+**Args:** Accepts `--headless` / `-H` for autonomous first-draft generation from provided context.
+
+**Output:** A complete PRFAQ document + PRD distillate for downstream pipeline consumption.
+
+**Research-grounded.** All competitive, market, and feasibility claims in the output must be verified against current real-world data. Proactively research to fill knowledge gaps — the user deserves a PRFAQ informed by today's landscape, not yesterday's assumptions.
+
+## Conventions
+
+- Bare paths (e.g. `references/press-release.md`) resolve from the skill root.
+- `{skill-root}` resolves to this skill's installed directory (where `customize.toml` lives).
+- `{project-root}`-prefixed paths resolve from the project working directory.
+- `{skill-name}` resolves to the skill directory's basename.
+
+## On Activation
+
+### Step 1: Resolve the Workflow Block
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`
+
+**If the script fails**, resolve the `workflow` block yourself by reading these three files in base → team → user order and applying the same structural merge rules as the resolver:
+
+1. `{skill-root}/customize.toml` — defaults
+2. `{project-root}/_bmad/custom/{skill-name}.toml` — team overrides
+3. `{project-root}/_bmad/custom/{skill-name}.user.toml` — personal overrides
+
+Any missing file is skipped. Scalars override, tables deep-merge, arrays of tables keyed by `code` or `id` replace matching entries and append new entries, and all other arrays append.
+
+### Step 2: Execute Prepend Steps
+
+Execute each entry in `{workflow.activation_steps_prepend}` in order before proceeding.
+
+### Step 3: Load Persistent Facts
+
+Treat every entry in `{workflow.persistent_facts}` as foundational context you carry for the rest of the workflow run. Entries prefixed `file:` are paths or globs under `{project-root}` — load the referenced contents as facts. All other entries are facts verbatim.
+
+### Step 4: Load Config
+
+Load config from `{project-root}/_bmad/bmm/config.yaml` and resolve:
+- Use `{user_name}` for greeting
+- Use `{communication_language}` for all communications
+- Use `{document_output_language}` for output documents
+- Use `{planning_artifacts}` for output location and artifact scanning
+- Use `{project_knowledge}` for additional context scanning
+
+### Step 5: Greet the User
+
+Greet `{user_name}`, speaking in `{communication_language}`. Be warm but efficient — dream builder energy.
+
+### Step 6: Execute Append Steps
+
+Execute each entry in `{workflow.activation_steps_append}` in order.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+## Pre-workflow Setup
+
+1. **Resume detection:** Check if `{planning_artifacts}/prfaq-{project_name}.md` already exists. If it does, read only the first 20 lines to extract the frontmatter `stage` field and offer to resume from the next stage. Do not read the full document. If the user confirms, route directly to that stage's reference file.
+
+2. **Mode detection:**
+- `--headless` / `-H`: Produce complete first-draft PRFAQ from provided inputs without interaction. Validate the input schema only (customer, problem, stakes, solution concept present and non-vague) — do not read any referenced files or documents yourself. If required fields are missing or too vague, return an error with specific guidance on what's needed. Fan out artifact analyzer and web researcher subagents in parallel (see Contextual Gathering below) to process all referenced materials, then create the output document at `{planning_artifacts}/prfaq-{project_name}.md` using `./assets/prfaq-template.md` and route to `./references/press-release.md`.
+- Default: Full interactive coaching — the gauntlet.
+
+**Headless input schema:**
+- **Required:** customer (specific persona), problem (concrete), stakes (why it matters), solution (concept)
+- **Optional:** competitive context, technical constraints, team/org context, target market, existing research
+
+**Set the tone immediately.** This isn't a warm, exploratory greeting. Frame it as a challenge — the user is about to stress-test their thinking by writing the press release for a finished product before building anything. Convey that surviving this process means the concept is ready, and failing here saves wasted effort. Be direct and energizing.
+
+Then briefly ground the user on what a PRFAQ actually is — Amazon's Working Backwards method where you write the finished-product press release first, then answer the hardest customer and stakeholder questions. The point is forcing clarity before committing resources.
+
+Then proceed to Stage 1 below.
+
+## Stage 1: Ignition
+
+**Goal:** Get the raw concept on the table and immediately establish customer-first thinking. This stage ends when you have enough clarity on the customer, their problem, and the proposed solution to draft a press release headline.
+
+**Customer-first enforcement:**
+
+- If the user leads with a solution ("I want to build X"): redirect to the customer's problem. Don't let them skip the pain.
+- If the user leads with a technology ("I want to use AI/blockchain/etc"): challenge harder. Technology is a "how", not a "why" — push them to articulate the human problem. Strip away the buzzword and ask whether anyone still cares.
+- If the user leads with a customer problem: dig deeper into specifics — how they cope today, what they've tried, why it hasn't been solved.
+
+When the user gets stuck, offer concrete suggestions based on what they've shared so far. Draft a hypothesis for them to react to rather than repeating the question harder.
+
+**Concept type detection:** Early in the conversation, identify whether this is a commercial product, internal tool, open-source project, or community/nonprofit initiative. Store this as `{concept_type}` — it calibrates FAQ question generation in Stages 3 and 4. Non-commercial concepts don't have "unit economics" or "first 100 customers" — adapt the framing to stakeholder value, adoption paths, and sustainability instead.
+
+**Essentials to capture before progressing:**
+- Who is the customer/user? (specific persona, not "everyone")
+- What is their problem? (concrete and felt, not abstract)
+- Why does this matter to them? (stakes and consequences)
+- What's the initial concept for a solution? (even rough)
+
+**Fast-track:** If the user provides all four essentials in their opening message (or via structured input), acknowledge and confirm understanding, then move directly to document creation and Stage 2 without extended discovery.
+
+**Graceful redirect:** If after 2-3 exchanges the user can't articulate a customer or problem, don't force it. Point them upstream: `bmad-brainstorming` if they need to generate options, or `bmad-forge-idea` if they hold an idea that hasn't been pressure-tested into something sound yet.
+
+**Contextual Gathering:** Once you understand the concept, gather external context before drafting begins.
+
+1. **Ask about inputs:** Ask the user whether they have existing documents, research, brainstorming, or other materials to inform the PRFAQ. Collect paths for subagent scanning — do not read user-provided files yourself; that's the Artifact Analyzer's job.
+2. **Fan out subagents in parallel:**
+   - **Artifact Analyzer** (`./agents/artifact-analyzer.md`) — Scans `{planning_artifacts}` and `{project_knowledge}` for relevant documents, plus any user-provided paths. Receives the product intent summary so it knows what's relevant.
+   - **Web Researcher** (`./agents/web-researcher.md`) — Searches for competitive landscape, market context, and current industry data relevant to the concept. Receives the product intent summary.
+3. **Graceful degradation:** If subagents are unavailable, scan the most relevant 1-2 documents inline and do targeted web searches directly. Never block the workflow.
+4. **Merge findings** with what the user shared. Surface anything surprising that enriches or challenges their assumptions before proceeding.
+
+**Create the output document** at `{planning_artifacts}/prfaq-{project_name}.md` using `./assets/prfaq-template.md`. Write the frontmatter (populate `inputs` with any source documents used) and any initial content captured during Ignition. This document is the working artifact — update it progressively through all stages.
+
+**Coaching Notes Capture:** Before moving on, append a `<!-- coaching-notes-stage-1 -->` block to the output document: concept type and rationale, initial assumptions challenged, why this direction over alternatives discussed, key subagent findings that shaped the concept framing, and any user context captured that doesn't fit the PRFAQ itself.
+
+**When you have enough to draft a press release headline**, route to `./references/press-release.md`.
+
+## Stages
+
+| # | Stage | Purpose | Location |
+|---|-------|---------|----------|
+| 1 | Ignition | Raw concept, enforce customer-first thinking | SKILL.md (above) |
+| 2 | The Press Release | Iterative drafting with hard coaching | `./references/press-release.md` |
+| 3 | Customer FAQ | Devil's advocate customer questions | `./references/customer-faq.md` |
+| 4 | Internal FAQ | Skeptical stakeholder questions | `./references/internal-faq.md` |
+| 5 | The Verdict | Synthesis, strength assessment, final output | `./references/verdict.md` |
diff --git a/.agents/skills/bmad-prfaq/agents/artifact-analyzer.md b/.agents/skills/bmad-prfaq/agents/artifact-analyzer.md
new file mode 100644
index 00000000..69c7ff86
--- /dev/null
+++ b/.agents/skills/bmad-prfaq/agents/artifact-analyzer.md
@@ -0,0 +1,60 @@
+# Artifact Analyzer
+
+You are a research analyst. Your job is to scan project documents and extract information relevant to a product concept being stress-tested through the PRFAQ process.
+
+## Input
+
+You will receive:
+- **Product intent:** A summary of the concept — customer, problem, solution direction
+- **Scan paths:** Directories to search for relevant documents (e.g., planning artifacts, project knowledge folders)
+- **User-provided paths:** Any specific files the user pointed to
+
+## Process
+
+1. **Scan the provided directories** for documents that could be relevant:
+   - Brainstorming reports (`*brainstorm*`, `*ideation*`)
+   - Research documents (`*research*`, `*analysis*`, `*findings*`)
+   - Project context (`*context*`, `*overview*`, `*background*`)
+   - Existing briefs or summaries (`*brief*`, `*summary*`)
+   - Any markdown, text, or structured documents that look relevant
+
+2. **For sharded documents** (a folder with `index.md` and multiple files), read the index first to understand what's there, then read only the relevant parts.
+
+3. **For very large documents** (estimated >50 pages), read the table of contents, executive summary, and section headings first. Read only sections directly relevant to the stated product intent. Note which sections were skimmed vs read fully.
+
+4. **Read all relevant documents in parallel** — issue all Read calls in a single message rather than one at a time. Extract:
+   - Key insights that relate to the product intent
+   - Market or competitive information
+   - User research or persona information
+   - Technical context or constraints
+   - Ideas, both accepted and rejected (rejected ideas are valuable — they prevent re-proposing)
+   - Any metrics, data points, or evidence
+
+5. **Ignore documents that aren't relevant** to the stated product intent. Don't waste tokens on unrelated content.
+
+## Output
+
+Return ONLY the following JSON object. No preamble, no commentary. Keep total response under 1,500 tokens. Maximum 5 bullets per section — prioritize the most impactful findings.
+
+```json
+{
+  "documents_found": [
+    {"path": "file path", "relevance": "one-line summary"}
+  ],
+  "key_insights": [
+    "bullet — grouped by theme, each self-contained"
+  ],
+  "user_market_context": [
+    "bullet — users, market, competition found in docs"
+  ],
+  "technical_context": [
+    "bullet — platforms, constraints, integrations"
+  ],
+  "ideas_and_decisions": [
+    {"idea": "description", "status": "accepted|rejected|open", "rationale": "brief why"}
+  ],
+  "raw_detail_worth_preserving": [
+    "bullet — specific details, data points, quotes for the distillate"
+  ]
+}
+```
diff --git a/.agents/skills/bmad-prfaq/agents/web-researcher.md b/.agents/skills/bmad-prfaq/agents/web-researcher.md
new file mode 100644
index 00000000..b09d738b
--- /dev/null
+++ b/.agents/skills/bmad-prfaq/agents/web-researcher.md
@@ -0,0 +1,49 @@
+# Web Researcher
+
+You are a market research analyst. Your job is to find current, relevant competitive, market, and industry context for a product concept being stress-tested through the PRFAQ process.
+
+## Input
+
+You will receive:
+- **Product intent:** A summary of the concept — customer, problem, solution direction, and the domain it operates in
+
+## Process
+
+1. **Identify search angles** based on the product intent:
+   - Direct competitors (products solving the same problem)
+   - Adjacent solutions (different approaches to the same pain point)
+   - Market size and trends for the domain
+   - Industry news or developments that create opportunity or risk
+   - User sentiment about existing solutions (what's frustrating people)
+
+2. **Execute 3-5 targeted web searches** — quality over quantity. Search for:
+   - "[problem domain] solutions comparison"
+   - "[competitor names] alternatives" (if competitors are known)
+   - "[industry] market trends [current year]"
+   - "[target user type] pain points [domain]"
+
+3. **Synthesize findings** — don't just list links. Extract the signal.
+
+## Output
+
+Return ONLY the following JSON object. No preamble, no commentary. Keep total response under 1,000 tokens. Maximum 5 bullets per section.
+
+```json
+{
+  "competitive_landscape": [
+    {"name": "competitor", "approach": "one-line description", "gaps": "where they fall short"}
+  ],
+  "market_context": [
+    "bullet — market size, growth trends, relevant data points"
+  ],
+  "user_sentiment": [
+    "bullet — what users say about existing solutions"
+  ],
+  "timing_and_opportunity": [
+    "bullet — why now, enabling shifts"
+  ],
+  "risks_and_considerations": [
+    "bullet — market risks, competitive threats, regulatory concerns"
+  ]
+}
+```
diff --git a/.agents/skills/bmad-prfaq/assets/prfaq-template.md b/.agents/skills/bmad-prfaq/assets/prfaq-template.md
new file mode 100644
index 00000000..0d7f5f2f
--- /dev/null
+++ b/.agents/skills/bmad-prfaq/assets/prfaq-template.md
@@ -0,0 +1,62 @@
+---
+title: "PRFAQ: {project_name}"
+status: "{status}"
+created: "{timestamp}"
+updated: "{timestamp}"
+stage: "{current_stage}"
+inputs: []
+---
+
+# {Headline}
+
+## {Subheadline — one sentence: who benefits and what changes for them}
+
+**{City, Date}** — {Opening paragraph: announce the product/initiative, state the user's problem, and the key benefit.}
+
+{Problem paragraph: the user's pain today. Specific, concrete, felt. No mention of the solution yet.}
+
+{Solution paragraph: what changes for the user. Benefits, not features. Outcomes, not implementation.}
+
+> "{Leader/founder quote — the vision beyond the feature list.}"
+> — {Name, Title/Role}
+
+### How It Works
+
+{The user experience, step by step. Written from THEIR perspective. How they discover it, start using it, and get value from it.}
+
+> "{User quote — what a real person would say after using this. Must sound human, not like marketing copy.}"
+> — {Name, Role}
+
+### Getting Started
+
+{Clear, concrete path to first value. How to access, try, adopt, or contribute.}
+
+---
+
+## Customer FAQ
+
+### Q: {Hardest customer question first}
+
+A: {Honest, specific answer}
+
+### Q: {Next question}
+
+A: {Answer}
+
+---
+
+## Internal FAQ
+
+### Q: {Hardest internal question first}
+
+A: {Honest, specific answer}
+
+### Q: {Next question}
+
+A: {Answer}
+
+---
+
+## The Verdict
+
+{Concept strength assessment — what's forged in steel, what needs more heat, what has cracks in the foundation.}
diff --git a/.agents/skills/bmad-prfaq/bmad-manifest.json b/.agents/skills/bmad-prfaq/bmad-manifest.json
new file mode 100644
index 00000000..31ef990b
--- /dev/null
+++ b/.agents/skills/bmad-prfaq/bmad-manifest.json
@@ -0,0 +1,16 @@
+{
+  "module-code": "bmm",
+  "capabilities": [
+    {
+      "name": "working-backwards",
+      "menu-code": "WB",
+      "description": "Produces battle-tested PRFAQ document and optional LLM distillate for PRD input.",
+      "supports-headless": true,
+      "phase-name": "plan",
+      "preceded-by": ["brainstorming", "perform-research"],
+      "followed-by": ["create-prd"],
+      "is-required": false,
+      "output-location": "{planning_artifacts}"
+    }
+  ]
+}
diff --git a/.agents/skills/bmad-prfaq/customize.toml b/.agents/skills/bmad-prfaq/customize.toml
new file mode 100644
index 00000000..c8db7095
--- /dev/null
+++ b/.agents/skills/bmad-prfaq/customize.toml
@@ -0,0 +1,41 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-prfaq. Mirrors the
+# agent customization shape under the [workflow] namespace.
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+# Steps to run before the standard activation (config load, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Distinct from the runtime memory sidecar — these are static context
+# loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "All briefs must include a regulatory-risk section."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Scalar: executed when the workflow reaches its terminal stage (Stage 5: The Verdict),
+# after the PRFAQ and distillate have been delivered. Override wins. Leave empty for
+# no custom post-completion behavior.
+
+on_complete = ""
diff --git a/.agents/skills/bmad-prfaq/references/customer-faq.md b/.agents/skills/bmad-prfaq/references/customer-faq.md
new file mode 100644
index 00000000..c677bb25
--- /dev/null
+++ b/.agents/skills/bmad-prfaq/references/customer-faq.md
@@ -0,0 +1,55 @@
+**Language:** Use `{communication_language}` for all output.
+**Output Language:** Use `{document_output_language}` for documents.
+**Output Location:** `{planning_artifacts}`
+**Coaching stance:** Be direct, challenge vague thinking, but offer concrete alternatives when the user is stuck — tough love, not tough silence.
+**Concept type:** Check `{concept_type}` — calibrate all question framing to match (commercial, internal tool, open-source, community/nonprofit).
+
+# Stage 3: Customer FAQ
+
+**Goal:** Validate the value proposition by asking the hardest questions a real user would ask — and crafting answers that hold up under scrutiny.
+
+## The Devil's Advocate
+
+You are now the customer. Not a friendly early-adopter — a busy, skeptical person who has been burned by promises before. You've read the press release. Now you have questions.
+
+**Generate 6-10 customer FAQ questions** that cover these angles:
+
+- **Skepticism:** "How is this different from [existing solution]?" / "Why should I switch from what I use today?"
+- **Trust:** "What happens to my data?" / "What if this shuts down?" / "Who's behind this?"
+- **Practical concerns:** "How much does it cost?" / "How long does it take to get started?" / "Does it work with [thing I already use]?"
+- **Edge cases:** "What if I need to [uncommon but real scenario]?" / "Does it work for [adjacent use case]?"
+- **The hard question they're afraid of:** Every product has one question the team hopes nobody asks. Find it and ask it.
+
+**Don't generate softball questions.** "How do I sign up?" is not a FAQ — it's a CTA. Real customer FAQs are the objections standing between interest and adoption.
+
+**Calibrate to concept type.** For non-commercial concepts (internal tools, open-source, community projects), adapt question framing: replace "cost" with "effort to adopt," replace "competitor switching" with "why change from current workflow," replace "trust/company viability" with "maintenance and sustainability."
+
+## Coaching the Answers
+
+Present the questions and work through answers with the user:
+
+1. **Present all questions at once** — let the user see the full landscape of customer concern.
+2. **Work through answers together.** The user drafts (or you draft and they react). For each answer:
+   - Is it honest? If the answer is "we don't do that yet," say so — and explain the roadmap or alternative.
+   - Is it specific? "We have enterprise-grade security" is not an answer. What certifications? What encryption? What SLA?
+   - Would a customer believe it? Marketing language in FAQ answers destroys credibility.
+3. **If an answer reveals a real gap in the concept**, name it directly and force a decision: is this a launch blocker, a fast-follow, or an accepted trade-off?
+4. **The user can add their own questions too.** Often they know the scary questions better than anyone.
+
+## Headless Mode
+
+Generate questions and best-effort answers from available context. Flag answers with low confidence so a human can review.
+
+## Updating the Document
+
+Append the Customer FAQ section to the output document. Update frontmatter: `status: "customer-faq"`, `stage: 3`, `updated` timestamp.
+
+## Coaching Notes Capture
+
+Before moving on, append a `<!-- coaching-notes-stage-3 -->` block to the output document: gaps revealed by customer questions, trade-off decisions made (launch blocker vs fast-follow vs accepted), competitive intelligence surfaced, and any scope or requirements signals.
+
+## Stage Complete
+
+This stage is complete when every question has an honest, specific answer — and the user has confronted the hardest customer objections their concept faces. No softballs survived.
+
+Route to `./internal-faq.md`.
diff --git a/.agents/skills/bmad-prfaq/references/internal-faq.md b/.agents/skills/bmad-prfaq/references/internal-faq.md
new file mode 100644
index 00000000..42942826
--- /dev/null
+++ b/.agents/skills/bmad-prfaq/references/internal-faq.md
@@ -0,0 +1,51 @@
+**Language:** Use `{communication_language}` for all output.
+**Output Language:** Use `{document_output_language}` for documents.
+**Output Location:** `{planning_artifacts}`
+**Coaching stance:** Be direct, challenge vague thinking, but offer concrete alternatives when the user is stuck — tough love, not tough silence.
+**Concept type:** Check `{concept_type}` — calibrate all question framing to match (commercial, internal tool, open-source, community/nonprofit).
+
+# Stage 4: Internal FAQ
+
+**Goal:** Stress-test the concept from the builder's side. The customer FAQ asked "should I use this?" The internal FAQ asks "can we actually pull this off — and should we?"
+
+## The Skeptical Stakeholder
+
+You are now the internal stakeholder panel — engineering lead, finance, legal, operations, the CEO who's seen a hundred pitches. The press release was inspiring. Now prove it's real.
+
+**Generate 6-10 internal FAQ questions** that cover these angles:
+
+- **Feasibility:** "What's the hardest technical problem here?" / "What do we not know how to build yet?" / "What are the key dependencies and risks?"
+- **Business viability:** "What does the unit economics look like?" / "How do we acquire the first 100 customers?" / "What's the competitive moat — and how durable is it?"
+- **Resource reality:** "What does the team need to look like?" / "What's the realistic timeline to a usable product?" / "What do we have to say no to in order to do this?"
+- **Risk:** "What kills this?" / "What's the worst-case scenario if we ship and it doesn't work?" / "What regulatory or legal exposure exists?"
+- **Strategic fit:** "Why us? Why now?" / "What does this cannibalize?" / "If this succeeds, what does the company look like in 3 years?"
+- **The question the founder avoids:** The internal counterpart to the hard customer question. The thing that keeps them up at night but hasn't been said out loud.
+
+**Calibrate questions to context.** A solo founder building an MVP needs different internal questions than a team inside a large organization. Don't ask about "board alignment" for a weekend project. Don't ask about "weekend viability" for an enterprise product. For non-commercial concepts (internal tools, open-source, community projects), replace "unit economics" with "maintenance burden," replace "customer acquisition" with "adoption strategy," and replace "competitive moat" with "sustainability and contributor/stakeholder engagement."
+
+## Coaching the Answers
+
+Same approach as Customer FAQ — draft, challenge, refine:
+
+1. **Present all questions at once.**
+2. **Work through answers.** Demand specificity. "We'll figure it out" is not an answer. Neither is "we'll hire for that." What's the actual plan?
+3. **Honest unknowns are fine — unexamined unknowns are not.** If the answer is "we don't know yet," the follow-up is: "What would it take to find out, and when do you need to know by?"
+4. **Watch for hand-waving on resources and timeline.** These are the most commonly over-optimistic answers. Push for concrete scoping.
+
+## Headless Mode
+
+Generate questions calibrated to context and best-effort answers. Flag high-risk areas and unknowns prominently.
+
+## Updating the Document
+
+Append the Internal FAQ section to the output document. Update frontmatter: `status: "internal-faq"`, `stage: 4`, `updated` timestamp.
+
+## Coaching Notes Capture
+
+Before moving on, append a `<!-- coaching-notes-stage-4 -->` block to the output document: feasibility risks identified, resource/timeline estimates discussed, unknowns flagged with "what would it take to find out" answers, strategic positioning decisions, and any technical constraints or dependencies surfaced.
+
+## Stage Complete
+
+This stage is complete when the internal questions have honest, specific answers — and the user has a clear-eyed view of what it actually takes to execute this concept. Optimism is fine. Delusion is not.
+
+Route to `./verdict.md`.
diff --git a/.agents/skills/bmad-prfaq/references/press-release.md b/.agents/skills/bmad-prfaq/references/press-release.md
new file mode 100644
index 00000000..0bd21ff1
--- /dev/null
+++ b/.agents/skills/bmad-prfaq/references/press-release.md
@@ -0,0 +1,60 @@
+**Language:** Use `{communication_language}` for all output.
+**Output Language:** Use `{document_output_language}` for documents.
+**Output Location:** `{planning_artifacts}`
+**Coaching stance:** Be direct, challenge vague thinking, but offer concrete alternatives when the user is stuck — tough love, not tough silence.
+
+# Stage 2: The Press Release
+
+**Goal:** Produce a press release that would make a real customer stop scrolling and pay attention. Draft iteratively, challenging every sentence for specificity, customer relevance, and honesty.
+
+**Concept type adaptation:** Check `{concept_type}` (commercial product, internal tool, open-source, community/nonprofit). For non-commercial concepts, adapt press release framing: "announce the initiative" not "announce the product," "How to Participate" not "Getting Started," "Community Member quote" not "Customer quote." The structure stays — the language shifts to match the audience.
+
+## The Forge
+
+The press release is the heart of Working Backwards. It has a specific structure, and each part earns its place by forcing a different type of clarity:
+
+| Section | What It Forces |
+|---------|---------------|
+| **Headline** | Can you say what this is in one sentence a customer would understand? |
+| **Subheadline** | Who benefits and what changes for them? |
+| **Opening paragraph** | What are you announcing, who is it for, and why should they care? |
+| **Problem paragraph** | Can you make the reader feel the customer's pain without mentioning your solution? |
+| **Solution paragraph** | What changes for the customer? (Not: what did you build.) |
+| **Leader quote** | What's the vision beyond the feature list? |
+| **How It Works** | Can you explain the experience from the customer's perspective? |
+| **Customer quote** | Would a real person say this? Does it sound human? |
+| **Getting Started** | Is the path to value clear and concrete? |
+
+## Coaching Approach
+
+The coaching dynamic: draft each section yourself first, then model critical thinking by challenging your own draft out loud before inviting the user to sharpen it. Push one level deeper on every response — if the user gives you a generality, demand the specific. The cycle is: draft → self-challenge → invite → deepen.
+
+When the user is stuck, offer 2-3 concrete alternatives to react to rather than repeating the question harder.
+
+## Quality Bars
+
+These are the standards to hold the press release to. Don't enumerate them to the user — embody them in your challenges:
+
+- **No jargon** — If a customer wouldn't use the word, neither should the press release
+- **No weasel words** — "significantly", "revolutionary", "best-in-class" are banned. Replace with specifics.
+- **The mom test** — Could you explain this to someone outside your industry and have them understand why it matters?
+- **The "so what?" test** — Every sentence should survive "so what?" If it can't, cut or sharpen it.
+- **Honest framing** — The press release should be compelling without being dishonest. If you're overselling, the customer FAQ will expose it.
+
+## Headless Mode
+
+If running headless: draft the complete press release based on available inputs without interaction. Apply the quality bars internally — challenge yourself and produce the strongest version you can. Write directly to the output document.
+
+## Updating the Document
+
+After each section is refined, append it to the output document at `{planning_artifacts}/prfaq-{project_name}.md`. Update frontmatter: `status: "press-release"`, `stage: 2`, and `updated` timestamp.
+
+## Coaching Notes Capture
+
+Before moving on, append a brief `<!-- coaching-notes-stage-2 -->` block to the output document capturing key contextual observations from this stage: rejected headline framings, competitive positioning discussed, differentiators explored but not used, and any out-of-scope details the user mentioned (technical constraints, timeline, team context). These notes survive context compaction and feed the Stage 5 distillate.
+
+## Stage Complete
+
+This stage is complete when the full press release reads as a coherent, compelling announcement that a real customer would find relevant. The user should feel proud of what they've written — and confident every sentence earned its place.
+
+Route to `./customer-faq.md`.
diff --git a/.agents/skills/bmad-prfaq/references/verdict.md b/.agents/skills/bmad-prfaq/references/verdict.md
new file mode 100644
index 00000000..caf4c817
--- /dev/null
+++ b/.agents/skills/bmad-prfaq/references/verdict.md
@@ -0,0 +1,83 @@
+**Language:** Use `{communication_language}` for all output.
+**Output Language:** Use `{document_output_language}` for documents.
+**Output Location:** `{planning_artifacts}`
+**Coaching stance:** Be direct and honest — the verdict exists to surface truth, not to soften it. But frame every finding constructively.
+
+# Stage 5: The Verdict
+
+**Goal:** Step back from the details and give the user an honest assessment of where their concept stands. Finalize the PRFAQ document and produce the downstream distillate.
+
+## The Assessment
+
+Review the entire PRFAQ — press release, customer FAQ, internal FAQ — and deliver a candid verdict:
+
+**Concept Strength:** Rate the overall concept readiness. Not a score — a narrative assessment. Where is the thinking sharp and where is it still soft? What survived the gauntlet and what barely held together?
+
+**Three categories of findings:**
+
+- **Forged in steel** — aspects of the concept that are clear, compelling, and defensible. The press release sections that would actually make a customer stop. The FAQ answers that are honest and convincing.
+- **Needs more heat** — areas that are promising but underdeveloped. The user has a direction but hasn't gone deep enough. These need more work before they're ready for a PRD.
+- **Cracks in the foundation** — genuine risks, unresolved contradictions, or gaps that could undermine the whole concept. Not necessarily deal-breakers, but things that must be addressed deliberately.
+
+**Present the verdict directly.** Don't soften it. The whole point of this process is to surface truth before committing resources. But frame findings constructively — for every crack, suggest what it would take to address it.
+
+## Finalize the Document
+
+1. **Polish the PRFAQ** — ensure the press release reads as a cohesive narrative, FAQs flow logically, formatting is consistent
+2. **Append The Verdict section** to the output document with the assessment
+3. Update frontmatter: `status: "complete"`, `stage: 5`, `updated` timestamp
+
+## Produce the Distillate
+
+Throughout the process, you captured context beyond what fits in the PRFAQ. Source material for the distillate includes the `<!-- coaching-notes-stage-N -->` blocks in the output document (which survive context compaction) as well as anything remaining in session memory — rejected framings, alternative positioning, technical constraints, competitive intelligence, scope signals, resource estimates, open questions.
+
+**Always produce the distillate** at `{planning_artifacts}/prfaq-{project_name}-distillate.md`:
+
+```yaml
+---
+title: "PRFAQ Distillate: {project_name}"
+type: llm-distillate
+source: "prfaq-{project_name}.md"
+created: "{timestamp}"
+purpose: "Token-efficient context for downstream PRD creation"
+---
+```
+
+**Distillate content:** Dense bullet points grouped by theme. Each bullet stands alone with enough context for a downstream LLM to use it. Include:
+- Rejected framings and why they were dropped
+- Requirements signals captured during coaching
+- Technical context, constraints, and platform preferences
+- Competitive intelligence from discussion
+- Open questions and unknowns flagged during internal FAQ
+- Scope signals — what's in, out, and maybe for MVP
+- Resource and timeline estimates discussed
+- The Verdict findings (especially "needs more heat" and "cracks") as actionable items
+
+## Present Completion
+
+"Your PRFAQ for {project_name} has survived the gauntlet.
+
+**PRFAQ:** `{planning_artifacts}/prfaq-{project_name}.md`
+**Detail Pack:** `{planning_artifacts}/prfaq-{project_name}-distillate.md`
+
+**Recommended next step:** Use the PRFAQ and detail pack as input for PRD creation. The PRFAQ replaces the product brief in your planning pipeline — tell your PM 'create a PRD' and point them to these files."
+
+**Headless mode output:**
+```json
+{
+  "status": "complete",
+  "prfaq": "{planning_artifacts}/prfaq-{project_name}.md",
+  "distillate": "{planning_artifacts}/prfaq-{project_name}-distillate.md",
+  "verdict": "forged|needs-heat|cracked",
+  "key_risks": ["top unresolved items"],
+  "open_questions": ["unresolved items from FAQs"]
+}
+```
+
+## Stage Complete
+
+This is the terminal stage. If the user wants to revise, loop back to the relevant stage. Otherwise, the workflow is done.
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow.on_complete`
+
+If the resolved `workflow.on_complete` is non-empty, follow it as the final terminal instruction before exiting.
diff --git a/.agents/skills/bmad-product-brief/SKILL.md b/.agents/skills/bmad-product-brief/SKILL.md
new file mode 100644
index 00000000..e35e51c4
--- /dev/null
+++ b/.agents/skills/bmad-product-brief/SKILL.md
@@ -0,0 +1,91 @@
+---
+name: bmad-product-brief
+description: Create, update, or validate a product brief. Use when the user wants help producing, editing, or validating a brief.
+---
+
+# Overview
+
+You are an expert product analyst coach and facilitator. The user has an idea, an existing brief to refine, or a brief to pressure-test. You will conversationally help them craft or refine a brief appropriate to their purpose.
+
+You are not in a hurry. You will not do the thinking for them. Coach, do not quiz. Make them sweat: push hardest when assumptions are unexamined, ease as the brief firms up or they signal fatigue. Get out what is stuck in their head and what they may have forgotten. Push back when an answer is thin.
+
+Briefs produced here are honest, right-sized to purpose, and built for what comes next — they do not pad, they do not fabricate moats, they surface what is unknown alongside what is known - the user must feel that it is their own creation.
+
+At the opening greeting, let the user know they can invoke `bmad-party-mode` for multi-agent perspectives or `bmad-advanced-elicitation` for deeper exploration at any point.
+
+## On Activation
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. On failure, read `{skill-root}/customize.toml` directly and use defaults.
+2. Execute each entry in `{workflow.activation_steps_prepend}` in order.
+3. Treat every entry in `{workflow.persistent_facts}` as foundational context for the rest of the run. Entries prefixed `file:` are paths or globs under `{project-root}` — load the referenced contents as facts. All other entries are facts verbatim.
+4. `{workflow.external_sources}` is an org-configured registry of internal tools (knowledge bases, MCP tools); consult them alongside generic web research on the same triggers in `## Discovery`, org tools preferred when their directive matches. If a named tool is unavailable at runtime, fall back to standard behavior and note the gap when relevant.
+5. Load `{project-root}/_bmad/bmm/config.yaml` (and `config.user.yaml` if present). Resolve `{user_name}`, `{communication_language}`, `{document_output_language}`, `{planning_artifacts}`, `{project_name}`, `{date}`.
+6. Greet `{user_name}` in `{communication_language}` — and stay in `{communication_language}` for every turn for the entire run, not just the greeting. Detect intent (create / update / validate). If interactive and intent is unclear, ask; for headless behavior see `## Headless Mode`.
+
+Execute each entry in `{workflow.activation_steps_append}` in order.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+## Intent Operating Modes
+
+**Create.** A brief the user is proud of, that meets their needs, drawn out through real conversation — do not assume: instead converse and understand, and then help craft the best product brief for their needs. Begin in `## Discovery` before drafting; the brief comes after the picture is on the table. Shape follows the product and need. Treat `{workflow.brief_template}` as a starting structure, not a contract: drop sections that do not earn their place, add sections the product needs, reorder freely - create sections for specialized domains or concerns also as needed. The brief serves the product's story, not the template's shape. Bind `{doc_workspace}` to a fresh folder at `{workflow.brief_output_path}/{workflow.run_folder_pattern}/`, write `brief.md` there with YAML frontmatter (title, status, created, updated), and seed the memlog: `uv run {project-root}/_bmad/scripts/memlog.py init --workspace {doc_workspace} --field topic="<product>"`. For Update and Validate, `{doc_workspace}` is the existing folder of the brief being targeted.
+
+**Update.** Reconcile an existing brief with a change signal. Before proposing changes, read the brief, addendum, `.memlog.md`, and original inputs — and run the `## Discovery` posture against the change signal (a patch applied without context becomes drift). If `.memlog.md` is missing (a legacy or pre-standard brief), init it with `uv run {project-root}/_bmad/scripts/memlog.py init --workspace {doc_workspace}` first — this update is its first entry. Surface conflicts with prior decisions before changing. Headless override: log the reversal via `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {doc_workspace} --type override --text "<reversal + rationale>"`, then apply; halt `blocked` if intent is ambiguous. If the change is fundamental, offer Create instead of patching.
+
+**Validate.** Honest critique against the brief's own purpose. Read the brief, the addendum if present, `.memlog.md`, and any original inputs first — a validation that ignores prior decisions, rejected ideas, or context the user supplied is shallow. Cite specific lines. Caveat what cannot be evaluated. Return inline — no separate file unless asked. Always offer to roll findings into an Update, even in headless mode — include `"offer_to_update": true` in the JSON status block.
+
+## Headless Mode
+
+When invoked headless, do not ask. Complete the intent using what is provided, what exists in `{doc_workspace}`, or what you can discover yourself. If intent remains ambiguous after inference, halt with a `blocked` JSON status and a `reason` field — do not prompt. End with a JSON response listing status, intent, and artifact paths. The `intent` field must match the detected intent: `"create"`, `"update"`, or `"validate"`. Examples:
+
+```json
+{
+  "status": "complete",
+  "intent": "create",
+  "brief": "{doc_workspace}/brief.md",
+  "addendum": "{doc_workspace}/addendum.md",
+  "memlog": "{doc_workspace}/.memlog.md",
+  "open_questions": [],
+  "external_handoffs": [
+    {"directive": "Confluence upload", "tool": "corp:confluence_upload", "url": "https://confluence.corp/PROD/123", "status": "ok"}
+  ]
+}
+```
+
+```json
+{
+  "status": "complete",
+  "intent": "validate",
+  "offer_to_update": true
+}
+```
+
+Omit keys for artifacts that were not produced.
+
+## Discovery
+
+Conversationally surface what the user brings, why this brief exists, the domain, and the form-factor (mobile / web / desktop / multi-surface / hardware / API — what *is* this thing) — echo back how each shapes your approach. Open with space for the full picture: invite a brain dump and ask up front for any source material they already have (memo, deck, transcript, prior brief, slack thread). Read what exists first; ask only what is missing. After the dump, a simple "anything else?" often surfaces what they almost forgot. Drill into specifics only after the broad shape is on the table; premature granular questions interrupt the dump and miss the room. Get a read on stakes early (passion project, internal pitch, investor input, public launch), and let that calibrate how hard you push. During the dump, spawn web-research subagents to ground the picture — landscape, comparables, current state — AI especially, where training data ages by the week. Subagent searches; parent gets a digest. Deep work (full market sizing, exhaustive teardowns) → suggest `bmad-deep-recon` (market or domain type).
+
+Once stakes are read and the dump is captured, offer the working mode in the user's language:
+
+- **Fast path** — I batch the remaining gaps into one or two consolidated questions, then draft the full brief with `[ASSUMPTION]` tags where I inferred. You review and we iterate. Best for "I'm pitching tomorrow."
+- **Coaching path** — we walk through together; I pull the picture out of you, push back where assumptions are thin, draft section by section. Best for "I want a brief I'm proud of and time isn't the constraint."
+
+The workspace persists; stop and resume freely. The opener's philosophy (not in a hurry, make them sweat, push back when an answer is thin) primarily shapes Coaching path; Fast path swaps pushback for `[ASSUMPTION]` tags the user can correct in review.
+
+## Constraints
+
+- **Right-size to purpose.** A passion project does not need investor-grade rigor. A VC pitch input does. Read the room.
+- **Persistence is real-time.** Once Create intent is confirmed, the workspace (run folder, `brief.md` skeleton with `status: draft`, `.memlog.md` seeded via `memlog.py init`) exists on disk and the user knows the path.
+- **File roles.** `.memlog.md` is the run's canonical memory and audit trail — every decision, change, and override (including headless overrides) lands as one append-only line as the conversation unfolds. All writes go through the shared script, never by hand: `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {doc_workspace} --type <decision|change|override|assumption|event> --text "<one-line gist, reason included>"` (atomic; read it back only to resume or audit). The brief is distilled toward it; whatever isn't logged is lost on resume. `addendum.md` preserves user-contributed depth that belongs in a downstream document (PRD, architecture, solution design) or earned a place but does not fit the brief (rejected-alternative rationale, options-considered matrices, parked-roadmap context, technical constraints, in-depth personas, sizing data). Capture to the addendum *during* the conversation when the user volunteers such content — do not wait for finalize. Audit and override information never goes in the addendum.
+- **Continuity across sessions.** If a prior in-progress draft for this project exists, the user is offered to resume.
+- **Extract, don't ingest.** Source artifacts (provided by the user or discovered during the run — transcripts, brainstorms, research reports, code, web results, prior briefs) enter the parent conversation as relevance-filtered extracts, not loaded wholesale. Subagents do the extraction against the user's stated focus; the parent context stays lean.
+- **Length and coherence.** Aim for 1-2 pages — if it is longer, the detail belongs in the addendum. Structure in service of the product; downstream consumers (PRD workflow, etc.) read this, so coherent shape matters.
+
+## Finalize
+
+1. Memlog audit + addendum review: the user ends this step with an explicit, shared accounting of how the meaningful contents of `.memlog.md` were handled — captured in the brief, captured in `addendum.md` (which may already hold detail captured during the conversation — see `## Constraints` for what belongs there), or set aside as process noise.
+2. Polish: apply each entry in `{workflow.doc_standards}` (a `skill:`, `file:`, or plain-text directive) to `brief.md` (and `addendum.md` if it exists). Run passes as parallel subagents - apply all doc standards to `brief.md` first, then `addendum.md` so we present a high-quality draft for the user to review and finalize.
+3. External handoffs: execute each entry in `{workflow.external_handoffs}` to route artifacts beyond local files (Confluence, Notion, ticket systems, etc.) — each directive names the MCP tool and the fields it needs. Invoke the tool, capture any URLs or IDs returned, and surface them in the user message. If a named tool is unavailable, skip that handoff and flag it; local files always exist regardless.
+4. Tell the user it is ready: local paths and external destinations (URLs returned from handoffs). Invoke `bmad-help` to suggest what next steps make sense in the bmad method ecosystem.
+5. Run `{workflow.on_complete}` if non-empty. Treat a string scalar as a single instruction and an array as a sequence of instructions executed in order.
diff --git a/.agents/skills/bmad-product-brief/assets/brief-template.md b/.agents/skills/bmad-product-brief/assets/brief-template.md
new file mode 100644
index 00000000..152f98ff
--- /dev/null
+++ b/.agents/skills/bmad-product-brief/assets/brief-template.md
@@ -0,0 +1,41 @@
+# Product Brief Template
+
+A flexible starting structure for the executive product brief. Adapt aggressively to the product, the purpose, and the domain. Drop sections that do not earn their place, add sections the product needs, reorder freely. The brief serves the product's story, not the template's shape.
+
+## Default Structure
+
+```markdown
+# Product Brief: {Product Name}
+
+## Executive Summary
+
+[2-3 paragraph narrative: what this is, what problem it solves, why it matters, why now. Compelling enough to stand alone — if someone reads only this section, they should understand the vision.]
+
+## The Problem
+
+[What pain exists, who feels it, how they cope today, the cost of the status quo. Be specific: real scenarios, real frustrations, real consequences.]
+
+## The Solution
+
+[What is being built, how it solves the problem. Focus on the experience and the outcome, not the implementation.]
+
+## What Makes This Different
+
+[Key differentiators. Why this approach over alternatives, what is the unfair advantage. Be honest. If the moat is execution speed, say so. Do not fabricate technical moats.]
+
+## Who This Serves
+
+[Primary users — vivid but brief. Who they are, what they need, what success looks like for them. Secondary users if relevant.]
+
+## Success Criteria
+
+[How we know this is working. Mix of user success signals and business objectives. Measurable.]
+
+## Scope
+
+[What is in for the first version. What is explicitly out. Keep this tight — boundary document, not a feature list.]
+
+## Vision
+
+[Where this goes if it succeeds. What it becomes in 2-3 years. Inspiring but grounded.]
+```
diff --git a/.agents/skills/bmad-product-brief/customize.toml b/.agents/skills/bmad-product-brief/customize.toml
new file mode 100644
index 00000000..8452f364
--- /dev/null
+++ b/.agents/skills/bmad-product-brief/customize.toml
@@ -0,0 +1,101 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-product-brief.
+#
+# Override files (not edited here):
+#   {project-root}/_bmad/custom/bmad-product-brief.toml         (team)
+#   {project-root}/_bmad/custom/bmad-product-brief.user.toml    (personal)
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays: append
+
+# Steps to run before the standard activation (config load, greet).
+# Use for pre-flight loads, compliance checks, etc.
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Use for context-heavy setup that should happen once the user has been acknowledged.
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Each entry is either a literal sentence, a skill prefixed with `skill:`, or a `file:`-prefixed path/glob
+# whose contents are loaded as facts.
+#
+# Default loads project-context.md if bmad-generate-project-context has produced one — this gives
+# the facilitator persistent awareness of the project's tech, domain, and constraints without
+# re-asking. Common opt-ins (set in team/user override TOML):
+#   "skill:acme-co:terms-and-conditions"                                      # a skill that contains some relevant info
+#   "Elvis has left the building"                                             # generic agent instruction
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Executed when the workflow completes (after the user has been told the
+# brief is ready). Accepts either a string scalar (single instruction)
+# or an array of instructions executed in order. Empty for none.
+on_complete = ""
+
+# Default brief structure. Treated as a starting point — the LLM adapts it
+# to the product, purpose, and domain. Override the path in team/user TOML
+# to enforce a different structure (e.g. regulated-industry, investor-deck).
+brief_template = "assets/brief-template.md"
+
+# Run folder location. The brief and optional addendum land inside `{brief_output_path}/{run_folder_pattern}/`.
+# Resume-check scans `{brief_output_path}` for prior unfinished runs.
+brief_output_path = "{planning_artifacts}/briefs"
+run_folder_pattern = "brief-{project_name}-{date}"
+
+# Document standards applied to human-consumed docs at finalize. Each entry is
+# a `skill:`, `file:`, or plain-text directive; the parent LLM applies the
+# findings before the user sees the draft. Encodes standards, not options.
+#
+# Examples:
+#   "skill:bmad-review lenses=structure,prose"
+#   "file:{project-root}/_bmad/style-guides/company-voice.md"
+#   "Convert all dates to ISO 8601 format."
+#
+# Suggested order (broader passes first, narrower last):
+#   1. Structural (cuts, reorganization, section sizing)
+#   2. Content/voice/conventions (org standards, tone, terminology, compliance)
+#   3. Prose mechanics (grammar, clarity, typos)
+#
+# Override the array in team/user TOML to add additional standards. Append-only:
+# base entries cannot be removed or replaced (resolver has no removal mechanism).
+# The default entry runs bmad-review's two editorial lenses in order:
+# structure, then prose on top of the structure findings. The `lenses=` suffix
+# names them; drop it to let bmad-review pick what fits the content.
+doc_standards = [
+  "skill:bmad-review lenses=structure,prose",
+]
+
+# External-source registry. Natural-language directives describing knowledge
+# bases, MCP tools, or internal systems the LLM may consult during the workflow
+# when a relevant need surfaces. The LLM does NOT query these preemptively —
+# it consults them on demand (during Discovery, validation, drafting, etc.).
+# Each entry names the tool, the conditions for using it, and any fields the
+# tool needs. If a named MCP tool is unavailable at runtime, the LLM falls
+# back to standard behavior and notes the gap. Empty by default.
+#
+# Examples (set in team/user override TOML):
+#   "When researching internal product context, consult corp:kb_search (database='product-docs') before web search."
+#   "For voice-of-customer signal during Discovery, query corp:feedback_search with project={project_name}."
+#   "When validating domain-compliance claims for a healthcare brief, cross-check against corp:hipaa_reference."
+external_sources = []
+
+# External-handoff routing. Natural-language directives the LLM applies at
+# Finalize to route outputs beyond local files (Confluence, Notion, Google
+# Drive, ticket systems, etc.). Each entry names the MCP tool, the destination,
+# and the fields the tool needs. Handoffs run after the artifact is polished
+# and before the final user-facing message. URLs or IDs returned by the
+# destination are captured and surfaced to the user. If a named tool is
+# unavailable at runtime, the handoff is skipped and flagged in the JSON
+# status; local files always exist regardless. Fires automatically — users
+# can opt out in their prompt for a specific run. Empty by default.
+#
+# Examples (set in team/user override TOML):
+#   "After finalize, upload brief.md and addendum.md to Confluence via corp:confluence_upload (space_key='PROD', parent_page='Product Briefs', label='brief', author={user_name})."
+#   "Post a ready-for-review ping to Slack via corp:slack_post (channel='#product', text='New brief: '+{confluence_url})."
+external_handoffs = []
diff --git a/.agents/skills/bmad-project-context/SKILL.md b/.agents/skills/bmad-project-context/SKILL.md
new file mode 100644
index 00000000..dcbf42a1
--- /dev/null
+++ b/.agents/skills/bmad-project-context/SKILL.md
@@ -0,0 +1,110 @@
+---
+name: bmad-project-context
+description: 'Set up, refresh, or audit a repository''s agent instructions (the AGENTS.md block) so AI agents work well in that repo. Also records observed agent mistakes as pitfall lines. Must be invoked by name.'
+---
+
+# Overview
+
+A conversation that produces a repository's agent instructions: a small verified block inside `AGENTS.md`. The user brings rules they want followed — governance, security, standards — and the repository supplies the rest, verified.
+
+Conversational always; the user approves every write.
+
+**Args:** intent (`setup` | `refresh` | `record` | `audit`); a target repo or path; extra source paths or URLs. Supplied values skip their questions.
+
+## Resolution rules
+
+- Bare paths and `{skill-root}` (e.g. `references/best-practices.md`) resolve from this skill's installed directory.
+- `{project-root}` → the project working directory.
+- **Target** → the repository being described, defaulting to `{project-root}`. If it resolves to more than one working tree, or to one the user cannot commit in, ask before writing.
+
+## On Activation
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. On failure, read `{skill-root}/customize.toml` directly and use defaults. Execute `{workflow.activation_steps_prepend}`; treat `{workflow.persistent_facts}` entries as standing context (`file:` = paths/globs to load, others verbatim).
+2. Config: if `{project-root}/_bmad` exists, `uv run {project-root}/_bmad/scripts/resolve_config.py --project-root {project-root}` and read `{user_name}`, `{communication_language}` (use it every turn), `{output_folder}`. Standalone: skip.
+3. **Load `references/best-practices.md` and `references/template.md` before anything else.** Every decision below is made against them.
+4. Detect intent and greet `{user_name}`: **setup** (no block in the target — the default), **refresh** (a block exists), **record** (the user reports a mistake agents made), **audit** (re-verify and prune). Fold `{workflow.external_sources}` into the source list. Execute `{workflow.activation_steps_append}`.
+
+## Setup and Refresh Steps
+
+No writes until step 5!
+
+### 1. Assess and report
+
+Read `AGENTS.md`, harness or agent specific rule files, docs folders, and any notes carrying lessons. Report what exists and how it measures up, per `best-practices.md`.
+
+If the target contains separable units — a workspace manifest listing members, or directories carrying their own build manifest — name them and ask whether this run covers the root only, all of them, or which. Absent that evidence, do not ask. Sibling repositories are not children; each is its own target, offered in turn.
+
+### 2. Ask what they bring
+
+Rules to follow regardless of what the repo does: governance, security and compliance, coding standards, style guides, frozen areas. Ask for outside documents too — handbooks, wikis, architecture docs, MCP knowledgebases. Note the paths; do not read them yet.
+
+Greenfield: this is the whole content. Brownfield: it is the half no scan reaches.
+
+### 3. Discover and verify
+
+Fan out with parallel subagents against what the sections need — executable config and CI for policy and for what they already state, tracked source for conventions and boundaries, targeted history for constraints whose reason must still hold.
+
+`package.json`, a `Makefile`, `pyproject.toml`, and CI config are read to know what the block must not repeat. Their caveats come from the human in step 4. Path-check every claim naming a file.
+
+Each child agreed in step 1 is scanned as its own scope, against its own manifests.
+
+### 4. Interview the gaps
+
+Only what no scan reaches: what agents keep getting wrong here, what is off limits, what a domain term means, why a constraint exists.
+
+- Never ask what a scan could answer. Asking the user to confirm a path-checked claim, or one a config file already states, is a defect.
+- Ask recall questions, not review lists. Never hand the user a selection problem a scan created.
+- A mistake this session made and caught is observed evidence — offer it.
+- A repeatable command spotted in anything read this session — a log, a doc, its own runs — whose correct form is not the obvious guess is a candidate line: offer it. E.g. `uv run pytest` where plain `pytest` looks right but runs outside the project environment.
+- Batches of at most eight; fewer is better. A batch yielding nothing new means write.
+- When the repo contradicts the user, show the evidence and ask. Never write the claim as given, never drop it silently.
+
+### 5. Show the block, then write it
+
+Compose against `template.md`. For each candidate, ask first whether a hook, lint rule, or CI check enforces it better than prose; if so propose the check, and the line becomes the fallback if they decline.
+
+**Show the complete block before writing it**, and every child block alongside it — one approval covers the set. On approval, splice between the markers, leaving everything outside them byte-identical. Fill each provenance line with today's date and the verified SHA.
+
+Where an instruction elsewhere contradicts the block in a way that changes behavior — a stale `CLAUDE.md` line, a retired command — propose the fix to that file. Two live contradictory instructions is a defect.
+
+Never commit.
+
+### 6. Close
+
+- What went in, and what was left out and why.
+- Why, in the user's terms, from `best-practices.md` — why it is small, why what the repo already states stays out, why a pitfall line stays until its cause is gone.
+- How it loads, and that other harness files can point at it.
+- Maintenance: re-run after significant change, `record` the moment an agent gets something wrong, prefer a check over a new line.
+- Rules repeating across their projects, or personal rather than the team's, belong in their global agent config.
+
+### Refresh
+
+Same steps, step 1 as a diff. Read the provenance line, re-verify every path and every caveat, and run `git log --diff-filter=DR --name-only` since the recorded SHA against every line — update or remove lines whose evidence is gone. Never re-ask what a prior run settled; the interview shrinks to what changed about how the team works. The block grows only on new evidence.
+
+### Greenfield
+
+Seeded from a spec or planning document, or interview alone. Commands that do not exist yet are written as explicit TODOs naming the decided stack, never a guessed invocation stated as fact, and verified on the first refresh after code exists. A genuinely contested design decision — real tradeoffs, multiple viable shapes — goes to `bmad-architecture`.
+
+### Migration
+
+If the target has a `project-context.md` from the retired skills, commonly under `{output_folder}`, read it in step 1 and offer to absorb its content. Do not delete it without agreement, and do not silently orphan it.
+
+## Record
+
+Capture one observed agent mistake as it happens — the only admissible source for a pitfall line.
+
+Take the task, the mistake, the correction, and its evidence. Check the block for a line already covering it. One occurrence is noted; a recurring or costly mistake earns a line now — an exact invocation under **Running and verifying** when it is a command error, otherwise a pitfall. Write it and show the diff. If it is mechanically preventable, propose the hook, lint rule, or CI check instead.
+
+## Audit
+
+Re-check every caveat, path-check every file, follow every pointer, and ask of every line whether removing it would change agent behavior. Check for contradictions with other instruction files.
+
+Failing lines move behind an observable trigger, get fixed, or are deleted — confirm deletions first. **A policy or pitfall line goes only when the thing it guards is gone or the user retires it; nothing failing lately is not grounds.** Audit ends smaller or equal.
+
+## Children
+
+A component, nested repository, or extracted rules file gets its own file under the same shape when work keeps landing there and its truths do not belong at the parent level. Rules bounded to a directory go in a nested `AGENTS.md` there, attached by location. Use a linked file only when the trigger is not a path.
+
+A chosen child that ends with nothing its parent does not already say gets no file. Say so and move on.
+
+List every child in the parent's **Where things are** with one line and its path. Discovery never depends on the harness finding it.
diff --git a/.agents/skills/bmad-project-context/customize.toml b/.agents/skills/bmad-project-context/customize.toml
new file mode 100644
index 00000000..bc9fe78a
--- /dev/null
+++ b/.agents/skills/bmad-project-context/customize.toml
@@ -0,0 +1,24 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-project-context.
+# Team overrides:     {project-root}/_bmad/custom/bmad-project-context.toml
+# Personal overrides: {project-root}/_bmad/custom/bmad-project-context.user.toml
+#
+# Merge rules: scalars override (last layer wins); arrays append.
+
+[workflow]
+
+# --- Universal defaults ---
+activation_steps_prepend = []
+activation_steps_append = []
+# Deliberately empty: this skill's own output (AGENTS.md) is loaded by the
+# harness, not through this array. Users append their own standing facts.
+persistent_facts = []
+on_complete = ""
+
+# Standing outside-the-repo sources offered at every setup/refresh run
+# (untrusted until verified against the repo or user-confirmed).
+# Append-only. Entries: "file:{project-root}/..." or "file:/abs/path" for
+# docs, "skill:name" to consult a skill, plain text for a standing fact,
+# "tool:name" for an MCP knowledgebase.
+external_sources = []
diff --git a/.agents/skills/bmad-project-context/references/best-practices.md b/.agents/skills/bmad-project-context/references/best-practices.md
new file mode 100644
index 00000000..0da7e6f8
--- /dev/null
+++ b/.agents/skills/bmad-project-context/references/best-practices.md
@@ -0,0 +1,65 @@
+# What belongs in a repo's agent instructions
+
+Rules for deciding what goes in the block, for judging what a repo already has, and for explaining both to the user.
+
+## The test
+
+Can an agent derive this by reading the repository? If yes, leave it out — a stored copy is a stale duplicate of something the agent reads more accurately first-hand, and it is charged on every session. Write down what the code cannot say.
+
+## Admit
+
+- **Policy the code cannot express** — branch rules, frozen and protected paths, generated files, secrets, security and compliance. Stated by a human or read off an enforcing config, never inferred.
+- **What a config file cannot say about running the project** — the root test script does nothing in this workspace, integration tests need a service up first, the suite takes eleven minutes so iterate on single files, the `Makefile` is the real entry point and `package.json` is vestigial, CI runs a typecheck the test script does not. The invocation itself is already stated in `package.json`, `Makefile`, `pyproject.toml`, or CI config and does not earn a line — the correction or the caveat does.
+- **Conventions that differ from ecosystem defaults.** An agent follows the norm unless told otherwise, so only the divergences earn a line. Command invocations count: when the obvious command is wrong here — a bare-repo prefix, a required wrapper — the exact working invocation earns a line, and no observed mistake is needed to admit it.
+- **Pitfalls with observed evidence** — a recorded lesson, the maintainer's recollection, the same mistake fixed repeatedly in history, or one this session made and caught. A repo yields hundreds of trap-looking facts and none of them predict real mistakes; only observed behavior does. A surprising scan finding is a question to ask, not a line to write.
+- **Runtime behavior invisible from the repo** — replaying webhooks, lying health endpoints, environment quirks — once a human confirms it.
+- **Entry points and pointers** to where work lands.
+
+Prefer prohibitions to advice, and name the permitted alternative in the same line.
+
+## Exclude
+
+| | Why |
+|---|---|
+| Repo overviews, directory trees, stack lists | Derived fresh, more accurately; stored copies rot |
+| Anything included for being interesting | Interest is not need |
+| Style rules an agent self-enforces | Belongs in a formatter, linter, hook, or CI check — propose the check instead |
+| Platitudes | Already the default |
+| Commands already stated in `package.json`, a `Makefile`, or CI config | Read from the source of truth; a copy drifts the moment a script is renamed |
+| Pasted code, changelog content, fast-changing facts | Stale immediately |
+| Aspirational state | Describe what is; intent belongs in specs |
+| History and edit narration | Git holds it; state present truth |
+
+## Retire
+
+A policy or pitfall line goes only when the thing it guards is gone, or the user retires it. Nothing failing lately is not evidence — a working rule erases its own evidence.
+
+Every other line faces one question at each write: would removing it change agent behavior? If no, cut it.
+
+## Size
+
+Every line is paid in every session, and instruction-following degrades as the loaded set grows. Count what other always-loaded files add. Over budget means cut the weakest lines or move them behind a trigger — never raise the budget. Ten lines of evidence means ten lines.
+
+## Retrieval
+
+An index the agent must choose to fetch gets skipped; one already in context does not. Keep everything load-bearing in the block. A pointer out of it names a trigger the agent can observe — a path, a file type, a named task — never one it must judge ("when the task is complex") or track about itself ("before your first edit").
+
+Rules bounded to a directory go in a nested `AGENTS.md` there, attached by location rather than by pointer. Use a linked file only when the trigger is not a path.
+
+## Maintain
+
+- Re-check that caveats still hold — a slow suite that got fast, a workaround for a bug that was fixed.
+- Diff deletions and renames since the verified SHA against every line.
+- Record provenance in the block so the next run knows what it is diffing from.
+- Capture mistakes when they happen, not at review time. One occurrence is a note; recurrence earns a line.
+- Route anything mechanically preventable to a hook, lint rule, or CI check. A check that lands deletes its line.
+
+## Repo or home directory
+
+This block belongs committed: shared by the team, consistent across machines, versioned with the code it constrains.
+
+Two things belong in the user's global agent config instead — rules repeating across all their projects, and personal preferences that are theirs rather than the team's.
+
+## Judging an existing file
+
+Report, in this order: what is derivable filler, what is unverifiable or stale, what is missing against the sections above, and what is already good. Keep recorded lessons by default — they are maintainer testimony, and are challenged only with evidence that the thing they name is gone or wrong.
diff --git a/.agents/skills/bmad-project-context/references/template.md b/.agents/skills/bmad-project-context/references/template.md
new file mode 100644
index 00000000..8a810894
--- /dev/null
+++ b/.agents/skills/bmad-project-context/references/template.md
@@ -0,0 +1,55 @@
+# Block shape
+
+Sections in this order. Omit any section with nothing that passes its rule — never write an empty one. Admission rules: `best-practices.md`.
+
+1. **Orientation** — three or four sentences: what this is, the stack, where planning and deeper docs live.
+2. **Policy** — what the org requires.
+3. **Where things are** — entry points, and pointers to children and linked files.
+4. **Running and verifying** — only what `package.json`, a `Makefile`, or CI config does not already say.
+5. **Conventions that differ from defaults**
+6. **Known pitfalls**
+
+Terse imperative lines under plain headings. No prose beyond Orientation, no introduction, no summary. A bare fact appears only as the justification clause of an instruction — "Exclude `vendor/` from searches, it is 60% of tracked files", never "`vendor/` is 60% of tracked files". A prohibition names the alternative. At most two emphasis markers in the whole block.
+
+## Worked example
+
+````markdown
+<!-- bmad:context -->
+<!-- Verified 2026-08-08 against a1b2c3d. Managed by bmad-project-context; edits inside this block are replaced on refresh. Keep anything you want preserved outside the markers. -->
+
+## acme-billing
+
+Payment processing for Acme storefronts. TypeScript/Node, pnpm, Postgres. Planning lives in `docs/planning/`, tickets in Linear (ACME board).
+
+## Policy
+
+- Never push to main; PRs only, one approval.
+- Never modify `legacy/` — frozen, being replaced. New work goes in `src/`.
+- Never hand-edit `src/generated/` — run `pnpm codegen`.
+
+## Where things are
+
+- Webhook handling: `src/routes/webhooks.ts`; conventions in `docs/webhooks.md`
+- Writing a migration? Read `docs/db-rules.md` first — ordering, transaction boundaries, pool limits.
+- Billing service has its own guide: `services/billing/AGENTS.md`
+
+## Running and verifying
+
+- Run single test files while iterating; the full suite takes ~11 minutes.
+- Integration tests need `docker compose up -d` first, and fail confusingly without it.
+- CI also runs `pnpm typecheck`, which `pnpm test` does not cover.
+
+## Conventions that differ from defaults
+
+- Money is integer cents (`amountCents`), never floats — `src/lib/money.ts`
+- All DB access goes through repositories in `src/repos/`; never call the client directly.
+
+## Known pitfalls
+
+- Stripe webhooks replay in staging every 6h — handlers must be idempotent.
+- Use vitest matchers, not jest — agents repeatedly add jest syntax here.
+
+<!-- /bmad:context -->
+````
+
+Fill the provenance line with the real date and the commit SHA verified against. Refresh diffs from that SHA.
diff --git a/.agents/skills/bmad-qa-generate-e2e-tests/SKILL.md b/.agents/skills/bmad-qa-generate-e2e-tests/SKILL.md
new file mode 100644
index 00000000..14bd3cd3
--- /dev/null
+++ b/.agents/skills/bmad-qa-generate-e2e-tests/SKILL.md
@@ -0,0 +1,176 @@
+---
+name: bmad-qa-generate-e2e-tests
+description: 'Generate end to end automated tests for existing features. Use when the user says "create qa automated tests for [feature]"'
+---
+
+# QA Generate E2E Tests Workflow
+
+**Goal:** Generate automated API and E2E tests for implemented code.
+
+**Your Role:** You are a QA automation engineer. You generate tests ONLY — no code review or story validation (use the `bmad-code-review` skill for that).
+
+## Conventions
+
+- Bare paths (e.g. `checklist.md`) resolve from the skill root.
+- `{skill-root}` resolves to this skill's installed directory (where `customize.toml` lives).
+- `{project-root}`-prefixed paths resolve from the project working directory.
+- `{skill-name}` resolves to the skill directory's basename.
+
+## On Activation
+
+### Step 1: Resolve the Workflow Block
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`
+
+**If the script fails**, resolve the `workflow` block yourself by reading these three files in base → team → user order and applying the same structural merge rules as the resolver:
+
+1. `{skill-root}/customize.toml` — defaults
+2. `{project-root}/_bmad/custom/{skill-name}.toml` — team overrides
+3. `{project-root}/_bmad/custom/{skill-name}.user.toml` — personal overrides
+
+Any missing file is skipped. Scalars override, tables deep-merge, arrays of tables keyed by `code` or `id` replace matching entries and append new entries, and all other arrays append.
+
+### Step 2: Execute Prepend Steps
+
+Execute each entry in `{workflow.activation_steps_prepend}` in order before proceeding.
+
+### Step 3: Load Persistent Facts
+
+Treat every entry in `{workflow.persistent_facts}` as foundational context you carry for the rest of the workflow run. Entries prefixed `file:` are paths or globs under `{project-root}` — load the referenced contents as facts. All other entries are facts verbatim.
+
+### Step 4: Load Config
+
+Load config from `{project-root}/_bmad/bmm/config.yaml` and resolve:
+
+- `project_name`, `user_name`
+- `communication_language`, `document_output_language`
+- `implementation_artifacts`
+- `date` as system-generated current datetime
+- YOU MUST ALWAYS SPEAK OUTPUT in your Agent communication style with the config `{communication_language}`
+
+### Step 5: Greet the User
+
+Greet `{user_name}`, speaking in `{communication_language}`.
+
+### Step 6: Execute Append Steps
+
+Execute each entry in `{workflow.activation_steps_append}` in order.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+## Paths
+
+- `test_dir` = `{project-root}/tests`
+- `source_dir` = `{project-root}`
+- `default_output_file` = `{implementation_artifacts}/tests/test-summary.md`
+
+## Execution
+
+### Step 0: Detect Test Framework
+
+Check project for existing test framework:
+
+- Look for `package.json` dependencies (playwright, jest, vitest, cypress, etc.)
+- Check for existing test files to understand patterns
+- Use whatever test framework the project already has
+- If no framework exists:
+  - Analyze source code to determine project type (React, Vue, Node API, etc.)
+  - Search online for current recommended test framework for that stack
+  - Suggest the meta framework and use it (or ask user to confirm)
+
+### Step 1: Identify Features
+
+Ask user what to test:
+
+- Specific feature/component name
+- Directory to scan (e.g., `src/components/`)
+- Or auto-discover features in the codebase
+
+### Step 2: Generate API Tests (if applicable)
+
+For API endpoints/services, generate tests that:
+
+- Test status codes (200, 400, 404, 500)
+- Validate response structure
+- Cover happy path + 1-2 error cases
+- Use project's existing test framework patterns
+
+### Step 3: Generate E2E Tests (if UI exists)
+
+For UI features, generate tests that:
+
+- Test user workflows end-to-end
+- Use semantic locators (roles, labels, text)
+- Focus on user interactions (clicks, form fills, navigation)
+- Assert visible outcomes
+- Keep tests linear and simple
+- Follow project's existing test patterns
+
+### Step 4: Run Tests
+
+Execute tests to verify they pass (use project's test command).
+
+If failures occur, fix them immediately.
+
+### Step 5: Create Summary
+
+Output markdown summary:
+
+```markdown
+# Test Automation Summary
+
+## Generated Tests
+
+### API Tests
+- [x] tests/api/endpoint.spec.ts - Endpoint validation
+
+### E2E Tests
+- [x] tests/e2e/feature.spec.ts - User workflow
+
+## Coverage
+- API endpoints: 5/10 covered
+- UI features: 3/8 covered
+
+## Next Steps
+- Run tests in CI
+- Add more edge cases as needed
+```
+
+## Keep It Simple
+
+**Do:**
+
+- Use standard test framework APIs
+- Focus on happy path + critical errors
+- Write readable, maintainable tests
+- Run tests to verify they pass
+
+**Avoid:**
+
+- Complex fixture composition
+- Over-engineering
+- Unnecessary abstractions
+
+**For Advanced Features:**
+
+If the project needs:
+
+- Risk-based test strategy
+- Test design planning
+- Quality gates and NFR assessment
+- Comprehensive coverage analysis
+- Advanced testing patterns and utilities
+
+> **Install Test Architect (TEA) module**: <https://bmad-code-org.github.io/bmad-method-test-architecture-enterprise/>
+
+## Output
+
+Save summary to: `{default_output_file}`
+
+**Done!** Tests generated and verified. Validate against `./checklist.md`.
+
+## On Complete
+
+Run: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow.on_complete`
+
+If the resolved `workflow.on_complete` is non-empty, follow it as the final terminal instruction before exiting.
diff --git a/.agents/skills/bmad-qa-generate-e2e-tests/checklist.md b/.agents/skills/bmad-qa-generate-e2e-tests/checklist.md
new file mode 100644
index 00000000..aa38ae89
--- /dev/null
+++ b/.agents/skills/bmad-qa-generate-e2e-tests/checklist.md
@@ -0,0 +1,33 @@
+# QA Automate - Validation Checklist
+
+## Test Generation
+
+- [ ] API tests generated (if applicable)
+- [ ] E2E tests generated (if UI exists)
+- [ ] Tests use standard test framework APIs
+- [ ] Tests cover happy path
+- [ ] Tests cover 1-2 critical error cases
+
+## Test Quality
+
+- [ ] All generated tests run successfully
+- [ ] Tests use proper locators (semantic, accessible)
+- [ ] Tests have clear descriptions
+- [ ] No hardcoded waits or sleeps
+- [ ] Tests are independent (no order dependency)
+
+## Output
+
+- [ ] Test summary created
+- [ ] Tests saved to appropriate directories
+- [ ] Summary includes coverage metrics
+
+## Validation
+
+Run the tests using your project's test command.
+
+**Expected**: All tests pass ✅
+
+---
+
+**Need more comprehensive testing?** Install [Test Architect (TEA)](https://bmad-code-org.github.io/bmad-method-test-architecture-enterprise/) for advanced workflows.
diff --git a/.agents/skills/bmad-qa-generate-e2e-tests/customize.toml b/.agents/skills/bmad-qa-generate-e2e-tests/customize.toml
new file mode 100644
index 00000000..0a2c6fec
--- /dev/null
+++ b/.agents/skills/bmad-qa-generate-e2e-tests/customize.toml
@@ -0,0 +1,41 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-qa-generate-e2e-tests. Mirrors the
+# agent customization shape under the [workflow] namespace.
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+# Steps to run before the standard activation (config load, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Distinct from the runtime memory sidecar — these are static context
+# loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "All tests must follow the project's existing test framework patterns."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Scalar: executed when the workflow reaches Step 5 (Create Summary),
+# after all tests pass and the summary document is saved. Override wins.
+# Leave empty for no custom post-completion behavior.
+
+on_complete = ""
diff --git a/.agents/skills/bmad-quick-dev/SKILL.md b/.agents/skills/bmad-quick-dev/SKILL.md
new file mode 100644
index 00000000..d57e4bf0
--- /dev/null
+++ b/.agents/skills/bmad-quick-dev/SKILL.md
@@ -0,0 +1,19 @@
+---
+name: bmad-quick-dev
+description: "Deprecated: forwards to bmad-build. Do not use unless invoked by name."
+---
+
+# Deprecated Build Alias
+
+## On Activation
+
+1. Check whether either legacy customization file exists:
+   - `{project-root}/_bmad/custom/bmad-quick-dev.toml`
+   - `{project-root}/_bmad/custom/bmad-quick-dev.user.toml`
+2. If neither legacy file exists, output exactly `bmad-quick-dev is deprecated. Redirecting to bmad-build. Please use bmad-build in the future.`, invoke `bmad-build` exactly once with the user's original input verbatim, then execute no further steps in this shim.
+3. For every legacy file that exists, use its matching new filename:
+   - `{project-root}/_bmad/custom/bmad-quick-dev.toml` becomes `{project-root}/_bmad/custom/bmad-build.toml`.
+   - `{project-root}/_bmad/custom/bmad-quick-dev.user.toml` becomes `{project-root}/_bmad/custom/bmad-build.user.toml`.
+4. If the matching new file does not exist, tell the user that the customization file uses the deprecated name and offer to rename it. Rename it only after explicit approval. If approval is declined or unavailable, or the rename fails, HALT and do not invoke any skill.
+5. If the matching new file already exists, do not overwrite it. Read both files, explain their differences, and propose the exact content for the new file. Resolve conflicting values with the user. Only after the user explicitly approves that content, save and verify the new file, then remove the legacy file. If approval is declined or unavailable, or any operation fails, HALT and do not invoke any skill.
+6. After every detected legacy file has been migrated successfully and no legacy file remains, output exactly `bmad-quick-dev is deprecated. Redirecting to bmad-build. Please use bmad-build in the future.`, invoke `bmad-build` exactly once with the user's original input verbatim, then execute no further steps in this shim.
diff --git a/.agents/skills/bmad-retrospective/SKILL.md b/.agents/skills/bmad-retrospective/SKILL.md
new file mode 100644
index 00000000..4786a018
--- /dev/null
+++ b/.agents/skills/bmad-retrospective/SKILL.md
@@ -0,0 +1,94 @@
+---
+name: bmad-retrospective
+description: 'Evidence-based epic retrospective — collect what the epic produced, verify findings against sources, render an acceptance verdict. Use when the user says "run a retrospective" or "lets retro the epic [epic]". Supports -H/--headless.'
+---
+
+# Retrospective
+
+Review a completed epic by reading the evidence it left — the epic spec, story files, the full diff, per-story commits, sprint status, and session logs when they exist. An unattended epic run leaves a record; this skill reads that record, surfaces the defects no single story could show, and judges the epic against the criteria it set for itself.
+
+Every finding you report carries a source reference (file, line, commit, or log). A claim you cannot point at — an invented root cause, a pattern the diff does not actually show — is not a finding. Drop it.
+
+## Resolution rules
+
+- Bare paths and `{skill-root}` (e.g. `references/aggregate-views.md`, `scripts/sprint_status.py`) resolve from this skill's installed directory.
+- `{project-root}` → the project working directory.
+- `{skill-name}` → the skill directory's basename.
+
+## Modes
+
+Interactive by default. With `-H`/`--headless`: skip every confirmation, take the epic from the invocation (falling back to detection only if none was supplied), never open the team discussion, render the verdict on the evidence alone, and record each assumption made without the user (which epic was selected, the machine verdict, each proposed item) into the retrospective document's Assumptions section so the audit trail survives. The Phase 4 acceptance fail-safe still applies in headless runs.
+
+For automation, `-H <epic>` — an explicit epic in headless mode — is the stable orchestrator-facing interface. Pass the same number to `detect-epic --epic <N>` so the unfinished-story gate is script-backed (see Inputs). Epic auto-detection is a human convenience, not an automation contract: unflagged `detect-epic` returns the highest epic with *any* `done` story, and stories-mode projects have no `sprint-status.yaml` to detect from.
+
+## On Activation
+
+Run these in order before the retrospective begins:
+
+1. **Resolve the workflow block.** Run `uv run --no-cache {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. If it fails, resolve `{workflow.*}` yourself by reading `{skill-root}/customize.toml`, then `{project-root}/_bmad/custom/{skill-name}.toml`, then `.user.toml` in that order, merging base → team → user (scalars override, keyed arrays-of-tables merge by `code`/`id`, other arrays append).
+2. **Run prepend steps** — execute each entry in `{workflow.activation_steps_prepend}` in order.
+3. **Load persistent facts** — treat every `{workflow.persistent_facts}` entry as standing context. `file:` entries are paths/globs under `{project-root}` whose contents load as facts; all others are literal facts.
+4. **Load config** from `{project-root}/_bmad/bmm/config.yaml`: `project_name`, `user_name`, `communication_language`, `document_output_language`, `user_skill_level`, `planning_artifacts`, `implementation_artifacts`, and `date` (system datetime), plus `output_folder` from `{project-root}/_bmad/core/config.yaml`. Speak all output in `{communication_language}`; write all documents in `{document_output_language}`. Never state time estimates — AI has changed development speed, so hour/day/week predictions are noise.
+5. **Greet and orient** (interactive only). Greet `{user_name}`, name the epic you are about to retro, and optionally invite their going-in concerns ("anything you want weighted — a story that felt rushed, a risky interaction between two stories?"). Use any answer to focus the Phase 1–2 analysis; it directs attention but never becomes a finding without a source.
+6. **Run append steps** — execute each entry in `{workflow.activation_steps_append}` in order.
+
+## Inputs
+
+| Input | Where | Use |
+|-------|-------|-----|
+| epic | invocation argument, or detected from sprint status | which epic to retro |
+| spec folder | invocation argument, or found under the spec roots | the stories-mode epic: `SPEC.md`, ordered `stories.yaml`, `stories/<id>-*.md` |
+| sprint status | `{implementation_artifacts}/sprint-status.yaml` | epic detection + final status update |
+| architecture / prd | `{planning_artifacts}/*architecture*`, `*prd*` | context for judging as-built vs intended |
+| previous retro (optional) | `{implementation_artifacts}/**/epic-{{prev}}-retro-*.md` | check whether last epic's actions landed |
+| session logs (optional) | conversation/session records for the epic's stories | process lessons; record the gap when absent |
+
+An epic reaches this skill in one of two shapes, and they are peers. **Sprint mode** reads `sprint-status.yaml`. **Stories mode** reads a spec folder holding `SPEC.md`, an ordered `stories.yaml`, and `stories/<id>-*.md` artifacts — the shape an unattended run leaves behind. Resolve which applies first: a named folder is stories mode whether or not sprint status exists; a named epic number is sprint mode; with neither, use sprint mode when `sprint-status.yaml` exists, and otherwise look for spec folders under `{output_folder}/specs`, `{planning_artifacts}`, and `{implementation_artifacts}`. Ask the user which to retro when there is more than one, and never choose silently; headless, stop and require an explicit folder.
+
+In stories mode, `stories.yaml` in list order is the story list — list order is authoritative, filename sort is not — and each story's `stories/<id>-*.md` frontmatter carries its `status`. `pending_stories` is the ids whose status is not `done`; apply the same completeness gate as below. Then skip to Phase 1: do not read or write sprint status for the rest of the run. The rest of this section is sprint mode.
+
+Determine the epic and its unfinished-story list from `sprint_status.py detect-epic` whenever `{implementation_artifacts}/sprint-status.yaml` is available:
+
+- **Epic supplied** (including the stable `-H <epic>` orchestrator path): run `uv run --no-cache {skill-root}/scripts/sprint_status.py detect-epic --file {implementation_artifacts}/sprint-status.yaml --epic <N>`. The script scopes `pending_stories` to that number even when auto-detect would have picked a different epic, and even when the epic has no `done` story yet. `story_count` is that same scoped count of the epic's story keys: `0` means the file has no such epic at all — a nonexistent epic returns the same empty `pending_stories` as a finished one, so treat `story_count: 0` as a likely mistyped epic number, confirm with the user, and headless, stop and report rather than proceeding.
+- **No epic supplied**: run the same command without `--epic` (returns the highest epic with a `done` story). Confirm the detected epic with the user and let them override; in headless mode accept it and record the assumption. If detection returns none, ask the user — or, headless, stop and report.
+
+If the script exits non-zero it emits `{"ok": false, "error": ...}` instead of a detection — the normal path for a stories-mode project with no `sprint-status.yaml`, and for a file that does not parse: surface that error verbatim — or, if the script produced no JSON at all, whatever it wrote to stderr — and ask the user which epic to retro; headless, stop and report. Without a readable sprint-status file there is no `pending_stories` list; record that the completeness check did not run and continue only if the user (or headless Assumptions trail) accepts proceeding without it.
+
+Then check the epic is actually finished before Phase 1. A successful detect carries `pending_stories` — the selected epic's story keys whose status is not `done`, in file order, scoped to that epic alone (an unfinished story in some *other* epic is out of scope for this retrospective). When the list is non-empty, interactively list those stories and ask whether to retro an unfinished epic: if the user declines, stop and report — do not enter Phase 1; if they accept, record the stories they accepted proceeding over in the document's Epic summary. Headless, proceed and record the same list in the Assumptions section — do not invent a confirmation. Either way the list sits in the document, and Phase 4's machine verdict is **rejected** when any story remained unfinished (see `references/acceptance-verdict.md`); a human may override interactively.
+
+## Working state and resumption
+
+The retrospective document is the working artifact, not only the final output. Once the epic is fixed, create it as a skeleton (`references/retro-document.md` names the sections) and write each phase's result into it as you finish — inventory, then findings with sources, then dispositions and verdict. Continuity is re-reading the file.
+
+If a retrospective document for this epic already exists, load it, reconcile its recorded state against the current evidence — the current evidence wins, since commits may have landed and questions may have been answered since — and resume at the first incomplete phase instead of redoing finished ones. In stories mode that document is `{spec-folder}/RETROSPECTIVE.md`, a fixed name so a resumed run finds it; sprint mode keeps its dated `{implementation_artifacts}` filename.
+
+## Flow
+
+Run the phases in order. A default run stops at a written evidence report and verdict; the team discussion in Phase 3 is opt-in.
+
+### Phase 1 — Gather
+
+Enumerate what the epic actually produced and record what is missing. Load `references/evidence-gathering.md` for the inventory checklist, the `git_evidence.py` pre-pass that derives the diff range and per-story commits, and the missing-evidence rule: each later analysis declares what it needs and records a narrowed scope when the evidence is absent, so a reader can always tell "checked and clean" from "never checked."
+
+### Phase 2 — Analyze
+
+Produce findings, each with a source reference, from three angles:
+
+- **Aggregate views** — the defects no single diff hunk shows: architecture delta, duplication map, god-class growth, pattern divergence, spec-to-implementation reconciliation. Load `references/aggregate-views.md` for the catalog and how to derive each (deterministic scripts first).
+- **Diff-scope review** — do not reimplement review. Invoke **`bmad-review`** on the epic's diff for the code lenses (adversarial, edge-case, verification-gap), weighting the boundaries between stories, where no single session ever saw both sides. Fold its findings in. If `bmad-review` is unavailable, run those lenses inline over the diff on a narrowed scope and record the narrowing.
+- **Behavior check (when the epic changed runtime behavior)** — exercise the changed flows end to end and record what you observed. Passing tests do not substitute for running the system.
+
+Consolidate: merge, dedupe, and provenance-link findings. Drop any finding you cannot tie to a source.
+
+### Phase 3 — Team Discussion (opt-in)
+
+Skip by default; never runs headless. When the user asks to "discuss it as a team," "run party mode," or similar, invoke the skill `bmad-party-mode` seeded with the Phase 2 findings so the installed agents react to real evidence — the god class the diff really grew, the verification gap that is actually there, the wins the evidence confirms. Load `references/team-discussion.md` for how to seed it and keep it grounded. If `bmad-party-mode` is unavailable, run the discussion inline over the Phase 2 findings and record the narrowing. The rule: agents speak only to findings with sources.
+
+### Phase 4 — Decide
+
+- **Action items** — compile fix-now findings and process lessons into specific, owned action items. Fixes and spec reconciliations are *proposed here*, not auto-applied; the human decides what to execute.
+- **Acceptance verdict** — judge the final state against the epic's declared acceptance criteria (profile it from the diff and stories if none were declared): **accepted**, **accepted-with-open-items**, or **rejected** — one spelling, everywhere a machine reads it. Unfinished stories in `pending_stories` force the machine verdict to **rejected**. A human decision always overrides. An epic that fails its criteria with no human decision is recorded as *not accepted* — never as silently accepted. Load `references/acceptance-verdict.md` for the rubric, the finding-routing dispositions, and the previous-retro follow-through record — the per-item evidence Phase 5's status offer reads.
+
+### Phase 5 — Finalize
+
+Finalize the retrospective document and update sprint status. Load `references/retro-document.md` for the document's sections and the exact `sprint_status.py update` invocation that marks the retro key `done`, appends the action items, and validates the write. Where the Phase 4 follow-through has evidence a *previous* epic's action item landed, offer `--set-action-status` and pass only the transitions the user confirms — the evidence justifies proposing a transition, and only the user's confirmation justifies writing it; a headless run records the transitions it would have proposed and does not pass the flag at all. In stories mode, finalize `{spec-folder}/RETROSPECTIVE.md` and stop there: no `sprint_status.py` call, no sprint-status file created, and no edits to `SPEC.md`, `stories.yaml`, or any story artifact. Then, if `{workflow.on_complete}` is non-empty, follow it as the final instruction.
diff --git a/.agents/skills/bmad-retrospective/customize.toml b/.agents/skills/bmad-retrospective/customize.toml
new file mode 100644
index 00000000..816e886f
--- /dev/null
+++ b/.agents/skills/bmad-retrospective/customize.toml
@@ -0,0 +1,41 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-retrospective. Mirrors the
+# agent customization shape under the [workflow] namespace.
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+# Steps to run before the standard activation (config load, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Distinct from the runtime memory sidecar — these are static context
+# loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "All retrospectives must produce SMART action items with named owners."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Scalar: executed at the end of Phase 5 (Close), after the retrospective
+# document is saved and sprint-status is updated. Override wins.
+# Leave empty for no custom post-completion behavior.
+
+on_complete = ""
diff --git a/.agents/skills/bmad-retrospective/references/acceptance-verdict.md b/.agents/skills/bmad-retrospective/references/acceptance-verdict.md
new file mode 100644
index 00000000..1d83dfa9
--- /dev/null
+++ b/.agents/skills/bmad-retrospective/references/acceptance-verdict.md
@@ -0,0 +1,55 @@
+# Decide: Routing and the Acceptance Verdict
+
+Phase 4. Turn the consolidated findings into two outputs: routed action items the human can act on, and an honest verdict on whether the epic met its acceptance criteria. This skill proposes; it does not auto-apply fixes or edit the project spec. The human decides what executes.
+
+## Route each finding
+
+Give every finding two independent dispositions:
+
+- **What to do about this instance** — *fix now*, *defer*, or *accept as-is*. Fix-now findings become action items. Deferred findings carry enough context to be acted on later without re-investigation. Accepted deviations are recorded so later retros stop re-flagging them.
+- **What would prevent the next one** — the upstream lesson: spec wording, story sizing, a missing convention or gate, or nothing. This is where a recurring finding becomes a process change rather than a one-off fix.
+
+Findings from sub-agents or the team discussion are unverified reports, not established facts. Before an action item relies on one, re-check it against the primary source — reopen the file, the commit, the spec. A finding whose source does not hold up is dropped, not routed.
+
+## Action items
+
+Compile fix-now findings and process lessons into specific, owned action items. Each names what to change and who owns it. Two kinds are *proposed, not applied* in this version:
+
+- **Remediation** — code fixes are written up as action items (or story-shaped work) for the normal dev loop to execute later. The retrospective does not run the dev loop itself.
+- **Spec reconciliation** — where the as-built diverges from the spec, propose the reconciliation as an action item with the evidence attached. The human applies it to the project contract; an uncertain interpretation is never written into the spec automatically.
+
+## Previous-retro follow-through
+
+When a prior retro exists, check whether the action items it committed to were completed. Read `action_items` in `{implementation_artifacts}/sprint-status.yaml` and, for every entry belonging to an earlier epic that is not already `done`, record one line in the retrospective document's Previous-retro follow-through section:
+
+- **How to address the item** — its `id`, exactly as the file spells it. Legacy entries written before ids existed have none; for those, record the item's `epic` (the integer in the file) plus its exact `action` text, character for character. One or the other is what Phase 5 needs to name the item at all.
+- **Whether it landed** — with the source that shows it: the commit, the file and line, the test. An item you cannot point at is "no evidence found", not "not done" — the reader must be able to tell a checked item from an unchecked one.
+- **The status it argues for** — `done` for a landed item, `in-progress` for one demonstrably underway, or nothing. A proposal, never a write.
+
+That record is exactly what Phase 5's `--set-action-status` offer reads: the selector becomes the JSON, the evidence is what the user is asked to confirm, and the proposed status is written only if they confirm it. A run with no prior retro, or one whose `sprint-status.yaml` is unreadable or carries no `action_items`, records that there was nothing to follow through on — and which of those it was, so a missing file is never mistaken for "no outstanding items."
+
+## The verdict
+
+Judge the final state against the epic's declared acceptance criteria. If the epic declared none, profile the criteria from the diff and stories and mark the verdict as **profiled** rather than declared. Weigh verification results (the Phase 2 behavior check) and unresolved findings. Render one of:
+
+- **Accepted** — criteria demonstrably met in the evidence, no blocking findings open, and **no unfinished stories** for this epic.
+- **Accepted-with-open-items** — criteria met, but named findings remain deferred and tracked — still only when every story of this epic is `done`.
+- **Rejected** — criteria not met, a blocking finding stands unresolved, **or any of this epic's stories is still not `done`**.
+
+### Unfinished stories
+
+`pending_stories` is authoritative for this epic's incomplete work, whichever mode produced it: sprint-status story keys in file order from `detect-epic`, or `stories.yaml` ids in list order whose artifact status is not `done`. When that list is non-empty:
+
+- The **machine** verdict is **rejected**. Name every unfinished story key in the Acceptance verdict section as the evidence. Do not soften this to accepted-with-open-items: unfinished delivery is not an open finding about a finished epic — the epic itself is incomplete.
+- Record the unfinished keys in Epic summary (interactive) or Assumptions (headless) as the Inputs section already requires.
+- Headless runs have no human at the console: the document's verdict is **rejected** when `pending_stories` was non-empty. Interactive runs may still let a human override (rule 1 below) after seeing the list.
+
+If the completeness check did not run (no readable `sprint-status.yaml`), do **not** render a rejected or accepted verdict from the absence of data — say the check was unavailable and weigh only the criteria and findings you have.
+
+Three hard rules:
+
+1. A human decision always overrides the machine verdict.
+2. An epic that fails its criteria with **no** human decision is recorded as **not accepted** — never as silently accepted.
+3. A non-empty `pending_stories` list makes the machine verdict **rejected**, including in headless mode.
+
+The verdict and its evidence carry into the Phase 5 document.
diff --git a/.agents/skills/bmad-retrospective/references/aggregate-views.md b/.agents/skills/bmad-retrospective/references/aggregate-views.md
new file mode 100644
index 00000000..1893e187
--- /dev/null
+++ b/.agents/skills/bmad-retrospective/references/aggregate-views.md
@@ -0,0 +1,17 @@
+# Aggregate Views
+
+Phase 2. An epic is many coding sessions, each validated in isolation; the defects that matter are the ones no single session — and no single diff hunk — could see. Nine sessions each added three hundred lines and none ever saw the 3,000-line class they collectively built. These views are properties of the *whole* change, derived across the full diff range from Phase 1.
+
+Prefer deterministic derivation: a script that measures the codebase is evidence; a model's impression is not. Where you compute a view inline instead of by script, record the narrowed scope. Every observation that becomes a finding carries a source reference — the file, the symbol, the commits. `references/evidence-gathering.md` is authoritative for what every `git_evidence.py` key means, including the commit-level `is_merge` and `stories` (every story id a subject names, so a commit spanning two counts for both) — read it there before deriving anything from the numbers.
+
+## The catalog
+
+- **Architecture delta** — how the dependency structure changed across the epic. Where a language-native dependency tool exists (dependency-cruiser, madge, pydeps, and the like), run it before and after the range and diff the graphs; otherwise derive the module/import graph from the changed files. Look for new cross-cutting dependencies, layering violations, and cycles introduced — structure the code's own conventions would forbid but no single story tripped.
+- **Duplication map** — the same problem solved more than one way across stories. Two sessions independently writing near-identical logic, or a helper reimplemented because the second session did not know the first existed.
+- **God-class / size growth** — files that grew past a healthy size *over the epic*, invisible per-commit because each session added only a little. The `git_evidence.py` pre-pass (Phase 1) reports `added` / `deleted` / `net` per path in `files` — *change volume*, not a file's absolute size or a per-commit growth rate. Those sums cover the range's **non-merge** commits only, and they are always integers: an unmeasurable revision is left out of them rather than nulling them. Rank on `files`, then open the top of the ranking and read each file's real current size and structure before calling anything a god-class — high net churn makes a file a candidate to inspect, not a verdict on its own. Three qualifiers say how far the ranking can be trusted: `binary_revisions` counts that path's revisions whose churn could not be measured, so its true volume is *at least* what the sums report; `merges_measured` short of `merge_count` means some merges were never measured at all, which caps how complete the ranking can be; and `merge_files` mostly restates churn `files` already counted, so summing the two double counts — but it is not redundant, because a merge's first-parent diff also carries whatever the conflict resolution itself added, code that lives in no non-merge commit and therefore appears in `files` nowhere. So read `merge_files` separately, for the paths whose churn shows up only there, rather than discarding it as double counting. Whether a flagged file is genuinely a god-class or legitimately large stays your judgment.
+- **Pattern divergence** — where the epic's code diverges from the conventions the surrounding codebase already established: naming, error handling, test structure, module boundaries. Agents learn conventions by pattern-matching the code, so divergence compounds.
+- **Spec-to-implementation reconciliation** — where the as-built diverges from what the epic spec and PRD/architecture described. Requirements silently dropped, added behavior nobody specified, intent reinterpreted between stories. Each divergence is either a defect (fix), an accepted deviation (record so later runs stop re-flagging it), or a spec that should be reconciled to reality (propose in Phase 4).
+
+## Delegation
+
+When sub-agents are available, delegate the derivation: each returns evidence with source refs and checked scope, never a verdict — the parent consolidates and decides. Give each a narrow view and an explicit return format. When sub-agents are unavailable, compute the highest-value views inline (architecture delta and spec reconciliation first) and record which views were narrowed or skipped.
diff --git a/.agents/skills/bmad-retrospective/references/evidence-gathering.md b/.agents/skills/bmad-retrospective/references/evidence-gathering.md
new file mode 100644
index 00000000..90e62134
--- /dev/null
+++ b/.agents/skills/bmad-retrospective/references/evidence-gathering.md
@@ -0,0 +1,30 @@
+# Evidence Gathering
+
+Phase 1 of the retrospective. Enumerate what the completed epic produced, so every later analysis works from real artifacts instead of memory. Output is an inventory: what exists, what is missing, and the diff range the rest of the retro will read.
+
+## Inventory checklist
+
+Collect what the epic produced and note the source path or range of each:
+
+- **Epic spec** — the epic file under `{planning_artifacts}`, including any declared acceptance criteria. If the spec declares how the epic will be judged, that governs Phase 4; if not, note that the verdict will be profiled from the diff.
+- **Story files** — the story specs implemented under this epic (`{implementation_artifacts}`), each carrying its intent and context. These mark the boundaries between coding sessions.
+- **Diff range and commits** — the full set of changes the epic introduced. Establish the range from the first and last story commits (or ask the user for it). The range must *include* the first story commit: `A..B` excludes `A`, so use the parent of the first commit as the left endpoint — `<first-commit>^..<last-commit>` — or the whole first story disappears from the diff, the commit attribution, and the verdict evidence. Then run `uv run --no-cache {skill-root}/scripts/git_evidence.py --repo {project-root} --range <range> --stories <story-ids>` to get, as JSON, the per-story commit attribution and the per-file change volume — added / deleted / net across the range — that Phase 2 reads. Record the range explicitly; Phase 2's aggregate views and the `bmad-review` pass both read it. When the range cannot be established, say so and narrow the scope rather than guessing. Read the output keys precisely: each commit carries `is_merge` and `stories` — *every* id its subject names, so a commit spanning two stories counts for both. `files` sums non-merge commits only. `merge_files` is each measured merge's diff against its first parent, so it *restates* the churn that merge brought in plus whatever the conflict resolution added — never add it into `files`, and never read it as merge-introduced work on its own. `merges_measured` counts the merges on the range head's first-parent spine; `merge_count` counts every merge in the range, so a gap between the two means merges went unmeasured. `binary_revisions` is unmeasured churn, not zero churn.
+- **Sprint status** — `{implementation_artifacts}/sprint-status.yaml`, for which stories are `done` and the current retro-key state.
+- **Previous retrospective** — the prior epic's retro doc, if one exists, so Phase 4 can check whether last epic's action items landed.
+- **Session logs** — conversation or session records for the epic's stories, when available. They are the only record of *why* a session took an unexpected turn — what was tried and abandoned. They are also the evidence most likely to be deleted or expire, so capture references now.
+
+## Stories mode
+
+A stories-mode epic is a spec folder. Map it onto the checklist above: `SPEC.md` is the epic spec; `stories.yaml` in list order is the story list, each entry's artifact being the single `stories/<id>-*.md` it names; there is no sprint status; the previous retrospective, when resuming, is `{spec-folder}/RETROSPECTIVE.md`; session logs are unchanged.
+
+The diff range differs. Each story records its own baseline in its artifact frontmatter — `baseline_revision` (deprecated) or `baseline_commit` — so there is no single epic-wide range. The range end is the next story's baseline in list order, which is exact because neither skill adds a commit of its own after the work. For the last story, when nothing records the end, derive it from the history — usually `HEAD`, though not always — and mark it inferred rather than recorded. A baseline that is absent or is not a revision leaves that story with no commit or diff evidence — record that too. Group the stories sharing an identical range and run `git_evidence.py` once per distinct range, passing that group's ids as one comma-separated `--stories` value. No `^` is needed here: unlike the sprint-mode range above, the recorded baseline is already the pre-change commit. Ranges may overlap or diverge; count a shared commit or file change once in the aggregate views while keeping each story's range as its provenance.
+
+## Missing evidence
+
+Evidence availability varies; never hide a gap. Each later analysis declares what it needs and, when that input is absent, records a narrowed scope rather than guessing. A reader of the final retro must always be able to tell **"checked and clean"** from **"never checked."**
+
+- Missing session logs → process-lesson analysis is skipped, and the retro says so.
+- No declared acceptance criteria → the verdict is profiled from the diff and stories, flagged as profiled rather than declared.
+- Sub-agents unavailable → analyses that would delegate run inline over a narrowed scope, and the narrowing is recorded.
+
+Carry the inventory forward into Phase 2 as the authoritative list of what is available to read.
diff --git a/.agents/skills/bmad-retrospective/references/retro-document.md b/.agents/skills/bmad-retrospective/references/retro-document.md
new file mode 100644
index 00000000..8fdcb9e1
--- /dev/null
+++ b/.agents/skills/bmad-retrospective/references/retro-document.md
@@ -0,0 +1,84 @@
+# Finalize: Retrospective Document and Sprint Status
+
+Phase 5. Finalize the retrospective and update sprint tracking. Two writes: the retrospective document, and the `sprint-status.yaml` update. Stories mode makes only the first.
+
+## The retrospective document
+
+This document is the run's working artifact: it is created as a skeleton once the epic is fixed and filled as each phase completes, so Phase 5 finalizes rather than writes it from scratch. It lives at `{implementation_artifacts}/epic-{{epic_number}}-retro-{date}.md`, in `{document_output_language}`, as readable markdown; ensure `{implementation_artifacts}` exists. In stories mode it lives at `{spec-folder}/RETROSPECTIVE.md` instead — a fixed name, so a resumed run finds it — and carries the same frontmatter without `epic`, which the folder already names.
+
+Open the document with YAML frontmatter a machine can read without parsing the prose — an epic gate or orchestrator keys off `verdict` to decide whether to hold the next epic:
+
+```
+---
+epic: {{epic_number}}
+date: {date}
+verdict: accepted | accepted-with-open-items | rejected
+criteria: declared | profiled
+headless: true | false
+---
+```
+
+Keep `verdict` in sync with the Acceptance verdict section below. Do not encode the verdict in the sprint-status retro key — that key's value stays `done` so the existing lifecycle consumers (sprint planning's `optional ↔ done` transition, status TUIs) keep working unchanged.
+
+That holds for a **rejected** epic too: the update below marks the retro key `done` whichever way the verdict went, because `done` there means *the retrospective ran*, not *the epic passed*. The script writes no verdict of any kind into `sprint-status.yaml` — there is no `retro_verdict` key and `--verdict` is only echoed back in the result JSON — so a gate or orchestrator that acts on the verdict **must** read this document's frontmatter. Reading sprint-status alone cannot tell a rejected epic from an accepted one.
+
+Sections:
+
+- **Epic summary** — which epic, the diff range, stories completed, any stories still unfinished (`pending_stories`) that the user accepted retro-ing over, the evidence inventory (what was available, what was missing). Unfinished stories force the machine acceptance verdict to **rejected** (see `references/acceptance-verdict.md`).
+- **Findings** — grouped by aggregate view and by lens, each with its source reference and disposition (fix now / defer / accept). This is the record; do not summarize away the provenance.
+- **Behavior verification** — what was exercised end to end and what was observed, or an explicit note that runtime behavior was not exercised.
+- **Previous-retro follow-through** — if a prior retro exists, whether its action items landed, with evidence, and the selector Phase 5 would need to act on each (`references/acceptance-verdict.md` specifies what to record).
+- **Action items** — the routed fix-now items and process lessons, each with an owner. Note which are proposed remediation or spec reconciliations awaiting human application.
+- **Acceptance verdict** — accepted / accepted-with-open-items / rejected, whether the criteria were declared or profiled, and the evidence behind the call.
+- **Open questions** — what a human answer would materially change, and anything the analyses could not resolve.
+- **Assumptions** — in headless runs, every choice made without the user: which epic was selected (invocation or auto-detect), the `detect-epic --epic <N>` (or unflagged) result including any non-empty `pending_stories`, a machine **rejected** verdict forced by unfinished stories or rendered with no human decision, each proposed item. Omit in interactive runs — an interactive run records the same facts where the user confirmed them, in Epic summary.
+
+Do not state time estimates anywhere in the document.
+
+## Sprint-status update
+
+Do not hand-edit `sprint-status.yaml` — its comment blocks and quoting are exactly the write that most often corrupts the file. Use the bundled script, which round-trips through a comment-preserving YAML parser, force-quotes values so punctuation (a leading `#`, a colon) cannot break parsing, and validates the result — restoring the original file untouched if the write does not verify:
+
+```
+uv run --no-cache {skill-root}/scripts/sprint_status.py update \
+  --file "{implementation_artifacts}/sprint-status.yaml" \
+  --epic {{epic_number}} --set-retro-done \
+  --add-action '[{"action":"...","owner":"..."}, ...]' \
+  --ref "{implementation_artifacts}/epic-{{epic_number}}-retro-{date}.md" \
+  --verdict "<accepted | accepted-with-open-items | rejected>" \
+  --date "{date}"
+```
+
+Keep every value quoted. `--date` is parsed as `MM-DD-YYYY HH:MM` and nothing else — unpadded spellings like `1-2-2026 9:05` are accepted and normalized to the padded form, but a value that does not parse is rejected with `ok: false`, `restored: true` and exit 1, before the file is touched, and the whole update is a no-op. So pass `{date}` only if it is already in that form; otherwise reformat it, or omit the flag entirely and let the script stamp the current time itself. That format carries a space, which is why the flag must be quoted: unquoted, `--date 07-28-2026 14:23` splits into two argv words and dies at argparse (`{"ok": false, "error": "argument error: unrecognized arguments: 14:23"}`, exit 2). `--file` and `--ref` are quoted for the same reason — an `{implementation_artifacts}` path containing a space breaks them exactly the same way.
+
+It sets `development_status["epic-{{epic_number}}-retrospective"]` to `done`, appends one `action_items` entry per proposed item, and bumps `last_updated`. Each appended item carries `status: open`, a stable `id` (`epic-<N>-retro-item-<n>-<slug>` derived from the action text, or the `id` you supply in the JSON), and a `ref` back to this retro document (from `--ref`, or a per-item `ref` in the JSON) — so an orchestrator can dedupe items across re-runs and dispatch each one to its full, sourced finding. `--verdict` is not written into the file; it is echoed back in the result JSON as a signal for consumers. It accepts exactly the frontmatter vocabulary — `accepted`, `accepted-with-open-items`, `rejected` — and any other spelling is rejected (`ok: false`, `restored: true`, exit 1) before the file is touched. Read the JSON it returns:
+
+- `ok: true` → report the retro-key transition, `action_items_added`, `action_items_updated`, and the echoed `verdict`.
+- `ok: false` → the file was left untouched (`restored: true`); surface the error, do not hand-edit. `restored: false` means the rollback write also failed and the file may be incomplete — warn the user explicitly.
+- `restored` speaks only for a command that may have written. Every `update` failure carries it; `detect-epic` never emits it, because it never writes; and an invocation the parser itself rejects (`argument error: ...`, exit 2) carries neither the key nor a file to speak about. Read a missing `restored` as "nothing was at risk", never as `false`.
+- `retro_key_found: false` → the retro key was absent, so nothing was marked done; the document still saved, but tell the user sprint-status needs a manual retro entry.
+- `retro_key_found: null` → `--set-retro-done` was not passed, so the key was never looked for. Distinct from `false`, which is a real absence the user needs to be told about.
+
+Moving a *previous* epic's action items off `open` is recorded in the retro document either way. When the Phase 4 follow-through has evidence an item landed, or the user says one did, offer to update the sprint-status entries too and run `--set-action-status` with exactly what the user confirms — that flag is the only supported way to change a status; hand-editing never is. It can be passed in the same invocation as the update above, or run on its own:
+
+```
+uv run --no-cache {skill-root}/scripts/sprint_status.py update \
+  --file "{implementation_artifacts}/sprint-status.yaml" \
+  --epic {{epic_number}} \
+  --set-action-status '[{"id":"epic-1-retro-item-1-add-error-handling","status":"done"},{"epic":1,"action":"Exact action text","status":"in-progress"}]'
+```
+
+Rules:
+
+- Select an item by its `id`, or — for legacy entries written before ids existed — by `epic` plus the item's exact `action` text. An entry carrying both uses the `id`. Matching is exact: no trimming, no case folding, and `epic` must be a JSON integer. The `--epic` flag does not scope selectors; it only names the retro key and the epic recorded on appended items, so items from any epic are addressable in one call.
+- The only statuses are `open`, `in-progress`, and `done`. `bmad-sprint-planning`'s status view counts both `open` and `in-progress` as open action items, so only `done` retires an item from the surfaced list — moving something to `in-progress` records progress, it does not quiet the dashboard.
+- Every selector must resolve to exactly one item already in the file. Matching nothing, matching more than one, or colliding with another entry in the same array aborts the whole invocation — `ok: false`, `restored: true`, the file byte-identical and nothing partially applied. "Whole invocation" includes any `--set-retro-done` and `--add-action` passed in the same call: one mistyped selector drops the entire update, so re-run the full command after fixing it rather than assuming the retro key was set.
+- Items appended by `--add-action` in the same run are not addressable in that run; they are always written as `open`.
+- Only ever apply a status the user confirmed, and in a headless run do not pass this flag at all.
+- Success reports `action_items_updated`.
+
+Only ever apply a status the user confirmed: the evidence justifies proposing a transition, and only the user's confirmation justifies writing it. In a headless run do not use this flag at all — record the transitions you would have proposed in the Previous-retro follow-through section and leave the prior items' statuses alone.
+
+## Finish
+
+Report where the document was saved, the verdict, and the action-item count. Then, if `{workflow.on_complete}` is non-empty, follow it as the final terminal instruction before exiting.
diff --git a/.agents/skills/bmad-retrospective/references/team-discussion.md b/.agents/skills/bmad-retrospective/references/team-discussion.md
new file mode 100644
index 00000000..9e6c9816
--- /dev/null
+++ b/.agents/skills/bmad-retrospective/references/team-discussion.md
@@ -0,0 +1,22 @@
+# Team Discussion (opt-in)
+
+An optional discussion layer over Phase 2's findings, off by default. It exists for users who want the retrospective discussed from multiple perspectives, the way a team would. One rule: **the team discusses evidence, never invention.** Agents speak only to findings that carry source references. No agent may describe an event that did not happen or report a pattern the diff does not show.
+
+## When to run it
+
+Only when asked — "discuss it as a team," "run party mode," or similar. A default run never enters this phase.
+
+## How to run it
+
+Invoke **`bmad-party-mode`**, seeded with the consolidated Phase 2 findings and the epic context, so the installed agents react as real subagents with independent thinking rather than a scripted dialogue. Seed it with:
+
+- The findings, each with its source reference, grouped by the aggregate view or lens that produced it.
+- The improvements the evidence confirms — real gains, patterns that worked — so positive observations are grounded in fact.
+- The epic's acceptance criteria (or the profiled stand-in), so the discussion can weigh the verdict.
+- The previous epic's action items and whether they landed, when a prior retro exists, so accountability is grounded in fact.
+
+If `bmad-party-mode` is unavailable, a discussion the user asked for must not silently fail to happen. Run it inline over the same seed — take each perspective yourself, hold every perspective to sourced findings — and record in the retrospective document that the discussion ran inline rather than through the installed agents. Record it as the narrowing it is: one model playing every role loses the independent disagreement that surfaces missed findings. State that in the document rather than omitting it.
+
+Facilitate in `{communication_language}`, keep the user an active participant, and steer toward systemic understanding over blame — the point is which process or convention would have prevented a finding, not who wrote the line. Capture anything the discussion surfaces that the analyses missed; a genuinely new observation becomes a finding only once you can tie it to a source, otherwise it is a question for Phase 4, not a conclusion.
+
+The discussion does not replace Phase 4. Its output feeds the action items and the verdict; it does not render them.
diff --git a/.agents/skills/bmad-retrospective/scripts/git_evidence.py b/.agents/skills/bmad-retrospective/scripts/git_evidence.py
new file mode 100644
index 00000000..4ff81162
--- /dev/null
+++ b/.agents/skills/bmad-retrospective/scripts/git_evidence.py
@@ -0,0 +1,304 @@
+# /// script
+# requires-python = ">=3.10"
+# ///
+"""Measure git commit and file-change evidence over a revision range.
+
+Prints ONLY JSON to stdout. Errors are emitted as JSON to stdout with a
+non-zero exit code: 2 for invalid arguments (rejected before git runs),
+1 for git or I/O failures. This script only MEASURES — it never judges
+acceleration or violations. The model interprets the numbers.
+
+Two git passes. The first lists every commit in the range (merges included)
+and sums the per-file churn of the non-merge commits, which is what `files`
+reports. The second runs only when the range contains merges and measures
+those merges alone, reported separately as `merge_files` — never folded into
+`files`, because a merge's diff against its first parent restates the churn
+of the commits it merged in, which the first pass already counted.
+"""
+
+import argparse
+import json
+import os
+import re
+import subprocess
+import sys
+
+UNIT_SEP = "\x1f"
+# sha, space-separated parents (empty for a root commit), subject.
+LOG_FORMAT = f"--format=%H{UNIT_SEP}%P{UNIT_SEP}%s"
+
+
+def _emit(obj, code=0):
+    sys.stdout.write(json.dumps(obj))
+    sys.exit(code)
+
+
+class JsonArgumentParser(argparse.ArgumentParser):
+    """Emit argparse failures on the JSON-only stdout contract, not usage text.
+
+    The parser is constructed with ``add_help=False``. The override below covers
+    ``error()``, but ``-h`` never reaches it: the built-in help action calls
+    ``print_help()`` and ``exit(0)`` directly, which would put plain usage text
+    on stdout with a zero exit and break the JSON-only contract. Removing the
+    action instead of intercepting it routes ``-h`` through the already-tested
+    ``error()`` path as an ordinary unrecognized argument. The cost is that the
+    ``help=`` strings are unreachable from the CLI; the skill's references carry
+    the usage a human needs.
+    """
+
+    def error(self, message):
+        _emit({"ok": False, "error": f"argument error: {message}"}, 2)
+
+
+def _parse_numstat_line(line):
+    # numstat lines: "<added>\t<deleted>\t<path>"; binary files use "-".
+    parts = line.split("\t")
+    if len(parts) < 3:
+        return None
+    added_raw, deleted_raw, path = parts[0], parts[1], "\t".join(parts[2:])
+    added = None if added_raw == "-" else int(added_raw)
+    deleted = None if deleted_raw == "-" else int(deleted_raw)
+    return added, deleted, path
+
+
+def _git_log(repo, extra_args, rng):
+    """Run one `git log --numstat` pass over `rng` and return its stdout.
+
+    `core.quotePath=false` keeps non-ASCII paths as real UTF-8 strings instead
+    of octal escapes, and `--no-renames` makes a rename an honest delete + add
+    instead of an unopenable "src/{a => b}" pseudo-path that splits one file's
+    churn across several keys. Both matter for every pass, so both live here.
+
+    `log.diffMerges=separate` is pinned on the command line because it is what
+    `-m` means: a user or repo config setting it to `off` makes pass 2 emit no
+    file rows at all, so `merge_files` would come back empty beside a non-zero
+    `merges_measured` and read as "the merges changed nothing".
+    """
+    cmd = [
+        "git",
+        "-c",
+        "core.quotePath=false",
+        "-c",
+        "log.diffMerges=separate",
+        "-C",
+        repo,
+        "log",
+        "--numstat",
+        "--no-renames",
+        *extra_args,
+        LOG_FORMAT,
+        rng,
+        "--",  # terminate rev parsing so the range can never match a pathspec
+    ]
+    try:
+        # Decode explicitly: git emits UTF-8 path bytes regardless of the
+        # caller's locale, and a C locale would otherwise decode them as ASCII.
+        # surrogateescape, not replace: replace maps every invalid byte to the
+        # same U+FFFD, so two distinct non-UTF-8 paths would collapse into one
+        # `files` key with their churn silently summed. Lone surrogates survive
+        # json.dumps (escaped as \udcXX under ensure_ascii) and json.loads.
+        proc = subprocess.run(
+            cmd,
+            capture_output=True,
+            text=True,
+            encoding="utf-8",
+            errors="surrogateescape",
+            env={k: v for k, v in os.environ.items() if not k.startswith("GIT_")},
+        )
+    except Exception as exc:  # noqa: BLE001
+        _emit({"ok": False, "error": str(exc)}, 1)
+
+    if proc.returncode != 0:
+        # stderr can be empty (a signal kill, a quiet failure); the exit code is
+        # then the only thing left to report, so never emit an empty error.
+        _emit(
+            {
+                "ok": False,
+                "error": proc.stderr.strip() or f"git exited {proc.returncode}",
+            },
+            1,
+        )
+    return proc.stdout
+
+
+def _parse_log(output, stories):
+    """Turn one pass's log output into (commits, files_map). Shared by both."""
+    commits = []
+    files = {}  # path -> {path, _added, _deleted, binary_revisions, commit_count}
+    seen = set()
+    counting = True
+
+    for raw in output.splitlines():
+        if UNIT_SEP in raw:
+            sha, parents, subject = raw.split(UNIT_SEP, 2)
+            # Under -m, git repeats a merge's header once per parent unless it
+            # also honours --first-parent (git 2.31+). Count only the first
+            # block for a sha — git emits parents in order, so that block is
+            # the first-parent diff either way, and no churn is double counted.
+            counting = sha not in seen
+            if not counting:
+                continue
+            seen.add(sha)
+            commits.append(
+                {
+                    "sha": sha,
+                    "subject": subject,
+                    # Every id the subject names, in --stories order: a commit
+                    # spanning two stories belongs to both. Word-boundary match
+                    # so a story id like "1-2" does not also match "11-2".
+                    "stories": [
+                        sid
+                        for sid in stories
+                        if re.search(rf"\b{re.escape(sid)}\b", subject)
+                    ],
+                    "is_merge": len(parents.split()) > 1,
+                }
+            )
+            continue
+
+        if not counting or not raw.strip():
+            continue
+
+        parsed = _parse_numstat_line(raw)
+        if parsed is None:
+            continue
+        added, deleted, path = parsed
+
+        entry = files.get(path)
+        if entry is None:
+            # _added/_deleted are running sums over the path's text revisions.
+            entry = {
+                "path": path,
+                "_added": 0,
+                "_deleted": 0,
+                "binary_revisions": 0,
+                "commit_count": 0,
+            }
+            files[path] = entry
+
+        entry["commit_count"] += 1
+        if added is None or deleted is None:
+            # A binary revision is unmeasurable, not zero — count it alongside
+            # the sums instead of nulling the path's real measured churn.
+            entry["binary_revisions"] += 1
+        else:
+            entry["_added"] += added
+            entry["_deleted"] += deleted
+
+    return commits, files
+
+
+def _file_list(files):
+    return [
+        {
+            "path": entry["path"],
+            "added": entry["_added"],
+            "deleted": entry["_deleted"],
+            "net": entry["_added"] - entry["_deleted"],
+            "commit_count": entry["commit_count"],
+            "binary_revisions": entry["binary_revisions"],
+        }
+        for entry in files.values()
+    ]
+
+
+def main(argv=None):
+    parser = JsonArgumentParser(
+        description=(
+            "Measure commit and per-file change evidence over a git revision "
+            "range. Measures only; does not judge."
+        ),
+        add_help=False,
+    )
+    parser.add_argument("--repo", default=".", help="Path to the git repo (default: .)")
+    parser.add_argument("--range", dest="range", help="Revision range REV..REV")
+    parser.add_argument(
+        "--stories",
+        help="Comma-separated story ids to match against commit subjects.",
+    )
+    args = parser.parse_args(argv)
+
+    stories = []
+    if args.stories:
+        # dict.fromkeys dedupes while keeping the caller's order: a repeated id
+        # would otherwise land twice in a commit's `stories`, double counting
+        # that commit in any per-story total built from the output.
+        stories = list(
+            dict.fromkeys(s.strip() for s in args.stories.split(",") if s.strip())
+        )
+
+    if not args.range:
+        _emit(
+            {
+                "range": None,
+                "note": "no range supplied",
+                "commits": [],
+                "files": [],
+            }
+        )
+
+    # Accept only an explicit REV..REV range. Anything else silently measures
+    # the wrong thing: a leading "-" is consumed by git as an option, a single
+    # rev logs all history up to it, a bare pathspec logs by path, an empty
+    # endpoint ("..", "a..", "..b") makes git default that side to HEAD, and a
+    # three-dot "A...B" is a symmetric difference — a different commit set
+    # entirely. partition splits at the FIRST "..", so any of those extra-dot
+    # shapes leaves `right` empty or dot-prefixed.
+    left, _, right = args.range.partition("..")
+    if (
+        args.range != args.range.strip()
+        or args.range.startswith("-")
+        or not left
+        or not right
+        or right.startswith(".")
+    ):
+        _emit(
+            {
+                "ok": False,
+                "error": f"invalid --range {args.range!r}: expected a revision range like REV..REV",
+            },
+            2,
+        )
+
+    # Pass 1 — the listing. No extra args, so full topology: every commit in
+    # the range including merges, which is what per-story attribution reads.
+    # Merges contribute no numstat rows here, so `files` is non-merge churn.
+    commits, files = _parse_log(_git_log(args.repo, [], args.range), stories)
+    merge_count = sum(1 for commit in commits if commit["is_merge"])
+
+    # Pass 2 — merge churn, only when there is any. `-m --first-parent
+    # --min-parents=2` walks the range head's first-parent spine and emits
+    # exactly one diff-against-first-parent block per merge sitting on it.
+    # Merges off that spine are counted in merge_count and never measured,
+    # which is precisely why merges_measured is a separate key: the gap
+    # between the two is a visible statement that some merges went
+    # unmeasured. This never folds into `files` — a merge's first-parent diff
+    # restates the churn of the commits it merged in, which pass 1 already
+    # counted, so adding it in would double count.
+    merge_commits, merge_files = [], {}
+    if merge_count:
+        merge_commits, merge_files = _parse_log(
+            _git_log(
+                args.repo,
+                ["-m", "--first-parent", "--min-parents=2"],
+                args.range,
+            ),
+            stories,
+        )
+
+    _emit(
+        {
+            "range": args.range,
+            "commit_count": len(commits),
+            "merge_count": merge_count,
+            "merges_measured": len(merge_commits),
+            "commits": commits,
+            "files": _file_list(files),
+            "merge_files": _file_list(merge_files),
+            "stories_supplied": stories,
+        }
+    )
+
+
+if __name__ == "__main__":
+    main()
diff --git a/.agents/skills/bmad-retrospective/scripts/sprint_status.py b/.agents/skills/bmad-retrospective/scripts/sprint_status.py
new file mode 100644
index 00000000..f41171be
--- /dev/null
+++ b/.agents/skills/bmad-retrospective/scripts/sprint_status.py
@@ -0,0 +1,746 @@
+# /// script
+# requires-python = ">=3.10"
+# dependencies = ["ruamel.yaml>=0.18"]
+# ///
+"""Detect the current retrospective epic and surgically update sprint-status.yaml.
+
+Prints ONLY JSON to stdout. Errors are emitted as JSON to stdout with a non-zero
+exit code. The ``update`` subcommand round-trips the YAML to preserve all comments
+and formatting, writes atomically (temp file + ``os.replace``), and restores the
+original file bytes on any validation failure.
+"""
+
+import argparse
+import hashlib
+import io
+import json
+import os
+import re
+import stat
+import sys
+import tempfile
+from collections import Counter
+from collections.abc import Mapping
+from datetime import datetime
+
+from ruamel.yaml import YAML
+from ruamel.yaml.scalarstring import DoubleQuotedScalarString
+
+STORY_RE = re.compile(r"^(\d+)-\d+[a-z]?-")  # trailing [a-z]? matches split-story keys like 2-6a-...
+DATE_FORMAT = "%m-%d-%Y %H:%M"
+# The authoritative action-item vocabulary, mirrored from bmad-sprint-planning's
+# SKILL.md. Anything outside it would render as unknown in the status dashboard.
+ACTION_STATUSES = ("open", "in-progress", "done")
+# The retro-document frontmatter vocabulary. --verdict is only echoed back, but
+# orchestrators branch on the echo, so a free-spelled value ("accepted with open
+# items") would silently fall through every branch they write.
+VERDICTS = ("accepted", "accepted-with-open-items", "rejected")
+
+
+def _load_yaml(path):
+    yaml = YAML(typ="rt")
+    yaml.preserve_quotes = True
+    # Pin the emitter to the indentation the sprint-status template ships with.
+    # Without this, ruamel re-dumps block sequences at its own default offset and
+    # every write silently de-indents pre-existing, untouched action_items.
+    yaml.indent(mapping=2, sequence=4, offset=2)
+    # Pin the dump encoding too: `_dump_bytes` serializes into a BytesIO, so the
+    # emitter -- not this module -- encodes the bytes that land in the user's
+    # file. utf-8 is ruamel's current default, but the file is read back as
+    # utf-8 unconditionally, so state it rather than inherit it.
+    yaml.encoding = "utf-8"
+    with open(path, "r", encoding="utf-8") as fh:
+        data = yaml.load(fh)
+    return yaml, data
+
+
+def _emit(obj, code=0):
+    sys.stdout.write(json.dumps(obj))
+    sys.exit(code)
+
+
+def _emit_error(message, code=1, restored=None):
+    """Emit a failure on the JSON-only contract.
+
+    ``restored`` is included only when the caller can speak to the state of the
+    target file; ``retro-document.md`` teaches callers to read it, so a write-path
+    failure must never omit it and a read-only subcommand must never invent it.
+    """
+    payload = {"ok": False, "error": message}
+    if restored is not None:
+        payload["restored"] = restored
+    _emit(payload, code)
+
+
+class JsonArgumentParser(argparse.ArgumentParser):
+    """Emit argparse failures on the JSON-only stdout contract, not usage text.
+
+    Every parser built from this class is constructed with ``add_help=False``.
+    The override below covers ``error()``, but ``-h`` never reaches it: the
+    built-in help action calls ``print_help()`` and ``exit(0)`` directly, which
+    would put plain usage text on stdout with a zero exit and break the
+    JSON-only contract for the machine consumer this script exists to serve.
+    Removing the action instead of intercepting it keeps the fix to one keyword
+    per parser and routes ``-h`` through the already-tested ``error()`` path as
+    an ordinary unrecognized argument. The cost is that the ``help=`` strings
+    are unreachable from the CLI; the skill's references carry the usage a
+    human needs.
+    """
+
+    def error(self, message):
+        _emit({"ok": False, "error": f"argument error: {message}"}, 2)
+
+
+def _slugify(text, maxlen=40):
+    text = str(text)
+    # Unicode-aware: a non-Latin action must keep its own characters in the id
+    # rather than collapsing to a single placeholder shared by every item.
+    slug = re.sub(r"[^\w]+", "-", text.lower(), flags=re.UNICODE).strip("-")
+    slug = slug[:maxlen].strip("-")
+    if not slug:
+        # Nothing sluggable (punctuation/emoji only): a short content hash keeps
+        # the id deterministic and distinct instead of a bare "item".
+        slug = hashlib.sha256(text.encode("utf-8")).hexdigest()[:8]
+    return slug
+
+
+def _selector_label(entry):
+    """Human-readable form of a --set-action-status selector, for error text.
+
+    An entry carrying both forms is described by its ``id``, because that is the
+    form resolution actually uses.
+    """
+    if isinstance(entry.get("id"), str):
+        return f"id={entry['id']!r}"
+    return f"epic={entry.get('epic')!r} action={entry.get('action')!r}"
+
+
+def _match_action_items(entry, items):
+    """Indices in ``items`` that the selector ``entry`` resolves to.
+
+    ``id`` wins whenever it is present: the epic/action pair is the fallback for
+    legacy items written before ids existed, so a caller that copied a whole item
+    through gets the precise match rather than a text comparison. Matching is
+    exact equality -- no normalization -- so a file that spells its epic as a
+    string simply does not match and the caller gets a "no match" error instead of
+    a silent write to the wrong item. ``bool`` is excluded on the file side too,
+    since ``True == 1`` in Python.
+    """
+    item_id = entry.get("id")
+    if isinstance(item_id, str):
+        return [
+            idx
+            for idx, item in enumerate(items)
+            if isinstance(item, Mapping) and item.get("id") == item_id
+        ]
+    epic_value = entry.get("epic")
+    action_value = entry.get("action")
+    return [
+        idx
+        for idx, item in enumerate(items)
+        if isinstance(item, Mapping)
+        and not isinstance(item.get("epic"), bool)
+        and item.get("epic") == epic_value
+        and item.get("action") == action_value
+    ]
+
+
+def _comment_counts(text):
+    """Multiset of the comment lines in ``text``, indentation included.
+
+    Keyed by the whole line so that a re-indented comment counts as a loss too:
+    the guarantee callers are given is comments *and formatting*, and ruamel
+    re-emits comments at their original column even when the block around them
+    is re-indented, so an exact key costs nothing in practice.
+    """
+    return Counter(
+        line for line in text.splitlines() if line.lstrip().startswith("#")
+    )
+
+
+def _load_document(path, restored=None):
+    """Load and shape-check the document, reporting every failure as JSON.
+
+    Returns ``(yaml, data, dev)``. ``dev`` is the live ``development_status``
+    mapping when the key exists, otherwise a detached empty mapping -- the key is
+    never inserted into the document as a side effect of loading.
+    """
+    try:
+        yaml, data = _load_yaml(path)
+    except UnicodeDecodeError as exc:
+        _emit_error(f"{path} is not valid UTF-8: {exc}", 1, restored)
+    except OSError as exc:
+        _emit_error(str(exc), 1, restored)
+    except Exception as exc:  # noqa: BLE001 - report any parse error as JSON
+        _emit_error(str(exc), 1, restored)
+
+    if data is not None and not isinstance(data, Mapping):
+        _emit_error("root document is not a mapping", 1, restored)
+
+    dev = data.get("development_status") if data is not None else None
+    if dev is None:
+        dev = {}
+    elif not isinstance(dev, Mapping):
+        _emit_error("development_status is not a mapping", 1, restored)
+
+    return yaml, data, dev
+
+
+def _retro_status(dev, retro_key, restored=None):
+    status_value = dev.get(retro_key)
+    if status_value is not None and not isinstance(status_value, str):
+        _emit_error(
+            f"{retro_key} status must be a string or null",
+            1,
+            restored,
+        )
+    return status_value
+
+
+def _dump_bytes(yaml, data):
+    """Serialize the document to bytes before any file is touched, so a dump
+    failure cannot leave a partial file anywhere."""
+    buf = io.BytesIO()
+    yaml.dump(data, buf)
+    return buf.getvalue()
+
+
+def _atomic_write(path, payload, mode=None):
+    """Replace ``path``'s contents with ``payload`` atomically.
+
+    The bytes land in a temp file alongside the target, are fsynced, take the
+    target's permission bits (mkstemp creates 0600, which would silently narrow
+    the file), and only then rename over it -- so a kill or a full disk leaves
+    the original file intact rather than truncated. ``path`` is resolved through
+    symlinks first: renaming onto a symlink would detach the link and leave the
+    real file stale while reporting success. The directory is fsynced too --
+    best-effort, see below -- so the rename survives a power loss and not just
+    the bytes.
+    """
+    path = os.path.realpath(path)
+    directory = os.path.dirname(path) or "."
+    fd, tmp_path = tempfile.mkstemp(
+        prefix=".sprint-status-", suffix=".tmp", dir=directory
+    )
+    try:
+        with os.fdopen(fd, "wb") as fh:
+            fh.write(payload)
+            fh.flush()
+            os.fsync(fh.fileno())
+        if mode is not None:
+            os.chmod(tmp_path, mode)
+        os.replace(tmp_path, path)
+    except BaseException:
+        try:
+            os.unlink(tmp_path)
+        except OSError:
+            pass
+        raise
+    # The directory sync sits outside the try because once os.replace has
+    # returned, the new bytes ARE the file: a failure past that point must not
+    # propagate as a write failure, or the caller would report the original
+    # "restored" about a write that in fact landed. Skipping it only risks the
+    # rename not surviving a hard power loss.
+    try:
+        dir_fd = os.open(directory, os.O_RDONLY)
+        try:
+            os.fsync(dir_fd)
+        finally:
+            os.close(dir_fd)
+    except OSError:
+        pass
+
+
+def cmd_detect_epic(args):
+    # detect-epic never writes, so it reports no "restored" key.
+    _, _, dev = _load_document(args.file)
+
+    done_stories = []
+    max_epic = None
+    # Every story key with its epic, in document order, so the pending list can
+    # be scoped to whichever epic detection lands on without a second pass over
+    # the mapping. Non-story keys (epic-2, epic-2-retrospective, ...) never enter
+    # here, because STORY_RE does not match them.
+    story_keys = []
+    for key, value in dev.items():
+        m = STORY_RE.match(str(key))
+        if not m:
+            continue
+        epic_num = int(m.group(1))
+        story_keys.append((epic_num, key, value))
+        if value == "done":
+            done_stories.append(key)
+            if max_epic is None or epic_num > max_epic:
+                max_epic = epic_num
+
+    # Optional --epic aims the gate at a supplied number (the -H <epic> path)
+    # instead of auto-picking the highest epic with a done story. Without it,
+    # behavior is unchanged: detect, then scope pending_stories to that epic.
+    if args.epic is not None:
+        if args.epic < 1:
+            _emit_error(
+                f"invalid --epic {args.epic} (expected a positive integer)",
+                1,
+            )
+        selected = args.epic
+    else:
+        selected = max_epic
+
+    if selected is None:
+        # Uniform shape: pending_stories is always present, even with no epic to
+        # scope it to, so a caller can read it without branching on epic first.
+        _emit(
+            {
+                "epic": None,
+                "story_count": 0,
+                "done_stories": done_stories,
+                "pending_stories": [],
+                "retro_key": None,
+                "retro_status": None,
+            }
+        )
+
+    # Scoped to the selected epic only -- deliberately unlike done_stories, which
+    # spans the whole file. A pending story in some *other* epic is not this
+    # retrospective's business.
+    selected_keys = [
+        (key, value) for epic_num, key, value in story_keys if epic_num == selected
+    ]
+    pending_stories = [key for key, value in selected_keys if value != "done"]
+
+    retro_key = f"epic-{selected}-retrospective"
+    retro_status = _retro_status(dev, retro_key)
+    _emit(
+        {
+            "epic": selected,
+            # An epic the file has never heard of returns the same empty
+            # pending_stories as a finished one; story_count is the key that
+            # separates "complete" from "nonexistent" (a typo'd --epic), so the
+            # unfinished-story gate can refuse to read silence as done.
+            "story_count": len(selected_keys),
+            "done_stories": done_stories,
+            "pending_stories": pending_stories,
+            "retro_key": retro_key,
+            "retro_status": retro_status,
+        }
+    )
+
+
+def cmd_update(args):
+    # Every failure below happens before the write is attempted, so the file is
+    # untouched and "restored": true is the honest report.
+    untouched = True
+
+    # 0. Validate the inputs before anything is mutated or written.
+    if args.epic < 1:
+        _emit_error(
+            f"invalid --epic {args.epic} (expected a positive integer)", 1, untouched
+        )
+
+    if args.date is not None:
+        try:
+            parsed_date = datetime.strptime(args.date, DATE_FORMAT)
+        except (ValueError, TypeError):
+            _emit_error(
+                f'invalid --date {args.date!r} (expected "MM-DD-YYYY HH:MM")',
+                1,
+                untouched,
+            )
+        # Normalize: strptime also accepts unpadded spellings like
+        # "1-2-2026 9:05", and writing those through would defeat the point of
+        # validating the format at all.
+        last_updated = parsed_date.strftime(DATE_FORMAT)
+    else:
+        last_updated = datetime.now().strftime(DATE_FORMAT)
+
+    if args.verdict is not None and args.verdict not in VERDICTS:
+        _emit_error(
+            f"invalid --verdict {args.verdict!r} (allowed: {', '.join(VERDICTS)})",
+            1,
+            untouched,
+        )
+
+    actions = []
+    if args.add_action:
+        try:
+            actions = json.loads(args.add_action)
+        except json.JSONDecodeError as exc:
+            _emit_error(f"invalid --add-action JSON: {exc}", 1, untouched)
+        if not isinstance(actions, list):
+            _emit_error("--add-action must be a JSON array", 1, untouched)
+        for item in actions:
+            if not isinstance(item, dict):
+                _emit_error(
+                    "each --add-action item must be an object", 1, untouched
+                )
+            action_value = item.get("action")
+            if not isinstance(action_value, str) or not action_value.strip():
+                # A JSON null/number/object would otherwise be str()'d into a
+                # literal "None"/"{...}" and written as a real action item.
+                _emit_error(
+                    "each --add-action item must have a non-empty string action",
+                    1,
+                    untouched,
+                )
+
+    status_updates = []
+    if args.set_action_status:
+        try:
+            status_updates = json.loads(args.set_action_status)
+        except json.JSONDecodeError as exc:
+            _emit_error(f"invalid --set-action-status JSON: {exc}", 1, untouched)
+        if not isinstance(status_updates, list):
+            _emit_error("--set-action-status must be a JSON array", 1, untouched)
+        for entry in status_updates:
+            if not isinstance(entry, dict):
+                _emit_error(
+                    "each --set-action-status entry must be an object", 1, untouched
+                )
+            status_value = entry.get("status")
+            if not isinstance(status_value, str) or status_value not in ACTION_STATUSES:
+                _emit_error(
+                    f"invalid --set-action-status status {status_value!r} "
+                    f"(allowed: {', '.join(ACTION_STATUSES)})",
+                    1,
+                    untouched,
+                )
+            item_id = entry.get("id")
+            if item_id is not None:
+                # Present but unusable is an input error, not a silent fallback to
+                # the epic/action form -- the caller meant to select by id.
+                if not isinstance(item_id, str) or not item_id.strip():
+                    _emit_error(
+                        "each --set-action-status id must be a non-empty string",
+                        1,
+                        untouched,
+                    )
+                continue
+            epic_value = entry.get("epic")
+            action_value = entry.get("action")
+            if isinstance(epic_value, bool) or not isinstance(epic_value, int):
+                _emit_error(
+                    "each --set-action-status entry must have a non-empty string id, "
+                    "or an integer epic and a non-empty string action",
+                    1,
+                    untouched,
+                )
+            if not isinstance(action_value, str) or not action_value.strip():
+                _emit_error(
+                    "each --set-action-status entry must have a non-empty string id, "
+                    "or an integer epic and a non-empty string action",
+                    1,
+                    untouched,
+                )
+
+    # 1. Keep original bytes for restore-on-failure, and the mode to write back
+    #    with -- taken from the open handle so an unlink mid-run cannot leave the
+    #    replacement silently narrowed to mkstemp's 0600.
+    try:
+        with open(args.file, "rb") as fh:
+            original_bytes = fh.read()
+            original_mode = stat.S_IMODE(os.fstat(fh.fileno()).st_mode)
+    except OSError as exc:
+        _emit_error(str(exc), 1, untouched)
+
+    try:
+        original_text = original_bytes.decode("utf-8")
+    except UnicodeDecodeError as exc:
+        _emit_error(f"{args.file} is not valid UTF-8: {exc}", 1, untouched)
+
+    # Every comment line in the file, not just the leading block: the template
+    # ships one above action_items, and losing it corrupts the document just the
+    # same as losing the header.
+    original_comments = _comment_counts(original_text)
+
+    yaml, data, dev = _load_document(args.file, restored=untouched)
+
+    if data is None:
+        _emit_error("empty or invalid YAML document", 1, untouched)
+
+    epic = args.epic
+    retro_key = f"epic-{epic}-retrospective"
+
+    # null distinguishes "the flag was not passed" from "the key was absent",
+    # which is the only case retro-document.md assigns "false" to.
+    retro_key_found = None
+    retro_status_before = None
+    retro_status_after = None
+
+    # 2. Optionally set the retrospective status to done (only if key exists).
+    if args.set_retro_done:
+        retro_key_found = retro_key in dev
+        if retro_key_found:
+            retro_status_before = _retro_status(dev, retro_key, restored=untouched)
+            dev[retro_key] = "done"
+            retro_status_after = "done"
+
+    # 3. Take the action_items sequence as loaded and shape-check it once; both
+    #    of the steps below operate on this same list.
+    existing_actions = data.get("action_items")
+    if existing_actions is not None and not isinstance(existing_actions, list):
+        # A hand-corrupted file must still fail on the JSON contract, not crash.
+        _emit_error("action_items in file is not a list", 1, untouched)
+    items_added = 0
+    original_action_len = len(existing_actions) if existing_actions is not None else 0
+
+    # 4. Optionally transition the status of items already in the file. Selectors
+    #    resolve against action_items *as loaded* and strictly before the
+    #    --add-action append below, which is what makes an item appended in the
+    #    same invocation unaddressable in that run. Every selector is resolved
+    #    before any is applied, so a rejected batch never leaves a partial edit --
+    #    and since nothing has been written yet, the file is still untouched.
+    status_targets = []
+    if status_updates:
+        pool = existing_actions if isinstance(existing_actions, list) else []
+        claimed = {}
+        for entry in status_updates:
+            label = _selector_label(entry)
+            matches = _match_action_items(entry, pool)
+            if not matches:
+                _emit_error(f"no action item matches {label}", 1, untouched)
+            if len(matches) > 1:
+                _emit_error(
+                    f"ambiguous --set-action-status selector {label}: "
+                    f"{len(matches)} matches",
+                    1,
+                    untouched,
+                )
+            idx = matches[0]
+            if idx in claimed:
+                # Applying both would overcount action_items_updated, and a
+                # conflicting pair would surface as a confusing post-write
+                # validation failure instead of the input error it is.
+                _emit_error(
+                    f"duplicate --set-action-status targets: {label} and "
+                    f"{claimed[idx]} resolve to the same action item",
+                    1,
+                    untouched,
+                )
+            claimed[idx] = label
+            status_targets.append((idx, entry["status"]))
+
+        for idx, new_status in status_targets:
+            # A plain assignment keeps the item's own scalar style: ruamel's
+            # CommentedMap re-applies the existing key's style on overwrite, for
+            # every ScalarString subclass. Pinned by the style tests.
+            pool[idx]["status"] = new_status
+
+    # 5. Optionally append action items.
+    if actions:
+        seq = data.get("action_items")
+        if seq is None:
+            seq = []
+            data["action_items"] = seq
+
+        for item in actions:
+            # Stable identity for orchestrator consumers: an id that lets a
+            # re-run dedupe against prior items, and a ref back to the sourced
+            # finding in the retro document. Both accept an explicit override.
+            seq_num = len(seq) + 1
+            action_text = str(item.get("action", ""))
+            item_id = item.get("id") or (
+                f"epic-{int(epic)}-retro-item-{seq_num}-{_slugify(action_text)}"
+            )
+            ref = item.get("ref") or (args.ref or "")
+            entry = {
+                "id": DoubleQuotedScalarString(str(item_id)),
+                "epic": int(epic),
+                "action": DoubleQuotedScalarString(action_text),
+                "owner": DoubleQuotedScalarString(str(item.get("owner", ""))),
+                "status": "open",
+                "ref": DoubleQuotedScalarString(str(ref)),
+            }
+            seq.append(entry)
+            items_added += 1
+
+    # 6. Update last_updated.
+    data["last_updated"] = last_updated
+
+    # 7. Serialize, then swap the file atomically.
+    try:
+        _atomic_write(args.file, _dump_bytes(yaml, data), original_mode)
+    except Exception as exc:  # noqa: BLE001
+        # The target is only ever touched by the final rename, so if the write
+        # raised, the original is still on disk byte-for-byte. Calling _restore
+        # here would rewrite a file that was never modified -- the one write in
+        # the program with nothing to gain and a truncated file to lose.
+        _emit({"ok": False, "error": f"write failed: {exc}", "restored": True}, 1)
+
+    # 8. Validate the written file; restore on any failure.
+    def _fail(msg):
+        restored = _restore(args.file, original_bytes, original_mode)
+        _emit({"ok": False, "error": msg, "restored": restored}, 1)
+
+    try:
+        _, reloaded = _load_yaml(args.file)
+    except Exception as exc:  # noqa: BLE001
+        _fail(f"re-parse failed after write: {exc}")
+
+    if reloaded is None:
+        _fail("re-parse produced empty document after write")
+
+    if not isinstance(reloaded, Mapping):
+        _fail("re-parse produced a non-mapping document after write")
+
+    rdev = reloaded.get("development_status") or {}
+    if args.set_retro_done and retro_key_found:
+        if not isinstance(rdev, Mapping) or rdev.get(retro_key) != "done":
+            _fail(f"validation: {retro_key} not set to done after write")
+
+    new_action_len = 0
+    if reloaded.get("action_items") is not None:
+        new_action_len = len(reloaded.get("action_items"))
+    if new_action_len != original_action_len + items_added:
+        _fail(
+            "validation: action_items length mismatch "
+            f"(expected {original_action_len + items_added}, got {new_action_len})"
+        )
+
+    if status_targets:
+        # The recorded indices are still valid: the only other mutation to the
+        # sequence is an append, and the length check above just confirmed it.
+        reloaded_actions = reloaded.get("action_items")
+        if not isinstance(reloaded_actions, list):
+            _fail("validation: action_items is not a list after write")
+        for idx, new_status in status_targets:
+            reloaded_item = reloaded_actions[idx]
+            if (
+                not isinstance(reloaded_item, Mapping)
+                or reloaded_item.get("status") != new_status
+            ):
+                _fail(
+                    f"validation: action item at index {idx} is not "
+                    f"{new_status!r} after write"
+                )
+
+    try:
+        with open(args.file, "r", encoding="utf-8") as fh:
+            new_text = fh.read()
+    except (OSError, UnicodeDecodeError) as exc:
+        _fail(f"re-read failed after write: {exc}")
+
+    # Loss-only: a comment may legitimately move or be added (a long quoted value
+    # can wrap onto a line that begins with '#'), but none may disappear.
+    lost = original_comments - _comment_counts(new_text)
+    if lost:
+        first = next(
+            (line for line in original_text.splitlines() if line in lost), None
+        )
+        _fail(f"validation: comment line lost after write: {first!r}")
+
+    _emit(
+        {
+            "ok": True,
+            "retro_key_found": retro_key_found,
+            "retro_status_before": retro_status_before,
+            "retro_status_after": retro_status_after,
+            "action_items_added": items_added,
+            "action_items_updated": len(status_targets),
+            "last_updated": last_updated,
+            "verdict": args.verdict,
+        }
+    )
+
+
+def _restore(path, original_bytes, mode=None):
+    """Best-effort restore of the original bytes. Returns True on success so a
+    caller can surface a restore failure instead of hiding a half-written file.
+
+    Atomic for the same reason the primary write is: a truncating rewrite that
+    dies halfway destroys the very bytes it was trying to put back.
+    """
+    try:
+        _atomic_write(path, original_bytes, mode)
+        return True
+    except Exception as exc:  # noqa: BLE001 - best-effort restore
+        sys.stderr.write(f"restore failed: {exc}\n")
+        return False
+
+
+def build_parser():
+    parser = JsonArgumentParser(
+        description=(
+            "Detect the current retrospective epic and surgically update "
+            "sprint-status.yaml while preserving comments and formatting."
+        ),
+        add_help=False,
+    )
+    sub = parser.add_subparsers(dest="command", required=True)
+
+    p_detect = sub.add_parser(
+        "detect-epic",
+        help=(
+            "Find the highest epic with a done story and its retrospective "
+            "status, or aim the same pending_stories gate at --epic N."
+        ),
+        add_help=False,
+    )
+    p_detect.add_argument("--file", required=True, help="Path to sprint-status.yaml")
+    p_detect.add_argument(
+        "--epic",
+        type=int,
+        default=None,
+        help=(
+            "Optional. Scope the response to this epic number instead of "
+            "auto-detecting the highest epic with a done story. Orchestrators "
+            "passing -H <epic> should pass the same number here so pending_stories "
+            "covers the epic they are about to retro."
+        ),
+    )
+    p_detect.set_defaults(func=cmd_detect_epic)
+
+    p_update = sub.add_parser(
+        "update",
+        help="Surgically update retro status and/or action items.",
+        add_help=False,
+    )
+    p_update.add_argument("--file", required=True, help="Path to sprint-status.yaml")
+    p_update.add_argument("--epic", required=True, type=int, help="Epic number")
+    p_update.add_argument(
+        "--set-retro-done",
+        action="store_true",
+        help="Set epic-<N>-retrospective to done if the key exists.",
+    )
+    p_update.add_argument(
+        "--add-action",
+        help='JSON array of {"action":str,"owner":str,"id"?:str,"ref"?:str} to append.',
+    )
+    p_update.add_argument(
+        "--set-action-status",
+        help=(
+            "JSON array of status transitions for action items already in the file. "
+            'Select each by id -- {"id":str,"status":"open|in-progress|done"} -- or, '
+            'for legacy items with no id, by epic plus exact action text: '
+            '{"epic":int,"action":str,"status":...}. An entry carrying both uses the '
+            "id. Every selector must match exactly one item; any failure aborts the "
+            "whole invocation and leaves the file untouched."
+        ),
+    )
+    p_update.add_argument(
+        "--ref",
+        help="Reference (e.g. the retro document path) recorded on each appended action item.",
+    )
+    p_update.add_argument(
+        "--verdict",
+        help=(
+            "Acceptance verdict echoed back in the JSON result for orchestrator "
+            f"consumers. One of: {', '.join(VERDICTS)}."
+        ),
+    )
+    p_update.add_argument(
+        "--date",
+        help='Value for last_updated (default: now as "MM-DD-YYYY HH:MM").',
+    )
+    p_update.set_defaults(func=cmd_update)
+
+    return parser
+
+
+def main(argv=None):
+    parser = build_parser()
+    args = parser.parse_args(argv)
+    args.func(args)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/.agents/skills/bmad-retrospective/scripts/tests/fixtures/sprint-status-template.yaml b/.agents/skills/bmad-retrospective/scripts/tests/fixtures/sprint-status-template.yaml
new file mode 100644
index 00000000..cc4dd7cf
--- /dev/null
+++ b/.agents/skills/bmad-retrospective/scripts/tests/fixtures/sprint-status-template.yaml
@@ -0,0 +1,71 @@
+# Sprint Status Template
+# This is an EXAMPLE showing the expected format
+# The actual file will be generated with all epics/stories from your epic files
+
+# generated: {date}
+# project: {project_name}
+# project_key: {project_key}
+# tracking_system: {tracking_system}
+# story_location: {story_location}
+
+# STATUS DEFINITIONS:
+# ==================
+# Epic Status:
+#   - backlog: Epic not yet started
+#   - in-progress: Epic actively being worked on
+#   - done: All stories in epic completed
+#
+# Story Status:
+#   - backlog: Story only exists in epic file
+#   - ready-for-dev: Story file created, ready for development
+#   - in-progress: Developer actively working on implementation
+#   - review: Implementation complete, ready for review
+#   - done: Story completed
+#
+# Retrospective Status:
+#   - optional: Can be completed but not required
+#   - done: Retrospective has been completed
+#
+# Action Item Status:
+#   - open: Committed during a retrospective, not yet addressed
+#   - in-progress: Actively being worked on
+#   - done: Completed
+#
+# WORKFLOW NOTES:
+# ===============
+# - Epic transitions to 'in-progress' automatically when its first story starts (via build's sprint sync)
+# - Stories can be worked in parallel if team capacity allows
+# - Developer typically creates the next story after the previous one is 'done' to incorporate learnings
+# - Dev moves story to 'review', then runs code-review (fresh context, different LLM recommended)
+# - Retrospective appends its action items to action_items; the status view surfaces open ones
+
+# EXAMPLE STRUCTURE (your actual epics/stories will replace these):
+# Timestamps use MM-DD-YYYY HH:MM.
+
+generated: 05-06-2025 21:30
+last_updated: 05-06-2025 21:30
+project: My Awesome Project
+project_key: NOKEY
+tracking_system: file-system
+story_location: "docs/stories"
+
+development_status:
+  epic-1: backlog
+  1-1-user-authentication: done
+  1-2-account-management: ready-for-dev
+  1-3-plant-data-model: backlog
+  1-4-add-plant-manual: backlog
+  epic-1-retrospective: optional
+
+  epic-2: backlog
+  2-1-personality-system: backlog
+  2-2-chat-interface: backlog
+  2-3-llm-integration: backlog
+  epic-2-retrospective: optional
+
+# Action items committed during retrospectives (section created by the retrospective workflow)
+action_items:
+  - epic: 1
+    action: "Add error-handling review to the code review checklist"
+    owner: "Charlie"
+    status: open
diff --git a/.agents/skills/bmad-retrospective/scripts/tests/test_git_evidence.py b/.agents/skills/bmad-retrospective/scripts/tests/test_git_evidence.py
new file mode 100644
index 00000000..32c4b5f3
--- /dev/null
+++ b/.agents/skills/bmad-retrospective/scripts/tests/test_git_evidence.py
@@ -0,0 +1,750 @@
+# /// script
+# requires-python = ">=3.10"
+# dependencies = ["pytest>=8.0"]
+# ///
+"""Tests for git_evidence.py — measurement over a real temp git repo.
+
+Run: uv run scripts/tests/test_git_evidence.py
+ or: uv run --with pytest -m pytest scripts/tests/test_git_evidence.py
+"""
+
+import json
+import os
+import shutil
+import subprocess
+import sys
+import unicodedata
+from pathlib import Path
+
+import pytest
+
+SCRIPT = Path(__file__).resolve().parents[1] / "git_evidence.py"
+
+# The fixture commit identity, shared by both git helpers below.
+_IDENT = {
+    "GIT_AUTHOR_NAME": "T",
+    "GIT_AUTHOR_EMAIL": "t@t",
+    "GIT_COMMITTER_NAME": "T",
+    "GIT_COMMITTER_EMAIL": "t@t",
+}
+
+
+def _git_env(repo):
+    """The environment every fixture git runs under, layered outward.
+
+    Inherit the real environment (PATH above all: git lives in /opt/homebrew,
+    /usr/local, or a nix store as readily as /usr/bin, and an env holding only
+    GIT_* vars sends execvp to os.defpath), then strip every ambient GIT_* var
+    -- GIT_DIR, GIT_WORK_TREE and GIT_CONFIG_COUNT would each silently redirect
+    or reconfigure the fixture -- and pin identity plus every source git reads
+    for settings, so nothing on the developer's machine can reach the fixture:
+
+    - gitconfig (commit.gpgsign, core.autocrlf, core.hooksPath,
+      init.defaultBranch). GIT_CONFIG_NOSYSTEM/GIT_CONFIG_GLOBAL cover
+      git >= 2.32; HOME and XDG_CONFIG_HOME cover older git, and are set to the
+      repo's parent directory -- always a per-test directory under pytest's
+      tmp_path -- so nothing is ever planted inside the working tree.
+    - gitattributes, a separate source GIT_CONFIG_NOSYSTEM does not cover: a
+      system `* -diff` rule would make numstat call every path binary and take
+      the churn assertions down with it. GIT_ATTR_NOSYSTEM shuts it out.
+    - the locale. LC_ALL/LANG are pinned to C, matching _run/_proc's existing
+      pin, so fixture git's text output cannot vary with the developer's
+      locale. Inheriting the environment is what makes this pin necessary:
+      the old four-variable env had no locale in it to inherit.
+    """
+    env = {k: v for k, v in os.environ.items() if not k.startswith("GIT_")}
+    env.update(_IDENT)
+    env["GIT_CONFIG_NOSYSTEM"] = "1"
+    env["GIT_ATTR_NOSYSTEM"] = "1"
+    env["GIT_CONFIG_GLOBAL"] = os.devnull
+    env["HOME"] = env["XDG_CONFIG_HOME"] = str(Path(repo).parent)
+    env["LC_ALL"] = env["LANG"] = "C"
+    return env
+
+
+def _json(proc):
+    """Parse the JSON-only stdout contract, surfacing a crash instead of hiding
+    it behind a JSONDecodeError."""
+    assert proc.stdout, f"empty stdout; stderr was: {proc.stderr}"
+    assert "Traceback" not in proc.stderr, proc.stderr
+    return json.loads(proc.stdout)
+
+
+def _run(*args):
+    # LC_ALL=C keeps git's error strings in English so assertions on them
+    # are stable across locales.
+    proc = subprocess.run(
+        ["uv", "run", str(SCRIPT), *args],
+        capture_output=True,
+        text=True,
+        env={**os.environ, "LC_ALL": "C", "LANG": "C"},
+    )
+    return proc.returncode, _json(proc)
+
+
+def _proc(*args, env=None):
+    """Run the script and return the raw process, so a test can assert on the
+    exit code and stderr together — and so `env` can carry an overlay (a fake
+    `git` earlier on PATH) that `_run` has no way to pass."""
+    overlay = {"LC_ALL": "C", "LANG": "C"}
+    if env:
+        overlay.update(env)
+    return subprocess.run(
+        ["uv", "run", str(SCRIPT), *args],
+        capture_output=True,
+        text=True,
+        env={**os.environ, **overlay},
+    )
+
+
+def _git(repo, *args):
+    subprocess.run(
+        ["git", "-C", str(repo), *args],
+        check=True,
+        capture_output=True,
+        env=_git_env(repo),
+    )
+
+
+def _git_unchecked(repo, *args):
+    """`git` that tolerates a non-zero exit — the conflicting merge in
+    `_merge_repo` is supposed to fail, and the hand resolution comes after it.
+    Same environment as `_git`, which runs with check=True."""
+    return subprocess.run(
+        ["git", "-C", str(repo), *args],
+        capture_output=True,
+        text=True,
+        env=_git_env(repo),
+    )
+
+
+def _make_repo(tmp_path):
+    repo = tmp_path / "repo"
+    repo.mkdir()
+    _git(repo, "init", "-q")
+    (repo / "a.py").write_text("one\ntwo\n")
+    _git(repo, "add", "-A")
+    _git(repo, "commit", "-qm", "epic-1-1 initial a")
+    (repo / "a.py").write_text("one\ntwo\nthree\nfour\n")
+    (repo / "b.py").write_text("x\n")
+    _git(repo, "add", "-A")
+    _git(repo, "commit", "-qm", "epic-1-2 grow a, add b")
+    return repo
+
+
+def test_no_range_returns_empty(tmp_path):
+    repo = _make_repo(tmp_path)
+    code, out = _run("--repo", str(repo))
+    assert code == 0
+    assert out["range"] is None
+    assert out["commits"] == [] and out["files"] == []
+
+
+def test_measures_commits_and_files_with_attribution(tmp_path):
+    repo = _make_repo(tmp_path)
+    code, out = _run(
+        "--repo", str(repo), "--range", "HEAD~1..HEAD", "--stories", "1-2,1-1"
+    )
+    assert code == 0
+    assert out["range"] == "HEAD~1..HEAD"
+    assert out["commit_count"] == 1
+    # The single commit in range is the second one; attributed to story "1-2".
+    assert out["commits"][0]["stories"] == ["1-2"]
+    files = {f["path"]: f for f in out["files"]}
+    # a.py grew by two lines, b.py added one — measured, not judged.
+    assert files["a.py"]["added"] == 2 and files["a.py"]["net"] == 2
+    assert files["b.py"]["added"] == 1
+
+
+def test_story_attribution_respects_word_boundary(tmp_path):
+    # Story id "1-2" must NOT match a commit subject mentioning "11-2".
+    repo = tmp_path / "repo"
+    repo.mkdir()
+    _git(repo, "init", "-q")
+    (repo / "f.py").write_text("a\n")
+    _git(repo, "add", "-A")
+    _git(repo, "commit", "-qm", "base")
+    (repo / "f.py").write_text("a\nb\n")
+    _git(repo, "add", "-A")
+    _git(repo, "commit", "-qm", "epic-11-2 unrelated story")
+    code, out = _run("--repo", str(repo), "--range", "HEAD~1..HEAD", "--stories", "1-2")
+    assert code == 0
+    assert out["commits"][0]["stories"] == []
+
+
+def test_bad_range_errors_as_json(tmp_path):
+    repo = _make_repo(tmp_path)
+    code, out = _run("--repo", str(repo), "--range", "nope..alsonope")
+    assert code == 1
+    assert out["ok"] is False and out["error"]
+
+
+def test_single_rev_range_rejected(tmp_path):
+    # A single rev is not a range: git would log ALL history up to it and the
+    # script would report the whole repo as the epic's evidence.
+    repo = _make_repo(tmp_path)
+    code, out = _run("--repo", str(repo), "--range", "HEAD")
+    assert code == 2
+    assert out["ok"] is False and "invalid --range" in out["error"]
+
+
+def test_pathspec_range_rejected(tmp_path):
+    # A path that exists must not be silently consumed as a pathspec.
+    repo = _make_repo(tmp_path)
+    code, out = _run("--repo", str(repo), "--range", "a.py")
+    assert code == 2
+    assert out["ok"] is False and "invalid --range" in out["error"]
+
+
+def test_range_shaped_pathspec_forced_to_rev_parse(tmp_path):
+    # A committed file literally named "a..b" passes the REV..REV shape check;
+    # without the trailing "--" in the git argv, git silently logs that FILE's
+    # history with exit 0. The "--" forces rev interpretation, so this must
+    # error instead of measuring the decoy.
+    repo = _make_repo(tmp_path)
+    (repo / "a..b").write_text("decoy\n")
+    _git(repo, "add", "-A")
+    _git(repo, "commit", "-qm", "add decoy file named like a range")
+    code, out = _run("--repo", str(repo), "--range", "a..b")
+    assert code == 1
+    assert out["ok"] is False and "bad revision" in out["error"]
+
+
+def test_option_like_range_rejected(tmp_path):
+    # A range starting with "-" must never reach git, where it would be
+    # consumed as an option (e.g. --output=... writes an arbitrary file).
+    repo = _make_repo(tmp_path)
+    code, out = _run("--repo", str(repo), "--range=--output=evil.txt")
+    assert code == 2
+    assert out["ok"] is False and "invalid --range" in out["error"]
+    assert not (repo / "evil.txt").exists()
+
+
+def test_degenerate_range_shapes_rejected(tmp_path):
+    # Shapes that contain ".." but are not REV..REV: git would silently
+    # default an empty endpoint to HEAD ("..", "a..", "..HEAD"), a leading
+    # dash must never reach git even when dots are present ("-3..HEAD"),
+    # unstripped values must not slip past the dash guard, and a three-dot
+    # range is a symmetric difference — git would measure commits reachable
+    # from either endpoint but not both, a different evidence set entirely.
+    repo = _make_repo(tmp_path)
+    for bad in (
+        "..",
+        "a..",
+        "..HEAD",
+        "-3..HEAD",
+        " HEAD~1..HEAD",
+        "HEAD~1...HEAD",
+        "a...b",
+    ):
+        code, out = _run("--repo", str(repo), f"--range={bad}")
+        assert code == 2, f"accepted {bad!r}"
+        assert out["ok"] is False and "invalid --range" in out["error"], bad
+
+
+def test_malformed_args_emit_json_not_usage(tmp_path):
+    # An unknown flag must still land on the JSON contract, not argparse's
+    # plain usage text on stderr.
+    code, out = _run("--bogus-flag")
+    assert code != 0
+    assert out["ok"] is False and out["error"]
+
+
+def test_help_flags_emit_json_not_usage():
+    # argparse's built-in help action bypasses the error() override entirely --
+    # it prints usage text on stdout and exits 0, which breaks the JSON-only
+    # contract for a machine consumer. add_help=False demotes -h to an ordinary
+    # unrecognized argument, which error() already handles.
+    for flag in ("-h", "--help"):
+        proc = _proc(flag)
+        # Exit 2 specifically: the module docstring reserves 2 for argument
+        # errors and 1 for git/I-O failures, so collapsing them must fail here.
+        assert proc.returncode == 2, flag
+        assert "usage:" not in proc.stdout, flag
+        out = _json(proc)
+        assert out["ok"] is False and out["error"], flag
+
+
+# --- helpers for the fixtures below -----------------------------------------
+
+
+def _rev(repo, ref):
+    return _git_unchecked(repo, "rev-parse", ref).stdout.strip()
+
+
+def _fake_git(tmp_path, body):
+    """Write a `git` shim and return the PATH overlay that puts it ahead of the
+    real binary for the script's own subprocesses (never for the fixtures,
+    which build their repos through `_git`'s own environment)."""
+    bindir = tmp_path / "fakebin"
+    bindir.mkdir()
+    shim = bindir / "git"
+    shim.write_text(body)
+    os.chmod(shim, 0o755)
+    return {"PATH": f"{bindir}{os.pathsep}{os.environ['PATH']}"}
+
+
+def _merge_repo(tmp_path):
+    """Two story branches merged into the mainline; the second merge conflicts
+    and is resolved by hand, adding a line neither branch had. Returns
+    (repo, base_sha) — `base_sha..HEAD` is the epic range."""
+    repo = tmp_path / "repo"
+    repo.mkdir()
+    _git(repo, "init", "-q", "-b", "main")
+    (repo / "s.py").write_text("l1\nl2\nl3\n")
+    _git(repo, "add", "-A")
+    _git(repo, "commit", "-qm", "base")
+    base = _rev(repo, "HEAD")
+
+    _git(repo, "checkout", "-q", "-b", "s1")
+    (repo / "s.py").write_text("l1\nA\nl3\n")
+    _git(repo, "commit", "-qam", "epic-1-2 story one")
+    _git(repo, "checkout", "-q", "main")
+    _git(repo, "merge", "-q", "--no-ff", "s1", "-m", "merge story 1-2")
+
+    _git(repo, "checkout", "-q", "-b", "s2", base)
+    (repo / "s.py").write_text("l1\nB\nl3\n")
+    _git(repo, "commit", "-qam", "epic-1-3 story two")
+    _git(repo, "checkout", "-q", "main")
+    conflicted = _git_unchecked(repo, "merge", "--no-ff", "s2", "-m", "merge story 1-3")
+    assert conflicted.returncode != 0, "fixture expected a merge conflict"
+    (repo / "s.py").write_text("l1\nAB\nl3\nl4\n")  # hand resolution
+    _git(repo, "add", "-A")
+    _git(repo, "commit", "-qm", "merge story 1-3")
+    return repo, base
+
+
+def test_rename_yields_two_openable_paths(tmp_path):
+    # git's default rename detection emits "src/{mod.py => renamed.py}" — an
+    # unopenable pseudo-path that also splits one file's churn across keys.
+    # --no-renames makes the rename an honest delete + add.
+    repo = tmp_path / "repo"
+    repo.mkdir()
+    _git(repo, "init", "-q", "-b", "main")
+    (repo / "src").mkdir()
+    (repo / "src" / "mod.py").write_text("l1\nl2\nl3\nl4\nl5\nl6\n")
+    _git(repo, "add", "-A")
+    _git(repo, "commit", "-qm", "base")
+    base = _rev(repo, "HEAD")
+    _git(repo, "mv", "src/mod.py", "src/renamed.py")
+    _git(repo, "commit", "-qm", "rename mod")
+    (repo / "src" / "renamed.py").write_text("l1\nl2\nl3\nl4\nl5\nl6\nl7\n")
+    _git(repo, "commit", "-qam", "add a line after the rename")
+
+    out = _json(_proc("--repo", str(repo), "--range", f"{base}..HEAD"))
+    files = {f["path"]: f for f in out["files"]}
+    assert not any("=>" in path for path in files), sorted(files)
+    assert files["src/mod.py"]["added"] == 0
+    assert files["src/mod.py"]["deleted"] == 6
+    assert files["src/renamed.py"]["added"] == 7
+    assert files["src/renamed.py"]["deleted"] == 0
+
+
+def _accented_repo(tmp_path):
+    repo = tmp_path / "repo"
+    repo.mkdir()
+    _git(repo, "init", "-q", "-b", "main")
+    (repo / "src").mkdir()
+    (repo / "src" / "café.py").write_text("ca\n")
+    _git(repo, "add", "-A")
+    _git(repo, "commit", "-qm", "base")
+    base = _rev(repo, "HEAD")
+    (repo / "src" / "café.py").write_text("ca\ncb\n")
+    _git(repo, "commit", "-qam", "touch the accented file")
+    return repo, base
+
+
+def _assert_accented_path(repo, out):
+    paths = [f["path"] for f in out["files"]]
+    assert len(paths) == 1
+    assert "\\" not in paths[0] and '"' not in paths[0], paths
+    assert unicodedata.normalize("NFC", paths[0]) == "src/café.py"
+    # The reported path is a real path: it opens under --repo.
+    assert (repo / paths[0]).read_text() == "ca\ncb\n"
+
+
+def test_non_ascii_path_is_a_real_string(tmp_path):
+    # Without core.quotePath=false git emits "src/caf\303\251.py" — quoted and
+    # octal-escaped, so the documented "open the ranked files" step cannot.
+    repo, base = _accented_repo(tmp_path)
+    _assert_accented_path(
+        repo, _json(_proc("--repo", str(repo), "--range", f"{base}..HEAD"))
+    )
+
+
+def test_non_ascii_path_survives_a_non_utf8_locale(tmp_path):
+    # Pins the explicit encoding="utf-8" on the subprocess. Modern CPython's
+    # UTF-8 mode hides its absence even under LC_ALL=C, so the pin only bites
+    # with UTF-8 mode and C-locale coercion both off — where the interpreter
+    # default is US-ASCII and git's UTF-8 path bytes fail to decode, taking the
+    # whole measurement down with them.
+    repo, base = _accented_repo(tmp_path)
+    out = _json(
+        _proc(
+            "--repo",
+            str(repo),
+            "--range",
+            f"{base}..HEAD",
+            env={"PYTHONUTF8": "0", "PYTHONCOERCECLOCALE": "0"},
+        )
+    )
+    _assert_accented_path(repo, out)
+
+
+def test_merge_churn_is_measured_and_counted(tmp_path):
+    repo, base = _merge_repo(tmp_path)
+    out = _json(_proc("--repo", str(repo), "--range", f"{base}..HEAD"))
+    assert out["commit_count"] == 4
+    assert out["merge_count"] == 2
+    assert out["merges_measured"] == 2
+    # Both merges' first-parent churn: 1/1 for the clean merge, 2/1 for the
+    # hand-resolved one (the resolution added a line neither branch had).
+    merge_files = {f["path"]: f for f in out["merge_files"]}
+    assert merge_files["s.py"]["added"] == 3
+    assert merge_files["s.py"]["deleted"] == 2
+    assert merge_files["s.py"]["net"] == 1
+    assert merge_files["s.py"]["commit_count"] == 2
+
+
+def test_merge_commits_listed_but_excluded_from_files(tmp_path):
+    repo, base = _merge_repo(tmp_path)
+    out = _json(_proc("--repo", str(repo), "--range", f"{base}..HEAD"))
+    by_subject = {c["subject"]: c for c in out["commits"]}
+    assert by_subject["merge story 1-2"]["is_merge"] is True
+    assert by_subject["merge story 1-3"]["is_merge"] is True
+    assert by_subject["epic-1-2 story one"]["is_merge"] is False
+    # `files` is the two story commits only — 1/1 each. Folding the merges in
+    # would double count: their diff restates the churn they merged.
+    files = {f["path"]: f for f in out["files"]}
+    assert files["s.py"]["added"] == 2
+    assert files["s.py"]["deleted"] == 2
+    assert files["s.py"]["commit_count"] == 2
+
+
+def test_story_attribution_survives_merges(tmp_path):
+    repo, base = _merge_repo(tmp_path)
+    out = _json(
+        _proc(
+            "--repo", str(repo), "--range", f"{base}..HEAD", "--stories", "1-2,1-3"
+        )
+    )
+    by_subject = {c["subject"]: c for c in out["commits"]}
+    assert by_subject["epic-1-2 story one"]["stories"] == ["1-2"]
+    assert by_subject["epic-1-3 story two"]["stories"] == ["1-3"]
+
+
+def test_off_spine_merge_is_counted_but_not_measured(tmp_path):
+    # A back-merge of the mainline into a story branch is a merge in the range,
+    # but its first-parent diff would restate unrelated mainline content as
+    # epic churn. It stays in merge_count and out of merges_measured, so the
+    # gap between the two says plainly that a merge went unmeasured.
+    repo = tmp_path / "repo"
+    repo.mkdir()
+    _git(repo, "init", "-q", "-b", "main")
+    (repo / "f.txt").write_text("base\n")
+    _git(repo, "add", "-A")
+    _git(repo, "commit", "-qm", "base")
+    base = _rev(repo, "HEAD")
+    (repo / "f.txt").write_text("base\nmain1\n")
+    _git(repo, "commit", "-qam", "mainline work")
+
+    _git(repo, "checkout", "-q", "-b", "s1", base)
+    (repo / "s1.txt").write_text("s1\n")
+    _git(repo, "add", "-A")
+    _git(repo, "commit", "-qm", "epic-1-2 story one")
+    _git(repo, "merge", "-q", "--no-ff", "main", "-m", "back-merge main into s1")
+    _git(repo, "checkout", "-q", "main")
+    _git(repo, "merge", "-q", "--no-ff", "s1", "-m", "merge story 1-2")
+
+    out = _json(_proc("--repo", str(repo), "--range", f"{base}..HEAD"))
+    assert out["merge_count"] == 2
+    assert out["merges_measured"] == 1
+    merge_files = {f["path"]: f for f in out["merge_files"]}
+    assert set(merge_files) == {"s1.txt"}
+
+
+def test_merge_pass_survives_a_hostile_log_diffmerges_config(tmp_path):
+    # `-m` means "whatever log.diffMerges says", so a user or repo config of
+    # `off` makes the merge pass emit no file rows: merge_files comes back
+    # empty beside a non-zero merges_measured and reads as "the merges changed
+    # nothing". The command-line -c pin beats the config.
+    repo, base = _merge_repo(tmp_path)
+    _git(repo, "config", "log.diffMerges", "off")
+    out = _json(_proc("--repo", str(repo), "--range", f"{base}..HEAD"))
+    assert out["merges_measured"] == 2
+    merge_files = {f["path"]: f for f in out["merge_files"]}
+    assert merge_files["s.py"]["added"] == 3
+    assert merge_files["s.py"]["deleted"] == 2
+
+
+def test_distinct_non_utf8_paths_stay_distinct(tmp_path):
+    # errors="replace" maps every invalid byte to the same U+FFFD, collapsing
+    # two different files into one `files` key with their churn summed —
+    # measurement corruption with nothing in the output admitting to it.
+    overlay = _fake_git(
+        tmp_path,
+        "#!/bin/sh\n"
+        "printf 'aaaa\\037\\037subj\\n\\n1\\t0\\tsrc/caf\\351.py\\n"
+        "2\\t0\\tsrc/caf\\377.py\\n'\n",
+    )
+    out = _json(_proc("--repo", str(tmp_path), "--range", "a..b", env=overlay))
+    files = {f["path"]: f for f in out["files"]}
+    assert len(files) == 2, files
+    assert sorted(f["added"] for f in files.values()) == [1, 2]
+
+
+def test_repeated_merge_headers_are_counted_once(tmp_path):
+    # Pins the dedupe guard in _parse_log. git before 2.31 does not honour
+    # --first-parent for `-m`'s diff format, so a merge's header repeats once
+    # per parent with a diff block under each. Without the guard that doubles
+    # the merge churn and pushes merges_measured above merge_count, inverting
+    # the invariant the reference documents. Inert on modern git, so it needs
+    # pre-2.31-shaped output to be exercised at all.
+    real_git = shutil.which("git")
+    assert real_git, "git must be on PATH"
+    repo, base = _merge_repo(tmp_path)
+    overlay = _fake_git(
+        tmp_path,
+        "#!/bin/sh\n"
+        'for a in "$@"; do\n'
+        '  if [ "$a" = "--min-parents=2" ]; then\n'
+        "    printf 'aaaa\\037p1 p2\\037merge story 1-3\\n\\n2\\t1\\ts.py\\n"
+        "aaaa\\037p1 p2\\037merge story 1-3\\n\\n5\\t4\\ts.py\\n'\n"
+        "    exit 0\n"
+        "  fi\n"
+        "done\n"
+        f'exec "{real_git}" "$@"\n',
+    )
+    out = _json(_proc("--repo", str(repo), "--range", f"{base}..HEAD", env=overlay))
+    # One merge sha, however many blocks git printed for it.
+    assert out["merges_measured"] == 1
+    assert out["merges_measured"] <= out["merge_count"]
+    merge_files = {f["path"]: f for f in out["merge_files"]}
+    # The first block is the first-parent diff on every git version; the
+    # repeat must not be added on top of it.
+    assert merge_files["s.py"]["added"] == 2
+    assert merge_files["s.py"]["deleted"] == 1
+    assert merge_files["s.py"]["commit_count"] == 1
+
+
+def test_valid_range_with_no_commits_keeps_the_full_shape(tmp_path):
+    # A mis-specified epic range is a valid, empty range. It must still answer
+    # with every documented key rather than a differently-shaped stub.
+    repo = _make_repo(tmp_path)
+    proc = _proc("--repo", str(repo), "--range", "HEAD..HEAD")
+    out = _json(proc)
+    assert proc.returncode == 0
+    assert set(out) == {
+        "range",
+        "commit_count",
+        "merge_count",
+        "merges_measured",
+        "commits",
+        "files",
+        "merge_files",
+        "stories_supplied",
+    }
+    assert out["commit_count"] == 0
+    assert out["merge_count"] == 0
+    assert out["merges_measured"] == 0
+    assert out["commits"] == [] and out["files"] == [] and out["merge_files"] == []
+
+
+def test_linear_history_reports_no_merges(tmp_path):
+    repo = _make_repo(tmp_path)
+    out = _json(_proc("--repo", str(repo), "--range", "HEAD~1..HEAD"))
+    assert out["merge_count"] == 0
+    assert out["merges_measured"] == 0
+    assert out["merge_files"] == []
+
+
+def _recording_git(tmp_path, calls):
+    """A `git` shim that records each invocation as one \\x1f-delimited record
+    (one field per argument, so argument boundaries survive) and then execs the
+    real git."""
+    real_git = shutil.which("git")
+    assert real_git, "git must be on PATH"
+    return _fake_git(
+        tmp_path,
+        "#!/bin/sh\n"
+        f"( printf '%s\\037' \"$@\"; printf '\\n' ) >> \"{calls}\"\n"
+        f'exec "{real_git}" "$@"\n',
+    )
+
+
+def _numstat_invocations(calls):
+    """The recorded `git log --numstat` invocations, each as an argument list."""
+    out = []
+    for line in calls.read_text().splitlines():
+        argv = [field for field in line.split("\x1f") if field]
+        if "--numstat" in argv:
+            out.append(argv)
+    return out
+
+
+def test_second_pass_runs_only_when_the_range_has_merges(tmp_path):
+    # The merge pass is skipped outright on linear history, so the common case
+    # still costs exactly one `git log` — and the two passes must differ in
+    # exactly the arguments the design depends on.
+    calls = tmp_path / "calls.log"
+    overlay = _recording_git(tmp_path, calls)
+    merge_args = {"-m", "--first-parent", "--min-parents=2"}
+
+    (tmp_path / "linear").mkdir()
+    (tmp_path / "merged").mkdir()
+
+    linear = _make_repo(tmp_path / "linear")
+    _json(_proc("--repo", str(linear), "--range", "HEAD~1..HEAD", env=overlay))
+    logs = _numstat_invocations(calls)
+    assert len(logs) == 1, logs
+
+    calls.write_text("")
+    merged, base = _merge_repo(tmp_path / "merged")
+    _json(_proc("--repo", str(merged), "--range", f"{base}..HEAD", env=overlay))
+    logs = _numstat_invocations(calls)
+    assert len(logs) == 2, logs
+    # Pass 1 keeps full topology: none of the merge-pass arguments may reach it,
+    # or the story-branch commits drop out of the listing and attribution dies.
+    assert merge_args.isdisjoint(logs[0]), logs[0]
+    assert merge_args.issubset(logs[1]), logs[1]
+
+
+def test_multi_story_subject_attributes_to_every_match(tmp_path):
+    # First-match-wins silently dropped the second story from the attribution.
+    repo = tmp_path / "repo"
+    repo.mkdir()
+    _git(repo, "init", "-q", "-b", "main")
+    (repo / "f.py").write_text("a\n")
+    _git(repo, "add", "-A")
+    _git(repo, "commit", "-qm", "base")
+    (repo / "f.py").write_text("a\nb\n")
+    _git(repo, "commit", "-qam", "fix seam between 1-2 and 1-3")
+
+    out = _json(
+        _proc(
+            "--repo", str(repo), "--range", "HEAD~1..HEAD", "--stories", "1-3,1-2"
+        )
+    )
+    # Every match, in --stories order — not whichever id was passed first.
+    assert out["commits"][0]["stories"] == ["1-3", "1-2"]
+
+    # A repeated id must not list the commit twice: any per-story total built
+    # from `stories` would count it twice.
+    repeated = _json(
+        _proc(
+            "--repo", str(repo), "--range", "HEAD~1..HEAD", "--stories", "1-3,1-2,1-3"
+        )
+    )
+    assert repeated["commits"][0]["stories"] == ["1-3", "1-2"]
+    assert repeated["stories_supplied"] == ["1-3", "1-2"]
+
+
+def test_subject_naming_no_story_gets_an_empty_list(tmp_path):
+    repo = tmp_path / "repo"
+    repo.mkdir()
+    _git(repo, "init", "-q", "-b", "main")
+    (repo / "f.py").write_text("a\n")
+    _git(repo, "add", "-A")
+    _git(repo, "commit", "-qm", "base")
+    (repo / "f.py").write_text("a\nb\n")
+    _git(repo, "commit", "-qam", "chore: tidy imports")
+
+    out = _json(
+        _proc(
+            "--repo", str(repo), "--range", "HEAD~1..HEAD", "--stories", "1-2,1-3"
+        )
+    )
+    assert out["commits"][0]["stories"] == []
+
+
+def test_git_failure_with_empty_stderr_reports_the_exit_code(tmp_path):
+    # A quiet git failure (signal kill, empty stderr) must not leave the caller
+    # with `"error": ""` and nothing to report.
+    overlay = _fake_git(tmp_path, "#!/bin/sh\nexit 3\n")
+    proc = _proc("--repo", str(tmp_path), "--range", "HEAD~1..HEAD", env=overlay)
+    out = _json(proc)
+    assert proc.returncode == 1
+    assert out["ok"] is False
+    assert out["error"] == "git exited 3"
+
+
+def _binary_repo(tmp_path):
+    """A binary-only path plus a path that is binary twice and text once."""
+    repo = tmp_path / "repo"
+    repo.mkdir()
+    _git(repo, "init", "-q", "-b", "main")
+    (repo / "keep.txt").write_text("keep\n")
+    _git(repo, "add", "-A")
+    _git(repo, "commit", "-qm", "base")
+    base = _rev(repo, "HEAD")
+
+    (repo / "src").mkdir()
+    (repo / "src" / "x.py").write_bytes(b"\x00\x01bin\n")
+    (repo / "blob.bin").write_bytes(b"\x00\x01blob\n")
+    _git(repo, "add", "-A")
+    _git(repo, "commit", "-qm", "add binary content")
+    (repo / "src" / "x.py").write_text("one\ntwo\n")  # binary -> text: still binary
+    _git(repo, "commit", "-qam", "x.py becomes text")
+    (repo / "src" / "x.py").write_text("one\ntwo\nthree\n")  # text -> text: measured
+    _git(repo, "commit", "-qam", "grow x.py")
+    (repo / "blob.bin").write_bytes(b"\x00\x02blob\n")
+    _git(repo, "commit", "-qam", "churn the blob")
+    return repo, base
+
+
+def test_binary_revisions_no_longer_erase_measured_text_churn(tmp_path):
+    repo, base = _binary_repo(tmp_path)
+    out = _json(_proc("--repo", str(repo), "--range", f"{base}..HEAD"))
+    x = {f["path"]: f for f in out["files"]}["src/x.py"]
+    assert x["added"] == 1 and x["deleted"] == 0 and x["net"] == 1
+    assert x["binary_revisions"] == 2
+    assert x["commit_count"] == 3
+
+
+def test_binary_only_path_reports_zero_sums_and_its_revision_count(tmp_path):
+    repo, base = _binary_repo(tmp_path)
+    out = _json(_proc("--repo", str(repo), "--range", f"{base}..HEAD"))
+    blob = {f["path"]: f for f in out["files"]}["blob.bin"]
+    assert blob["added"] == 0 and blob["deleted"] == 0 and blob["net"] == 0
+    assert blob["binary_revisions"] == 2
+    assert blob["commit_count"] == 2
+
+
+def test_success_shape_carries_every_documented_key(tmp_path):
+    repo = _make_repo(tmp_path)
+    out = _json(_proc("--repo", str(repo), "--range", "HEAD~1..HEAD"))
+    assert set(out) >= {
+        "range",
+        "commit_count",
+        "merge_count",
+        "merges_measured",
+        "commits",
+        "files",
+        "merge_files",
+        "stories_supplied",
+    }
+    assert set(out["commits"][0]) == {"sha", "subject", "stories", "is_merge"}
+    assert set(out["files"][0]) == {
+        "path",
+        "added",
+        "deleted",
+        "net",
+        "commit_count",
+        "binary_revisions",
+    }
+
+
+def test_explicit_repo_ignores_ambient_git_dir(tmp_path):
+    repo = _make_repo(tmp_path)
+    proc = _proc(
+        "--repo",
+        str(repo),
+        "--range",
+        "HEAD~1..HEAD",
+        env={"GIT_DIR": str(tmp_path / "wrong-git-dir")},
+    )
+    assert _json(proc)["commit_count"] == 1
+
+
+if __name__ == "__main__":
+    sys.exit(pytest.main([__file__, "-q"]))
diff --git a/.agents/skills/bmad-retrospective/scripts/tests/test_sprint_status.py b/.agents/skills/bmad-retrospective/scripts/tests/test_sprint_status.py
new file mode 100644
index 00000000..590215f5
--- /dev/null
+++ b/.agents/skills/bmad-retrospective/scripts/tests/test_sprint_status.py
@@ -0,0 +1,1579 @@
+# /// script
+# requires-python = ">=3.10"
+# dependencies = ["pytest>=8.0", "ruamel.yaml>=0.18"]
+# ///
+"""Corruption-critical tests for sprint-status.py.
+
+Each test runs the script as a subprocess via ``uv run`` against a temp copy of
+an inline fixture, then re-reads the file to assert comments and formatting
+survive and punctuation-heavy action values round-trip intact.
+
+Run: uv run scripts/tests/test_sprint_status.py
+ or: uv run --with pytest --with ruamel.yaml -m pytest scripts/tests/test_sprint_status.py
+"""
+
+import importlib.util
+import json
+import os
+import re
+import stat
+import subprocess
+import sys
+from pathlib import Path
+
+import pytest
+from ruamel.yaml import YAML
+
+SCRIPT = Path(__file__).resolve().parents[1] / "sprint_status.py"
+# Vendored copy of bmad-sprint-planning's sprint-status-template.yaml — skills
+# must not path into each other's directories (PATH-05). The repo-level
+# test/test-template-sync.js keeps this fixture identical to the source.
+TEMPLATE = Path(__file__).resolve().parent / "fixtures" / "sprint-status-template.yaml"
+
+FIXTURE = """\
+# Sprint Status Tracking
+# STATUS DEFINITIONS:
+#   backlog        - not yet started
+#   ready-for-dev  - ready to be implemented
+#   done           - completed
+generated: "01-01-2026 09:00"
+last_updated: "01-01-2026 09:00"
+project: "Demo Project"
+project_key: "DEMO"
+tracking_system: "file"
+story_location: "docs/stories"
+development_status:
+  epic-1: backlog
+  1-1-user-authentication: done
+  1-2-account-management: done
+  epic-1-retrospective: optional
+  epic-2: backlog
+  2-1-dashboard: backlog
+"""
+
+
+# Two epics' worth of items, because the flag's headline use is epic N's retro
+# closing epic N-1's items: nothing may scope a selector to --epic. The two
+# "Scripted item" entries share their action text and differ only by epic, so the
+# legacy epic+action selector has to discriminate on the epic to resolve at all.
+ACTION_FIXTURE = """\
+# Sprint Status Tracking
+# STATUS DEFINITIONS:
+#   open           - committed during a retrospective, not yet addressed
+generated: "01-01-2026 09:00"
+last_updated: "01-01-2026 09:00"
+development_status:
+  1-1-a: done
+  epic-1-retrospective: optional
+  2-1-b: done
+  epic-2-retrospective: optional
+
+# Action items committed during retrospectives
+action_items:
+  - id: "epic-1-retro-item-1-x"
+    epic: 1
+    action: "Scripted item"
+    owner: "Amelia"
+    status: "open"
+    ref: "docs/epic-1-retro.md"
+  - epic: 1
+    action: "Pre-existing item"
+    owner: "Charlie"
+    status: open
+  - id: "epic-2-retro-item-1-y"
+    epic: 2
+    action: "Scripted item"
+    owner: "Dana"
+    status: "in-progress"
+    ref: "docs/epic-2-retro.md"
+"""
+
+# Three spellings of the same scalar, to pin that a status write never re-styles
+# the line it lands on.
+STYLE_FIXTURE = """\
+last_updated: "01-01-2026 09:00"
+development_status:
+  1-1-a: done
+  epic-1-retrospective: optional
+action_items:
+  - id: "double"
+    epic: 1
+    action: "Double"
+    status: "open"
+  - id: "single"
+    epic: 1
+    action: "Single"
+    status: 'open'
+  - id: "plain"
+    epic: 1
+    action: "Plain"
+    status: open
+"""
+
+
+def _run(args):
+    cmd = ["uv", "run", str(SCRIPT), *args]
+    # LC_ALL=C keeps os.strerror text stable so error-string assertions do not
+    # depend on the developer's locale.
+    return subprocess.run(
+        cmd, capture_output=True, text=True, env={**os.environ, "LC_ALL": "C"}
+    )
+
+
+def _module():
+    """Import the script as a module, for the few properties that cannot be
+    triggered through the CLI (a failure after the temp file already exists)."""
+    spec = importlib.util.spec_from_file_location("sprint_status", SCRIPT)
+    mod = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(mod)
+    return mod
+
+
+def _write_fixture(tmp_path):
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(FIXTURE, encoding="utf-8")
+    return target
+
+
+def _write_action_fixture(tmp_path):
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(ACTION_FIXTURE, encoding="utf-8")
+    return target
+
+
+def _load(path):
+    yaml = YAML(typ="rt")
+    with open(path, "r", encoding="utf-8") as fh:
+        return yaml.load(fh)
+
+
+def _json(proc):
+    """Parse the JSON-only stdout contract, surfacing a crash instead of hiding
+    it behind a JSONDecodeError."""
+    assert proc.stdout, f"empty stdout; stderr was: {proc.stderr}"
+    assert "Traceback" not in proc.stderr, proc.stderr
+    return json.loads(proc.stdout)
+
+
+def test_detect_epic(tmp_path):
+    target = _write_fixture(tmp_path)
+    proc = _run(["detect-epic", "--file", str(target)])
+    assert proc.returncode == 0, proc.stderr
+    out = json.loads(proc.stdout)
+    assert out["epic"] == 1
+    assert out["story_count"] == 2
+    assert out["retro_key"] == "epic-1-retrospective"
+    assert out["retro_status"] == "optional"
+    assert set(out["done_stories"]) == {
+        "1-1-user-authentication",
+        "1-2-account-management",
+    }
+
+
+def test_detect_epic_rejects_typed_retrospective_status_as_json(tmp_path):
+    fixture = (
+        "development_status:\n"
+        "  1-1-a: done\n"
+        "  epic-1-retrospective: 2026-01-01\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+
+    proc = _run(["detect-epic", "--file", str(target)])
+
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["error"] == "epic-1-retrospective status must be a string or null"
+    assert "restored" not in out
+    assert target.read_text(encoding="utf-8") == fixture
+
+
+# --- pending_stories: the unfinished-epic gate --------------------------------
+
+
+def test_pending_stories_lists_the_selected_epics_unfinished_keys(tmp_path):
+    # The gate the skill branches on before Phase 1: an epic whose highest done
+    # story selected it, but which is not actually finished. Document order, so
+    # the listing the user confirms matches the file they can open.
+    fixture = (
+        "development_status:\n"
+        "  epic-2: backlog\n"
+        "  2-1-a: done\n"
+        "  2-2-b: backlog\n"
+        "  2-3-c: ready-for-dev\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(["detect-epic", "--file", str(target)])
+    assert proc.returncode == 0, proc.stderr
+    out = _json(proc)
+    assert out["epic"] == 2
+    assert out["pending_stories"] == ["2-2-b", "2-3-c"]
+    # A non-story key sitting beside them never leaks in: STORY_RE gates entry.
+    assert "epic-2" not in out["pending_stories"]
+
+
+def test_pending_stories_is_empty_for_a_complete_epic(tmp_path):
+    fixture = (
+        "development_status:\n"
+        "  2-1-a: done\n"
+        "  2-2-b: done\n"
+        "  epic-2-retrospective: optional\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(["detect-epic", "--file", str(target)])
+    assert proc.returncode == 0, proc.stderr
+    out = _json(proc)
+    assert out["epic"] == 2
+    assert out["pending_stories"] == []
+
+
+def test_pending_stories_ignores_other_epics(tmp_path):
+    # FIXTURE detects epic 1 while 2-1-dashboard sits at backlog. The key is
+    # scoped to the *selected* epic -- unlike done_stories, which spans the whole
+    # file -- so another epic's unfinished work must never block this retro.
+    target = _write_fixture(tmp_path)
+    proc = _run(["detect-epic", "--file", str(target)])
+    assert proc.returncode == 0, proc.stderr
+    out = _json(proc)
+    assert out["epic"] == 1
+    assert out["pending_stories"] == []
+    # done_stories keeps its whole-file scope, unchanged.
+    assert set(out["done_stories"]) == {
+        "1-1-user-authentication",
+        "1-2-account-management",
+    }
+
+
+def test_pending_stories_present_when_no_epic_is_detected(tmp_path):
+    # No done story anywhere: the shape stays uniform so a caller can read
+    # pending_stories without first branching on epic.
+    fixture = (
+        "development_status:\n"
+        "  1-1-a: backlog\n"
+        "  2-1-b: ready-for-dev\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(["detect-epic", "--file", str(target)])
+    assert proc.returncode == 0, proc.stderr
+    out = _json(proc)
+    assert out["epic"] is None
+    assert out["story_count"] == 0
+    assert out["pending_stories"] == []
+
+
+def test_detect_epic_flag_aims_pending_stories_at_a_supplied_epic(tmp_path):
+    # Auto-detect would pick epic 2 (highest with a done story). --epic 1 aims
+    # the unfinished-epic gate at the orchestrator's explicit choice instead —
+    # the -H <epic> path that previously had no pending_stories at all.
+    fixture = (
+        "development_status:\n"
+        "  1-1-a: done\n"
+        "  1-2-b: backlog\n"
+        "  1-3-c: review\n"
+        "  2-1-a: done\n"
+        "  2-2-b: done\n"
+        "  epic-1-retrospective: optional\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(["detect-epic", "--file", str(target), "--epic", "1"])
+    assert proc.returncode == 0, proc.stderr
+    out = _json(proc)
+    assert out["epic"] == 1
+    assert out["retro_key"] == "epic-1-retrospective"
+    assert out["retro_status"] == "optional"
+    assert out["pending_stories"] == ["1-2-b", "1-3-c"]
+    # done_stories keeps its whole-file scope.
+    assert set(out["done_stories"]) == {"1-1-a", "2-1-a", "2-2-b"}
+
+
+def test_detect_epic_flag_lists_pending_when_no_story_is_done(tmp_path):
+    # Without --epic, no done story means epic is null. With --epic, an
+    # unfinished epic that never landed a done story is still addressable —
+    # every story key of that epic is pending.
+    fixture = (
+        "development_status:\n"
+        "  3-1-a: backlog\n"
+        "  3-2-b: ready-for-dev\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(["detect-epic", "--file", str(target), "--epic", "3"])
+    assert proc.returncode == 0, proc.stderr
+    out = _json(proc)
+    assert out["epic"] == 3
+    assert out["pending_stories"] == ["3-1-a", "3-2-b"]
+    assert out["retro_key"] == "epic-3-retrospective"
+    assert out["retro_status"] is None
+
+
+def test_detect_epic_flag_empty_pending_for_a_complete_supplied_epic(tmp_path):
+    fixture = (
+        "development_status:\n"
+        "  1-1-a: done\n"
+        "  1-2-b: done\n"
+        "  2-1-a: backlog\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(["detect-epic", "--file", str(target), "--epic", "1"])
+    assert proc.returncode == 0, proc.stderr
+    out = _json(proc)
+    assert out["epic"] == 1
+    assert out["pending_stories"] == []
+
+
+def test_detect_epic_flag_zero_story_count_marks_a_nonexistent_epic(tmp_path):
+    # --epic 9 against a file that has no epic-9 stories: pending_stories is
+    # empty exactly as it is for a finished epic, so story_count is the only
+    # signal separating "complete" from "typo'd". The gate reads 0 as suspect,
+    # never as done.
+    fixture = (
+        "development_status:\n"
+        "  1-1-a: done\n"
+        "  1-2-b: done\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(["detect-epic", "--file", str(target), "--epic", "9"])
+    assert proc.returncode == 0, proc.stderr
+    out = _json(proc)
+    assert out["epic"] == 9
+    assert out["story_count"] == 0
+    assert out["pending_stories"] == []
+    assert out["retro_status"] is None
+    # The finished epic it could be confused with reports its real count.
+    proc = _run(["detect-epic", "--file", str(target), "--epic", "1"])
+    out = _json(proc)
+    assert out["story_count"] == 2
+    assert out["pending_stories"] == []
+
+
+def test_detect_epic_flag_rejects_non_positive_epic(tmp_path):
+    target = _write_fixture(tmp_path)
+    for bad in ("0", "-3"):
+        proc = _run(["detect-epic", "--file", str(target), "--epic", bad])
+        assert proc.returncode == 1, proc.stderr
+        out = _json(proc)
+        assert out["ok"] is False
+        assert "epic" in out["error"]
+        assert "restored" not in out
+
+
+def test_update_rejects_non_positive_epic(tmp_path):
+    target = _write_fixture(tmp_path)
+    for bad in ("0", "-3"):
+        proc = _run(["update", "--file", str(target), "--epic", bad])
+        assert proc.returncode == 1, proc.stderr
+        assert "positive integer" in _json(proc)["error"]
+        assert target.read_text(encoding="utf-8") == FIXTURE
+
+
+# --- The JSON-only contract covers the help paths ----------------------------
+
+
+@pytest.mark.parametrize(
+    "args",
+    [["-h"], ["--help"], ["detect-epic", "-h"], ["detect-epic", "--help"]],
+    ids=["top-short", "top-long", "sub-short", "sub-long"],
+)
+def test_help_flags_emit_json_not_usage(args):
+    # argparse's built-in help action bypasses error() -- it prints usage text
+    # to stdout and exits 0, which is exactly the contract this script sells.
+    # add_help=False turns -h into an ordinary unrecognized argument instead.
+    proc = _run(args)
+    # Exit 2 specifically: the module docstring reserves 2 for argument errors
+    # and 1 for I/O failures, and retro-document.md teaches callers to tell the
+    # two apart, so collapsing them must fail here.
+    assert proc.returncode == 2
+    assert "usage:" not in proc.stdout
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["error"]
+    # An argparse rejection speaks for no file, so it carries no "restored" --
+    # the same rule test_only_the_write_path_reports_restored pins for exit 1.
+    assert "restored" not in out
+
+
+@pytest.mark.parametrize("flag", ["-h", "--help"])
+def test_update_help_flag_emits_json_not_usage(tmp_path, flag):
+    # The update subparser too, driven with its required arguments present so
+    # nothing but the help flag itself can be what argparse objects to.
+    target = _write_fixture(tmp_path)
+    proc = _run(["update", "--file", str(target), "--epic", "1", flag])
+    assert proc.returncode == 2
+    assert "usage:" not in proc.stdout
+    out = _json(proc)
+    assert out["ok"] is False
+    assert flag in out["error"]
+    assert "restored" not in out
+    # A rejected invocation must not have written anything.
+    assert target.read_text(encoding="utf-8") == FIXTURE
+
+
+def test_update_sets_retro_and_appends_action(tmp_path):
+    target = _write_fixture(tmp_path)
+    payload = '[{"action":"Fix #42: colons: and # hashes","owner":"Amelia"}]'
+    proc = _run(
+        [
+            "update",
+            "--file",
+            str(target),
+            "--epic",
+            "1",
+            "--set-retro-done",
+            "--add-action",
+            payload,
+        ]
+    )
+    assert proc.returncode == 0, proc.stderr
+    out = json.loads(proc.stdout)
+    assert out["ok"] is True
+    assert out["retro_key_found"] is True
+    assert out["retro_status_after"] == "done"
+    assert out["action_items_added"] == 1
+
+    # File must still parse cleanly (punctuation did not corrupt it).
+    data = _load(target)
+    assert data is not None
+
+    # STATUS DEFINITIONS comment survived.
+    raw = target.read_text(encoding="utf-8")
+    assert "STATUS DEFINITIONS" in raw
+
+    # Retro status flipped to done.
+    assert data["development_status"]["epic-1-retrospective"] == "done"
+
+    # The action value round-trips with literal '#' and ':' intact.
+    action = data["action_items"][0]
+    assert action["action"] == "Fix #42: colons: and # hashes"
+    assert action["owner"] == "Amelia"
+    assert action["epic"] == 1
+    assert action["status"] == "open"
+
+
+def test_update_rejects_typed_retrospective_status_before_writing(tmp_path):
+    fixture = (
+        "last_updated: 01-01-2026 09:00\n"
+        "development_status:\n"
+        "  1-1-a: done\n"
+        "  epic-1-retrospective: 2026-01-01\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+
+    proc = _run(
+        [
+            "update",
+            "--file",
+            str(target),
+            "--epic",
+            "1",
+            "--set-retro-done",
+            "--add-action",
+            '[{"action":"Must not be appended","owner":"Amelia"}]',
+        ]
+    )
+
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert out["error"] == "epic-1-retrospective status must be a string or null"
+    assert target.read_text(encoding="utf-8") == fixture
+
+
+def test_detect_epic_matches_split_story_keys(tmp_path):
+    # A split-story key like 2-6a-... is first-class in BMAD (an oversized story
+    # split into 2-6a / 2-6b) and must not be invisible to detection — otherwise
+    # an epic whose only done stories are splits is silently skipped.
+    fixture = (
+        "development_status:\n"
+        "  1-1-first: done\n"
+        "  2-6a-split-auth: done\n"
+        "  epic-2-retrospective: optional\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(["detect-epic", "--file", str(target)])
+    assert proc.returncode == 0, proc.stderr
+    out = json.loads(proc.stdout)
+    assert out["epic"] == 2
+    assert "2-6a-split-auth" in out["done_stories"]
+    assert out["retro_key"] == "epic-2-retrospective"
+
+
+def test_update_rejects_non_list_action_items(tmp_path):
+    # A hand-corrupted action_items must fail on the JSON contract, not crash.
+    fixture = (
+        "development_status:\n"
+        "  1-1-a: done\n"
+        "  epic-1-retrospective: optional\n"
+        'action_items: "oops-not-a-list"\n'
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1",
+         "--add-action", '[{"action":"x","owner":"y"}]']
+    )
+    assert proc.returncode == 1
+    out = json.loads(proc.stdout)  # must be JSON, not a traceback
+    assert out["ok"] is False
+    assert "action_items" in out["error"]
+
+
+def test_appended_items_carry_id_and_ref(tmp_path):
+    target = _write_fixture(tmp_path)
+    ref = "docs/stories/epic-1-retro-2026-07-21.md"
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1", "--set-retro-done",
+         "--add-action", '[{"action":"Fix the seam","owner":"Amelia"}]',
+         "--ref", ref, "--verdict", "accepted-with-open-items"]
+    )
+    assert proc.returncode == 0, proc.stderr
+    out = json.loads(proc.stdout)
+    assert out["verdict"] == "accepted-with-open-items"  # echoed, not written to a key
+
+    item = _load(target)["action_items"][0]
+    assert item["id"].startswith("epic-1-retro-item-1-")
+    assert item["ref"] == ref
+    # The retro key value stays "done" — verdict is not encoded into it.
+    assert _load(target)["development_status"]["epic-1-retrospective"] == "done"
+
+
+def test_free_spelled_verdict_is_rejected_before_the_file_is_touched(tmp_path):
+    # The SKILL's prose verdict ("accepted with open items") and the frontmatter
+    # token (accepted-with-open-items) used to be two spellings of one value; an
+    # orchestrator branching on the echo would fall through both. Only the
+    # frontmatter vocabulary passes; anything else fails with the file intact.
+    target = _write_fixture(tmp_path)
+    for bad in ("accepted with open items", "ship it", "ACCEPTED"):
+        proc = _run(
+            ["update", "--file", str(target), "--epic", "1", "--set-retro-done",
+             "--verdict", bad]
+        )
+        assert proc.returncode == 1, f"accepted {bad!r}"
+        out = _json(proc)
+        assert out["ok"] is False and "--verdict" in out["error"]
+        assert out["restored"] is True
+    assert target.read_text(encoding="utf-8") == FIXTURE
+
+
+def test_explicit_item_id_is_preserved(tmp_path):
+    target = _write_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1",
+         "--add-action", '[{"action":"a","owner":"o","id":"custom-id-7"}]']
+    )
+    assert proc.returncode == 0, proc.stderr
+    assert _load(target)["action_items"][0]["id"] == "custom-id-7"
+
+
+@pytest.mark.skipif(
+    hasattr(os, "geteuid") and os.geteuid() == 0,
+    reason="root bypasses file permission bits",
+)
+def test_write_failure_reports_restore_status(tmp_path):
+    # If the write cannot happen, the caller must be told whether the original
+    # was restored — a silent failure defeats the script's core guarantee.
+    #
+    # The write is atomic (temp file + os.replace), and os.replace needs write
+    # permission on the *directory*, not on the target — a read-only target is
+    # now replaceable. Making the containing directory read-only is what blocks
+    # the write: mkstemp fails, while the restore write to the still-writable
+    # target succeeds.
+    holder = tmp_path / "holder"
+    holder.mkdir()
+    target = _write_fixture(holder)
+    os.chmod(holder, 0o555)
+    try:
+        proc = _run(
+            ["update", "--file", str(target), "--epic", "1", "--set-retro-done"]
+        )
+    finally:
+        os.chmod(holder, 0o755)
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    # The file was never touched: the temp file could not even be created.
+    assert target.read_text(encoding="utf-8") == FIXTURE
+    assert [p.name for p in holder.iterdir()] == ["sprint-status.yaml"]
+
+
+def test_punctuation_does_not_corrupt_file(tmp_path):
+    # Explicit re-parse guarantee for YAML-breaking punctuation.
+    target = _write_fixture(tmp_path)
+    payload = '[{"action":"weird: value # with: hashes","owner":"Bob # Smith"}]'
+    proc = _run(
+        [
+            "update",
+            "--file",
+            str(target),
+            "--epic",
+            "1",
+            "--add-action",
+            payload,
+        ]
+    )
+    assert proc.returncode == 0, proc.stderr
+    # Re-parse must succeed and preserve the literal punctuation.
+    data = _load(target)
+    assert data["action_items"][0]["action"] == "weird: value # with: hashes"
+    assert data["action_items"][0]["owner"] == "Bob # Smith"
+
+
+# --- Formatting fidelity -----------------------------------------------------
+
+
+def test_template_round_trip_changes_only_last_updated(tmp_path):
+    # The repo's own sprint-status template is the shape every generated file
+    # inherits: 2-space sequence indent and a mid-file comment above
+    # action_items. An update must touch nothing but last_updated — a re-indent
+    # of a pre-existing, untouched entry defeats the preservation guarantee that
+    # motivates "do not hand-edit this file".
+    source = TEMPLATE.read_text(encoding="utf-8")
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(source, encoding="utf-8")
+
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1", "--date", "01-01-2026 09:00"]
+    )
+    assert proc.returncode == 0, proc.stderr
+    assert _json(proc)["ok"] is True
+
+    before = source.splitlines()
+    after = target.read_text(encoding="utf-8").splitlines()
+    assert len(before) == len(after)
+    changed = [(b, a) for b, a in zip(before, after) if b != a]
+    assert len(changed) == 1, changed
+    assert changed[0][1] == "last_updated: 01-01-2026 09:00"
+    # The pre-existing action item keeps its 2-space sequence indent.
+    assert "  - epic: 1" in after
+
+
+def test_mid_file_comment_survives_update(tmp_path):
+    fixture = (
+        "# header\n"
+        "development_status:\n"
+        "  1-1-a: done\n"
+        "  epic-1-retrospective: optional\n"
+        "\n"
+        "# Action items committed during retrospectives\n"
+        "action_items:\n"
+        "  - epic: 1\n"
+        '    action: "Pre-existing item"\n'
+        '    owner: "Charlie"\n'
+        "    status: open\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1", "--set-retro-done",
+         "--add-action", '[{"action":"New item","owner":"Amelia"}]']
+    )
+    assert proc.returncode == 0, proc.stderr
+    raw = target.read_text(encoding="utf-8")
+    assert "# Action items committed during retrospectives" in raw
+    assert "# header" in raw
+    assert '    action: "Pre-existing item"' in raw
+
+
+def test_legacy_offset_zero_file_is_canonicalized(tmp_path):
+    # Files the previous version of this script wrote carry action_items at
+    # column 0. The indent pin re-indents them to the template's shape on the
+    # next write. That is a deliberate one-time canonicalization, not a silent
+    # failure: the update still succeeds and no comment is lost.
+    fixture = (
+        "# header\n"
+        "development_status:\n"
+        "  1-1-a: done\n"
+        "  epic-1-retrospective: optional\n"
+        "action_items:\n"
+        '- id: "legacy"\n'
+        "  epic: 1\n"
+        '  action: "written by the old code"\n'
+        "  status: open\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(["update", "--file", str(target), "--epic", "1", "--set-retro-done"])
+    assert proc.returncode == 0, proc.stderr
+    raw = target.read_text(encoding="utf-8")
+    assert '  - id: "legacy"' in raw
+    assert '    action: "written by the old code"' in raw
+    assert "# header" in raw
+
+
+def test_lost_comment_fails_with_restore(tmp_path):
+    # A standalone comment inside a flow collection is genuinely dropped by the
+    # round-trip. The leading-block check never saw it; the full multiset does,
+    # and the original bytes must come back.
+    fixture = (
+        "# header\n"
+        "development_status:\n"
+        "  1-1-a: done\n"
+        "  epic-1-retrospective: optional\n"
+        "tags: [\n"
+        "  # a standalone comment the round-trip drops\n"
+        '  "alpha",\n'
+        '  "beta",\n'
+        "]\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1", "--set-retro-done"]
+    )
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert "comment line lost" in out["error"]
+    assert "a standalone comment the round-trip drops" in out["error"]
+    assert target.read_text(encoding="utf-8") == fixture
+
+
+# --- Malformed input stays on the JSON contract ------------------------------
+
+
+@pytest.mark.parametrize("command", ["detect-epic", "update"])
+def test_non_mapping_root_is_json_error(tmp_path, command):
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text("- a\n- b\n", encoding="utf-8")
+    args = ["--file", str(target)] + (["--epic", "1"] if command == "update" else [])
+    proc = _run([command, *args])
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert "root document is not a mapping" in out["error"]
+    assert target.read_text(encoding="utf-8") == "- a\n- b\n"
+
+
+@pytest.mark.parametrize("command", ["detect-epic", "update"])
+@pytest.mark.parametrize(
+    "body",
+    ['development_status: "not-a-mapping"\n', "development_status:\n  - a\n  - b\n"],
+    ids=["scalar", "list"],
+)
+def test_non_mapping_development_status_is_json_error(tmp_path, command, body):
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(body, encoding="utf-8")
+    args = ["--file", str(target)] + (["--epic", "1"] if command == "update" else [])
+    proc = _run([command, *args])
+    # update used to report ok:true here while silently doing nothing.
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert "development_status is not a mapping" in out["error"]
+
+
+@pytest.mark.parametrize("command", ["detect-epic", "update"])
+def test_directory_target_is_json_error(tmp_path, command):
+    target = tmp_path / "a-directory"
+    target.mkdir()
+    args = ["--file", str(target)] + (["--epic", "1"] if command == "update" else [])
+    proc = _run([command, *args])
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["error"]
+
+
+@pytest.mark.skipif(
+    hasattr(os, "geteuid") and os.geteuid() == 0,
+    reason="root bypasses file permission bits",
+)
+@pytest.mark.parametrize("command", ["detect-epic", "update"])
+def test_unreadable_target_is_json_error(tmp_path, command):
+    # The other half of the OSError widening: PermissionError, not just
+    # IsADirectoryError, has to stay on the JSON contract.
+    target = _write_fixture(tmp_path)
+    os.chmod(target, 0o000)
+    args = ["--file", str(target)] + (["--epic", "1"] if command == "update" else [])
+    try:
+        proc = _run([command, *args])
+    finally:
+        os.chmod(target, 0o644)
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert "denied" in out["error"].lower()
+
+
+@pytest.mark.parametrize("command", ["detect-epic", "update"])
+def test_invalid_utf8_is_json_error(tmp_path, command):
+    target = tmp_path / "sprint-status.yaml"
+    target.write_bytes(b"development_status:\n  1-1-a: d\xffone\n")
+    args = ["--file", str(target)] + (["--epic", "1"] if command == "update" else [])
+    proc = _run([command, *args])
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert "utf-8" in out["error"].lower()
+
+
+# --- Atomic write ------------------------------------------------------------
+
+
+def test_atomic_write_failure_leaves_target_byte_identical(tmp_path, monkeypatch):
+    # The failure the atomic write exists for: something goes wrong after the
+    # temp file has been written. Nothing may reach the target and no temp file
+    # may survive. No CLI path reaches here -- a read-only directory fails at
+    # mkstemp instead -- so this drives the helper directly.
+    mod = _module()
+    target = _write_fixture(tmp_path)
+
+    def boom(*args, **kwargs):
+        raise OSError(28, "No space left on device")
+
+    monkeypatch.setattr(mod.os, "replace", boom)
+    with pytest.raises(OSError):
+        mod._atomic_write(str(target), b"replacement bytes\n", 0o644)
+    assert target.read_text(encoding="utf-8") == FIXTURE
+    assert [p.name for p in tmp_path.iterdir()] == ["sprint-status.yaml"]
+
+
+def test_dir_fsync_failure_after_rename_is_not_a_write_failure(tmp_path, monkeypatch):
+    # Once os.replace has returned, the new bytes ARE the file. The directory
+    # fsync that follows is durability polish; if it raised, cmd_update would
+    # emit "restored": true about a write that in fact landed — the one lie the
+    # restored contract exists to prevent. Deny opening the directory (the only
+    # thing _atomic_write opens by path after the rename) and require success.
+    mod = _module()
+    target = _write_fixture(tmp_path)
+    directory = os.path.dirname(os.path.realpath(str(target)))
+    real_open = os.open
+
+    def deny_directory_open(p, *args, **kwargs):
+        if p == directory:
+            raise OSError(5, "Input/output error")
+        return real_open(p, *args, **kwargs)
+
+    monkeypatch.setattr(mod.os, "open", deny_directory_open)
+    mod._atomic_write(str(target), b"replacement bytes\n", 0o644)
+    assert target.read_bytes() == b"replacement bytes\n"
+    assert [p.name for p in tmp_path.iterdir()] == ["sprint-status.yaml"]
+
+
+def test_restore_is_atomic(tmp_path, monkeypatch):
+    # _restore is the rollback the reference sells as the safety net. A
+    # truncating rewrite that dies halfway would destroy the very bytes it is
+    # putting back, which is how a full disk used to corrupt the file.
+    mod = _module()
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text("damaged\n", encoding="utf-8")
+
+    def boom(*args, **kwargs):
+        raise OSError(28, "No space left on device")
+
+    monkeypatch.setattr(mod.os, "replace", boom)
+    assert mod._restore(str(target), FIXTURE.encode("utf-8"), 0o644) is False
+    # It reported failure honestly and left the file no worse than it found it.
+    assert target.read_text(encoding="utf-8") == "damaged\n"
+    assert [p.name for p in tmp_path.iterdir()] == ["sprint-status.yaml"]
+
+
+def test_symlinked_target_is_written_through(tmp_path):
+    # os.replace onto a symlink would detach the link and leave the real file
+    # stale while reporting ok:true.
+    real = tmp_path / "real-sprint-status.yaml"
+    real.write_text(FIXTURE, encoding="utf-8")
+    link = tmp_path / "sprint-status.yaml"
+    link.symlink_to(real)
+    proc = _run(["update", "--file", str(link), "--epic", "1", "--set-retro-done"])
+    assert proc.returncode == 0, proc.stderr
+    assert link.is_symlink(), "the symlink was replaced by a regular file"
+    assert _load(real)["development_status"]["epic-1-retrospective"] == "done"
+
+
+def test_atomic_write_preserves_mode_and_leaves_no_temp_file(tmp_path):
+    # mkstemp creates 0600; without carrying the target's mode over, every
+    # update would silently narrow the file.
+    holder = tmp_path / "holder"
+    holder.mkdir()
+    target = _write_fixture(holder)
+    os.chmod(target, 0o640)
+    proc = _run(["update", "--file", str(target), "--epic", "1", "--set-retro-done"])
+    assert proc.returncode == 0, proc.stderr
+    assert stat.S_IMODE(target.stat().st_mode) == 0o640
+    assert [p.name for p in holder.iterdir()] == ["sprint-status.yaml"]
+
+
+# --- Result-JSON precision ---------------------------------------------------
+
+
+def test_retro_key_found_is_null_without_the_flag(tmp_path):
+    # No development_status key at all: the update must not conjure one, and
+    # retro_key_found must say "not asked" rather than "absent".
+    fixture = 'project: "Demo"\nlast_updated: "01-01-2026 09:00"\n'
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(["update", "--file", str(target), "--epic", "1"])
+    assert proc.returncode == 0, proc.stderr
+    out = _json(proc)
+    assert out["ok"] is True
+    assert out["retro_key_found"] is None
+    assert "development_status" not in target.read_text(encoding="utf-8")
+
+
+def test_retro_key_found_is_false_when_the_key_is_absent(tmp_path):
+    target = _write_fixture(tmp_path)
+    proc = _run(["update", "--file", str(target), "--epic", "2", "--set-retro-done"])
+    assert proc.returncode == 0, proc.stderr
+    out = _json(proc)
+    assert out["ok"] is True
+    assert out["retro_key_found"] is False
+    # Nothing was written into the mapping.
+    assert "epic-2-retrospective" not in _load(target)["development_status"]
+
+
+@pytest.mark.parametrize("command", ["detect-epic", "update"])
+def test_only_the_write_path_reports_restored(tmp_path, command):
+    # "restored" speaks to the state of a file the command may have written.
+    # detect-epic never writes, so inventing the key there would mislead callers
+    # that branch on it.
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text("- a\n- b\n", encoding="utf-8")
+    args = ["--file", str(target)] + (["--epic", "1"] if command == "update" else [])
+    out = _json(_run([command, *args]))
+    assert out["ok"] is False
+    assert ("restored" in out) is (command == "update")
+
+
+def test_pre_write_failure_reports_restored(tmp_path):
+    # retro-document.md teaches callers that ok:false carries restored:true;
+    # a failure before the write must not read as "the file may be incomplete".
+    target = _write_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1", "--add-action", "{not json"]
+    )
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert target.read_text(encoding="utf-8") == FIXTURE
+
+
+# --- Action-item validation and identity -------------------------------------
+
+
+def test_non_latin_action_keeps_its_text_in_the_id(tmp_path):
+    target = _write_fixture(tmp_path)
+    payload = json.dumps(
+        [{"action": "Улучшить обработку ошибок", "owner": "Amelia"}],
+        ensure_ascii=False,
+    )
+    proc = _run(["update", "--file", str(target), "--epic", "1", "--add-action", payload])
+    assert proc.returncode == 0, proc.stderr
+    item_id = _load(target)["action_items"][0]["id"]
+    assert item_id == "epic-1-retro-item-1-улучшить-обработку-ошибок"
+
+
+def test_unsluggable_action_falls_back_to_a_hash(tmp_path):
+    target = _write_fixture(tmp_path)
+    payload = json.dumps([{"action": "!!! 🎉", "owner": "Amelia"}], ensure_ascii=False)
+    proc = _run(["update", "--file", str(target), "--epic", "1", "--add-action", payload])
+    assert proc.returncode == 0, proc.stderr
+    item_id = _load(target)["action_items"][0]["id"]
+    assert not item_id.endswith("-item")
+    assert re.fullmatch(r"epic-1-retro-item-1-[0-9a-f]{8}", item_id), item_id
+
+
+def test_empty_action_is_rejected(tmp_path):
+    target = _write_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1",
+         "--add-action", '[{"action":"   ","owner":"x"}]']
+    )
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert "action" in out["error"]
+    assert target.read_text(encoding="utf-8") == FIXTURE
+
+
+def test_non_string_action_is_rejected(tmp_path):
+    # A JSON null would otherwise be str()'d into a literal "None" and written
+    # as a real action item, which the new emptiness check alone lets through.
+    target = _write_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1",
+         "--add-action", '[{"action":null,"owner":"x"}]']
+    )
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert target.read_text(encoding="utf-8") == FIXTURE
+
+
+def test_date_is_normalized_to_the_canonical_format(tmp_path):
+    # strptime accepts unpadded spellings; writing those through would defeat
+    # the point of validating the format.
+    target = _write_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1", "--date", "1-2-2026 9:05"]
+    )
+    assert proc.returncode == 0, proc.stderr
+    assert _json(proc)["last_updated"] == "01-02-2026 09:05"
+    assert _load(target)["last_updated"] == "01-02-2026 09:05"
+
+
+def test_malformed_date_is_rejected(tmp_path):
+    target = _write_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1", "--date", "not-a-date"]
+    )
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert "--date" in out["error"]
+    assert target.read_text(encoding="utf-8") == FIXTURE
+
+
+# --- Action-item status transitions ------------------------------------------
+
+
+def test_set_action_status_applies_both_selector_forms(tmp_path):
+    # The whole point of the flag: an item written by this script (selected by
+    # id) and a legacy item that predates ids (selected by epic + exact action
+    # text) both move off "open" in a single call.
+    target = _write_action_fixture(tmp_path)
+    payload = json.dumps(
+        [
+            {"id": "epic-1-retro-item-1-x", "status": "done"},
+            {"epic": 1, "action": "Pre-existing item", "status": "in-progress"},
+        ]
+    )
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1", "--set-action-status", payload]
+    )
+    assert proc.returncode == 0, proc.stderr
+    out = _json(proc)
+    assert out["ok"] is True
+    assert out["action_items_updated"] == 2
+    assert out["action_items_added"] == 0
+
+    items = _load(target)["action_items"]
+    assert items[0]["status"] == "done"
+    assert items[1]["status"] == "in-progress"
+    # Every other key of both items survived untouched, in place.
+    assert items[0]["id"] == "epic-1-retro-item-1-x"
+    assert items[0]["epic"] == 1
+    assert items[0]["action"] == "Scripted item"
+    assert items[0]["owner"] == "Amelia"
+    assert items[0]["ref"] == "docs/epic-1-retro.md"
+    assert "id" not in items[1]
+    assert items[1]["epic"] == 1
+    assert items[1]["action"] == "Pre-existing item"
+    assert items[1]["owner"] == "Charlie"
+    # The epic-2 item shares its action text with items[0]; the epic-1 selector
+    # must not have touched it.
+    assert items[2]["status"] == "in-progress"
+    assert items[2]["id"] == "epic-2-retro-item-1-y"
+    assert len(items) == 3
+
+
+def test_set_action_status_changes_exactly_one_line(tmp_path):
+    # The status write is surgical: it must not re-style neighbouring lines, and
+    # an item whose status was quoted keeps its quoting. last_updated is rewritten
+    # with the same text it already held, so the whole file differs by one line.
+    target = _write_action_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1", "--date", "01-01-2026 09:00",
+         "--set-action-status", '[{"id":"epic-1-retro-item-1-x","status":"done"}]']
+    )
+    assert proc.returncode == 0, proc.stderr
+
+    before = ACTION_FIXTURE.splitlines()
+    after = target.read_text(encoding="utf-8").splitlines()
+    assert len(before) == len(after)
+    changed = [(b, a) for b, a in zip(before, after) if b != a]
+    assert changed == [('    status: "open"', '    status: "done"')], changed
+
+
+def test_set_action_status_composes_with_retro_done_and_add_action(tmp_path):
+    target = _write_action_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1", "--set-retro-done",
+         "--add-action", '[{"action":"Brand new item","owner":"Amelia"}]',
+         "--set-action-status",
+         '[{"epic":1,"action":"Pre-existing item","status":"done"}]']
+    )
+    assert proc.returncode == 0, proc.stderr
+    out = _json(proc)
+    assert out["ok"] is True
+    assert out["retro_status_after"] == "done"
+    assert out["action_items_added"] == 1
+    assert out["action_items_updated"] == 1
+
+    data = _load(target)
+    assert data["development_status"]["epic-1-retrospective"] == "done"
+    items = data["action_items"]
+    assert len(items) == 4
+    assert items[1]["status"] == "done"  # the targeted pre-existing item
+    assert items[3]["action"] == "Brand new item"
+    assert items[3]["status"] == "open"  # the appended item is always open
+
+
+def test_set_action_status_cannot_target_an_item_added_in_the_same_run(tmp_path):
+    # Selectors resolve against action_items as loaded, so the append cannot be
+    # observed by the same invocation. Silently succeeding here would make the
+    # flag a back door for writing a non-open status onto a brand-new item.
+    target = _write_action_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1",
+         "--add-action", '[{"action":"Brand new","owner":"A","id":"brand-new"}]',
+         "--set-action-status", '[{"id":"brand-new","status":"done"}]']
+    )
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert "no action item matches" in out["error"]
+    assert target.read_text(encoding="utf-8") == ACTION_FIXTURE
+
+
+def test_set_action_status_rejects_unknown_id(tmp_path):
+    target = _write_action_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1",
+         "--set-action-status", '[{"id":"not-in-the-file","status":"done"}]']
+    )
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert "not-in-the-file" in out["error"]
+    assert target.read_text(encoding="utf-8") == ACTION_FIXTURE
+
+
+def test_set_action_status_rejects_selector_when_action_items_is_absent(tmp_path):
+    # No action_items key at all must read as "no match", not as a crash.
+    target = _write_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1",
+         "--set-action-status", '[{"id":"anything","status":"done"}]']
+    )
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert "no action item matches" in out["error"]
+    assert target.read_text(encoding="utf-8") == FIXTURE
+
+
+def test_set_action_status_rejects_ambiguous_selector(tmp_path):
+    # Two legacy items with identical epic + action text: guessing between them
+    # would write the wrong row half the time.
+    fixture = (
+        "development_status:\n"
+        "  1-1-a: done\n"
+        "  epic-1-retrospective: optional\n"
+        "action_items:\n"
+        "  - epic: 1\n"
+        '    action: "Same text"\n'
+        '    owner: "Charlie"\n'
+        "    status: open\n"
+        "  - epic: 1\n"
+        '    action: "Same text"\n'
+        '    owner: "Dana"\n'
+        "    status: open\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1",
+         "--set-action-status", '[{"epic":1,"action":"Same text","status":"done"}]']
+    )
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert "ambiguous" in out["error"]
+    assert "Same text" in out["error"]
+    assert "2 matches" in out["error"]
+    assert target.read_text(encoding="utf-8") == fixture
+
+
+def test_set_action_status_rejects_two_entries_hitting_the_same_item(tmp_path):
+    # The id form and the epic/action form can name the same row; applying both
+    # would overcount action_items_updated and hide a conflicting pair.
+    target = _write_action_fixture(tmp_path)
+    payload = json.dumps(
+        [
+            {"id": "epic-1-retro-item-1-x", "status": "done"},
+            {"epic": 1, "action": "Scripted item", "status": "in-progress"},
+        ]
+    )
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1", "--set-action-status", payload]
+    )
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert "same action item" in out["error"]
+    assert target.read_text(encoding="utf-8") == ACTION_FIXTURE
+
+
+def test_entry_with_both_selector_forms_uses_the_id(tmp_path):
+    # A caller that copied a whole item through supplies both. The id is the
+    # precise form and wins; the extra keys are ignored, not rejected. The two
+    # forms are pointed at *different* rows so the precedence is observable:
+    # the id names items[0], the epic/action pair names items[1].
+    target = _write_action_fixture(tmp_path)
+    payload = json.dumps(
+        [
+            {
+                "id": "epic-1-retro-item-1-x",
+                "epic": 1,
+                "action": "Pre-existing item",
+                "owner": "Charlie",
+                "status": "done",
+            }
+        ]
+    )
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1", "--set-action-status", payload]
+    )
+    assert proc.returncode == 0, proc.stderr
+    assert _json(proc)["action_items_updated"] == 1
+    items = _load(target)["action_items"]
+    assert items[0]["status"] == "done"  # the id's item
+    assert items[1]["status"] == "open"  # the epic/action item, untouched
+
+
+def test_set_action_status_rejects_invalid_status(tmp_path):
+    target = _write_action_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1",
+         "--set-action-status", '[{"id":"epic-1-retro-item-1-x","status":"closed"}]']
+    )
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert "closed" in out["error"]
+    # The allowed vocabulary is named so the caller can correct the call.
+    assert "open, in-progress, done" in out["error"]
+    assert target.read_text(encoding="utf-8") == ACTION_FIXTURE
+
+
+def test_set_action_status_rejects_malformed_json(tmp_path):
+    target = _write_action_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1",
+         "--set-action-status", "{not json"]
+    )
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert "invalid --set-action-status JSON" in out["error"]
+    assert target.read_text(encoding="utf-8") == ACTION_FIXTURE
+
+
+_SELECTOR_SHAPE_ERROR = (
+    "each --set-action-status entry must have a non-empty string id, "
+    "or an integer epic and a non-empty string action"
+)
+
+
+@pytest.mark.parametrize(
+    ("payload", "expected_error"),
+    [
+        (
+            '{"id":"epic-1-retro-item-1-x","status":"done"}',
+            "--set-action-status must be a JSON array",
+        ),
+        (
+            '["epic-1-retro-item-1-x"]',
+            "each --set-action-status entry must be an object",
+        ),
+        ('[{"status":"done"}]', _SELECTOR_SHAPE_ERROR),
+        (
+            '[{"id":"","status":"done"}]',
+            "each --set-action-status id must be a non-empty string",
+        ),
+        (
+            '[{"id":42,"status":"done"}]',
+            "each --set-action-status id must be a non-empty string",
+        ),
+        (
+            '[{"epic":"1","action":"Pre-existing item","status":"done"}]',
+            _SELECTOR_SHAPE_ERROR,
+        ),
+        (
+            '[{"epic":true,"action":"Pre-existing item","status":"done"}]',
+            _SELECTOR_SHAPE_ERROR,
+        ),
+        ('[{"epic":1,"action":"   ","status":"done"}]', _SELECTOR_SHAPE_ERROR),
+        (
+            # No status key at all: the status validator runs before the selector
+            # validator, so this is the status branch, not the selector branch.
+            '[{"epic":1,"action":"Pre-existing item"}]',
+            "invalid --set-action-status status None",
+        ),
+        (
+            '[{"id":"epic-1-retro-item-1-x","status":3}]',
+            "invalid --set-action-status status 3",
+        ),
+    ],
+    ids=[
+        "not-a-list",
+        "entry-not-an-object",
+        "no-selector",
+        "empty-id",
+        "non-string-id",
+        "string-epic",
+        "bool-epic",
+        "blank-action",
+        "no-status-key",
+        "non-string-status",
+    ],
+)
+def test_set_action_status_rejects_bad_shapes(tmp_path, payload, expected_error):
+    # Each case pins its own message: collapsing the branches into one generic
+    # error would leave a caller unable to tell which part of the array is wrong.
+    target = _write_action_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1",
+         "--set-action-status", payload]
+    )
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert expected_error in out["error"], out["error"]
+    assert target.read_text(encoding="utf-8") == ACTION_FIXTURE
+
+
+def test_selectors_are_not_scoped_to_the_epic_flag(tmp_path):
+    # The flag's headline use: epic 2's retro closing epic 1's items. --epic only
+    # names the retro key and stamps appended items; scoping selectors to it would
+    # silently break the documented cross-epic workflow while every same-epic test
+    # kept passing.
+    target = _write_action_fixture(tmp_path)
+    payload = json.dumps(
+        [
+            {"id": "epic-1-retro-item-1-x", "status": "done"},
+            {"epic": 1, "action": "Pre-existing item", "status": "done"},
+        ]
+    )
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "2", "--set-retro-done",
+         "--set-action-status", payload]
+    )
+    assert proc.returncode == 0, proc.stderr
+    assert _json(proc)["action_items_updated"] == 2
+
+    data = _load(target)
+    assert data["development_status"]["epic-2-retrospective"] == "done"
+    items = data["action_items"]
+    assert items[0]["status"] == "done"
+    assert items[1]["status"] == "done"
+    # Epic 2's own item is not swept along.
+    assert items[2]["status"] == "in-progress"
+
+
+def test_legacy_selector_discriminates_on_the_epic(tmp_path):
+    # Two items share the action text "Scripted item" and differ only by epic, so
+    # an epic-blind text match would be ambiguous -- or worse, silently pick one.
+    target = _write_action_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1", "--set-action-status",
+         '[{"epic":2,"action":"Scripted item","status":"done"}]']
+    )
+    assert proc.returncode == 0, proc.stderr
+    assert _json(proc)["action_items_updated"] == 1
+    items = _load(target)["action_items"]
+    assert items[2]["status"] == "done"
+    assert items[0]["status"] == "open"  # the epic-1 namesake, untouched
+
+
+def test_non_mapping_action_item_does_not_crash_the_selector(tmp_path):
+    # A hand-edited scalar in the action_items list must be skipped, not
+    # AttributeError'd into an empty stdout with a traceback.
+    fixture = (
+        "development_status:\n"
+        "  1-1-a: done\n"
+        "  epic-1-retrospective: optional\n"
+        "action_items:\n"
+        '  - "a bare string someone hand-edited in"\n'
+        '  - id: "real"\n'
+        "    epic: 1\n"
+        '    action: "Real item"\n'
+        "    status: open\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1",
+         "--set-action-status", '[{"id":"real","status":"done"}]']
+    )
+    assert proc.returncode == 0, proc.stderr
+    assert _json(proc)["action_items_updated"] == 1
+    items = _load(target)["action_items"]
+    assert items[0] == "a bare string someone hand-edited in"
+    assert items[1]["status"] == "done"
+
+
+def test_non_mapping_action_item_stays_on_the_json_contract_when_unmatched(tmp_path):
+    # Same guard, reject path: the scalar must not be dereferenced while looking
+    # for a selector that is not there.
+    fixture = (
+        "development_status:\n"
+        "  1-1-a: done\n"
+        "action_items:\n"
+        "  - 42\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1",
+         "--set-action-status", '[{"epic":1,"action":"Nothing","status":"done"}]']
+    )
+    assert proc.returncode == 1
+    out = _json(proc)  # asserts stdout is JSON and stderr carries no traceback
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert "no action item matches" in out["error"]
+    assert target.read_text(encoding="utf-8") == fixture
+
+
+def test_boolean_epic_in_the_file_does_not_match_epic_one(tmp_path):
+    # True == 1 in Python, so without the file-side bool guard a hand-edited
+    # "epic: true" row would be silently rewritten by a selector aimed at epic 1.
+    fixture = (
+        "development_status:\n"
+        "  1-1-a: done\n"
+        "action_items:\n"
+        "  - epic: true\n"
+        '    action: "Boolean epic"\n'
+        "    status: open\n"
+    )
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(fixture, encoding="utf-8")
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1", "--set-action-status",
+         '[{"epic":1,"action":"Boolean epic","status":"done"}]']
+    )
+    assert proc.returncode == 1
+    out = _json(proc)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert "no action item matches" in out["error"]
+    assert target.read_text(encoding="utf-8") == fixture
+
+
+def test_status_write_preserves_every_scalar_style(tmp_path):
+    # The status write must land on the line without re-styling it, whichever way
+    # the file spells the scalar.
+    target = tmp_path / "sprint-status.yaml"
+    target.write_text(STYLE_FIXTURE, encoding="utf-8")
+    payload = json.dumps(
+        [
+            {"id": "double", "status": "done"},
+            {"id": "single", "status": "done"},
+            {"id": "plain", "status": "done"},
+        ]
+    )
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1", "--date",
+         "01-01-2026 09:00", "--set-action-status", payload]
+    )
+    assert proc.returncode == 0, proc.stderr
+    assert _json(proc)["action_items_updated"] == 3
+
+    before = STYLE_FIXTURE.splitlines()
+    after = target.read_text(encoding="utf-8").splitlines()
+    assert len(before) == len(after)
+    changed = [(b, a) for b, a in zip(before, after) if b != a]
+    assert changed == [
+        ('    status: "open"', '    status: "done"'),
+        ("    status: 'open'", "    status: 'done'"),
+        ("    status: open", "    status: done"),
+    ], changed
+
+
+def test_action_status_vocabulary_is_exactly_the_three():
+    # bmad-sprint-planning is the authority. Widening this tuple would let the
+    # script write a value sprint-planning's status view reports as illegal.
+    assert _module().ACTION_STATUSES == ("open", "in-progress", "done")
+
+
+def test_empty_status_array_is_a_no_op(tmp_path):
+    target = _write_action_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "1", "--date",
+         "01-01-2026 09:00", "--set-action-status", "[]"]
+    )
+    assert proc.returncode == 0, proc.stderr
+    assert _json(proc)["action_items_updated"] == 0
+    assert target.read_text(encoding="utf-8") == ACTION_FIXTURE
+
+
+def test_in_progress_item_transitions_to_done(tmp_path):
+    # Every other success case starts from "open"; the middle of the lifecycle
+    # has to work too.
+    target = _write_action_fixture(tmp_path)
+    proc = _run(
+        ["update", "--file", str(target), "--epic", "2", "--set-action-status",
+         '[{"id":"epic-2-retro-item-1-y","status":"done"}]']
+    )
+    assert proc.returncode == 0, proc.stderr
+    assert _json(proc)["action_items_updated"] == 1
+    assert _load(target)["action_items"][2]["status"] == "done"
+
+
+def test_action_items_updated_is_always_reported(tmp_path):
+    # Consumers read the counter unconditionally, so it must be present even when
+    # the flag was not passed.
+    target = _write_fixture(tmp_path)
+    proc = _run(["update", "--file", str(target), "--epic", "1", "--set-retro-done"])
+    assert proc.returncode == 0, proc.stderr
+    assert _json(proc)["action_items_updated"] == 0
+
+
+def test_post_write_status_mismatch_restores(tmp_path, monkeypatch, capsys):
+    # The last line of defence: the written file is re-parsed and every targeted
+    # item is checked. No CLI path can fake a mismatch, so the re-parse is
+    # doctored directly.
+    mod = _module()
+    target = _write_action_fixture(tmp_path)
+    real_load_yaml = mod._load_yaml
+    calls = {"n": 0}
+
+    def flaky(path):
+        yaml, data = real_load_yaml(path)
+        calls["n"] += 1
+        if calls["n"] == 2:  # the post-write re-parse
+            data["action_items"][0]["status"] = "open"
+        return yaml, data
+
+    monkeypatch.setattr(mod, "_load_yaml", flaky)
+    args = mod.build_parser().parse_args(
+        ["update", "--file", str(target), "--epic", "1", "--date", "01-01-2026 09:00",
+         "--set-action-status", '[{"id":"epic-1-retro-item-1-x","status":"done"}]']
+    )
+    with pytest.raises(SystemExit) as excinfo:
+        mod.cmd_update(args)
+    assert excinfo.value.code == 1
+    out = json.loads(capsys.readouterr().out)
+    assert out["ok"] is False
+    assert out["restored"] is True
+    assert "after write" in out["error"]
+    assert target.read_text(encoding="utf-8") == ACTION_FIXTURE
+
+
+if __name__ == "__main__":
+    sys.exit(pytest.main([__file__, "-q"]))
diff --git a/.agents/skills/bmad-review-adversarial-general/SKILL.md b/.agents/skills/bmad-review-adversarial-general/SKILL.md
new file mode 100644
index 00000000..ec9f0256
--- /dev/null
+++ b/.agents/skills/bmad-review-adversarial-general/SKILL.md
@@ -0,0 +1,6 @@
+---
+name: bmad-review-adversarial-general
+description: 'Deprecated — forwards to bmad-review.'
+---
+
+Merged into `bmad-review`. Invoke the `bmad-review` skill on the same content with only the `adversarial` lens, passing through any `also_consider` areas. Present the findings as a Markdown list — descriptions only, no severity, priority, or ranking; no JSON block.
diff --git a/.agents/skills/bmad-review-edge-case-hunter/SKILL.md b/.agents/skills/bmad-review-edge-case-hunter/SKILL.md
new file mode 100644
index 00000000..85bc024f
--- /dev/null
+++ b/.agents/skills/bmad-review-edge-case-hunter/SKILL.md
@@ -0,0 +1,6 @@
+---
+name: bmad-review-edge-case-hunter
+description: 'Deprecated — forwards to bmad-review.'
+---
+
+Merged into `bmad-review`. Invoke the `bmad-review` skill on the same content with only the `edge-case-hunter` lens, passing through any `also_consider` areas. Output ONLY the raw findings JSON array in the legacy shape: the four standard fields (plus `kind`/`confidence` on deletion findings), no `lens` field, no markdown wrapping, no extra text. `[]` is valid when nothing is found.
diff --git a/.agents/skills/bmad-review-verification-gap/SKILL.md b/.agents/skills/bmad-review-verification-gap/SKILL.md
new file mode 100644
index 00000000..9ada5686
--- /dev/null
+++ b/.agents/skills/bmad-review-verification-gap/SKILL.md
@@ -0,0 +1,6 @@
+---
+name: bmad-review-verification-gap
+description: 'Deprecated — forwards to bmad-review.'
+---
+
+Merged into `bmad-review`. Invoke the `bmad-review` skill on the same content with only the `verification-gap` lens. Present the markdown rendering only (no JSON block), listing any `gap_shape: "other"` findings under an `## Other findings` heading. When there are no findings at all, output exactly this single line: `No verification gaps found.`
diff --git a/.agents/skills/bmad-review/SKILL.md b/.agents/skills/bmad-review/SKILL.md
new file mode 100644
index 00000000..0d168db8
--- /dev/null
+++ b/.agents/skills/bmad-review/SKILL.md
@@ -0,0 +1,49 @@
+---
+name: bmad-review
+description: 'Multi-lens review over any diff, doc, spec, or artifact — whichever installed lenses fit the content, run singly or together. Shipped lenses include adversarial, edge-case, verification-gap, structure, and prose. Use when the user says "review this", "critical review", "editorial review", "hunt edge cases", "review the structure", or "review the prose".'
+---
+
+# BMad Review
+
+Review content through lenses — each a distinct method and stance — and report findings in one canonical shape. Report what is real — never pad to look thorough. Each lens sets its own stance toward the content and toward zero findings: for most an empty result is valid; the adversarial lens requires at least ten concrete findings and treats an empty list as a signal to re-check; the editorial lenses hold content sacrosanct and critique only how it is organized and expressed.
+
+The lens set is whatever `{workflow.lenses}` resolves to, not a fixed list — overrides add lenses and replace shipped ones. Never claim a capability from this file; read the resolved lenses and work from those.
+
+## Inputs
+
+- **content** — what to review: a diff, branch, uncommitted changes, file, spec, story, or any document. Args: `[path]`.
+- **lenses** (optional) — one or more lens codes or names, however the caller expresses them: a spoken request, or a directive of the form `skill:bmad-review lenses=<code>[,<code>...]` (the form bmm's `doc_standards` uses). Default: every applicable lens (a full review).
+- **also_consider** (optional) — areas to keep in mind alongside each lens's normal analysis.
+- **pre-resolved customization** (optional) — `[workflow]` field values supplied by a forwarding caller. See Execution step 1.
+
+## Conventions
+
+- Bare paths (e.g. `references/lens-edge-case-hunter.md`) resolve from `{skill-root}` — this skill's installed directory, where `customize.toml` lives. `{project-root}` resolves to the project working directory.
+- `{workflow.<name>}` resolves to fields in `customize.toml`'s `[workflow]` table (overrides win per BMad merge rules).
+- In `style_guide`, `review_guidance`, and `persistent_facts`, a value prefixed `file:` is a path or glob — load that file's contents. If a `file:` value cannot be read, name the failed file in the output header and continue: the shipped baseline for `style_guide`, the remaining entries otherwise.
+
+## Execution
+
+1. **Resolve customization:** `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. On failure, read `{skill-root}/customize.toml` directly and use defaults. **Forwarded activation:** if a caller invoked you with pre-resolved customization fields (e.g. the `bmad-editorial-review` shim), honor them verbatim for those named fields — they already carry the user's overrides — and resolve only the remaining fields from your own `customize.toml`. Then execute each `{workflow.activation_steps_prepend}` entry in order, hold `{workflow.persistent_facts}` as standing context for the session, and treat `{workflow.review_guidance}` entries as standing review directives for every lens.
+2. **Load the content.** If it is empty or cannot be decoded as text: when the caller expects the raw findings JSON array (e.g. the legacy edge-case forwarder), return `[{"location":"N/A","trigger_condition":"Input empty or undecodable","guard_snippet":"Provide valid content to review","potential_consequence":"Review skipped — no analysis performed"}]` (no `lens` field) and stop; otherwise say what's wrong and ask for reviewable content. Classify the content — diff, source file, function, or document — and whether it is **code** or **docs**; scope rules and lens applicability both depend on it. A document that defines behavior (spec, requirements, plan, story) is `docs` that a behavioral lens may still apply to; judge by `when`.
+3. **Select lenses** from `{workflow.lenses}`. A lens with an empty `instruction` is disabled. If the user or caller named lenses, run exactly those only — `applies_to` and `when` do not filter an explicit request. Otherwise run every enabled lens whose `applies_to` covers the content class (`any` always covers) and whose `when` applies.
+4. **Announce the plan** in one line before running anything: the content class, the lenses about to run, and — when any lens has `after` set — that it runs on top of the named lens's findings. Skip the announcement entirely when the caller pinned an exact output contract (the legacy forwarders that demand raw JSON or one exact line) — their contract covers everything you emit, not just the findings block. Then execute each `{workflow.activation_steps_append}` entry in order.
+5. **Run the independent lenses** — every selected lens without `after`. Each sees the content and `also_consider`, never another lens's findings. Follow each lens's `instruction`; the shipped lenses load their reference file just-in-time, so load only what runs. When subagents are available, spawn one per lens in parallel: give it the lens `instruction` with `{skill-root}` and paths resolved absolute, the content or where to read it, any `also_consider` areas, the standing review directives, and the constraint "Return ONLY your findings — no other output." Otherwise run the lenses sequentially yourself, completing one before starting the next.
+6. **Run the dependent lenses** — every selected lens with `after`, once the lens it names has completed, passing that lens's findings in. A lens whose `after` target was not selected or produced nothing still runs, with no prior findings. Dependent lenses that name different targets are independent of each other and may run in parallel.
+7. **Assemble and present** per Output below. Keep every lens's findings — overlap between lenses is signal, not duplication; note it in the markdown report rather than deduping. Execute `{workflow.on_complete}` if set.
+
+## Output
+
+One JSON array holding every finding from every lens. Each finding carries:
+
+- `lens` — the code of the lens that produced it
+- `location` — where in the content (file:line-range for code, section for documents)
+- `trigger_condition` — the problem, or the condition that exposes it, in one line
+- `guard_snippet` — the concrete fix, guard, or missing check
+- `potential_consequence` — what goes wrong if it ships as-is
+
+Each lens file refines these semantics for its findings and may add lens-specific fields (e.g. `kind`/`confidence` on deletion findings, `gap_shape`/`consumer`/`evidence` on verification-gap findings). A lens file may instead declare its own findings shape and rendering — the editorial lenses render a findings table — and that shape wins for that lens's findings. `[]` is valid when nothing is found. No severity, priority, or ranking anywhere.
+
+Present per `{workflow.output_format}` — `"json"` (the raw array in a fenced json block), `"markdown"`, or `"both"` — unless the caller requested a specific shape; a legacy forwarder's output contract always wins, and governs everything you emit rather than the findings block alone. The markdown report groups findings by lens, each rendered in its declared shape: a short block per finding rendering the fields plus any extras worth surfacing, one line for a lens that found nothing, and a plain clean statement when the whole review is clean. Shape the report per `{workflow.output_preferences}`.
+
+When `{workflow.report_path}` is set, write the report there; otherwise present it in chat.
diff --git a/.agents/skills/bmad-review/customize.toml b/.agents/skills/bmad-review/customize.toml
new file mode 100644
index 00000000..e9f8e36c
--- /dev/null
+++ b/.agents/skills/bmad-review/customize.toml
@@ -0,0 +1,141 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-review.
+#
+# Override files (not edited here):
+#   {project-root}/_bmad/custom/bmad-review.toml         (team)
+#   {project-root}/_bmad/custom/bmad-review.user.toml    (personal)
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins
+#   arrays (persistent_facts, activation_steps_*, review_guidance): append
+#   arrays of tables keyed by `code`: matching key replaces, new keys append
+
+# Steps executed on activation: prepend runs before the skill's own
+# activation flow, append runs after the lens plan is settled and before the
+# lenses run. Each entry is a literal instruction.
+activation_steps_prepend = []
+activation_steps_append = []
+
+# Standing context held for every review, code and document alike. Entries
+# prefixed `file:` are paths or globs whose contents load as facts; all others
+# are literal facts. The shipped entry is a project-wide glob — set it to []
+# if you don't want every review scanning for it.
+persistent_facts = ["file:{project-root}/**/project-context.md"]
+
+# Standing review directives applied on every run alongside each lens's own
+# method. Each entry is a literal sentence or a `file:`-prefixed path/glob
+# whose contents load as directives.
+#
+# Examples:
+#   "Flag passive voice in headings."
+#   "Second-person imperative is the house voice; never suggest changing it."
+#   "file:{project-root}/docs/terminology.md"
+review_guidance = []
+
+# Executed after the findings are delivered. Freeform directive; empty = the
+# review ends with the findings.
+#
+# Example:
+#   on_complete = "Append a one-line review summary to {project-root}/docs/review-log.md"
+on_complete = ""
+
+# How findings are presented when the caller doesn't say: "json" (the raw
+# findings array only), "markdown" (the human report only), or "both". A lens
+# that declares its own rendering keeps it for its own findings.
+output_format = "both"
+
+# Where to write the review report. Empty = present in chat only. Accepts
+# {project-root}-prefixed paths.
+report_path = ""
+
+# How findings are presented — shaping, not destination. Freeform directive;
+# empty = each lens's default ordering and rollup.
+#
+# Example:
+#   output_preferences = "Cap output at the 20 highest-impact findings."
+output_preferences = ""
+
+# --- Editorial lens settings (used by the structure and prose lenses) ---
+
+# Default reader the editorial lenses calibrate for when the request doesn't
+# say:
+#   "humans"  clarity, flow, comprehension aids preserved
+#   "llm"     precision, consistent terminology, no hedging
+# A reader type stated in the request wins for that run.
+reader_type = "humans"
+
+# The baseline style guide for every editorial review: the name of a guide the
+# model knows well, a `file:`-prefixed path to a style guide document, or the
+# rules inline as text. A style guide stated in the request wins for that run.
+# Where the guide in effect conflicts with the lens's generic principles, the
+# guide wins — except content is sacrosanct.
+#
+# Examples (set in team/user override TOML):
+#   style_guide = "file:{project-root}/_bmad/style-guides/company-voice.md"
+#   style_guide = "Sentence-case headings. No Oxford comma. Address the reader as 'you'."
+style_guide = "Microsoft Writing Style Guide"
+
+# ---------------------------------------------------------------------------
+# Review lenses. Each lens is a pass over the content with its own method and
+# stance. `instruction` is the lens's whole execution recipe — the shipped
+# lenses load a reference file from the skill root, but an override may inline
+# any prompt.
+#
+# `applies_to` is the content this lens can review: "code", "docs", or "any".
+# It is the first filter — a lens never joins a default review for content it
+# does not apply to. `when` (optional) refines that judgement in prose. An
+# explicitly requested lens always runs, whatever both say.
+#
+# `after` (optional) names a lens this one builds on: it runs once that lens
+# has completed and receives its findings, instead of running independently.
+#
+# Empty `instruction` disables a lens. Keyed by `code`: an override with a
+# matching code replaces the shipped lens, a new code appends.
+#
+# Example (add an org-specific lens in team/user override TOML):
+#   [[workflow.lenses]]
+#   code = "accessibility"
+#   name = "Accessibility"
+#   applies_to = "any"
+#   when = "UI code or user-facing documents."
+#   instruction = "Review against WCAG 2.2 AA. Emit findings in the canonical fields."
+# ---------------------------------------------------------------------------
+
+[[workflow.lenses]]
+code = "adversarial"
+name = "Adversarial"
+applies_to = "any"
+when = "always"
+instruction = "Load `references/lens-adversarial.md` from the skill root and follow it."
+
+[[workflow.lenses]]
+code = "edge-case-hunter"
+name = "Edge-Case Hunter"
+applies_to = "any"
+when = "Content with behavior to trace: code, diffs, and the specs, requirements, plans, and stories that define behavior. Skip for prose documents with no behavioral surface."
+instruction = "Load `references/lens-edge-case-hunter.md` from the skill root and follow it."
+
+[[workflow.lenses]]
+code = "verification-gap"
+name = "Verification Gap"
+applies_to = "code"
+when = "Reviewed inside a repo where tests can be searched and read."
+instruction = "Load `references/lens-verification-gap.md` from the skill root and follow it."
+
+[[workflow.lenses]]
+code = "structure"
+name = "Editorial Structure"
+applies_to = "docs"
+when = "Documents whose shape is the author's to change."
+instruction = "Load `references/lens-structure.md` from the skill root and follow it."
+
+[[workflow.lenses]]
+code = "prose"
+name = "Editorial Prose"
+applies_to = "docs"
+after = "structure"
+when = "Documents being copy-edited."
+instruction = "Load `references/lens-prose.md` from the skill root and follow it."
diff --git a/.agents/skills/bmad-review/references/editorial-common.md b/.agents/skills/bmad-review/references/editorial-common.md
new file mode 100644
index 00000000..a5fb1046
--- /dev/null
+++ b/.agents/skills/bmad-review/references/editorial-common.md
@@ -0,0 +1,56 @@
+# Editorial Lenses — Common Ground
+
+Shared by the `structure` and `prose` lenses. Load this once; when both lenses run, the setup below is done once and serves both.
+
+## Stance
+
+Review a document as a clinical editor and return suggested fixes the author can accept or reject row by row. Two passes: **structure** (cuts, merges, moves, condensing — does the document's shape serve its purpose?) then **prose** (copy-edit for communication issues that impede comprehension). Which of the two run, and in what order, is decided by lens selection — see the skill's Execution section.
+
+**Content is sacrosanct.** Never challenge ideas — only how they're organized and expressed. Propose, don't execute: the author decides what to accept.
+
+The baseline style guide is `{workflow.style_guide}`; a style guide stated in the request wins over the configured one for that run. Where the style guide in effect conflicts with a generic principle here — including the reader calibration — the style guide wins. Nothing overrides content being sacrosanct.
+
+## Setup
+
+1. Gather inputs: the content (required — a path or pasted text), plus whatever the request states: purpose, target audience, length target, reader type, style guide. If no reviewable content was provided, say so and stop. Request-level values win; `{workflow.reader_type}` and `{workflow.style_guide}` fill what the request leaves unstated. Treat `{workflow.review_guidance}` entries as standing review directives.
+2. When the content is a file, get exact word counts — document total and per heading section — via `uv run {skill-root}/scripts/word_metrics.py <path>` (`--help` documents the output), and ground every word-impact estimate and the reduction summary in those numbers. If the content was pasted or the script cannot run, estimate and mark the numbers as estimates.
+3. Infer purpose and audience from the content and standing context when not provided, and open the output with your one-sentence read — "this document exists to help [audience] accomplish [goal]" — so the author can correct a wrong premise before acting on the findings.
+
+## Reader calibration
+
+Calibrate every finding to the reader type — stated in the request, else `{workflow.reader_type}`.
+
+**humans** (default) — optimize for clarity, flow, and natural progression. These elements serve comprehension and engagement; preserve them unless clearly wasteful, and flag any recommendation that would cut one:
+
+- Visual aids: diagrams, images, and flowcharts anchor understanding
+- Expectation-setting: "What You'll Learn" helps readers confirm they're in the right place
+- Reader's journey: organize content as a linear progression, not a database
+- Mental models: overview before details prevents cognitive overload
+- Warmth: encouraging tone reduces anxiety for new users
+- Whitespace: admonitions and callouts provide visual breathing room
+- Summaries: recaps help retention; they're reinforcement, not redundancy
+- Examples: concrete illustrations make abstract concepts accessible
+- Engagement: flow techniques (transitions, variety) are functional, not fluff — they maintain attention
+
+**llm** — optimize for precision and unambiguity. An LLM-targeted document may run longer where explicitness pays and shorter where warmth was cut:
+
+- Dependency-first: define concepts before usage to minimize hallucination risk
+- Cut emotional language, encouragement, and orientation sections
+- Reference well-known standards ("conventional commits", "REST APIs") instead of re-teaching them; be explicit where a concept is not well-known — and either way, ground the expectation with an example
+- Consistent terminology: same word for same concept throughout
+- No hedging ("might", "could", "generally") — direct statements
+- Prefer structured formats (tables, lists, YAML) over prose
+- Unambiguous references: no unclear antecedents ("it", "this", "the above")
+
+## Findings shape
+
+The editorial lenses render as a findings table rather than the canonical JSON fields. One findings table serves both passes:
+
+| Pass      | Original Text                                         | Revised Text                                  | Changes                                                              |
+| --------- | ----------------------------------------------------- | --------------------------------------------- | -------------------------------------------------------------------- |
+| structure | §Setup — full section (~180 words)                    | MERGE into §Installation                      | Duplicates the install steps; one source of truth (saves ~150 words) |
+| prose     | The system will processes data and it handles errors. | The system processes data and handles errors. | Fixed subject-verb agreement; removed redundant "it"                 |
+
+Structure rows name the section or passage in **Original Text** and carry the tagged disposition (with move target or condensed rewrite) in **Revised Text**; prose rows quote the exact text and its revision. Order rows by comprehension impact; when a long document would produce more rows than an author can realistically act on, present the highest-impact rows and roll the rest into one closing line — "N further minor fixes; ask to expand." Above the table, give the purpose/audience read plus — when the structure pass ran — the chosen structure model. When the structure pass ran, close with a summary: total recommendations, estimated reduction (words and % of original, computed from the word-metrics counts) if all are accepted, whether a provided length target is met, and any comprehension trade-offs (cuts that sacrifice reader engagement for brevity). A pass that finds nothing is a valid result; say so.
+
+Shape the table per `{workflow.output_preferences}`.
diff --git a/.agents/skills/bmad-review/references/lens-adversarial.md b/.agents/skills/bmad-review/references/lens-adversarial.md
new file mode 100644
index 00000000..cd417e73
--- /dev/null
+++ b/.agents/skills/bmad-review/references/lens-adversarial.md
@@ -0,0 +1,19 @@
+# Adversarial Lens
+
+Conduct a review of the provided content.
+Look for what's missing, not only what's wrong.
+Find at least ten issues to fix or improve.
+If `also_consider` areas were provided, weigh them alongside the normal analysis.
+If the content is empty, stop and say so.
+If you have zero findings, re-check and keep thinking; do not stop with an empty list.
+
+## Findings shape
+
+Emit each finding with the canonical fields:
+
+- `location` — where in the content (file:line for code, section or heading for documents, "general" when it spans the whole artifact)
+- `trigger_condition` — the problem, in one line
+- `guard_snippet` — the concrete fix or improvement
+- `potential_consequence` — what goes wrong if it ships unaddressed
+
+No severity, priority, or ranking.
diff --git a/.agents/skills/bmad-review/references/lens-edge-case-hunter.md b/.agents/skills/bmad-review/references/lens-edge-case-hunter.md
new file mode 100644
index 00000000..5ee24301
--- /dev/null
+++ b/.agents/skills/bmad-review/references/lens-edge-case-hunter.md
@@ -0,0 +1,54 @@
+# Edge-Case Lens
+
+You are a pure path tracer. Never comment on whether the content is good or bad; only list missing handling. Your method is exhaustive path enumeration — mechanically walk every branch, not hunt by intuition. Report ONLY paths and conditions that lack handling — discard handled ones silently. Do not editorialize or add filler.
+
+**MANDATORY: Execute the steps below IN EXACT ORDER. DO NOT skip steps or change the sequence. Each action within a step is a REQUIRED action to complete that step.**
+
+**Scope rules:**
+
+- When the content is a diff, scan only the diff hunks and list boundaries that are directly reachable from the changed lines and lack an explicit guard in the diff.
+- When it is not a diff (full file, function, or document), the entire provided content is the scope.
+- Ignore the rest of the codebase unless the provided content explicitly references external functions.
+
+## Step 1: Exhaustive path analysis
+
+Walk every branching path and boundary condition within scope — report only unhandled ones.
+
+- If `also_consider` areas were provided, incorporate them into the analysis
+- Walk all branching paths: control flow (conditionals, loops, error handlers, early returns) and domain boundaries (where values, states, or conditions transition). Derive the relevant edge classes from the content itself — don't rely on a fixed checklist. Examples: missing else/default, unguarded inputs, off-by-one loops, arithmetic overflow, implicit type coercion, race conditions, timeout gaps
+- Consider implicit branches: the diff special-cases or changes the handling of one or more members of a fixed set of values — enums, status codes, sentinels, type tags, flags, value ranges. The rest of the set is implicit branches (e.g. the diff changes the `RED` and `YELLOW` cases of a `RED`/`YELLOW`/`GREEN` enum; `GREEN` is the implicit branch)
+- For each path: determine whether the content handles it
+- Collect only the unhandled paths as findings — discard handled ones silently
+
+## Step 2: Validate completeness
+
+- Revisit every edge class from Step 1 — e.g., missing else/default, null/empty inputs, off-by-one loops, arithmetic overflow, implicit type coercion, race conditions, timeout gaps
+- Add any newly found unhandled paths to findings; discard confirmed-handled ones
+
+## Step 3: Deletion check
+
+Runs only when the diff removed or replaced meaningful code (ignore pure renames and whitespace). Subordinate to the edge-case pass; findings are usually few or none.
+
+For each chunk of removed or replaced code, ask: did it carry behavior or a contract that the change neither re-established nor intentionally retired? Add a finding for any resulting regression, orphaned reference, or newly-dead code. Skip anything already covered by your edge-case findings. Add nothing if nothing qualifies.
+
+Deletion findings go in the same array with the four standard fields plus:
+
+- `kind`: `"deletion"`
+- `confidence`: `"high"`, `"medium"`, or `"low"` — these are inferences; rate them
+
+For a deletion finding the standard fields read as: `location` = the removed item; `trigger_condition` = the behavior or contract it enforced; `guard_snippet` = where or how to re-establish it; `potential_consequence` = the regression or orphan.
+
+## Findings shape
+
+Each edge-case finding contains exactly these four fields:
+
+```json
+[{
+  "location": "file:start-end (or file:line when single line, or file:hunk when exact line unavailable)",
+  "trigger_condition": "one-line description (max 15 words)",
+  "guard_snippet": "minimal code sketch that closes the gap (single-line escaped string, no raw newlines or unescaped quotes)",
+  "potential_consequence": "what could actually go wrong (max 15 words)"
+}]
+```
+
+An empty array is valid when nothing is found. Do not assign severity labels, rankings, or priority levels.
diff --git a/.agents/skills/bmad-review/references/lens-prose.md b/.agents/skills/bmad-review/references/lens-prose.md
new file mode 100644
index 00000000..9483bbe5
--- /dev/null
+++ b/.agents/skills/bmad-review/references/lens-prose.md
@@ -0,0 +1,7 @@
+# Prose Lens
+
+Load `references/editorial-common.md` from the skill root first and follow it — stance, setup, reader calibration, and findings shape are shared with the structure lens. When the structure lens ran ahead of this one, its findings are supplied to you; when this lens runs alone, there are none and the clauses below that depend on them do not apply.
+
+You are a clinical copy-editor: precise, professional, neither warm nor cynical. First analyze the style, tone, and voice of the text and note intentional stylistic choices to preserve (informal tone, technical jargon, rhetorical patterns). Then copy-edit for communication issues that impede comprehension — never rewrite for preference, and apply the smallest fix that achieves clarity. Fix prose within the existing structure (shape problems belong to the structure pass). Skip code blocks, frontmatter, and structural markup. Preserve the author's voice and the stylistic choices you noted. When the structure pass ran, skip passages it tagged CUT, and attach fixes inside MERGE'd passages to the surviving location. Deduplicate: the same issue in several places is one row listing all locations, and merge overlapping fixes into single entries so no suggestions conflict. Phrase uncertain fixes as "Consider: …?" rather than definitive changes.
+
+Emit rows with `Pass` = `prose`.
diff --git a/.agents/skills/bmad-review/references/lens-structure.md b/.agents/skills/bmad-review/references/lens-structure.md
new file mode 100644
index 00000000..02b3f6c6
--- /dev/null
+++ b/.agents/skills/bmad-review/references/lens-structure.md
@@ -0,0 +1,9 @@
+# Structure Lens
+
+Load `references/editorial-common.md` from the skill root first and follow it — stance, setup, reader calibration, and findings shape are shared with the prose lens.
+
+You are a structural editor focused on high-value density. Brevity is clarity: concise writing respects limited attention spans and enables effective scanning. Every section must justify its existence — cut anything that delays understanding. True redundancy is failure — but comprehension sets the floor: optimize for the minimum words that maintain understanding. Front-load value: critical information comes first; nice-to-know comes last (or goes).
+
+Load `references/structure-models.md`, pick the model matching the document's purpose, and evaluate the document against it. Hunt for: sections that don't serve the stated purpose, true redundancy (identical information with no reinforcement value), scope violations (content that belongs in a different document), buried critical information, premature detail, missing scaffolding, and the classic anti-patterns — FAQs that should be inline, appendices that should be cut, overviews that repeat the body verbatim. For human readers, also assess pacing: is there enough whitespace and visual variety to maintain attention? Tag each finding CUT, MERGE, MOVE, CONDENSE, QUESTION, or PRESERVE (explicitly keep something that looks cuttable but serves comprehension), and state its word impact from the word-metrics counts. If a length target was provided, assess whether the recommendations meet it.
+
+Emit rows with `Pass` = `structure`.
diff --git a/.agents/skills/bmad-review/references/lens-verification-gap.md b/.agents/skills/bmad-review/references/lens-verification-gap.md
new file mode 100644
index 00000000..88c580cf
--- /dev/null
+++ b/.agents/skills/bmad-review/references/lens-verification-gap.md
@@ -0,0 +1,92 @@
+# Verification-Gap Lens
+
+**Goal:** Find changed behavior that could break without reliable verification catching it. Ask one question — "if the behavior this change is supposed to produce broke where it's actually used, would verification fail?" Do not hunt for correctness bugs, but report genuine problems you notice while tracing verification.
+
+The main verification gap shapes are:
+
+1. **Regression gap:** the changed code regresses where it's used, and no test covering that use would fail.
+2. **Missing-adoption gap:** a place that should now use the new behavior doesn't; it handles the same case its own way, or not at all, and no test would flag the omission.
+3. **Broken-verification gap:** a test appears to cover the changed behavior, but would not actually protect it because it is skipped, flaky, not run in the normal verification path, or too weak to observe the regression.
+
+## Evidence rules
+
+- Read a test before claiming what it covers, runs, asserts, or misses.
+- Before claiming no test exists, search the whole repo by the symbol under test and by import references; expected file locations are not enough.
+- Never assert what you did not verify. If a finding cannot be grounded, drop it.
+- In a finding, say what you actually checked — "none of the tests I read cover this" — and show how far you looked. Say a test doesn't exist anywhere only when the symbol/import-reference search actually shows that.
+- Do not assign severity, confidence, priority, or ranking.
+
+## Review sequence
+
+### Step 1: Screen for behavioral change
+
+Before applying the non-behavioral stop to a test-only change, check whether it removes or weakens verification of deterministic behavior. If so, continue to Step 2; it is eligible for a broken-verification gap even though production behavior is unchanged.
+
+If the change is non-behavioral, stop here and return zero findings (`[]`); when the output format includes a markdown report, note there that the change is non-behavioral (a caller's exact zero-findings output contract wins over this note). Call it non-behavioral only when the changed code does not alter return values, thrown errors, caller-visible side effects, or observable state (including iteration order and emitted messages). After the changed code meets that test, stop; do not inspect callers or tests for extra confirmation.
+
+Common non-behavioral examples: formatting, comments, whitespace; pure renames; trivial getters/setters and pass-throughs; type-only or compiler-enforced changes with no runtime effect; etc.
+
+### Step 2: Find the behavior that changed
+
+Identify what behavior changed compared to the previous version: output, side effect, branch, error path, schema/event shape, config default, validation/authorization rule, external contract, etc. If the change affects more than one behavior, handle each separately.
+
+Treat broad-impact changes as behavioral even when no single changed line looks important: dependency, toolchain, build/config, data-file, etc.
+
+Seek verification of behavior, not the literal text of implementation or documentation artifacts. Tests may assert exact content or structure when they execute deterministic construction or transformation and inspect its output, including generated prompts and request payloads. Do not seek phrase-existence assertions over hand-authored prompts, skills, documents, or source files.
+
+For LLM-backed behavior, stop at the inference boundary: do not require invoking a model or judging its semantic response. Deterministic request construction and response handling remain eligible without live inference; existing inference tests are not precedent for more.
+
+### Step 3: Trace where that behavior is used
+
+Trace the changed behavior to the places that observe it. Start with direct callers and registered entry points (routes, commands, DI), contract consumers (schemas, events, APIs, database readers), and reverse-dependency info if already available.
+
+Follow a path only while the changed behavior is reachable and unverified. Stop when a test at that boundary would fail, the consumer does not observe the changed behavior, or the next hop is guesswork (dynamic dispatch, reflection, outside-repo consumers, etc.). Prefer the nearest observable boundary, often one to three hops away, especially across contract, integration, or service edges. If there are more than five similar consumers, group obvious repeats and check representative paths; expand only when a consumer observes the behavior differently.
+
+### Step 4: Qualify the consumer, then check its test
+
+For each consumer, name the smallest realistic regression this consumer would observe: invert the branch, drop the default, omit the field, return the old error code, skip the integration call, etc. This is the Demonstration. If no such regression exists, drop the path; untested downstream code is not a finding.
+
+A `Missing-adoption gap` qualifies not by the adoption failure alone but by a supersession signal: the change gives clear evidence the new behavior is meant to replace the local one — PR intent, naming or docs, a replaced sibling site, deleted duplicate logic, or a test defining the new rule — and the local site shares the same observable contract. Without a supersession signal and a shared observable contract, it is a refactor suggestion, not a verification-gap finding. Once both hold, check whether any test for that site would flag the non-adoption; missing coverage of the non-adoption is the gap itself, not a disqualifier.
+
+Find and read the relevant test. Ask whether the Demonstration would make an assertion fail.
+
+- If yes, the behavior is verified. No finding.
+- For a regression-style Demonstration: if no test runs the path, the test is skipped/flaky/not run normally, or the test runs the code without checking the changed result, report a `Regression gap` or `Broken-verification gap`.
+- For a qualifying Missing-adoption case: if none of the site tests you found assert it adopts the new behavior, report a `Missing-adoption gap`.
+
+A test counts only if it runs normally and an assertion observes the changed output, branch, or contract. These do not count: no execution; success/no-throw/snapshot-only checks; mock/log-call checks; human-only checks; tests that mock away the integration; e2e tests that pass through without checking the changed output; stale assertions or fixtures.
+
+For example, `expect(x ?? DEFAULT).toBe(DEFAULT)` passes when `x` is missing.
+
+Common patterns:
+
+- **Caller-path gap** — helper test covers the branch, but caller values skip it.
+- **Contract drift** — payload/schema/event changes must be verified at the consumer.
+- **Migration compatibility** — tests only create new-format rows or fresh schemas.
+- **Phantom exception** — handled partial-failure path has no test.
+- **Missing-adoption gap** — sibling site should use the new rule/helper and does not.
+- **Removed verification** — deleted test or weakened assertion leaves behavior unpinned.
+
+### Step 5: Confirm each finding is real
+
+Before writing a finding, re-open the specific tests or search results the finding relies on. Verify the Demonstration would not make any test you checked fail, or that the absence claim is backed by the symbol/import-reference search. Do not claim more than you verified; drop any finding you cannot ground.
+
+Explain why the test misses the bug using what the test sets up and checks.
+
+Do not report: compiler/type-checker-enforced cases; behavior already verified by an integration, contract, or e2e test; implementation-detail or mock-only tests; low coverage or a missing test file by itself; legacy untested code the change did not affect.
+
+Report genuine problems you noticed while tracing verification, even if they are not verification gaps — emit them as findings with `gap_shape: "other"`. This permits reporting what you already reached, not extra hunting.
+
+## Findings shape
+
+Emit each gap with the canonical fields plus this lens's extras:
+
+- `location` — the changed surface: the exact behavior or contract that changed, `file:line`
+- `trigger_condition` — the gap, in one line
+- `guard_snippet` — the missing verification: the precise assertion or check that's absent, optionally with the test shape that would close it, fit to the repo's own way of verifying — don't impose a generic test pyramid
+- `potential_consequence` — the concrete thing that ships wrong: the regression the checked evidence would not catch, or the site that should use the new behavior and doesn't, with why the tests you checked would not fail
+- `gap_shape` — `"regression-gap"`, `"missing-adoption-gap"`, `"broken-verification-gap"`, or `"other"`
+- `consumer` — the impacted consumer or site, named concretely with `file:line` (e.g. "the `createInvoice` mutation used by the billing dashboard at `billing/dashboard.ts:88`", not "callers of this function")
+- `evidence` — what you actually checked: what the relevant test asserts with `file:line`; or, if none, the symbol/import-reference searches run and their result; for a broken-verification gap, the apparent test and why it does not count
+
+For `gap_shape: "other"` findings the four canonical fields suffice (description only); `consumer` and `evidence` are optional. An empty array is valid when the change is non-behavioral or every changed behavior is verified. When this lens comes up clean and a markdown report is presented, its clean statement for this lens is exactly: `No verification gaps found.`
diff --git a/.agents/skills/bmad-review/references/structure-models.md b/.agents/skills/bmad-review/references/structure-models.md
new file mode 100644
index 00000000..d520d929
--- /dev/null
+++ b/.agents/skills/bmad-review/references/structure-models.md
@@ -0,0 +1,44 @@
+# Structure Models
+
+Reference shapes for the structure pass. Pick the one matching the document's purpose and evaluate the document against its rules; a document that fits none cleanly is judged against the closest model, with the mismatch itself noted as a finding when the shape fights the purpose.
+
+## Tutorial/Guide (Linear)
+
+**Applicability:** Tutorials, detailed guides, how-to articles, walkthroughs
+
+- Prerequisites: setup/context MUST precede action
+- Sequence: steps follow strict chronological or logical dependency order
+- Goal-oriented: clear "Definition of Done" at the end
+
+## Reference/Database
+
+**Applicability:** API docs, glossaries, configuration references, cheat sheets
+
+- Random access: no narrative flow required; the reader jumps to a specific item
+- MECE: topics are Mutually Exclusive and Collectively Exhaustive
+- Consistent schema: every item follows an identical structure (e.g., Signature → Params → Returns)
+
+## Explanation (Conceptual)
+
+**Applicability:** Deep dives, architecture overviews, conceptual guides, whitepapers, project context
+
+- Abstract to concrete: Definition → Context → Implementation/Example
+- Scaffolding: complex ideas built on established foundations
+
+## Prompt/Task Definition (Functional)
+
+**Applicability:** BMad skills and workflows, prompts, system instructions, agent definitions
+
+- Meta-first: inputs, usage constraints, and context defined before instructions
+- Separation of concerns: instructions (logic) separate from data (content)
+- Explicit flow: execution order is stated, never implied
+
+## Strategic/Context (Pyramid)
+
+**Applicability:** PRDs, research reports, proposals, decision records
+
+- Top-down: conclusion/status/recommendation starts the document
+- Grouping: supporting context grouped logically below the headline
+- Ordering: most critical information first
+- MECE: arguments/groups are Mutually Exclusive and Collectively Exhaustive
+- Evidence: data supports arguments, never leads
diff --git a/.agents/skills/bmad-review/scripts/tests/test_word_metrics.py b/.agents/skills/bmad-review/scripts/tests/test_word_metrics.py
new file mode 100644
index 00000000..be293749
--- /dev/null
+++ b/.agents/skills/bmad-review/scripts/tests/test_word_metrics.py
@@ -0,0 +1,62 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.10"
+# ///
+"""Tests for word_metrics.py."""
+
+import sys
+import unittest
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
+
+from word_metrics import section_metrics, word_count
+
+DOC = """Intro line before any heading.
+
+# Title
+
+Two words here indeed.
+
+## Section A
+
+Alpha beta gamma.
+
+```
+# not a heading
+fenced words ignored as headings
+```
+
+## Section B
+
+Delta epsilon.
+"""
+
+
+class WordMetricsTest(unittest.TestCase):
+    def test_word_count(self):
+        self.assertEqual(word_count("one two  three\nfour"), 4)
+        self.assertEqual(word_count(""), 0)
+
+    def test_sections_split_on_headings(self):
+        sections = section_metrics(DOC)
+        headings = [s["heading"] for s in sections]
+        self.assertEqual(headings, ["(preamble)", "Title", "Section A", "Section B"])
+
+    def test_fenced_heading_not_a_section(self):
+        sections = section_metrics(DOC)
+        self.assertNotIn("not a heading", [s["heading"] for s in sections])
+
+    def test_section_words_counted(self):
+        sections = {s["heading"]: s["words"] for s in section_metrics(DOC)}
+        self.assertEqual(sections["Section B"], 2)
+        # Section A body includes the fenced block's tokens
+        self.assertGreater(sections["Section A"], 3)
+
+    def test_empty_preamble_dropped(self):
+        sections = section_metrics("# Only\n\nwords here\n")
+        self.assertEqual([s["heading"] for s in sections], ["Only"])
+
+
+if __name__ == "__main__":
+    unittest.main()
diff --git a/.agents/skills/bmad-review/scripts/word_metrics.py b/.agents/skills/bmad-review/scripts/word_metrics.py
new file mode 100644
index 00000000..9e8b1524
--- /dev/null
+++ b/.agents/skills/bmad-review/scripts/word_metrics.py
@@ -0,0 +1,102 @@
+#!/usr/bin/env python3
+# /// script
+# requires-python = ">=3.10"
+# ///
+"""Exact word counts for a document, as JSON.
+
+Emits the document's total word count and a per-heading-section breakdown so
+an editorial review can ground word-impact estimates and reduction
+percentages in real numbers instead of guessing. Sections are delimited by
+markdown headings (# through ######); heading markers inside fenced code
+blocks are ignored (fences pair CommonMark-style: a fence closes only on a
+run of the same character at least as long, so ```` fences may embed ```
+examples). A word is any whitespace-separated token, plus one word per CJK
+character since those scripts do not space-delimit words. For non-markdown
+input the result is a single section holding the full text.
+"""
+
+import argparse
+import json
+import re
+import sys
+from pathlib import Path
+
+HEADING = re.compile(r"^(#{1,6})\s+(\S.*)$")
+FENCE = re.compile(r"^ {0,3}(`{3,}|~{3,})")
+CJK = re.compile(r"[぀-ヿ㐀-䶿一-鿿豈-﫿가-힯ｦ-ﾟ]")
+
+
+def word_count(text: str) -> int:
+    cjk = len(CJK.findall(text))
+    return cjk + len(CJK.sub(" ", text).split())
+
+
+def section_metrics(text: str) -> list[dict]:
+    sections = []
+    current = {"heading": "(preamble)", "level": 0, "body": []}
+    open_fence = None  # (char, length) while inside a fenced block
+    for line in text.splitlines():
+        fence = FENCE.match(line)
+        if fence:
+            marker = fence.group(1)
+            if open_fence is None:
+                open_fence = (marker[0], len(marker))
+            elif marker[0] == open_fence[0] and len(marker) >= open_fence[1] and line.strip() == marker:
+                open_fence = None
+            current["body"].append(line)
+            continue
+        match = None if open_fence else HEADING.match(line)
+        if match:
+            sections.append(current)
+            current = {
+                "heading": match.group(2).strip(),
+                "level": len(match.group(1)),
+                "body": [],
+            }
+        else:
+            current["body"].append(line)
+    sections.append(current)
+
+    out = []
+    for section in sections:
+        words = word_count("\n".join(section["body"]))
+        if section["heading"] == "(preamble)" and words == 0:
+            continue
+        out.append(
+            {"heading": section["heading"], "level": section["level"], "words": words}
+        )
+    return out
+
+
+def metrics(path: Path) -> dict:
+    text = path.read_text(encoding="utf-8", errors="replace")
+    return {
+        "file": str(path),
+        "total_words": word_count(text),
+        "sections": section_metrics(text),
+    }
+
+
+def main() -> int:
+    if hasattr(sys.stdout, "reconfigure"):
+        sys.stdout.reconfigure(encoding="utf-8")  # JSON is UTF-8 regardless of locale code page
+    parser = argparse.ArgumentParser(description=__doc__.splitlines()[0])
+    parser.add_argument("path", help="document to measure")
+    parser.add_argument("-o", "--output", help="write JSON here (default: stdout)")
+    args = parser.parse_args()
+
+    path = Path(args.path)
+    if not path.is_file():
+        print(f"error: not a readable file: {path}", file=sys.stderr)
+        return 2
+
+    result = json.dumps(metrics(path), indent=2, ensure_ascii=False)
+    if args.output:
+        Path(args.output).write_text(result + "\n", encoding="utf-8")
+    else:
+        print(result)
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/.agents/skills/bmad-spec/SKILL.md b/.agents/skills/bmad-spec/SKILL.md
new file mode 100644
index 00000000..d9252bec
--- /dev/null
+++ b/.agents/skills/bmad-spec/SKILL.md
@@ -0,0 +1,160 @@
+---
+name: bmad-spec
+description: Distill any intent input into the SPEC kernel + companions — the canonical, preservation-validated machine contract for downstream work. Use when the user says "create a spec", "distill this into a spec", "validate this spec", "update the spec", or "break this into stories".
+---
+
+# BMad Spec
+## Overview
+
+Canonical transformer for the BMad spec-kernel ecosystem. Takes any intent input — vague idea, brain dump, PRD, GDD, RFC, brief, Slack thread, customer email, meeting transcript, mockups, mixed multi-source — and produces **SPEC.md** carrying the five-field kernel (Why, Capabilities, Constraints, Non-goals, Success signal) plus companion files for load-bearing content that does not fit or would bloat the kernel with expansive line-item detail. Together they are the machine contract every downstream BMad skill consumes.
+
+Multiple skills may call to update the same spec over time.
+
+## Conventions
+
+- Bare paths (e.g. `assets/spec-template.md`) resolve from the skill root.
+- `{skill-root}` is this skill's install dir; `{project-root}` is the working dir.
+- `{workflow.<name>}` resolves to fields in `customize.toml`.
+
+## On Activation
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. On failure, read `{skill-root}/customize.toml` directly.
+2. Run `{workflow.activation_steps_prepend}`. Treat `{workflow.persistent_facts}` as foundational context (`file:` entries are loaded).
+3. Resolve config: `uv run {project-root}/_bmad/scripts/resolve_config.py --project-root {project-root}` (merges `_bmad/config.toml`, `_bmad/config.user.toml`, and the `_bmad/custom/` overrides). From the merged JSON resolve `{user_name}`, `{communication_language}`, `{document_output_language}`, `{project_name}`, `{output_folder}` (under `core`), and `{date}`.
+4. Detect mode. **Headless** when any of: no TTY, programmatic caller (another skill or non-interactive runner), or the first message pre-supplies all inputs and asks for an artifact path back. **Interactive** otherwise. In interactive mode, greet by `{user_name}` in `{communication_language}`, stay in that language, and mention that `bmad-party-mode` and `bmad-advanced-elicitation` are available for deeper exploration on any field.
+
+Run `{workflow.activation_steps_append}`.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+## Workspace
+
+The spec is **always a folder** named `{workflow.spec_output_path}/{workflow.run_folder_pattern}`, resolving by default to `{output_folder}/specs/spec-{slug}/`.
+
+`{slug}` describes the thing being specced, not the input shape:
+
+- Source artifact already carries a slug (e.g., `prd-foo-bar-2026-05-23/`): inherit (`foo-bar`).
+- Sparse, in-chat, or multi-source input: interactive asks; headless caller provides it as part of the input. If absent and underivable, headless blocks with `error_code: "missing_slug"`.
+- Same slug = same folder. A second invocation with the same `{slug}` lands at the existing spec folder and updates in place, preserving capability IDs.
+
+**No input.** Interactive: ask the user to share a file path, paste content, explain the idea in detail, or point to a source. Headless: respond with JSON containing `error_code: "insufficient_intent"`.
+
+Inside the spec folder:
+
+```
+<spec-folder>/
+  SPEC.md                  ← uppercase, the kernel — DERIVED from .memlog.md, never hand-edited
+  <companion-1>.md         ← optional, content-typed (e.g. glossary.md); spec-authored ones are derived too
+  <companion-2>.md
+  stories.yaml             ← optional, written only by Story Breakdown — fixed name, never in companions:
+  .memlog.md               ← canonical, append-only memory; what SPEC.md is distilled from
+```
+
+## Memory and derivation
+
+`.memlog.md` is canonical — an append-only, chronological record of every decision, constraint, capability (with its stable `CAP-N`), assumption, open question, and bit of user direction, one line each in the order it happened, never edited or reordered. `SPEC.md` and every spec-authored companion are **derived on each run** from the memlog (the decision-of-record) plus the sources it cites for raw content — never hand-patched.
+
+Deriving the contract from a living log instead of editing the contract in place is what lets the steps around the spec (PRD, UX, architecture, epics) run in any order and feed the same spec without merge drift: the log only accumulates, the artifact is re-rendered. So the spec is updated *only* by re-deriving it here — bmad-spec is its single writer; a hand-edit to `SPEC.md` from outside is unsupported and is overwritten on the next derive.
+
+Writes go through the shared script — `{project-root}/_bmad/scripts/memlog.py`, the same location as `resolve_customization.py` (atomic; never read it back except to resume):
+
+- `uv run {project-root}/_bmad/scripts/memlog.py init --workspace {spec-folder} --field topic="<what is being specced>"` — once, at create.
+- `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {spec-folder} --type <decision|constraint|capability|assumption|question|direction|note|event> --text "<one-line gist, reason included>"` — as each lands.
+- Terminal moments (a validation verdict, "spec finalized") are `--type event` entries; the memlog carries no status field.
+
+## The Operation
+
+Read the input and its ancillary linked materials. If there is no input, follow the no-input branch in **Workspace** (ask or block). If a prior `.memlog.md` exists at the target folder, read it — the operation becomes an update, and the memlog (not the rendered `SPEC.md`) is the authority on what was decided and on capability IDs. Preserve those IDs; new capabilities get the next unused `CAP-N`; never reuse retired IDs. Otherwise this is a create, and the first move is `memlog.py init`.
+
+When the input is structured and pre-sorted (a PRD with an addendum, a GDD, a brief produced by an upstream BMad skill), trust the authored separation: lift kernel-fitting content into SPEC.md, lift overflow into appropriately-named companions. When the input is mixed (a brain dump, a transcript, an RFC, a customer email), do the sorting yourself: walk each claim, apply the three-lens load-bearing test (Spec Law rule 7), and route to the kernel field or a companion.
+
+Distill the input into the five-field kernel using `{workflow.spec_template}` as the skeleton. When input is rich, extract directly — no elicitation. When input is sparse, choose: **express** (best-effort distill, every gap becomes an `open_questions[]` entry) or **guided** (walk the five fields with the user one at a time). Headless defaults to express and logs the choice. Interactive asks.
+
+A recognized domain implication the input leaves unaddressed *is* such a gap — name it as an `open_questions[]` entry (healthcare input silent on PHI/HIPAA, payments silent on PCI, control systems silent on fail-safe) and move on. Flag it; never invent the answer or coach toward it. If these dominate, the input is too thin — suggest `bmad-prd`.
+
+Write lean from the first pass: every sentence must earn its place. Decoration costs tokens and dilutes downstream readers.
+
+Log each decision, capability, constraint, and accepted change to `.memlog.md` as it is made — that running record is what the render reads. Because the log is append-only, a later entry supersedes an earlier one on the same point while the history stays intact. When two currently-live sources or companions disagree on the same field, or an either/or never got resolved, surface it to the user rather than silently choosing — the resolution is itself a new memlog entry.
+
+If the input is genuinely too thin to distill (e.g. "an app for hikers" with no surrounding context), stop and suggest `bmad-prd` (or sibling ceremony skill). This skill distills; it does not coach.
+
+## Load-bearing
+
+A claim is **load-bearing** if any consumer (downstream skill, implementing agent, verification pass) would change a decision without it.
+
+## Companions
+
+When load-bearing content does not fit the five-field kernel, it lives in a companion. The kernel cites it; the companion holds it. Companions are part of the contract; every consumer reads `companions:` in SPEC.md frontmatter to discover them. Companions follow the same lean discipline as SPEC.md (Spec Law rule 8).
+
+**Spawn a companion when the content needs more than one kernel-shape line:** multi-item catalogs (per-entity matrices like archetypes, drinks, modes, routes), tables, diagrams (always), editorial voice rules, long-form reference material the kernel cites by name (glossary, brownfield notes, project conventions). Single-line decision-benders stay in Constraints; intent+success pairs stay in Capabilities. If a kernel field is starting to bullet into sub-bullets, the content has outgrown the kernel and wants a companion.
+
+Companions are either:
+
+- **Spec-authored** companions are written by bmad-spec and live as **siblings of SPEC.md** (e.g., `glossary.md`, `patron-archetypes.md`). bmad-spec owns them and may edit them on update operations.
+- **Adopted** companions are load-bearing artifacts written by an upstream skill that downstream still needs to read. bmad-spec references them into `companions:` by relative path but does NOT edit them (e.g., a `DESIGN.md` or `EXPERIENCE.md` from a UX run, an integration partner's API spec). The originating skill owns them.
+
+Two rules govern companions:
+
+1. **Name spec-authored companions for the content type they hold.** `glossary.md`, `<entity-class>.md` (e.g. `patron-archetypes.md`, `medication-routes.md`, `flight-modes.md`), `stack.md`, `conventions.md`, `brownfield.md`, `architecture-diagrams.md`, `state-machines.md`, `failure-modes.md`, `compliance-references.md`. The principle: "a reader should know what is inside before opening it." Adopted companions keep whatever name their originating skill gave them.
+2. **Diagrams always land in a companion**, regardless of size. SPEC.md kernel holds prose only. Mermaid blocks, ASCII diagrams, and image references all live in a companion (e.g. `architecture-diagrams.md`), with sibling image files referenced from there.
+
+Pre-existing project-wide docs (e.g. `project-context.md`) that downstream needs are listed as **adopted companions**, never duplicated into SPEC.md or a spec-authored companion.
+
+`stories.yaml`, when produced, is spec-authored but deliberately **not** a companion — see Story Breakdown below.
+
+## Spec Law
+
+Every spec must satisfy these eight rules. The operation aims for them; the self-validate sweep enforces them.
+
+1. **Each capability has both `intent` and `success`.** Missing either = not a capability.
+2. **Intents describe WHAT, not HOW.** Implementation prescription belongs in a companion (stack, conventions).
+3. **Constraints actually bend design decisions.** A "constraint" that rules nothing out is decoration.
+4. **Non-goals are explicit.** At least one. Absence means downstream skills fill the vacuum.
+5. **Success signal is concrete enough to test or demonstrate against.** "Users love it" doesn't qualify.
+6. **Capability IDs are stable and unique.** Never reused, never renumbered.
+7. **Preservation.** Every load-bearing source claim lands in SPEC.md or a companion. Wrapper ceremony does not.
+8. **Lean prose.** Every sentence carries load-bearing content. Cut decoration, hedges, backstory, throat-clearing. Applies to SPEC.md, companions, and `.memlog.md`.
+
+## Self-Validate
+
+After every create or update, sweep the resulting artifact in **two passes** before presenting.
+
+**Pass 1 — Coherence.** Judge the spec against Spec Law rules 1–6 and 8. For anything that fails or feels weak, attempt to fix it without inventing content the input did not support. Calls made without direct confirmation become `assumptions[]`; gaps that could not be filled become `open_questions[]`.
+
+**Pass 2 — Preservation.** Walk the source claim by claim. Confirm each load-bearing claim landed in SPEC.md or a companion. Wrapper-ceremony drops are logged under "Wrapper-only content" so the drop is on the record, not silent.
+
+Record the verdict for each pass to `.memlog.md` (`append --type event`). In interactive mode, review it with the user. In headless mode, `.memlog.md` is one of the files returned, so the caller (or its downstream LLM) reads the verdict there.
+
+## Spec with no change signal
+
+When the user points the skill at an existing spec folder (or its SPEC.md) with no change signal, offer to review assumptions or open questions, or determine what they want to do.
+
+## Story Breakdown (optional, interactive-only)
+
+Requires `SPEC.md` on disk — run the normal Operation first if it doesn't exist yet. Headless runs never do this, even when the invocation text asks for it: if mode detection (On Activation, step 4) resolved headless, skip this section entirely and proceed with the normal headless response. In interactive mode, offer it at most once per run when the input reads as multiple independently shippable slices; a decline ends the offer for this run, not forever. Also run it on direct request ("break this into stories") whenever `SPEC.md` exists. When a spec update runs and `stories.yaml` exists, check the story descriptions against the updated spec; if any no longer matches, say so and offer to re-run Story Breakdown. The update itself never rewrites `stories.yaml`.
+
+Either way, walk the capabilities and constraints with the user and propose a story per independently reviewable slice — this is a conversation, not a silent render. For each story, ask the user for `spec_checkpoint`, `done_checkpoint`, and any `invoke_dev_with` note rather than defaulting them silently; capturing that human judgment is what the fields are for. If the conversation surfaces load-bearing detail beyond dispatch notes (a constraint, a design decision), route it into SPEC.md or a companion — `invoke_dev_with` carries dispatch notes only (Spec Law rule 7 still applies).
+
+The output is `stories.yaml`, a sibling of `SPEC.md` inside the spec folder, discovered by that fixed name — same convention as `SPEC.md` and `.memlog.md`. Never list it in `companions:` and never point a frontmatter key at it: companions carry the what-to-build contract every consumer reads; `stories.yaml` is input for whichever tool dispatches the stories.
+
+Field definitions, the validity rules, and a worked example live in `assets/stories-schema.md`. Before writing or re-writing the file, check every entry against those rules; fix violations rather than presenting a file that fails them. Record the check's verdict to `.memlog.md` (`append --type event`), the same discipline as Self-Validate.
+
+Derive `stories.yaml` from `.memlog.md` exactly like any other spec-authored artifact: log each proposed story (`--type decision`) as the user agrees to it, then render. On a later run against the same spec folder, re-derive the same way, handling ids per the schema's update semantics.
+
+## Output
+
+**Interactive** — share the spec folder path conversationally. Name the capability count, the companions produced, and the verdict in one or two sentences. Name the story count too if `stories.yaml` was written this run. If `assumptions[]` or `open_questions[]` are non-empty, list them (short — one line each) and invite the user to walk through them. Make clear that addressing them can update the source input (if it was a file), the spec, or both — whichever combination the user prefers. Do not dump JSON or present a wall of output.
+
+**Headless** — return JSON per `assets/headless-schemas.md`.
+
+Run `{workflow.on_complete}` if set.
+
+## After Spec is Output
+
+Any update to the spec — resolved assumptions, answered open questions, other changes — is appended to `.memlog.md` as it happens. When a change overrides something that came from a source input, offer to update that source too, so upstream and the spec don't silently diverge.
+
+## Frontmatter conventions
+
+- `companions:` array of `.md` files downstream MUST read alongside SPEC.md to have the full contract. Paths may point inside the spec folder (spec-authored companions like `glossary.md`) or outside it (adopted companions like `../planning-artifacts/ux-designs/ux-foo-bar-2026-05-23/DESIGN.md`). The split between spec-authored and adopted is implicit by path; downstream treats both the same.
+- `sources:` array of paths to files that were **fully absorbed** into the SPEC, with no remaining downstream value (e.g., a PRD whose every load-bearing claim is now in the kernel). Listed for audit and for bmad-spec to re-read on update. Downstream does NOT read these. Files that downstream still needs to read belong in `companions:`, not here.
+- **Do not list** the memlog, README files, organizational artifacts, or any operational record of how upstream skills produced their artifacts. Those are not source content; they are process metadata that downstream consumers don't need.
diff --git a/.agents/skills/bmad-spec/assets/headless-schemas.md b/.agents/skills/bmad-spec/assets/headless-schemas.md
new file mode 100644
index 00000000..8e2093be
--- /dev/null
+++ b/.agents/skills/bmad-spec/assets/headless-schemas.md
@@ -0,0 +1,33 @@
+# Headless JSON Response
+
+The default invocation is headless: input goes in, JSON comes out. The contract is intentionally tiny — return the outcome and the files touched. Anything else a caller needs is inside those files (SPEC.md, companions, `.memlog.md`).
+
+## Success
+
+```json
+{
+  "status": "complete",
+  "files": [
+    "_bmad-output/specs/spec-quarter-drop/SPEC.md",
+    "_bmad-output/specs/spec-quarter-drop/glossary.md",
+    "_bmad-output/specs/spec-quarter-drop/.memlog.md"
+  ]
+}
+```
+
+`files` lists every file written or modified in this run, in any order. The spec folder, kernel filename, memlog location, capabilities, companions, and verdict are all readable from those files; no need to re-encode them in the response.
+
+## Blocked
+
+```json
+{
+  "status": "blocked",
+  "error_code": "insufficient_intent",
+  "reason": "Input was a one-line idea with no surrounding context; too thin to distill. Suggest bmad-prd to draw the vision out first."
+}
+```
+
+Defined `error_code` values:
+
+- `insufficient_intent` — input too thin to distill into a kernel.
+- `missing_slug` — input is sparse or multi-source and no slug was provided by the caller or derivable from a source path.
diff --git a/.agents/skills/bmad-spec/assets/spec-template.md b/.agents/skills/bmad-spec/assets/spec-template.md
new file mode 100644
index 00000000..d7568fa8
--- /dev/null
+++ b/.agents/skills/bmad-spec/assets/spec-template.md
@@ -0,0 +1,49 @@
+---
+id: SPEC-{slug}
+companions: []     # files downstream MUST read alongside SPEC.md. Paths may point inside the spec folder (spec-authored) or outside it (adopted from an upstream skill).
+sources: []        # files fully absorbed into the SPEC (audit only; downstream does NOT read these). Never the memlog.
+---
+
+> **Canonical contract.** This SPEC and the files in `companions:` are the complete, preservation-validated contract for what to build, test, and validate. Source documents listed in frontmatter are for traceability — consult them only if you need narrative rationale or prose color this contract intentionally omits.
+
+# {Spec Title}
+
+## Why
+
+{One paragraph naming the force behind this work. A spec can exist for any of:
+  - **a pain to solve** — a user or operator is stuck on a specific gap;
+  - **an opportunity to capture** — something newly possible we want to claim;
+  - **a vision to realize** — a thing we want to make exist because we want it to exist;
+  - **a mandate to meet** — a regulation, deprecation, deadline, or contractual obligation.
+
+Name which (or which combination) applies, who is affected, and the backdrop that makes it matter now. This is the anchor every downstream trade-off resolves against.}
+
+## Capabilities
+
+- **CAP-1**
+  - **intent:** {One sentence. "User or system can do X to achieve Y." WHAT, not HOW.}
+  - **success:** {Testable or demonstrable criterion. Something a test or a real demonstration can decide.}
+
+## Constraints
+
+- {A non-negotiable that bends design. If it doesn't rule anything out, it doesn't belong.}
+
+## Non-goals
+
+- {Explicit out-of-scope item. At least one. Stops downstream from filling the vacuum.}
+
+## Success signal
+
+- {One or two sentences. World-change moment, not dashboard. Concrete enough to write a test or run a demonstration against.}
+
+## Assumptions
+
+<!-- Optional. Omit this section entirely if empty. Inferred calls made without direct confirmation from the input. -->
+
+- {Statement of fact the Spec proceeded under, e.g. "Assumed mobile-first since input mentioned GPS but no platform."}
+
+## Open Questions
+
+<!-- Optional. Omit this section entirely if empty. Gaps the input did not resolve that need a human decision before downstream skills consume the Spec. -->
+
+- {Question phrased so a human can answer it, e.g. "Is offline playback in scope for CAP-2?"}
diff --git a/.agents/skills/bmad-spec/assets/stories-schema.md b/.agents/skills/bmad-spec/assets/stories-schema.md
new file mode 100644
index 00000000..0b1aaaf9
--- /dev/null
+++ b/.agents/skills/bmad-spec/assets/stories-schema.md
@@ -0,0 +1,44 @@
+# Stories schema
+
+`stories.yaml` — the optional output of Story Breakdown: a top-level YAML list, one entry per story, in execution order — stories run top to bottom. Sibling of `SPEC.md`, discovered by its fixed filename (same convention as `SPEC.md` and `.memlog.md`); never listed in `companions:`, never referenced from frontmatter.
+
+## Fields
+
+| Field | Type | Required | Meaning |
+|---|---|---|---|
+| `id` | string | yes | Identity, unique within the file. Compared for equality and used as a filename prefix; carries no other meaning. Use unpadded integers (`"1"`), or composites (`"3-2"`) when the epic sits inside a larger project spec. No zero-padding — execution order is the list order, not filename sort. Pinned once the story's spec file exists (see Update semantics); until then it may be renumbered. |
+| `title` | string, one line | yes | Display name; becomes the story's heading downstream. |
+| `description` | string | yes | What this story covers, pointing into `SPEC.md` — not a story spec. Keep it to two sentences; a writing rule, not something tooling enforces. |
+| `spec_checkpoint` | boolean | no (default `false`) | Set by the human at breakdown time; read only by the dispatching caller, never by the implementing dev skill. When true, a human reviews the story spec between planning and implementation. |
+| `done_checkpoint` | boolean | no (default `false`) | Caller-only, like `spec_checkpoint`. When true, dispatch pauses after this story completes, before anything further runs. |
+| `invoke_dev_with` | string | no (default `""`) | Free text appended verbatim to the prompt that dispatches this story; the implementing dev skill reads it as part of its prompt, and nothing else interprets it. If the text needs structure, put it inside the string. Which dev skill to invoke is the caller's configuration, never data in this file. |
+
+## Validity rules
+
+1. Every entry parses with all required fields; ids unique.
+2. Ids are prefix-free under the `<id>-` filename-matching convention: no id may equal another id plus a dash-suffix (`"3"` and `"3-2"` cannot coexist).
+3. No `status` field, ever.
+4. Ids are YAML strings, always quoted, containing only letters, digits, and dashes. An unquoted `id: 1` parses as a number and breaks string comparison; characters like `/` or `*` break the filename match.
+
+## Example
+
+```yaml
+- id: "1"
+  title: Add rate limiting to the public API
+  description: >-
+    Introduce a token-bucket limiter in front of the public endpoints;
+    return 429 with a Retry-After header on limit breach.
+  spec_checkpoint: true
+  invoke_dev_with: >-
+    Rate limit state must be shared across instances; use the existing
+    Redis client, not in-process memory.
+- id: "2"
+  title: Expose limiter metrics to the ops dashboard
+  description: >-
+    Emit per-route accept/reject counters the existing dashboard can
+    scrape; no new dashboard panels in this story.
+```
+
+## Update semantics
+
+Updates to `stories.yaml` go through Story Breakdown: append the change to `.memlog.md`, then re-derive. An id is pinned once its story spec file exists (any `stories/<id>-*.md` in the spec folder): a pinned story keeps its id through edits, its removal retires the id, and retired ids are never reassigned. Stories with no spec file yet may be renumbered, reordered, or removed freely on re-derive — typically so ids keep following list order. Never give a story an id that collides with a `stories/` file belonging to a different story.
diff --git a/.agents/skills/bmad-spec/customize.toml b/.agents/skills/bmad-spec/customize.toml
new file mode 100644
index 00000000..742ab8ac
--- /dev/null
+++ b/.agents/skills/bmad-spec/customize.toml
@@ -0,0 +1,52 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-spec.
+#
+# Override files (not edited here):
+#   {project-root}/_bmad/custom/bmad-spec.toml        (team)
+#   {project-root}/_bmad/custom/bmad-spec.user.toml   (personal)
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays: append
+
+# Steps to run before the standard activation (config load, greet).
+activation_steps_prepend = []
+
+# Steps to run after greet but before the operation begins.
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run.
+# Each entry is either a literal sentence, a skill prefixed with `skill:`,
+# or a `file:`-prefixed path/glob whose contents are loaded as facts.
+# Default points to a single top-level file; override in team/user TOML
+# to widen the scope (e.g. `_bmad/**/project-context.md`) if needed.
+persistent_facts = [
+  "file:{project-root}/project-context.md",
+]
+
+# Executed when the workflow completes. Scalar or array of instructions.
+on_complete = ""
+
+# Spec template. The five-field kernel skeleton. Override the path in
+# team/user TOML to enforce a different shape (e.g. a hypothesis field
+# for research initiatives, or a mechanics field for games).
+spec_template = "assets/spec-template.md"
+
+# Canonical filename for the kernel artifact inside the spec folder.
+# Uppercase by convention to signal "the central source of truth."
+spec_filename = "SPEC.md"
+
+# Output path for spec folders. Lands directly under {output_folder},
+# matching the long-term BMad direction of grouping artifacts as siblings
+# under {output_folder}/<type>/ rather than nested inside planning vs
+# implementation folders.
+spec_output_path = "{output_folder}/specs"
+
+# Run-folder pattern inside spec_output_path. Resolved against the
+# input-derived slug at activation. Same slug = same folder, so a
+# second invocation updates the existing spec in place (capability
+# IDs preserved). Override to add {date} or other components if a
+# fresh dated history is preferred.
+run_folder_pattern = "spec-{slug}"
diff --git a/.agents/skills/bmad-sprint-planning/SKILL.md b/.agents/skills/bmad-sprint-planning/SKILL.md
new file mode 100644
index 00000000..a2fec42e
--- /dev/null
+++ b/.agents/skills/bmad-sprint-planning/SKILL.md
@@ -0,0 +1,62 @@
+---
+name: bmad-sprint-planning
+description: 'Gate planning readiness, generate sprint status tracking from epics, summarize sprint progress, and validate or repair the tracking file. Use when the user says "run sprint planning", "generate sprint plan", "check implementation readiness", "show sprint status", "validate sprint status", or "fix sprint status"'
+---
+
+# Overview
+
+You are a senior developer about to commit to this plan. Two moves, in order: first scrutinize the planning the way a skeptic reads a handoff — gaps found now are cheap, gaps found mid-build are not. Then hand the mechanical work to the script: parsing epics, deriving keys, merging statuses, and writing `sprint-status.yaml` are deterministic jobs, not judgment calls. Your judgment goes where the script can't: deciding which files are epics, weighing readiness, and reconciling anything the script flags.
+
+## On Activation
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. On failure, read `{skill-root}/customize.toml` directly and use defaults.
+2. Execute each entry in `{workflow.activation_steps_prepend}` in order.
+3. Treat every entry in `{workflow.persistent_facts}` as foundational context for the rest of the run. Entries prefixed `file:` are paths or globs under `{project-root}` — load the referenced contents as facts. All other entries are facts verbatim.
+4. Load `{project-root}/_bmad/bmm/config.yaml` (and `config.user.yaml` if present). Resolve `{user_name}`, `{communication_language}`, `{document_output_language}`, `{project_name}`, `{planning_artifacts}`, `{implementation_artifacts}`, `{project_knowledge}` (skip gracefully if unset), `{date}`. Stay in `{communication_language}` for every turn, not just the greeting.
+5. Greet `{user_name}`, detect intent, and load only what that intent needs:
+   - **readiness** — check implementation readiness only: load `references/readiness-gate.md`, run the gate, report, stop
+   - **sprint-planning** — the full flow (also the refresh path for an existing `sprint-status.yaml`): load `references/readiness-gate.md`, then on PASS `references/generate-tracking.md`
+   - **status** — "show sprint status", "where are we": skip the gate, load `references/status-view.md`
+   - **validate** — check the tracking file's format: load `references/validate.md`
+   - **fix** — repair or rebuild a broken `sprint-status.yaml`: load `references/fix-sprint-status.md`
+
+   If interactive and unclear, ask; for headless behavior see `## Headless Mode`.
+
+Execute each entry in `{workflow.activation_steps_append}` in order.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding.
+
+## If the Script Fails
+
+This rule covers every intent: when `sprint_plan.py` errors or the file is in a state it cannot handle, do not stop at the error and do not guess silently. Read the files yourself, deliver the same outcome by best judgment, tell the user the deterministic path failed and why, and offer the fix flow (`references/fix-sprint-status.md`) to restore a file the script can work with.
+
+## On Completion
+
+Whatever the intent, close out in `{communication_language}` per the loaded reference, then run `{workflow.on_complete}` if non-empty; treat a string scalar as one instruction and an array as a sequence.
+
+## Headless Mode
+
+When invoked headless, do not ask. Run the gate and, unless intent was readiness-only, generate tracking. Ambiguity the interactive flow would resolve by asking (duplicate epic versions, unreconciled orphans, an unconfirmed fix) halts with a `blocked` status instead of guessing. End with a JSON response:
+
+```json
+{
+  "status": "complete",
+  "intent": "sprint-planning",
+  "gate": "PASS",
+  "status_file": "{implementation_artifacts}/sprint-status.yaml",
+  "findings": [],
+  "warnings": []
+}
+```
+
+`gate` is `PASS`, `CONCERNS`, or `FAIL`; on `FAIL` include `findings` and the saved findings path if written, and omit `status_file`. `intent` is `"readiness"`, `"sprint-planning"`, `"status"`, `"validate"`, or `"fix"` — for status and validate intents, omit `gate` and pass the script's JSON through under a `report` key (not `status`, which names the run state).
+
+## References
+
+- `scripts/sprint_plan.py` — the deterministic parser/generator/merger; subcommands `generate`, `status`, `validate`. Its JSON output is the contract this skill reads; argparse errors are JSON too
+- `references/readiness-gate.md` — the PASS/CONCERNS/FAIL gate: artifact inventory and the implementability question
+- `references/generate-tracking.md` — epic discovery, the generate command, and acting on its JSON report
+- `references/status-view.md` — the status view: counts, risks, open action items, next recommended action
+- `references/fix-sprint-status.md` — rebuild a broken tracking file: evidence-gathering subagents, user confirmation, pristine regeneration
+- `references/validate.md` — format validation of an existing `sprint-status.yaml`
+- `sprint-status-template.yaml` — the documented file format and status vocabulary; the script embeds the same block and the test suite pins the two copies together
diff --git a/.agents/skills/bmad-sprint-planning/customize.toml b/.agents/skills/bmad-sprint-planning/customize.toml
new file mode 100644
index 00000000..bc89e823
--- /dev/null
+++ b/.agents/skills/bmad-sprint-planning/customize.toml
@@ -0,0 +1,41 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-sprint-planning. Mirrors the
+# agent customization shape under the [workflow] namespace.
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+# Steps to run before the standard activation (config load, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Distinct from the runtime memory sidecar — these are static context
+# loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "All stories must include testable acceptance criteria."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Scalar: executed when the workflow reaches its final step,
+# after sprint-status.yaml is generated and validated. Override wins.
+# Leave empty for no custom post-completion behavior.
+
+on_complete = ""
diff --git a/.agents/skills/bmad-sprint-planning/references/fix-sprint-status.md b/.agents/skills/bmad-sprint-planning/references/fix-sprint-status.md
new file mode 100644
index 00000000..8aeb8c39
--- /dev/null
+++ b/.agents/skills/bmad-sprint-planning/references/fix-sprint-status.md
@@ -0,0 +1,30 @@
+# Fix Sprint Status
+
+Rebuild `sprint-status.yaml` to a pristine, script-valid state when it is broken, hand-mangled, drifted from reality, or the user simply asks to fix it. Inference determines what the state *should* be; the user confirms it; the script writes it. Never write without the confirmation.
+
+1. **Scope the damage.** Run `sprint_plan.py validate` and share what it found. If even the epic files are missing or unparseable, say so — there is nothing to rebuild tracking against until planning artifacts exist.
+
+2. **Determine the true state by inference.** This is judgment work — fan out subagents in parallel, each gathering one kind of evidence, and have each return proposed `key=status` pairs with the evidence behind them:
+   - **Epics** — read the epic files in `{planning_artifacts}`: the authoritative work breakdown (which epics, stories, and retrospectives should exist at all)
+   - **Story files** — scan `{implementation_artifacts}`: which stories have files on disk, and what their content says about progress (acceptance criteria checked off, completion notes, review sections)
+   - **Code evidence** — git history and the codebase: commits or merged work referencing story keys are evidence a story is done or in progress
+   - **The current file** — salvage everything credible from the existing `sprint-status.yaml`, especially `action_items`, even when its structure is broken
+
+3. **Reconcile into one proposed state.** Merge the evidence into a single table: key → proposed status, evidence, and anything uncertain. When evidence conflicts or is thin, prefer the lower status and flag it — a false `done` costs more than a false `in-progress`.
+
+4. **Confirm with `{user_name}`.** Show the table. Highlight every entry that differs from the current file — especially downgrades — and every low-confidence call. Adjust to their corrections. Headless: halt with `blocked` instead of confirming.
+
+5. **Write pristine.** One command, from the confirmed table:
+
+   ```
+   uv run {skill-root}/scripts/sprint_plan.py generate \
+     --epic-file <path> [...] \
+     --status-file {implementation_artifacts}/sprint-status.yaml \
+     --stories-dir {implementation_artifacts} \
+     --project "{project_name}" --date "{date}" \
+     --fresh --set <key>=<status> [--set <key>=<status> ...]
+   ```
+
+   `--fresh` rebuilds the document cleanly (canonical vocabulary, standard header) while still carrying `action_items`; `--set` applies the confirmed statuses and is the one path allowed to downgrade. Only confirmed entries that differ from the fresh defaults need a `--set`.
+
+6. **Verify.** Run `validate` again (expect `valid: true`) and present the status view summary so the user sees the repaired state.
diff --git a/.agents/skills/bmad-sprint-planning/references/generate-tracking.md b/.agents/skills/bmad-sprint-planning/references/generate-tracking.md
new file mode 100644
index 00000000..7633b905
--- /dev/null
+++ b/.agents/skills/bmad-sprint-planning/references/generate-tracking.md
@@ -0,0 +1,25 @@
+# Generate Tracking
+
+Discovery is your call; everything after it is the script's.
+
+1. Identify the epic files. The gate inventory already surfaced them — typically `epics.md`, `epic-*.md`, or a sharded `epics/` folder in `{planning_artifacts}`, but trust content over filename. If both a whole document and a sharded version exist, ask which is current rather than guessing.
+2. Run the script, passing every epic file:
+
+   ```
+   uv run {skill-root}/scripts/sprint_plan.py generate \
+     --epic-file <path> [--epic-file <path> ...] \
+     --status-file {implementation_artifacts}/sprint-status.yaml \
+     --stories-dir {implementation_artifacts} \
+     --project "{project_name}" --date "{date}"
+   ```
+
+   `{date}` must be `MM-DD-YYYY HH:MM` — the format the staleness check parses. The script owns parsing (`## Epic N:` / `### Story N.M: Title` → kebab-case keys; fenced code blocks ignored), ordering (epic, its stories, its retrospective), merging with any existing file (preserve advanced statuses, never downgrade; legacy v6 values like `drafted`/`contexted` are normalized to their modern meaning, never reset; `action_items`, custom keys, and user comments carried through; `project_key`/`tracking_system`/`story_location` kept from the existing file unless overridden by flag), story-file detection (a story file on disk floors its status at `ready-for-dev`), atomic writes, and post-write validation. It prints a JSON report. Add `--dry-run` to preview — the report's `in_sync`, `new_entries`, `dropped_orphans`, `illegal`, and `legacy_mapped` fields answer "is tracking in sync?" without writing.
+
+3. Read the JSON report and act on it — this is where judgment re-enters:
+   - `warnings` about unparsed Epic/Story-like headings mean the epic file deviates from the standard format. Show the user, fix the headings together (or accept the omission), and rerun.
+   - `dropped_orphans` are entries that existed in the old status file but match nothing in the epics — usually renames. Each carries its old status; reconcile with the user, then transplant by rerunning with `--set <new-key>=<old-status>`.
+   - If the epics defeat the parser entirely (a format the regexes can't see), fall back to building the file yourself against `sprint-status-template.yaml`, and tell the user the deterministic path didn't apply.
+
+## Report
+
+Present the result from the script's JSON in `{communication_language}`: file path, epic/story counts, status breakdown, anything upgraded from disk. Suggest next steps — review the file, `bmad-build` to start the first story, rerun this skill anytime to refresh after epics change.
diff --git a/.agents/skills/bmad-sprint-planning/references/readiness-gate.md b/.agents/skills/bmad-sprint-planning/references/readiness-gate.md
new file mode 100644
index 00000000..ce08a5aa
--- /dev/null
+++ b/.agents/skills/bmad-sprint-planning/references/readiness-gate.md
@@ -0,0 +1,20 @@
+# Readiness Gate
+
+Before generating any tracking, judge whether the plan can actually be built. If the user only asked to check readiness, this gate is the deliverable — report the verdict and stop.
+
+Inventory what planning actually exists: scan `{planning_artifacts}` and `{project_knowledge}` for intent and planning artifacts — briefs, PRFAQs, PRDs, specs, UX outputs, architecture, epics and stories. Identify documents by reading what they are, not by filename patterns; projects arrive with different artifact mixes and naming.
+
+Assess the plan as a whole against one question: **could a developer implement these epics without inventing decisions nothing records?**
+
+- Requirements and decisions in the intent artifacts trace forward into stories; stories trace back to recorded intent — flag orphans in both directions
+- Epics deliver user value and carry no forward dependencies; stories are independently completable
+- Architecture and UX decisions the stories rely on are recorded somewhere, not assumed
+- Conflicts between artifacts (a spec and an epic disagreeing) are surfaced, not silently resolved
+
+A missing document type is only a finding if stories depend on decisions nothing records — a project with no UX artifact and no UI stories is fine.
+
+Deliver a verdict:
+
+- **PASS** — state it in one line; for the full sprint-planning intent, continue with `generate-tracking.md`
+- **CONCERNS** — list them briefly with where each gap lives; ask `{user_name}` whether to proceed anyway or fix first
+- **FAIL** — the plan is not implementable as recorded. Present findings ordered by severity, name the skill that fixes each (the relevant plan skill, or `bmad-correct-course` for cross-cutting changes), offer to save the findings to `{planning_artifacts}/implementation-readiness.md`, and stop
diff --git a/.agents/skills/bmad-sprint-planning/references/status-view.md b/.agents/skills/bmad-sprint-planning/references/status-view.md
new file mode 100644
index 00000000..7ca84e72
--- /dev/null
+++ b/.agents/skills/bmad-sprint-planning/references/status-view.md
@@ -0,0 +1,14 @@
+# Status View
+
+When the user wants to know where the sprint stands ("show sprint status", "where are we"), run:
+
+```
+uv run {skill-root}/scripts/sprint_plan.py status \
+  --status-file {implementation_artifacts}/sprint-status.yaml --date "{date}"
+```
+
+`{date}` is `MM-DD-YYYY HH:MM`. The script computes everything: counts by status (legacy values like `drafted` mapped transparently and reported in `legacy_mapped`), risk flags (stale file, orphaned stories, in-progress epics without stories, stories waiting in review, unrecognized keys), open action items from retrospectives, and the next recommended action by fixed priority — resume in-progress → review what's in review → start the next ready or backlog story → run an open retrospective → all done. If the file is missing, the script says so — offer to run sprint planning to create it.
+
+Render the JSON as a compact summary in `{communication_language}`: counts, risks, open action items, and the recommendation with its story key. Offer to run the recommended skill. Surface `illegal` and `unrecognized` entries and the script's `warnings` (malformed action items land there); if the user gives corrections, apply them via the fix flow rather than ad-hoc edits. No time estimates — status, risks, and next steps only.
+
+If the script errors — malformed YAML, a hand-edited structure it can't parse, anything — do not stop at the error. Read `sprint-status.yaml` yourself, apply best judgment to give the user the same summary (counts, risks, next recommended action), tell them the deterministic path failed and why, and offer the fix flow (`fix-sprint-status.md`) so the script works next time.
diff --git a/.agents/skills/bmad-sprint-planning/references/validate.md b/.agents/skills/bmad-sprint-planning/references/validate.md
new file mode 100644
index 00000000..c0cc7d41
--- /dev/null
+++ b/.agents/skills/bmad-sprint-planning/references/validate.md
@@ -0,0 +1,10 @@
+# Validate
+
+When the user asks whether `sprint-status.yaml` is well-formed, run:
+
+```
+uv run {skill-root}/scripts/sprint_plan.py validate \
+  --status-file {implementation_artifacts}/sprint-status.yaml
+```
+
+Never writes; exits 0 whether valid or not. Report `valid` in one line. If `problems` is non-empty, list them plainly (each names the key or field at fault) and offer the fix flow (`fix-sprint-status.md`). If `legacy_mapped` is non-empty, note the file still uses v6 status names and that any regenerate will rewrite them to the modern vocabulary — progress is preserved either way.
diff --git a/.agents/skills/bmad-sprint-planning/scripts/sprint_plan.py b/.agents/skills/bmad-sprint-planning/scripts/sprint_plan.py
new file mode 100644
index 00000000..bc486ba8
--- /dev/null
+++ b/.agents/skills/bmad-sprint-planning/scripts/sprint_plan.py
@@ -0,0 +1,697 @@
+# /// script
+# requires-python = ">=3.10"
+# dependencies = ["ruamel.yaml>=0.18"]
+# ///
+"""Parse epic files and deterministically generate or refresh sprint-status.yaml.
+
+Prints ONLY JSON to stdout — argparse failures included. Errors are emitted as
+JSON to stdout with a non-zero exit code. Writes are atomic (temp file, fsync,
+``os.replace``) and the original file bytes are restored atomically if
+post-write validation fails.
+
+Subcommands:
+  generate  Parse epics, merge with any existing status file, write the result.
+            --dry-run reports (including drift: in_sync, illegal, orphans)
+            without writing. --fresh ignores existing statuses for a pristine
+            rebuild; --set key=status applies explicit, user-confirmed statuses
+            on top — the repair path is allowed to downgrade.
+  status    Summarize an existing status file: counts, risks, open action
+            items, and the next recommended action. No writes.
+  validate  Report whether an existing status file is structurally valid:
+            parseable, recognized keys, legal statuses, well-formed
+            action_items. No writes; exit 0 whether valid or not.
+
+The LLM decides *which* files are epics (discovery is judgment); this script
+owns everything after that decision: parsing, key derivation, ordering, status
+preservation, story-file detection, action-item carry-over, and validation.
+Legacy v6 statuses (drafted, contexted) are normalized on read everywhere, so
+they merge and count by their modern meaning and are reported, never reset.
+"""
+
+import argparse
+import hashlib
+import io
+import json
+import os
+import re
+import sys
+import tempfile
+from pathlib import Path
+
+from ruamel.yaml import YAML
+from ruamel.yaml.comments import CommentedMap
+
+EPIC_RE = re.compile(r"^#{1,3}\s*Epic\s+(\d+)\s*:?\s*(.*?)\s*#*\s*$", re.IGNORECASE)
+STORY_RE = re.compile(
+    r"^#{2,4}\s*Story\s+(\d+)\.(\d+[a-z]?)\s*:?\s*(.*?)\s*#*\s*$", re.IGNORECASE
+)
+# Heading lines that mention Epic/Story but failed the strict patterns above.
+SUSPECT_RE = re.compile(r"^#{1,4}\s.*\b(?:epic|story)\b", re.IGNORECASE)
+FENCE_RE = re.compile(r"^\s{0,3}(?:```|~~~)")
+
+# The key grammar for sprint-status.yaml. The trailing [a-z]? matches
+# split-story keys like 2-6a-...; bmad-retrospective's sprint_status.py reads
+# the same file with the same grammar.
+EPIC_KEY_RE = re.compile(r"^epic-(\d+)$")
+RETRO_KEY_RE = re.compile(r"^epic-(\d+)-retrospective$")
+STORY_KEY_RE = re.compile(r"^(\d+)-(\d+)([a-z]?)-.+")
+
+STORY_RANK = {"backlog": 0, "ready-for-dev": 1, "in-progress": 2, "review": 3, "done": 4}
+EPIC_RANK = {"backlog": 0, "in-progress": 1, "done": 2}
+RETRO_RANK = {"optional": 0, "done": 1}
+RANKS = {"epic": EPIC_RANK, "story": STORY_RANK, "retro": RETRO_RANK}
+ACTION_STATUSES = ("open", "in-progress", "done")
+
+# v6 wrote these; they still exist in the wild (v6-shims/bmad-create-story
+# actively writes 'contexted'). Normalized on every read so no subcommand ever
+# treats a valid legacy file as illegal or resets its progress.
+LEGACY_STATUS = {"drafted": "ready-for-dev", "contexted": "in-progress"}
+
+STALE_DAYS_DEFAULT = 7
+DATE_FORMAT = "%m-%d-%Y %H:%M"
+# Hand-edited files drift toward ISO stamps; accept them rather than silently
+# disabling the staleness check.
+STAMP_FORMATS = (DATE_FORMAT, "%Y-%m-%d %H:%M", "%Y-%m-%d")
+
+# Kept byte-identical (modulo the leading "# ") with the STATUS DEFINITIONS
+# block in sprint-status-template.yaml; test_sprint_plan.py asserts the two
+# never drift.
+HEADER_COMMENT = """\
+STATUS DEFINITIONS:
+==================
+Epic Status:
+  - backlog: Epic not yet started
+  - in-progress: Epic actively being worked on
+  - done: All stories in epic completed
+
+Story Status:
+  - backlog: Story only exists in epic file
+  - ready-for-dev: Story file created, ready for development
+  - in-progress: Developer actively working on implementation
+  - review: Implementation complete, ready for review
+  - done: Story completed
+
+Retrospective Status:
+  - optional: Can be completed but not required
+  - done: Retrospective has been completed
+
+Action Item Status:
+  - open: Committed during a retrospective, not yet addressed
+  - in-progress: Actively being worked on
+  - done: Completed
+
+WORKFLOW NOTES:
+===============
+- Epic transitions to 'in-progress' automatically when its first story starts (via build's sprint sync)
+- Stories can be worked in parallel if team capacity allows
+- Developer typically creates the next story after the previous one is 'done' to incorporate learnings
+- Dev moves story to 'review', then runs code-review (fresh context, different LLM recommended)
+- Retrospective appends its action items to action_items; the status view surfaces open ones
+"""
+
+
+def _fail(message, **extra):
+    print(json.dumps({"ok": False, "error": message, **extra}, default=str))
+    sys.exit(1)
+
+
+class JsonArgumentParser(argparse.ArgumentParser):
+    """Emit argparse failures on the JSON-only stdout contract, not usage text.
+
+    Built with ``add_help=False`` everywhere: the built-in help action prints
+    plain usage to stdout with exit 0, which would break the machine consumer
+    this script serves. ``-h`` therefore routes through ``error()`` as an
+    ordinary unrecognized argument; the skill's SKILL.md carries the usage a
+    human needs.
+    """
+
+    def error(self, message):
+        print(json.dumps({"ok": False, "error": f"argument error: {message}"}))
+        sys.exit(2)
+
+
+def _slug(text, maxlen=60):
+    # Unicode-aware: a non-Latin title must keep its own characters in the key
+    # rather than every such story collapsing onto one shared placeholder.
+    slug = re.sub(r"[^\w]+", "-", str(text).lower(), flags=re.UNICODE).strip("-")
+    slug = slug[:maxlen].strip("-")
+    if not slug:
+        # Nothing sluggable (punctuation/emoji only): a short content hash keeps
+        # the key deterministic and distinct instead of a bare "untitled".
+        slug = hashlib.sha256(str(text).encode("utf-8")).hexdigest()[:8]
+    return slug
+
+
+def classify_key(key):
+    """Return (kind, epic_num) for a recognized key, else None."""
+    m = RETRO_KEY_RE.match(key)
+    if m:
+        return "retro", int(m.group(1))
+    m = EPIC_KEY_RE.match(key)
+    if m:
+        return "epic", int(m.group(1))
+    m = STORY_KEY_RE.match(key)
+    if m:
+        return "story", int(m.group(1))
+    return None
+
+
+def _story_sort_key(key):
+    m = STORY_KEY_RE.match(key)
+    if not m:
+        return (10**9, 10**9, "", key)
+    return (int(m.group(1)), int(m.group(2)), m.group(3), key)
+
+
+def _normalize(raw):
+    """Map a raw status through the legacy vocabulary. Returns (status, was_legacy)."""
+    status = LEGACY_STATUS.get(raw, raw)
+    return status, raw in LEGACY_STATUS
+
+
+def parse_epics(paths):
+    """Return (entries, warnings). Entries are (key, kind, epic_num) in file order."""
+    epics = {}  # epic_num -> [story keys in order]
+    warnings = []
+    for path in paths:
+        try:
+            lines = Path(path).read_text(encoding="utf-8").splitlines()
+        except OSError as exc:
+            _fail(f"cannot read epic file {path}: {exc}")
+        in_fence = False
+        for lineno, line in enumerate(lines, 1):
+            if FENCE_RE.match(line):
+                in_fence = not in_fence
+                continue
+            if in_fence:
+                continue
+            epic_m = EPIC_RE.match(line)
+            if epic_m:
+                epics.setdefault(int(epic_m.group(1)), [])
+                continue
+            story_m = STORY_RE.match(line)
+            if story_m:
+                epic_num = int(story_m.group(1))
+                story_num = story_m.group(2)
+                key = f"{epic_num}-{story_num}-{_slug(story_m.group(3))}"
+                stories = epics.setdefault(epic_num, [])
+                if key in stories:
+                    warnings.append(f"duplicate story heading '{key}' at {path}:{lineno}")
+                else:
+                    stories.append(key)
+                continue
+            if SUSPECT_RE.match(line):
+                warnings.append(f"unparsed Epic/Story-like heading at {path}:{lineno}: {line.strip()}")
+    entries = []
+    for epic_num in sorted(epics):
+        entries.append((f"epic-{epic_num}", "epic", epic_num))
+        for story_key in epics[epic_num]:
+            entries.append((story_key, "story", epic_num))
+        entries.append((f"epic-{epic_num}-retrospective", "retro", epic_num))
+    return entries, warnings
+
+
+def _make_yaml():
+    yaml = YAML(typ="rt")
+    yaml.preserve_quotes = True
+    # Pin the emitter to the indentation the sprint-status template ships with.
+    # Without this, ruamel re-dumps block sequences at its own default offset and
+    # every write silently de-indents pre-existing, untouched action_items.
+    yaml.indent(mapping=2, sequence=4, offset=2)
+    yaml.encoding = "utf-8"
+    return yaml
+
+
+def _load_existing(path):
+    yaml = _make_yaml()
+    if not Path(path).exists():
+        return yaml, None
+    try:
+        with io.open(path, "r", encoding="utf-8") as fh:
+            data = yaml.load(fh)
+    except Exception as exc:
+        _fail(f"existing status file is not valid YAML: {exc}", status_file=str(path))
+    if data is not None and not isinstance(data, dict):
+        _fail(
+            f"existing status file is valid YAML but not a mapping (got {type(data).__name__})",
+            status_file=str(path),
+        )
+    return yaml, data
+
+
+def _merge_status(kind, computed, existing_raw, key, warnings, report):
+    """Return the higher-ranked of computed/existing; never downgrade."""
+    rank = RANKS[kind]
+    if existing_raw is None:
+        return computed
+    existing, was_legacy = _normalize(existing_raw)
+    if was_legacy:
+        report["legacy_mapped"].append({"key": key, "from": existing_raw, "to": existing})
+    if existing not in rank:
+        warnings.append(f"illegal status '{existing_raw}' on '{key}' replaced with '{computed}'")
+        report["illegal"].append({"key": key, "status": existing_raw})
+        return computed
+    return existing if rank[existing] >= rank[computed] else computed
+
+
+def build_status(entries, existing_data, stories_dir, warnings):
+    """Return (development_status CommentedMap, merge report dict)."""
+    existing_status = {}
+    if existing_data is not None:
+        existing_status = dict(existing_data.get("development_status") or {})
+    report = {
+        "new_entries": [],
+        "preserved": 0,
+        "changed": 0,
+        "upgraded_from_disk": [],
+        "dropped_orphans": [],
+        "legacy_mapped": [],
+        "illegal": [],
+    }
+    # One directory scan instead of a stat() per story.
+    story_files = set()
+    if stories_dir and Path(stories_dir).is_dir():
+        story_files = {p.name for p in Path(stories_dir).glob("*.md")}
+    dev = CommentedMap()
+    first_epic = True
+    for key, kind, _epic_num in entries:
+        default = {"epic": "backlog", "story": "backlog", "retro": "optional"}[kind]
+        computed = default
+        if kind == "story" and f"{key}.md" in story_files:
+            computed = "ready-for-dev"
+        merged = _merge_status(kind, computed, existing_status.get(key), key, warnings, report)
+        if key not in existing_status:
+            report["new_entries"].append(key)
+        elif merged == _normalize(existing_status[key])[0]:
+            report["preserved"] += 1
+        else:
+            report["changed"] += 1
+        if kind == "story" and computed == "ready-for-dev" and existing_status.get(key) in (None, "backlog"):
+            report["upgraded_from_disk"].append(key)
+        dev[key] = merged
+        if kind == "epic" and not first_epic:
+            dev.yaml_set_comment_before_after_key(key, before="\n")
+        if kind == "epic":
+            first_epic = False
+    computed_keys = {key for key, _, _ in entries}
+    # Old statuses ride along so the LLM can transplant them after a rename —
+    # the values would otherwise be destroyed by the write.
+    report["dropped_orphans"] = [
+        {"key": k, "status": existing_status[k]}
+        for k in existing_status
+        if k not in computed_keys
+    ]
+    report["in_sync"] = (
+        not report["new_entries"]
+        and not report["dropped_orphans"]
+        and not report["illegal"]
+        and not report["legacy_mapped"]
+        and report["changed"] == 0
+    )
+    return dev, report
+
+
+def _counts(dev):
+    counts = {}
+    for value in dev.values():
+        counts[value] = counts.get(value, 0) + 1
+    return counts
+
+
+def _dump_bytes(yaml, doc):
+    """Serialize before any file is touched, so a dump failure cannot leave a
+    partial file anywhere."""
+    buf = io.BytesIO()
+    yaml.dump(doc, buf)
+    return buf.getvalue()
+
+
+def _atomic_write(path, payload, mode=None):
+    """Replace ``path``'s contents with ``payload`` atomically.
+
+    Temp file alongside the target, fsynced, taking the target's permission
+    bits (mkstemp creates 0600, which would silently narrow the file), then
+    renamed over it. ``path`` is resolved through symlinks first: renaming onto
+    a symlink would detach the link and leave the real file stale.
+    """
+    path = os.path.realpath(path)
+    directory = os.path.dirname(path) or "."
+    os.makedirs(directory, exist_ok=True)
+    fd, tmp = tempfile.mkstemp(prefix=".sprint-status-", suffix=".tmp", dir=directory)
+    try:
+        with os.fdopen(fd, "wb") as fh:
+            fh.write(payload)
+            fh.flush()
+            os.fsync(fh.fileno())
+        if mode is not None:
+            os.chmod(tmp, mode)
+        os.replace(tmp, path)
+    except BaseException:
+        try:
+            os.unlink(tmp)
+        except OSError:
+            pass
+        raise
+
+
+def _parse_sets(pairs, valid_keys):
+    """Validate --set key=status pairs against the generated plan and vocabulary."""
+    parsed = []
+    for pair in pairs:
+        key, sep, status = pair.partition("=")
+        if not sep or not key or not status:
+            _fail(f"--set expects key=status, got '{pair}'")
+        if key not in valid_keys:
+            _fail(f"--set key '{key}' is not in the generated plan", valid_keys=sorted(valid_keys))
+        kind, _ = classify_key(key)
+        if status not in RANKS[kind]:
+            _fail(
+                f"--set status '{status}' is not legal for {kind} '{key}'",
+                legal=sorted(RANKS[kind]),
+            )
+        parsed.append((key, status))
+    return parsed
+
+
+def cmd_generate(args):
+    entries, warnings = parse_epics(args.epic_file)
+    if not entries:
+        _fail("no epics or stories parsed from the given epic files", epic_files=args.epic_file)
+    yaml, existing = _load_existing(args.status_file)
+    status_path = Path(args.status_file)
+    original_bytes = status_path.read_bytes() if status_path.exists() else None
+    original_mode = (os.stat(status_path).st_mode & 0o777) if status_path.exists() else None
+
+    merge_source = None if args.fresh else existing
+    dev, report = build_status(entries, merge_source, args.stories_dir, warnings)
+
+    # Explicit, user-confirmed statuses (the fix flow). Applied last: repair is
+    # the one path allowed to downgrade.
+    explicit = _parse_sets(args.set or [], set(dev.keys()))
+    for key, status in explicit:
+        dev[key] = status
+    report["explicit_set"] = [f"{k}={s}" for k, s in explicit]
+
+    def _meta(field, arg_value, default):
+        if arg_value is not None:
+            return arg_value
+        if existing is not None and existing.get(field):
+            return str(existing[field])
+        return default
+
+    generated = args.date
+    if existing is not None and existing.get("generated"):
+        generated = str(existing["generated"])
+
+    if existing is not None and not args.fresh:
+        # Round-trip the existing document: unknown top-level keys and their
+        # comments survive; only the managed fields and development_status are
+        # replaced.
+        doc = existing
+    else:
+        doc = CommentedMap()
+        doc.yaml_set_start_comment(HEADER_COMMENT)
+    doc["generated"] = generated
+    doc["last_updated"] = args.date
+    doc["project"] = args.project
+    doc["project_key"] = _meta("project_key", args.project_key, "NOKEY")
+    doc["tracking_system"] = _meta("tracking_system", args.tracking_system, "file-system")
+    doc["story_location"] = _meta("story_location", args.story_location, args.stories_dir)
+    doc["development_status"] = dev
+    if "action_items" not in doc and existing is not None and existing.get("action_items") is not None:
+        doc["action_items"] = existing["action_items"]
+        doc.yaml_set_comment_before_after_key(
+            "action_items",
+            before="\nAction items committed during retrospectives (section created by the retrospective workflow)",
+        )
+
+    result = {
+        "ok": True,
+        "action": "generate",
+        "status_file": str(args.status_file),
+        "dry_run": bool(args.dry_run),
+        "fresh": bool(args.fresh),
+        "epics": sum(1 for _, kind, _ in entries if kind == "epic"),
+        "stories": sum(1 for _, kind, _ in entries if kind == "story"),
+        "counts": _counts(dev),
+        "generated": generated,
+        "last_updated": args.date,
+        "warnings": warnings,
+        **report,
+    }
+
+    if args.dry_run:
+        print(json.dumps(result, default=str))
+        return
+
+    try:
+        payload = _dump_bytes(yaml, doc)
+        _atomic_write(args.status_file, payload, original_mode)
+        verify_yaml = _make_yaml()
+        with io.open(args.status_file, "r", encoding="utf-8") as fh:
+            reread = verify_yaml.load(fh)
+        if dict(reread.get("development_status") or {}) != {k: v for k, v in dev.items()}:
+            raise ValueError("development_status mismatch after write")
+        for field in ("generated", "last_updated", "project"):
+            if str(reread.get(field)) != str(doc[field]):
+                raise ValueError(f"{field} mismatch after write")
+    except Exception as exc:
+        if original_bytes is not None:
+            try:
+                _atomic_write(args.status_file, original_bytes, original_mode)
+                restored = True
+            except Exception:
+                restored = False
+        else:
+            Path(args.status_file).unlink(missing_ok=True)
+            restored = True
+        _fail(f"write or validation failed, original {'restored' if restored else 'NOT restored'}: {exc}",
+              restored=restored)
+    print(json.dumps(result, default=str))
+
+
+def _parse_stamp(value):
+    from datetime import datetime
+
+    for fmt in STAMP_FORMATS:
+        try:
+            return datetime.strptime(str(value), fmt)
+        except ValueError:
+            continue
+    return None
+
+
+def cmd_status(args):
+    from datetime import timedelta
+
+    _, data = _load_existing(args.status_file)
+    if data is None:
+        _fail("status file does not exist — run sprint planning to generate it",
+              status_file=str(args.status_file))
+    dev = dict(data.get("development_status") or {})
+    if not dev:
+        _fail("development_status missing or empty — re-run sprint planning",
+              status_file=str(args.status_file))
+
+    warnings = []
+    counts = {"story": {}, "epic": {}, "retro": {}}
+    by_status = {}
+    legacy_mapped, illegal, unrecognized = [], [], []
+    epic_nums, story_epic_nums = set(), set()
+    epic_status, retro_status = {}, {}
+    for key, raw in dev.items():
+        key = str(key)
+        parsed = classify_key(key)
+        if parsed is None:
+            unrecognized.append({"key": key, "status": raw})
+            continue
+        kind, epic_num = parsed
+        status, was_legacy = _normalize(raw)
+        if was_legacy:
+            legacy_mapped.append({"key": key, "from": raw, "to": status})
+        if status not in RANKS[kind]:
+            illegal.append({"key": key, "status": raw})
+            continue
+        counts[kind][status] = counts[kind].get(status, 0) + 1
+        if kind == "story":
+            by_status.setdefault(status, []).append(key)
+            story_epic_nums.add(epic_num)
+        elif kind == "epic":
+            epic_nums.add(epic_num)
+            epic_status[epic_num] = status
+        else:
+            retro_status[epic_num] = status
+    for stories in by_status.values():
+        stories.sort(key=_story_sort_key)
+
+    action_items = data.get("action_items") or []
+    open_items = []
+    for i, item in enumerate(action_items):
+        if not isinstance(item, dict):
+            warnings.append(f"action_items[{i}] is not a mapping and was skipped: {item!r}")
+            continue
+        status = item.get("status")
+        if status not in ACTION_STATUSES:
+            warnings.append(f"action_items[{i}] has a missing or unknown status ({status!r})")
+            continue
+        if status in ("open", "in-progress"):
+            open_items.append({k: item.get(k) for k in ("epic", "action", "owner", "status")})
+
+    risks = []
+    stamp = data.get("last_updated") or data.get("generated")
+    if args.date and stamp:
+        now, then = _parse_stamp(args.date), _parse_stamp(stamp)
+        if now is None or then is None:
+            warnings.append(
+                f"timestamp format not recognized (--date {args.date!r}, file {stamp!r}); "
+                "staleness check skipped"
+            )
+        elif now - then > timedelta(days=args.stale_days):
+            risks.append(f"sprint-status.yaml may be stale (last updated {stamp})")
+    for stories in by_status.values():
+        for key in stories:
+            num = classify_key(key)[1]
+            if num not in epic_nums:
+                risks.append(f"orphaned story '{key}' has no epic-{num} entry")
+    for num, status in epic_status.items():
+        if status == "in-progress" and num not in story_epic_nums:
+            risks.append(f"in-progress epic 'epic-{num}' has no stories")
+    if by_status.get("review"):
+        risks.append(f"{len(by_status['review'])} story(ies) in review — run bmad-code-review")
+    if unrecognized:
+        risks.append(f"{len(unrecognized)} unrecognized key(s) in development_status — run validate")
+
+    recommendation = None
+    if by_status.get("in-progress"):
+        recommendation = {"skill": "bmad-build", "story_key": by_status["in-progress"][0],
+                          "reason": "resume the in-progress story"}
+    elif by_status.get("review"):
+        recommendation = {"skill": "bmad-code-review", "story_key": by_status["review"][0],
+                          "reason": "review the completed implementation"}
+    elif by_status.get("ready-for-dev"):
+        recommendation = {"skill": "bmad-build", "story_key": by_status["ready-for-dev"][0],
+                          "reason": "start the next ready story"}
+    elif by_status.get("backlog"):
+        recommendation = {"skill": "bmad-build", "story_key": by_status["backlog"][0],
+                          "reason": "start the first backlog story"}
+    else:
+        optional_retros = sorted(num for num, status in retro_status.items() if status == "optional")
+        if optional_retros:
+            recommendation = {"skill": "bmad-retrospective", "story_key": None,
+                              "reason": f"all stories done — epic-{optional_retros[0]}-retrospective is still open"}
+
+    print(json.dumps({
+        "ok": True, "action": "status", "status_file": str(args.status_file),
+        "project": data.get("project"), "project_key": data.get("project_key"),
+        "tracking_system": data.get("tracking_system"),
+        "generated": data.get("generated"), "last_updated": data.get("last_updated"),
+        "stories": counts["story"], "epics": counts["epic"], "retrospectives": counts["retro"],
+        "legacy_mapped": legacy_mapped, "illegal": illegal, "unrecognized": unrecognized,
+        "open_action_items": open_items, "risks": risks, "warnings": warnings,
+        "recommendation": recommendation,
+        "all_done": recommendation is None,
+    }, default=str))
+
+
+def cmd_validate(args):
+    problems = []
+    legacy_mapped = []
+    path = Path(args.status_file)
+    if not path.exists():
+        print(json.dumps({
+            "ok": True, "action": "validate", "status_file": str(args.status_file),
+            "valid": False, "problems": ["status file does not exist"], "legacy_mapped": [],
+        }))
+        return
+    yaml = _make_yaml()
+    try:
+        with io.open(args.status_file, "r", encoding="utf-8") as fh:
+            data = yaml.load(fh)
+    except Exception as exc:
+        print(json.dumps({
+            "ok": True, "action": "validate", "status_file": str(args.status_file),
+            "valid": False, "problems": [f"not valid YAML: {exc}"], "legacy_mapped": [],
+        }, default=str))
+        return
+    if not isinstance(data, dict):
+        problems.append(f"top level is not a mapping (got {type(data).__name__})")
+    else:
+        for field in ("generated", "last_updated", "project", "development_status"):
+            if data.get(field) is None:
+                problems.append(f"missing required key '{field}'")
+        for field in ("generated", "last_updated"):
+            value = data.get(field)
+            if value is not None and _parse_stamp(value) is None:
+                problems.append(f"'{field}' timestamp {str(value)!r} does not match '{DATE_FORMAT}'")
+        dev = data.get("development_status")
+        if dev is not None and not isinstance(dev, dict):
+            problems.append("development_status is not a mapping")
+        elif dev:
+            for key, raw in dev.items():
+                parsed = classify_key(str(key))
+                if parsed is None:
+                    problems.append(f"unrecognized key '{key}' (expected epic-N, N-M-slug, or epic-N-retrospective)")
+                    continue
+                kind, _ = parsed
+                status, was_legacy = _normalize(raw)
+                if was_legacy:
+                    legacy_mapped.append({"key": str(key), "from": raw, "to": status})
+                if status not in RANKS[kind]:
+                    problems.append(f"illegal {kind} status {str(raw)!r} on '{key}'")
+        elif isinstance(data.get("development_status"), dict):
+            problems.append("development_status is empty")
+        items = data.get("action_items")
+        if items is not None:
+            if not isinstance(items, list):
+                problems.append("action_items is not a list")
+            else:
+                for i, item in enumerate(items):
+                    if not isinstance(item, dict):
+                        problems.append(f"action_items[{i}] is not a mapping")
+                    elif item.get("status") not in ACTION_STATUSES:
+                        problems.append(
+                            f"action_items[{i}] has a missing or unknown status ({item.get('status')!r})"
+                        )
+    print(json.dumps({
+        "ok": True, "action": "validate", "status_file": str(args.status_file),
+        "valid": not problems, "problems": problems, "legacy_mapped": legacy_mapped,
+    }, default=str))
+
+
+def build_parser():
+    parser = JsonArgumentParser(prog="sprint_plan.py", add_help=False)
+    sub = parser.add_subparsers(dest="command", required=True, parser_class=JsonArgumentParser)
+
+    gen = sub.add_parser("generate", add_help=False)
+    gen.add_argument("--epic-file", action="append", required=True)
+    gen.add_argument("--status-file", required=True)
+    gen.add_argument("--stories-dir", required=True)
+    gen.add_argument("--project", required=True)
+    gen.add_argument("--date", required=True)
+    gen.add_argument("--project-key", default=None)
+    gen.add_argument("--tracking-system", default=None)
+    gen.add_argument("--story-location", default=None)
+    gen.add_argument("--dry-run", action="store_true")
+    gen.add_argument("--fresh", action="store_true")
+    gen.add_argument("--set", action="append", metavar="KEY=STATUS")
+    gen.set_defaults(func=cmd_generate)
+
+    st = sub.add_parser("status", add_help=False)
+    st.add_argument("--status-file", required=True)
+    st.add_argument("--date", default=None)
+    st.add_argument("--stale-days", type=int, default=STALE_DAYS_DEFAULT)
+    st.set_defaults(func=cmd_status)
+
+    val = sub.add_parser("validate", add_help=False)
+    val.add_argument("--status-file", required=True)
+    val.set_defaults(func=cmd_validate)
+    return parser
+
+
+def main(argv=None):
+    args = build_parser().parse_args(argv)
+    args.func(args)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/.agents/skills/bmad-sprint-planning/scripts/tests/test_sprint_plan.py b/.agents/skills/bmad-sprint-planning/scripts/tests/test_sprint_plan.py
new file mode 100644
index 00000000..253b2207
--- /dev/null
+++ b/.agents/skills/bmad-sprint-planning/scripts/tests/test_sprint_plan.py
@@ -0,0 +1,524 @@
+# /// script
+# requires-python = ">=3.10"
+# dependencies = ["pytest>=8.0", "ruamel.yaml>=0.18"]
+# ///
+"""Tests for sprint_plan.py — deterministic sprint-status generation.
+
+Run: uv run scripts/tests/test_sprint_plan.py
+ or: uv run --with pytest --with ruamel.yaml -m pytest scripts/tests/test_sprint_plan.py
+"""
+
+import importlib.util
+import json
+import sys
+from pathlib import Path
+
+import pytest
+from ruamel.yaml import YAML
+
+SCRIPT = Path(__file__).resolve().parents[1] / "sprint_plan.py"
+TEMPLATE = Path(__file__).resolve().parents[2] / "sprint-status-template.yaml"
+
+spec = importlib.util.spec_from_file_location("sprint_plan", SCRIPT)
+mod = importlib.util.module_from_spec(spec)
+spec.loader.exec_module(mod)
+
+EPICS_FIXTURE = """\
+# Project Epics
+
+## Epic 1: Foundation
+Some prose.
+
+### Story 1.1: User Authentication
+Acceptance criteria...
+
+### Story 1.2: Account Management
+
+## Epic 2: Chat
+### Story 2.1: Personality System
+### Story 2.6a: Split Story, With Punctuation!
+"""
+
+DATE = "08-01-2026 14:30"
+
+
+def run_generate(tmp_path, epics_text=EPICS_FIXTURE, existing=None, stories=(), extra=()):
+    epic_file = tmp_path / "epics.md"
+    epic_file.write_text(epics_text, encoding="utf-8")
+    status_file = tmp_path / "impl" / "sprint-status.yaml"
+    if existing is not None:
+        status_file.parent.mkdir(parents=True, exist_ok=True)
+        status_file.write_text(existing, encoding="utf-8")
+    stories_dir = tmp_path / "impl"
+    stories_dir.mkdir(parents=True, exist_ok=True)
+    for name in stories:
+        (stories_dir / f"{name}.md").write_text("story", encoding="utf-8")
+    argv = [
+        "generate", "--epic-file", str(epic_file), "--status-file", str(status_file),
+        "--stories-dir", str(stories_dir), "--project", "My Project", "--date", DATE,
+        *extra,
+    ]
+    mod.main(argv)
+    return status_file
+
+
+def load(status_file):
+    yaml = YAML()
+    with open(status_file, encoding="utf-8") as fh:
+        return yaml.load(fh)
+
+
+def out_json(capsys):
+    return json.loads(capsys.readouterr().out)
+
+
+def test_fresh_generate_orders_and_defaults(tmp_path, capsys):
+    status_file = run_generate(tmp_path)
+    result = out_json(capsys)
+    data = load(status_file)
+    keys = list(data["development_status"].keys())
+    assert keys == [
+        "epic-1", "1-1-user-authentication", "1-2-account-management", "epic-1-retrospective",
+        "epic-2", "2-1-personality-system", "2-6a-split-story-with-punctuation", "epic-2-retrospective",
+    ]
+    assert data["development_status"]["epic-1"] == "backlog"
+    assert data["development_status"]["1-1-user-authentication"] == "backlog"
+    assert data["development_status"]["epic-1-retrospective"] == "optional"
+    assert data["project"] == "My Project"
+    assert data["generated"] == DATE and data["last_updated"] == DATE
+    assert result["ok"] and result["epics"] == 2 and result["stories"] == 4
+    text = status_file.read_text(encoding="utf-8")
+    assert "STATUS DEFINITIONS" in text
+    assert "\n\n  epic-2:" in text  # blank line between epic groups
+
+
+def test_header_comment_matches_template():
+    """The template's STATUS DEFINITIONS block and the script's HEADER_COMMENT
+    are two copies of one contract; this pins them together."""
+    lines = TEMPLATE.read_text(encoding="utf-8").splitlines()
+    start = lines.index("# STATUS DEFINITIONS:")
+    block = []
+    for line in lines[start:]:
+        if not line.startswith("#"):
+            break
+        block.append(line[2:] if line.startswith("# ") else line[1:])
+    assert "\n".join(block) + "\n" == mod.HEADER_COMMENT
+
+
+EXISTING = """\
+generated: 01-01-2026 09:00
+last_updated: 01-01-2026 09:00
+project: My Project
+project_key: NOKEY
+tracking_system: file-system
+story_location: impl
+
+development_status:
+  epic-1: in-progress
+  1-1-user-authentication: done
+  1-2-account-management: backlog
+  epic-1-retrospective: optional
+  9-9-ghost-story: done
+
+action_items:
+  - epic: 1
+    action: "Add error-handling review; watch: quotes, commas"
+    owner: "Charlie"
+    status: open
+"""
+
+
+def test_merge_preserves_and_never_downgrades(tmp_path, capsys):
+    status_file = run_generate(tmp_path, existing=EXISTING)
+    result = out_json(capsys)
+    data = load(status_file)
+    assert data["development_status"]["epic-1"] == "in-progress"
+    assert data["development_status"]["1-1-user-authentication"] == "done"
+    assert data["generated"] == "01-01-2026 09:00"
+    assert data["last_updated"] == DATE
+    assert result["dropped_orphans"] == [{"key": "9-9-ghost-story", "status": "done"}]
+    assert "9-9-ghost-story" not in data["development_status"]
+
+
+def test_legacy_statuses_merge_by_meaning_not_reset(tmp_path, capsys):
+    existing = (EXISTING
+                .replace("1-2-account-management: backlog", "1-2-account-management: drafted")
+                .replace("epic-1: in-progress", "epic-1: contexted"))
+    status_file = run_generate(tmp_path, existing=existing)
+    result = out_json(capsys)
+    data = load(status_file)
+    assert data["development_status"]["1-2-account-management"] == "ready-for-dev"
+    assert data["development_status"]["epic-1"] == "in-progress"
+    assert {"key": "1-2-account-management", "from": "drafted", "to": "ready-for-dev"} in result["legacy_mapped"]
+    assert not any("illegal" in w for w in result["warnings"])
+    assert result["illegal"] == []
+
+
+def test_metadata_preserved_when_flags_omitted(tmp_path, capsys):
+    existing = (EXISTING
+                .replace("project_key: NOKEY", "project_key: JIRA-PROJ")
+                .replace("tracking_system: file-system", "tracking_system: jira")
+                .replace("story_location: impl", "story_location: /custom/stories"))
+    status_file = run_generate(tmp_path, existing=existing)
+    out_json(capsys)
+    data = load(status_file)
+    assert data["project_key"] == "JIRA-PROJ"
+    assert data["tracking_system"] == "jira"
+    assert data["story_location"] == "/custom/stories"
+
+
+def test_unknown_keys_and_comments_survive_regenerate(tmp_path, capsys):
+    existing = EXISTING + "\n# my own note\nsprint_goal: Ship the beta\n"
+    status_file = run_generate(tmp_path, existing=existing)
+    out_json(capsys)
+    data = load(status_file)
+    assert data["sprint_goal"] == "Ship the beta"
+    assert "# my own note" in status_file.read_text(encoding="utf-8")
+
+
+def test_fresh_rebuild_ignores_existing_statuses(tmp_path, capsys):
+    status_file = run_generate(tmp_path, existing=EXISTING, extra=("--fresh",))
+    result = out_json(capsys)
+    data = load(status_file)
+    assert data["development_status"]["1-1-user-authentication"] == "backlog"
+    assert data["development_status"]["epic-1"] == "backlog"
+    assert result["fresh"] is True
+    # action_items are retro history, not tracking state — carried even on --fresh
+    assert data["action_items"][0]["status"] == "open"
+
+
+def test_set_applies_explicit_statuses_even_downgrades(tmp_path, capsys):
+    status_file = run_generate(
+        tmp_path, existing=EXISTING,
+        extra=("--fresh", "--set", "1-1-user-authentication=in-progress", "--set", "epic-1=in-progress"),
+    )
+    result = out_json(capsys)
+    data = load(status_file)
+    assert data["development_status"]["1-1-user-authentication"] == "in-progress"
+    assert data["development_status"]["epic-1"] == "in-progress"
+    assert "1-1-user-authentication=in-progress" in result["explicit_set"]
+
+
+def test_set_rejects_unknown_key_and_illegal_status(tmp_path, capsys):
+    with pytest.raises(SystemExit) as excinfo:
+        run_generate(tmp_path, extra=("--set", "9-9-nope=done"))
+    assert excinfo.value.code == 1
+    assert out_json(capsys)["ok"] is False
+    with pytest.raises(SystemExit):
+        run_generate(tmp_path, extra=("--set", "epic-1=review"))
+    assert out_json(capsys)["ok"] is False
+
+
+def test_action_items_carried_verbatim(tmp_path):
+    status_file = run_generate(tmp_path, existing=EXISTING)
+    data = load(status_file)
+    assert data["action_items"][0]["action"] == "Add error-handling review; watch: quotes, commas"
+    assert data["action_items"][0]["status"] == "open"
+
+
+def test_story_file_on_disk_floors_ready_for_dev(tmp_path, capsys):
+    status_file = run_generate(tmp_path, stories=["1-2-account-management"])
+    result = out_json(capsys)
+    data = load(status_file)
+    assert data["development_status"]["1-2-account-management"] == "ready-for-dev"
+    assert data["development_status"]["1-1-user-authentication"] == "backlog"
+    assert result["upgraded_from_disk"] == ["1-2-account-management"]
+
+
+def test_story_file_never_downgrades_done(tmp_path):
+    status_file = run_generate(tmp_path, existing=EXISTING, stories=["1-1-user-authentication"])
+    data = load(status_file)
+    assert data["development_status"]["1-1-user-authentication"] == "done"
+
+
+def test_illegal_existing_status_warns_and_resets(tmp_path, capsys):
+    existing = EXISTING.replace("1-2-account-management: backlog", "1-2-account-management: shipped")
+    status_file = run_generate(tmp_path, existing=existing)
+    result = out_json(capsys)
+    data = load(status_file)
+    assert data["development_status"]["1-2-account-management"] == "backlog"
+    assert any("illegal status 'shipped'" in w for w in result["warnings"])
+    assert {"key": "1-2-account-management", "status": "shipped"} in result["illegal"]
+
+
+def test_fenced_code_blocks_are_not_parsed(tmp_path, capsys):
+    text = EPICS_FIXTURE + "\n```\n## Epic 9: Example Format\n### Story 9.1: Sample\n```\n"
+    status_file = run_generate(tmp_path, epics_text=text)
+    result = out_json(capsys)
+    data = load(status_file)
+    assert "epic-9" not in data["development_status"]
+    assert result["epics"] == 2
+    assert not any("Epic 9" in w for w in result["warnings"])
+
+
+def test_non_ascii_titles_keep_distinct_keys(tmp_path, capsys):
+    text = "## Epic 1: 基础\n### Story 1.1: 用户认证\n### Story 1.2: 账户管理\n"
+    status_file = run_generate(tmp_path, epics_text=text)
+    out_json(capsys)
+    keys = list(load(status_file)["development_status"].keys())
+    assert "1-1-用户认证" in keys and "1-2-账户管理" in keys
+
+
+def test_suspect_heading_is_reported(tmp_path, capsys):
+    text = EPICS_FIXTURE + "\n### Story Two point one: Bad Format\n"
+    run_generate(tmp_path, epics_text=text)
+    result = out_json(capsys)
+    assert any("unparsed Epic/Story-like heading" in w for w in result["warnings"])
+
+
+def test_dry_run_writes_nothing_and_reports_drift(tmp_path, capsys):
+    epic_file = tmp_path / "epics.md"
+    epic_file.write_text(EPICS_FIXTURE, encoding="utf-8")
+    status_file = tmp_path / "sprint-status.yaml"
+    status_file.write_text(EXISTING, encoding="utf-8")
+    original = status_file.read_bytes()
+    mod.main([
+        "generate", "--epic-file", str(epic_file), "--status-file", str(status_file),
+        "--stories-dir", str(tmp_path), "--project", "P", "--date", DATE, "--dry-run",
+    ])
+    result = out_json(capsys)
+    assert result["dry_run"] is True and result["ok"] is True
+    assert result["in_sync"] is False
+    assert "epic-2" in result["new_entries"]
+    assert result["dropped_orphans"] == [{"key": "9-9-ghost-story", "status": "done"}]
+    assert status_file.read_bytes() == original
+
+
+def test_dry_run_in_sync_after_generate(tmp_path, capsys):
+    status_file = run_generate(tmp_path)
+    capsys.readouterr()
+    mod.main([
+        "generate", "--epic-file", str(tmp_path / "epics.md"), "--status-file", str(status_file),
+        "--stories-dir", str(tmp_path / "impl"), "--project", "My Project", "--date", DATE,
+        "--dry-run",
+    ])
+    result = out_json(capsys)
+    assert result["in_sync"] is True
+    assert result["new_entries"] == [] and result["dropped_orphans"] == [] and result["illegal"] == []
+
+
+def test_no_epics_fails_with_json(tmp_path, capsys):
+    epic_file = tmp_path / "notes.md"
+    epic_file.write_text("just prose, no epics", encoding="utf-8")
+    with pytest.raises(SystemExit) as excinfo:
+        mod.main([
+            "generate", "--epic-file", str(epic_file), "--status-file", str(tmp_path / "s.yaml"),
+            "--stories-dir", str(tmp_path), "--project", "P", "--date", DATE,
+        ])
+    assert excinfo.value.code == 1
+    assert out_json(capsys)["ok"] is False
+
+
+def test_non_mapping_yaml_fails_with_json(tmp_path, capsys):
+    epic_file = tmp_path / "epics.md"
+    epic_file.write_text(EPICS_FIXTURE, encoding="utf-8")
+    status_file = tmp_path / "sprint-status.yaml"
+    status_file.write_text("- just\n- a\n- list\n", encoding="utf-8")
+    with pytest.raises(SystemExit) as excinfo:
+        mod.main([
+            "generate", "--epic-file", str(epic_file), "--status-file", str(status_file),
+            "--stories-dir", str(tmp_path), "--project", "P", "--date", DATE,
+        ])
+    assert excinfo.value.code == 1
+    assert "not a mapping" in out_json(capsys)["error"]
+    with pytest.raises(SystemExit):
+        mod.main(["status", "--status-file", str(status_file)])
+    assert "not a mapping" in out_json(capsys)["error"]
+
+
+def test_argument_errors_emit_json(capsys):
+    with pytest.raises(SystemExit) as excinfo:
+        mod.main(["generate"])
+    assert excinfo.value.code == 2
+    assert out_json(capsys)["ok"] is False
+    with pytest.raises(SystemExit) as excinfo:
+        mod.main(["-h"])
+    assert excinfo.value.code == 2
+    assert out_json(capsys)["ok"] is False
+
+
+STATUS_FIXTURE = """\
+generated: 01-01-2026 09:00
+last_updated: 07-30-2026 09:00
+project: My Project
+project_key: NOKEY
+tracking_system: file-system
+story_location: impl
+
+development_status:
+  epic-1: in-progress
+  1-1-user-authentication: done
+  1-2-account-management: drafted
+  epic-1-retrospective: optional
+  epic-2: backlog
+  2-1-personality-system: backlog
+  epic-2-retrospective: optional
+
+action_items:
+  - epic: 1
+    action: "Tighten error handling"
+    owner: "Charlie"
+    status: open
+  - epic: 1
+    action: "Old item"
+    owner: "Charlie"
+    status: done
+"""
+
+
+def run_status(tmp_path, capsys, fixture=STATUS_FIXTURE, extra=()):
+    status_file = tmp_path / "sprint-status.yaml"
+    status_file.write_text(fixture, encoding="utf-8")
+    mod.main(["status", "--status-file", str(status_file), "--date", DATE, *extra])
+    return json.loads(capsys.readouterr().out)
+
+
+def test_status_counts_and_recommendation(tmp_path, capsys):
+    result = run_status(tmp_path, capsys)
+    assert result["stories"] == {"done": 1, "ready-for-dev": 1, "backlog": 1}
+    assert result["epics"] == {"in-progress": 1, "backlog": 1}
+    assert result["retrospectives"] == {"optional": 2}
+    assert result["recommendation"]["skill"] == "bmad-build"
+    assert result["recommendation"]["story_key"] == "1-2-account-management"
+    assert result["all_done"] is False
+
+
+def test_status_maps_legacy_values(tmp_path, capsys):
+    result = run_status(tmp_path, capsys)
+    assert {"key": "1-2-account-management", "from": "drafted", "to": "ready-for-dev"} in result["legacy_mapped"]
+
+
+def test_status_open_action_items(tmp_path, capsys):
+    result = run_status(tmp_path, capsys)
+    assert len(result["open_action_items"]) == 1
+    assert result["open_action_items"][0]["action"] == "Tighten error handling"
+
+
+def test_status_malformed_action_items_are_flagged_not_dropped(tmp_path, capsys):
+    fixture = STATUS_FIXTURE + "  - \"just a string\"\n  - epic: 2\n    action: \"No status\"\n"
+    result = run_status(tmp_path, capsys, fixture=fixture)
+    assert any("not a mapping" in w for w in result["warnings"])
+    assert any("missing or unknown status" in w for w in result["warnings"])
+    assert len(result["open_action_items"]) == 1
+
+
+def test_status_review_beats_ready(tmp_path, capsys):
+    fixture = STATUS_FIXTURE.replace("2-1-personality-system: backlog", "2-1-personality-system: review")
+    result = run_status(tmp_path, capsys, fixture=fixture)
+    assert result["recommendation"]["skill"] == "bmad-code-review"
+    assert result["recommendation"]["story_key"] == "2-1-personality-system"
+    assert any("review" in r for r in result["risks"])
+
+
+def test_status_in_progress_beats_all(tmp_path, capsys):
+    fixture = STATUS_FIXTURE.replace("2-1-personality-system: backlog", "2-1-personality-system: in-progress")
+    result = run_status(tmp_path, capsys, fixture=fixture)
+    assert result["recommendation"]["skill"] == "bmad-build"
+    assert result["recommendation"]["story_key"] == "2-1-personality-system"
+    assert result["recommendation"]["reason"] == "resume the in-progress story"
+
+
+def test_status_staleness_and_orphan_risks(tmp_path, capsys):
+    fixture = (STATUS_FIXTURE
+               .replace("last_updated: 07-30-2026 09:00", "last_updated: 01-02-2026 09:00")
+               .replace("  epic-2-retrospective: optional",
+                        "  epic-2-retrospective: optional\n  5-1-ghost: backlog"))
+    result = run_status(tmp_path, capsys, fixture=fixture)
+    assert any("stale" in r for r in result["risks"])
+    assert any("orphaned story '5-1-ghost'" in r for r in result["risks"])
+
+
+def test_status_unparseable_timestamp_warns_instead_of_silence(tmp_path, capsys):
+    fixture = STATUS_FIXTURE.replace("last_updated: 07-30-2026 09:00", "last_updated: whenever")
+    result = run_status(tmp_path, capsys, fixture=fixture)
+    assert any("staleness check skipped" in w for w in result["warnings"])
+
+
+def test_status_iso_and_date_typed_stamps_do_not_crash(tmp_path, capsys):
+    fixture = (STATUS_FIXTURE
+               .replace("generated: 01-01-2026 09:00", "generated: 2026-01-01")
+               .replace("last_updated: 07-30-2026 09:00", "last_updated: 2026-01-02"))
+    result = run_status(tmp_path, capsys, fixture=fixture)
+    assert result["ok"] is True
+    assert result["generated"] == "2026-01-01"
+    assert any("stale" in r for r in result["risks"])  # ISO stamp still parses
+
+
+def test_status_all_done_recommends_retro_then_nothing(tmp_path, capsys):
+    fixture = (STATUS_FIXTURE
+               .replace("1-2-account-management: drafted", "1-2-account-management: done")
+               .replace("2-1-personality-system: backlog", "2-1-personality-system: done"))
+    result = run_status(tmp_path, capsys, fixture=fixture)
+    assert result["recommendation"]["skill"] == "bmad-retrospective"
+    assert "epic-1-retrospective" in result["recommendation"]["reason"]
+    fixture_done = fixture.replace("epic-1-retrospective: optional", "epic-1-retrospective: done") \
+                          .replace("epic-2-retrospective: optional", "epic-2-retrospective: done")
+    result = run_status(tmp_path, capsys, fixture=fixture_done)
+    assert result["all_done"] is True and result["recommendation"] is None
+
+
+def test_status_odd_retro_key_reports_instead_of_crashing(tmp_path, capsys):
+    fixture = (STATUS_FIXTURE
+               .replace("1-2-account-management: drafted", "1-2-account-management: done")
+               .replace("2-1-personality-system: backlog", "2-1-personality-system: done")
+               .replace("epic-1-retrospective: optional", "epic-1-retrospective: done")
+               .replace("epic-2-retrospective: optional",
+                        "epic-2-retrospective: done\n  epic-abc-retrospective: optional"))
+    result = run_status(tmp_path, capsys, fixture=fixture)
+    assert result["ok"] is True
+    assert {"key": "epic-abc-retrospective", "status": "optional"} in result["unrecognized"]
+    assert any("unrecognized key" in r for r in result["risks"])
+
+
+def test_status_illegal_status_reported(tmp_path, capsys):
+    fixture = STATUS_FIXTURE.replace("2-1-personality-system: backlog", "2-1-personality-system: shipped")
+    result = run_status(tmp_path, capsys, fixture=fixture)
+    assert {"key": "2-1-personality-system", "status": "shipped"} in result["illegal"]
+
+
+def test_status_missing_file_fails_json(tmp_path, capsys):
+    with pytest.raises(SystemExit) as excinfo:
+        mod.main(["status", "--status-file", str(tmp_path / "nope.yaml")])
+    assert excinfo.value.code == 1
+    assert json.loads(capsys.readouterr().out)["ok"] is False
+
+
+def run_validate(tmp_path, capsys, content):
+    status_file = tmp_path / "sprint-status.yaml"
+    if content is not None:
+        status_file.write_text(content, encoding="utf-8")
+    mod.main(["validate", "--status-file", str(status_file)])
+    return json.loads(capsys.readouterr().out)
+
+
+def test_validate_clean_file(tmp_path, capsys):
+    clean = STATUS_FIXTURE.replace("1-2-account-management: drafted", "1-2-account-management: backlog")
+    result = run_validate(tmp_path, capsys, clean)
+    assert result["valid"] is True and result["problems"] == []
+
+
+def test_validate_reports_problems_without_crashing(tmp_path, capsys):
+    broken = (STATUS_FIXTURE
+              .replace("2-1-personality-system: backlog", "2-1-personality-system: shipped")
+              .replace("epic-2-retrospective: optional",
+                       "epic-2-retrospective: optional\n  weird-key: done")
+              .replace("last_updated: 07-30-2026 09:00", "last_updated: whenever"))
+    result = run_validate(tmp_path, capsys, broken)
+    assert result["valid"] is False
+    assert any("illegal story status 'shipped'" in p for p in result["problems"])
+    assert any("unrecognized key 'weird-key'" in p for p in result["problems"])
+    assert any("'last_updated' timestamp" in p for p in result["problems"])
+    assert {"key": "1-2-account-management", "from": "drafted", "to": "ready-for-dev"} in result["legacy_mapped"]
+
+
+def test_validate_missing_file_and_bad_yaml(tmp_path, capsys):
+    result = run_validate(tmp_path, capsys, None)
+    assert result["valid"] is False and "does not exist" in result["problems"][0]
+    result = run_validate(tmp_path, capsys, "development_status: [unclosed\n")
+    assert result["valid"] is False and "not valid YAML" in result["problems"][0]
+    result = run_validate(tmp_path, capsys, "- a\n- b\n")
+    assert result["valid"] is False and any("not a mapping" in p for p in result["problems"])
+
+
+if __name__ == "__main__":
+    sys.exit(pytest.main([__file__, "-q"]))
diff --git a/.agents/skills/bmad-sprint-planning/sprint-status-template.yaml b/.agents/skills/bmad-sprint-planning/sprint-status-template.yaml
new file mode 100644
index 00000000..cc4dd7cf
--- /dev/null
+++ b/.agents/skills/bmad-sprint-planning/sprint-status-template.yaml
@@ -0,0 +1,71 @@
+# Sprint Status Template
+# This is an EXAMPLE showing the expected format
+# The actual file will be generated with all epics/stories from your epic files
+
+# generated: {date}
+# project: {project_name}
+# project_key: {project_key}
+# tracking_system: {tracking_system}
+# story_location: {story_location}
+
+# STATUS DEFINITIONS:
+# ==================
+# Epic Status:
+#   - backlog: Epic not yet started
+#   - in-progress: Epic actively being worked on
+#   - done: All stories in epic completed
+#
+# Story Status:
+#   - backlog: Story only exists in epic file
+#   - ready-for-dev: Story file created, ready for development
+#   - in-progress: Developer actively working on implementation
+#   - review: Implementation complete, ready for review
+#   - done: Story completed
+#
+# Retrospective Status:
+#   - optional: Can be completed but not required
+#   - done: Retrospective has been completed
+#
+# Action Item Status:
+#   - open: Committed during a retrospective, not yet addressed
+#   - in-progress: Actively being worked on
+#   - done: Completed
+#
+# WORKFLOW NOTES:
+# ===============
+# - Epic transitions to 'in-progress' automatically when its first story starts (via build's sprint sync)
+# - Stories can be worked in parallel if team capacity allows
+# - Developer typically creates the next story after the previous one is 'done' to incorporate learnings
+# - Dev moves story to 'review', then runs code-review (fresh context, different LLM recommended)
+# - Retrospective appends its action items to action_items; the status view surfaces open ones
+
+# EXAMPLE STRUCTURE (your actual epics/stories will replace these):
+# Timestamps use MM-DD-YYYY HH:MM.
+
+generated: 05-06-2025 21:30
+last_updated: 05-06-2025 21:30
+project: My Awesome Project
+project_key: NOKEY
+tracking_system: file-system
+story_location: "docs/stories"
+
+development_status:
+  epic-1: backlog
+  1-1-user-authentication: done
+  1-2-account-management: ready-for-dev
+  1-3-plant-data-model: backlog
+  1-4-add-plant-manual: backlog
+  epic-1-retrospective: optional
+
+  epic-2: backlog
+  2-1-personality-system: backlog
+  2-2-chat-interface: backlog
+  2-3-llm-integration: backlog
+  epic-2-retrospective: optional
+
+# Action items committed during retrospectives (section created by the retrospective workflow)
+action_items:
+  - epic: 1
+    action: "Add error-handling review to the code review checklist"
+    owner: "Charlie"
+    status: open
diff --git a/.agents/skills/bmad-sprint-status/SKILL.md b/.agents/skills/bmad-sprint-status/SKILL.md
new file mode 100644
index 00000000..3c3cbec6
--- /dev/null
+++ b/.agents/skills/bmad-sprint-status/SKILL.md
@@ -0,0 +1,26 @@
+---
+name: bmad-sprint-status
+description: 'Deprecated — forwards to bmad-sprint-planning (status view).'
+---
+
+# DEPRECATED — forwards to bmad-sprint-planning (status view)
+
+This skill was consolidated into `bmad-sprint-planning`, which now owns the whole sprint-status artifact: gate it, generate it, view it. It is retained as a thin compatibility shim so existing invocations by name and `_bmad/custom/bmad-sprint-status.toml` override files keep working. New work should invoke `bmad-sprint-planning` directly — "show sprint status" routes straight to the status view.
+
+## On Activation
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. This picks up any `{project-root}/_bmad/custom/bmad-sprint-status.toml` and `bmad-sprint-status.user.toml` overrides for the legacy fields (`activation_steps_prepend`, `activation_steps_append`, `persistent_facts`, `on_complete`).
+
+2. Load `{project-root}/_bmad/bmm/config.yaml` (and `config.user.yaml` if present) to resolve `{user_name}` and `{communication_language}`.
+
+3. Emit a deprecation notice to the user in `{communication_language}`:
+
+   > Notice: `bmad-sprint-status` is deprecated and will be removed in a future release. It now forwards to `bmad-sprint-planning`, whose status view covers everything this skill did. To silence this notice, invoke `bmad-sprint-planning` directly next time (e.g. "show sprint status") and migrate any `_bmad/custom/bmad-sprint-status.toml` overrides to `_bmad/custom/bmad-sprint-planning.toml`.
+
+4. Invoke `bmad-sprint-planning` with the following context. Pass these as the activating context so it honors them instead of resolving its own customization from scratch:
+
+   - **Intent:** `status view` — skip `bmad-sprint-planning`'s usual intent detection and its readiness gate.
+   - **Pre-resolved legacy customization** — use these in place of resolving from `bmad-sprint-planning`'s own `customize.toml` for the four legacy fields: `activation_steps_prepend`, `activation_steps_append`, `persistent_facts`, and `on_complete` = the resolved values from step 1.
+   - **Original user input:** forward whatever the user said when invoking this skill verbatim.
+
+   `bmad-sprint-planning` takes the workflow from here. Do not execute any further steps in this shim.
diff --git a/.agents/skills/bmad-sprint-status/customize.toml b/.agents/skills/bmad-sprint-status/customize.toml
new file mode 100644
index 00000000..c3c5600c
--- /dev/null
+++ b/.agents/skills/bmad-sprint-status/customize.toml
@@ -0,0 +1,41 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-sprint-status. Mirrors the
+# agent customization shape under the [workflow] namespace.
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+# Steps to run before the standard activation (config load, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Distinct from the runtime memory sidecar — these are static context
+# loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "All stories must include testable acceptance criteria."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Scalar: executed when the workflow reaches its final step,
+# after sprint status is summarized and risks are surfaced. Override wins.
+# Leave empty for no custom post-completion behavior.
+
+on_complete = ""
diff --git a/.agents/skills/bmad-technical-research/SKILL.md b/.agents/skills/bmad-technical-research/SKILL.md
new file mode 100644
index 00000000..6fe34686
--- /dev/null
+++ b/.agents/skills/bmad-technical-research/SKILL.md
@@ -0,0 +1,14 @@
+---
+name: bmad-technical-research
+description: 'Deprecated — forwards to bmad-deep-recon (technical type).'
+---
+
+# DEPRECATED — forwards to bmad-deep-recon (technical type)
+
+This skill was consolidated into `bmad-deep-recon`. It is retained as a thin compatibility shim so existing invocations by name and `_bmad/custom/bmad-technical-research.toml` override files keep working. New work should invoke `bmad-deep-recon` directly — it drafts deep-research prompts for outside tools, processes finished reports into downstream-ready summaries, and runs research directly, across market, domain, technical, competitive, user-voice, and academic-lit types (plus a select shape for choose-between decisions and custom types).
+
+## On Activation
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. This picks up any `{project-root}/_bmad/custom/bmad-technical-research.toml` and `bmad-technical-research.user.toml` overrides for the legacy fields (`activation_steps_prepend`, `activation_steps_append`, `persistent_facts`, `on_complete`).
+2. Emit a deprecation notice to the user (in their configured communication language): `bmad-technical-research` is deprecated and forwards to `bmad-deep-recon` with the technical type. To silence this notice and access the full new surface (draft/process/run modes, research types, verification levels, HTML briefing, handoffs), migrate `_bmad/custom/bmad-technical-research.toml` to `_bmad/custom/bmad-deep-recon.toml` and invoke `bmad-deep-recon` directly.
+3. Invoke `bmad-deep-recon` with: **research type** `technical` (skip its type inference), the four legacy fields above as pre-resolved values, and the user's original input verbatim. `bmad-deep-recon` takes the workflow from here — do not execute any further steps in this shim.
diff --git a/.agents/skills/bmad-ux/SKILL.md b/.agents/skills/bmad-ux/SKILL.md
new file mode 100644
index 00000000..2450ec82
--- /dev/null
+++ b/.agents/skills/bmad-ux/SKILL.md
@@ -0,0 +1,90 @@
+---
+name: bmad-ux
+description: Plan UX patterns and design specifications. Use when the user says "lets create UX design" or "create UX specifications" or "help me plan the UX"
+---
+# BMad UX
+
+## Overview
+
+You are a master UX facilitator. **Elicit and capture** the user's vision, never impose yours. Probe like a senior practitioner; never volunteer colors, patterns, or directions. Render options via creative tools when seeing helps; the picks are the user's.
+
+Produce two peer contracts: **`DESIGN.md`** (visual identity per the [Google Labs spec](https://github.com/google-labs-code/design.md) — owns *how it looks*) and **`EXPERIENCE.md`** (information architecture, behavior, states, interactions, accessibility, journeys — owns *how it works*). EXPERIENCE.md cross-references DESIGN.md tokens by name using `{path.to.token}` syntax. Both spines win on conflict with any mock, wireframe, or import.
+
+## The DESIGN.md spine
+
+Per the [Google Labs spec](https://github.com/google-labs-code/design.md). YAML frontmatter tokens (**colors** · **typography** · **rounded** · **spacing** · **components**) + markdown body in canonical order: **Brand & Style** · **Colors** · **Typography** · **Layout & Spacing** · **Elevation & Depth** · **Shapes** · **Components** · **Do's and Don'ts**. Sections omittable; order locked when present. Spec rules: `references/design-md-spec.md`. Shape: read every entry in `{workflow.design_md_examples}`.
+
+## The EXPERIENCE.md spine
+
+Always: **Foundation** (form-factor, UI system when present; DESIGN.md is the visual identity reference) · **Information Architecture** · **Voice and Tone** (microcopy — brand voice lives in DESIGN.md.Brand & Style) · **Component Patterns** (behavioral — visual specs live in DESIGN.md.Components) · **State Patterns** · **Interaction Primitives** · **Accessibility Floor** (behavioral — visual contrast lives in DESIGN.md) · **Key Flows** (named-protagonist journeys with a climax beat).
+
+When triggered: **Inspiration & Anti-patterns** · **Responsive & Platform**.
+
+Invent sections for product-specific concerns. Shape: read every entry in `{workflow.experience_md_examples}`.
+
+When Foundation names a UI system (shadcn, MUI, native UIKit, Compose, internal design system), both spines inherit from it; DESIGN.md tokens reference or extend the system's defaults, EXPERIENCE.md specifies only the behavioral delta.
+
+## Sources
+
+UX may lead, follow, or stand alone. Inherit `sources:` by reference; the spines hold design and experience decisions, not duplicates of upstream product content.
+
+## On Activation
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. On failure, read `{skill-root}/customize.toml` directly and use defaults.
+2. Run `{workflow.activation_steps_prepend}`. Treat `{workflow.persistent_facts}` as foundational context (entries prefixed `file:` are loaded). `{workflow.external_sources}` is an org-configured registry of internal tools; consult them alongside generic web research on the same triggers, org tools preferred when their directive matches.
+3. Load `{project-root}/_bmad/bmm/config.yaml` (+ `config.user.yaml` if present). Resolve `{user_name}`, `{communication_language}`, `{document_output_language}`, `{planning_artifacts}`, `{project_name}`, `{date}`. Missing keys → neutral defaults; never block.
+4. If headless, follow `references/headless.md` for the whole run. Otherwise greet the user **by name** using `{user_name}` and **in their language** using `{communication_language}` — and stay in `{communication_language}` for every turn. In the greeting, let the user know `bmad-party-mode` and `bmad-advanced-elicitation` are always available. Then scan for misroute on the first message: PRD → `bmad-prd`; architecture → `bmad-architecture`; game UX → BMad GDS; agent/skill → `bmad-workflow-builder`; brief → `bmad-product-brief`.
+5. Detect intent: **Create**, **Update**, **Validate**. For Create, before binding a fresh workspace, scan `{workflow.ux_output_path}` for prior in-progress runs (folders matching `{workflow.run_folder_pattern}` whose `DESIGN.md` frontmatter `status` is not `final`) and offer to resume rather than starting over.
+
+Run `{workflow.activation_steps_append}`.
+
+Activation is complete. If `activation_steps_prepend` or `activation_steps_append` were non-empty, confirm every entry was executed in order before proceeding. Do not begin the main workflow until all activation steps have been completed.
+
+## Modes
+
+**Create.** Bind `{doc_workspace}` to `{workflow.ux_output_path}/{workflow.run_folder_pattern}/`. Create `.working/` and `imports/`; seed the memlog with `uv run {project-root}/_bmad/scripts/memlog.py init --workspace {doc_workspace} --field topic="<product/UX>"`; create `DESIGN.md` (frontmatter only) and `EXPERIENCE.md` (frontmatter only). Run Discovery → Finalize.
+
+**Update.** Read spines + memlog + sources. If `.memlog.md` is missing, init it with `uv run {project-root}/_bmad/scripts/memlog.py init --workspace {doc_workspace}` — this update is entry one. Surface conflicts with prior decisions. Run Finalize.
+
+**Validate.** See `references/validate.md`.
+
+## Discovery
+
+**Capture; do not author.** The spines are distilled at Finalize toward the memlog. Decisions → `.memlog.md` (canonical), each appended via `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {doc_workspace} --type <decision|change|override|assumption|event> --text "…"` — never hand-edited; a resume reloads it. Creative-tool artifacts → `.working/`. User-supplied visuals (Figma, sketches, brand decks, image folders) → `imports/`, one `memlog.py append` per item. Spines win on conflict.
+
+**Source scan.** Glob `{planning_artifacts}/` for candidate input paths; surface paths only — never read content in the parent. User confirms which apply or adds others; subagent-extracts on confirm.
+
+Brain dump first — even when the user opens with paragraphs (that's intake). Subagent-extract big docs. One "anything else?" probe. Stakes: hobby / internal / consumer / regulated.
+
+Working mode:
+
+- **Fast path** — batch gaps, draft both spines with `[ASSUMPTION]` tags, skip creative tools.
+- **Coaching path** — walk decisions; creative tools woven in.
+- **Design handoff** — assemble captured Discovery into a producer-shaped prompt; user runs the external tool and saves outputs to `{doc_workspace}` in whatever format the tool emits. Producer registry: `{workflow.design_handoffs}` (default: Google Stitch). EXPERIENCE.md can follow via Update mode when ready.
+
+Creative tools — scan `{workflow.creative_tools}`, invoke when seeing helps. Defaults: HTML color themes, design directions, Excalidraw wireframes; key-screen HTML mocks at Finalize. See `references/creative-tools.md`. Research subagents on demand; consult `{workflow.external_sources}` when entries match.
+
+Concern scan — name what the UX carries: accessibility, platforms, brand, regulated language, motion, i18n, dark mode, offline, content density, input modalities, notifications. Open list; drives invented sections.
+
+Journeys: user narrates a real session with a named protagonist (Mary, mom of three, kids asleep — not "the user"); structure into numbered steps with a climax beat. Mirror source-spec names verbatim when defined.
+
+Form-factor: mobile / web / desktop / multi-surface must resolve before IA closes. Named-protagonist journeys often derive it (Pary on iPad implies an iPad surface; Skeeter on Android adds a multi-surface need); when journeys don't disambiguate, probe.
+
+Surface closure: stated needs become screens through journeys. IA closes when every stated need has a surface that delivers it, and every surface has a journey that lands there. When closure fails, probe — never invent the missing piece.
+
+## Reviewer Gate
+
+Used by Validate and Finalize. **Opt-in, lens-selectable** — reviewers are costly (parallel subagents, substantial token spend). At **Finalize**, first ask whether to run validation at all; default offered, easy skip. At **Validate** intent the user already opted in — skip that question. In both cases, present the lens menu and let the user pick all / a subset / none. Menu: rubric walker (`references/validate.md`) + `{workflow.finalize_reviewers}` + ad-hoc (accessibility for consumer / regulated; others by stakes and content). Picked lenses dispatch as parallel subagents → each writes `review-{slug}.md`, returns a compact summary. If any lens ran, run the synthesis pipeline in `references/validate.md`.
+
+## Finalize
+
+Outcomes, in order:
+
+- **Spines distilled.** Subagent reads `.memlog.md`, `.working/`, `imports/`, sources; produces `DESIGN.md` against `## The DESIGN.md spine` + `{workflow.design_md_examples}` and `EXPERIENCE.md` against `## The EXPERIENCE.md spine` + `{workflow.experience_md_examples}`. Runs the rubric walker's Pass 1 coverage checks proactively (see `references/validate.md`). Surface gaps; never invent.
+- **Inputs reconciled.** Subagent per user-supplied input → `reconcile-{slug}.md`. Surface dropped qualitative ideas.
+- **Reviewer Gate offered.** Ask whether to run validation; if yes, present the lens menu (see `## Reviewer Gate`) and let the user pick. If any lens ran, resolve findings before polish; otherwise proceed.
+- **Open items triaged.** Open Questions, `[ASSUMPTION]`, `[NOTE FOR UX]`. Phase-blockers one at a time; non-blockers → `memlog.py append`.
+- **Key-screen mocks rendered.** Key-screens tool → `.working/` for surfaces where layout drives behavior or anchors visual language.
+- **Mock coverage confirmed.** Walk every IA surface; classify *mocked* vs *spine-only*. Ask: *"These will be built from spine tables alone — any need a visual reference?"* Render more if named; log spine-only choices.
+- **Layout extracted, artifacts promoted.** Distill subagent re-reads each `.working/` and `imports/` artifact; lifts visual decisions into DESIGN.md and behavioral decisions into EXPERIENCE.md. Promote `.working/` keepers to `mockups/` (HTML) or `wireframes/` (Excalidraw); imports stay. Inline relative links at relevant spine sections; state spines-win-on-conflict once.
+- **Polished, handed off, closed.** Apply `{workflow.doc_standards}` in order. Execute `{workflow.external_handoffs}`; surface URLs. Set both files' `status: final`, `updated: {date}`. Log finalization via `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {doc_workspace} --type event --text "spines finalized"`. Share paths. Common next: `bmad-architecture`, `bmad-create-epics-and-stories`, `bmad-build`. Run `{workflow.on_complete}`.
diff --git a/.agents/skills/bmad-ux/assets/color-themes.md b/.agents/skills/bmad-ux/assets/color-themes.md
new file mode 100644
index 00000000..2b23b466
--- /dev/null
+++ b/.agents/skills/bmad-ux/assets/color-themes.md
@@ -0,0 +1,9 @@
+# Color Themes Renderer
+
+Subagent prompt. Produce one self-contained HTML page at the supplied `.working/color-themes-{n}.html` path showing 4-6 distinct theme variations side by side so the user can pick.
+
+Each variation: header (name + one-line emotional register), token chips for every semantic role decided so far, and one realistic UI snippet using the palette (content drawn from the conversation, not lorem). Include light and dark side-by-side when both modes are in scope. Avoid near-identical pastels — variations must differ in register, not just hue.
+
+Inline CSS only, system font stack, no JS, no network. Document concrete hex values in `<style>` comments per variation so the user can lift them if they pick that theme. The spine itself stays semantic.
+
+Return to the parent: file path, one-line per variation, mode coverage. Do not dump HTML into the parent context. If interactive, open the file with the platform opener — `open "PATH"` on macOS, `xdg-open` on Linux, `start ""` on Windows, path always double-quoted. On failure, give the user the path instead.
diff --git a/.agents/skills/bmad-ux/assets/design-directions.md b/.agents/skills/bmad-ux/assets/design-directions.md
new file mode 100644
index 00000000..289b270a
--- /dev/null
+++ b/.agents/skills/bmad-ux/assets/design-directions.md
@@ -0,0 +1,9 @@
+# Design Directions Renderer
+
+Subagent prompt. Produce 3-6 distinct visual directions for the product's hero screen, each a separate self-contained HTML file at `.working/direction-{slug}.html` (or one combined `directions-{n}.html` if the parent's intent says side-by-side).
+
+Each direction is a *complete visual personality* applied to the same key screen — not a palette swap. Differ on density, type weight, motion implication, brand register. Each file: 2-3 sentence rationale, near-1:1 hero screen mockup in a phone or browser frame, ideally a secondary screen, at least one state variant visible (aging row, empty state, etc).
+
+Use real product content from the conversation. Voice/tone from `.memlog.md` applied to every visible string — no lorem. Inline CSS, system fonts, no JS or network. Document hex values in `<style>` comments per direction.
+
+Return to the parent: file paths, one-line personality summary per direction, what hero screen was depicted. Do not dump HTML into parent context. If interactive, open each file in the browser.
diff --git a/.agents/skills/bmad-ux/assets/design-example-editorial.md b/.agents/skills/bmad-ux/assets/design-example-editorial.md
new file mode 100644
index 00000000..e72dc7d1
--- /dev/null
+++ b/.agents/skills/bmad-ux/assets/design-example-editorial.md
@@ -0,0 +1,158 @@
+---
+name: Linen & Logic
+colors:
+  surface: '#fbf9f4'
+  surface-dim: '#dbdad5'
+  surface-bright: '#fbf9f4'
+  surface-container-lowest: '#ffffff'
+  surface-container-low: '#f5f3ee'
+  surface-container: '#f0eee9'
+  surface-container-high: '#eae8e3'
+  surface-container-highest: '#e4e2dd'
+  on-surface: '#1b1c19'
+  on-surface-variant: '#4e453d'
+  inverse-surface: '#30312e'
+  inverse-on-surface: '#f2f1ec'
+  outline: '#80756b'
+  outline-variant: '#d1c4b9'
+  surface-tint: '#715a3f'
+  primary: '#59452b'
+  on-primary: '#ffffff'
+  primary-container: '#735c41'
+  on-primary-container: '#f5d6b4'
+  inverse-primary: '#e0c1a1'
+  secondary: '#a43b2c'
+  on-secondary: '#ffffff'
+  secondary-container: '#fd7d69'
+  on-secondary-container: '#71160b'
+  tertiary: '#374a5f'
+  on-tertiary: '#ffffff'
+  tertiary-container: '#4f6278'
+  on-tertiary-container: '#caddf8'
+  error: '#ba1a1a'
+  on-error: '#ffffff'
+  error-container: '#ffdad6'
+  on-error-container: '#93000a'
+  primary-fixed: '#fdddbb'
+  primary-fixed-dim: '#e0c1a1'
+  on-primary-fixed: '#281804'
+  on-primary-fixed-variant: '#58432a'
+  secondary-fixed: '#ffdad4'
+  secondary-fixed-dim: '#ffb4a7'
+  on-secondary-fixed: '#400200'
+  on-secondary-fixed-variant: '#842417'
+  tertiary-fixed: '#d0e4ff'
+  tertiary-fixed-dim: '#b4c8e2'
+  on-tertiary-fixed: '#071d30'
+  on-tertiary-fixed-variant: '#35485d'
+  background: '#fbf9f4'
+  on-background: '#1b1c19'
+  surface-variant: '#e4e2dd'
+typography:
+  display-lg:
+    fontFamily: Libre Caslon Text
+    fontSize: 48px
+    fontWeight: '400'
+    lineHeight: '1.1'
+    letterSpacing: -0.02em
+  display-lg-mobile:
+    fontFamily: Libre Caslon Text
+    fontSize: 36px
+    fontWeight: '400'
+    lineHeight: '1.1'
+  headline-md:
+    fontFamily: Libre Caslon Text
+    fontSize: 32px
+    fontWeight: '400'
+    lineHeight: '1.2'
+  headline-sm:
+    fontFamily: Libre Caslon Text
+    fontSize: 24px
+    fontWeight: '400'
+    lineHeight: '1.3'
+  body-lg:
+    fontFamily: DM Sans
+    fontSize: 18px
+    fontWeight: '400'
+    lineHeight: '1.6'
+    letterSpacing: 0.01em
+  body-md:
+    fontFamily: DM Sans
+    fontSize: 16px
+    fontWeight: '400'
+    lineHeight: '1.6'
+  label-caps:
+    fontFamily: DM Sans
+    fontSize: 12px
+    fontWeight: '500'
+    lineHeight: '1.4'
+    letterSpacing: 0.1em
+  caption:
+    fontFamily: DM Sans
+    fontSize: 13px
+    fontWeight: '400'
+    lineHeight: '1.4'
+rounded:
+  sm: 0.125rem
+  DEFAULT: 0.25rem
+  md: 0.375rem
+  lg: 0.5rem
+  xl: 0.75rem
+  full: 9999px
+spacing:
+  unit: 8px
+  gutter: 24px
+  margin-mobile: 20px
+  margin-desktop: 64px
+  editorial-gap: 80px
+---
+
+## Brand & Style
+
+The design system is rooted in the philosophy of "Slow Design"—an intentional departure from the frantic pace of fast fashion. It evokes a tactile, "linen-weight" sensation through high-end editorial layouts and a restrained aesthetic. The target audience values provenance over presence, seeking a reflective and sophisticated discovery experience that feels as much like a boutique magazine as a digital marketplace.
+
+The style is **Editorial Minimalism** with **Tactile** accents. It prioritizes breathable white space, asymmetrical layouts that mimic printed lookbooks, and a soft, sun-faded palette. Every interaction is designed to be deliberate and "anti-hype," eschewing aggressive animations for subtle transitions and quiet confidence.
+
+## Colors
+
+The palette is inspired by natural fibers and weathered landscapes. 
+- **Warm White (#F9F7F2)** serves as the primary canvas, providing a soft, non-clinical background that reduces eye strain.
+- **Bone (#E3DED1)** and **Dust (#C2B9A7)** are used for structural depth, subtle dividers, and secondary surfaces.
+- **Tobacco (#735C41)** is the primary ink color, used for high-contrast typography and essential UI elements.
+- **Sun-faded Red (#B84A39)** and **Wool Blanket Blue (#4A5D73)** are used sparingly as "organic accents"—highlighting editorial picks or signifying subtle state changes without disrupting the tranquil atmosphere.
+
+## Typography
+
+Typography is the primary vehicle for the brand’s sophisticated voice. 
+- **Libre Caslon Text** is the voice of the curator. Its classic proportions and elegant serifs provide the editorial weight required for discovery and storytelling.
+- **DM Sans** provides a quiet, functional counterpoint. It is used for body copy and navigational elements, ensuring clarity without competing with the headlines. 
+
+Large display titles should often use "optical sizing" logic—tighter leading and slightly negative letter spacing to create a cohesive visual block. Labels are always tracked out (0.1em) to maintain a sense of airy premiumness.
+
+## Layout & Spacing
+
+This design system employs a **Fluid Editorial Grid**. While it follows a 12-column structure on desktop, it encourages "asymmetrical breathing room"—intentionally leaving columns empty to direct focus toward high-quality imagery.
+
+Spacing is generous. The `editorial-gap` (80px+) should be used between major content sections to allow the user to pause and reflect. Mobile layouts should maintain a minimum of 20px side margins to ensure the content feels framed like a page, rather than bleeding to the edges of the device. Elements should lean toward vertical stacks to mimic the scroll of a digital journal.
+
+## Elevation & Depth
+
+Depth is communicated through **Tonal Layering** and **Ambient Shadows** rather than sharp borders.
+- **Surfaces:** Use the "Bone" color to define containers against the "Warm White" base. 
+- **Shadows:** Shadows are highly diffused and tinted with the "Tobacco" hue (`rgba(115, 92, 65, 0.08)`). They should feel like a soft glow of light hitting fabric, with large blur radii (20px+) and very low opacity.
+- **Borders:** When borders are necessary, they are 1px thick and rendered in "Dust," creating a "ghost" outline that barely separates elements from the background.
+
+## Shapes
+
+The shape language is **Soft (0.25rem)**. While a sharp edge feels too aggressive and a pill-shape feels too digital/tech-heavy, a subtle rounding of corners mimics the natural softening of woven textiles over time. 
+
+Larger containers (Cards, Modals) may use `rounded-lg` (0.5rem) to emphasize their tactile, object-like quality. Imagery should always follow these corner radii to maintain a cohesive, "framed" appearance.
+
+## Components
+
+- **Buttons:** Primary buttons use a solid "Tobacco" fill with "Warm White" text. Secondary buttons are "Bone" with "Tobacco" text or simply "Tobacco" text with a 1px "Dust" border. Padding is generous horizontally to create an elegant, elongated silhouette.
+- **Cards:** Editorial cards feature large imagery, a "headline-sm" title, and a "caption" subline. Shadows are only applied on hover to simulate a gentle lift.
+- **Inputs:** Minimalist underlines in "Dust" that transition to "Tobacco" on focus. Label text remains in "label-caps" above the field.
+- **Chips/Tags:** Used for material types (e.g., "100% Linen"). These are rendered in "Bone" backgrounds with "Tobacco" text, using the "Soft" corner radius.
+- **Icons:** Must be "Hand-drawn" or "Fine-line" style. Lines should have slight imperfections and vary in weight to reinforce the tactile, artisanal nature of the fashion being discovered.
+- **Navigation:** A simple, centered bottom bar or a top-weighted "Ghost" header that disappears on scroll to maximize the editorial viewport.
\ No newline at end of file
diff --git a/.agents/skills/bmad-ux/assets/design-example-mobile.md b/.agents/skills/bmad-ux/assets/design-example-mobile.md
new file mode 100644
index 00000000..1c301c82
--- /dev/null
+++ b/.agents/skills/bmad-ux/assets/design-example-mobile.md
@@ -0,0 +1,93 @@
+---
+name: Quill
+description: Daily writing companion. Calm, intentional, dark-mode-by-default. No streaks, no gamification.
+colors:
+  surface-base: '#FAF9F7'
+  surface-raised: '#FFFFFF'
+  ink-primary: '#1A1B1F'
+  ink-secondary: '#6B655A'
+  ink-disabled: '#B5AFA5'
+  accent: '#A87434'
+  border-hairline: '#E8E4DD'
+  surface-base-dark: '#1A1B1F'
+  surface-raised-dark: '#23252B'
+  ink-primary-dark: '#F0EDE8'
+  ink-secondary-dark: '#A39E94'
+  ink-disabled-dark: '#5E5A53'
+  accent-dark: '#D4A574'
+  border-hairline-dark: '#2E3036'
+typography:
+  title:
+    note: 'Platform native — iOS Title 1 · Android Headline Small'
+  body:
+    note: 'Platform native — iOS Body · Android Body Large'
+  meta:
+    note: 'Platform native — iOS Footnote · Android Body Small'
+rounded:
+  sm: 6px
+  md: 12px
+spacing:
+  '1': 4px
+  '2': 8px
+  '3': 12px
+  '4': 16px
+  '5': 24px
+  '6': 32px
+---
+
+## Brand & Style
+
+Quill is designed against the grain of contemporary habit apps. Where most products weaponize the user's calendar with streak counters and re-engagement nudges, Quill insists on something quieter — a daily prompt, a place to write, and the unspoken assurance that today's entry is enough. Showing up is the point, not the streak.
+
+The visual language follows. Calm surfaces in warm off-white (light) or deep ink (dark, the default). Generous breathing room. No chromatic color competing for attention except a single warm tobacco that signals save-and-send. Text-first. Hand-on-paper, not buzz-on-screen.
+
+## Colors
+
+The palette is restrained on purpose — a writing surface should not compete with the writing.
+
+- **Warm White (`#FAF9F7`)** is the primary canvas in light mode. Slightly warm to reduce eye strain and keep the surface from feeling clinical.
+- **Deep Ink (`#1A1B1F`)** is the dark-mode canvas and the primary body text color in light mode. Quill defaults to dark because most writing happens at night.
+- **Tobacco (`#A87434` light / `#D4A574` dark)** is the only chromatic color. Used exclusively for the save indicator and primary action — never for decoration, never for state badges.
+- **Hairline (`#E8E4DD` light / `#2E3036` dark)** separates list items at the lowest possible contrast. Anything heavier feels like UI rather than paper.
+
+Avoid: red error fills (Quill is a journal, not a form), gradients (the surface is paper), and saturated accent variants — one accent, used sparingly.
+
+## Typography
+
+Platform conventions are the spec. iOS uses Title 1 / Body / Footnote; Android uses Headline Small / Body Large / Body Small. Dynamic type honored at every level — the largest accessibility setting must still render legibly without truncation.
+
+Headlines are rare. The Today prompt is set in `title`; everything else is `body` or `meta`. No display sizes, no all-caps labels.
+
+## Layout & Spacing
+
+Scale: 4 / 8 / 12 / 16 / 24 / 32 px. The largest gaps land between major surfaces; the smallest sit between tightly related elements. Vertical rhythm follows a hard rule: composer breathes, list items don't.
+
+Mobile margins follow platform conventions (iOS 16pt, Android 16dp). Single-column always; modal stacks one level deep, never two.
+
+## Elevation & Depth
+
+Quill avoids elevation as a visual device. Cards and composer surfaces sit on `surface-raised`, distinguished from `surface-base` only by tone. Shadows are reserved for the rare moment of literal physical metaphor — never for hierarchy. Hierarchy comes from layout and typography, not shadow.
+
+## Shapes
+
+`rounded/sm` (6px) for inputs, list rows, and small surfaces. `rounded/md` (12px) for cards and the composer. Nothing fully rounded; no pills, no perfect circles for surfaces. The aesthetic is paper-with-soft-corners, not iOS-button-pill.
+
+Imagery follows container corners exactly.
+
+## Components
+
+- **Prompt card** — `surface-raised`. One per day. Today's prompt in `title`. Tap to open composer. No icon, no decoration; the prompt itself is the affordance.
+- **Composer** — Full-screen text view. Clean text field, generous vertical padding, single-line save indicator in the header.
+- **Save indicator** — Text only. Uses `ink-secondary`, never a checkmark icon, never a colored badge.
+- **Entry row** (Library) — Date in `meta`, first line of body in `body` (truncated to one line). Hairline divider only, no fill.
+- **Settings row** — Label left, value or chevron right. Tobacco accent only on destructive confirmations.
+
+## Do's and Don'ts
+
+| Do | Don't |
+|---|---|
+| Single accent color, used sparingly on save & primary action | Color-code by sentiment, mood, or category |
+| Text-only state indicators (`Saved.`) | Iconography for state (✓, ⚠, ●) |
+| Hairline dividers at lowest legible contrast | Card shadows, gradient fills, accent fills behind text |
+| Generous vertical rhythm in composer | Compress to fit more on screen |
+| Honor platform conventions for navigation | Override platform nav with custom drawer or hamburger |
diff --git a/.agents/skills/bmad-ux/assets/design-example-shadcn.md b/.agents/skills/bmad-ux/assets/design-example-shadcn.md
new file mode 100644
index 00000000..a7147f8e
--- /dev/null
+++ b/.agents/skills/bmad-ux/assets/design-example-shadcn.md
@@ -0,0 +1,109 @@
+---
+name: Drift
+description: Focused task tracker for solo founders and small async teams. shadcn/ui on Next.js + Tailwind; this DESIGN.md specifies the brand-layer delta only.
+colors:
+  # Brand overrides on top of shadcn defaults. All unlisted tokens inherit
+  # from shadcn (background, foreground, muted, muted-foreground, popover,
+  # popover-foreground, card, card-foreground, border, input, ring, destructive).
+  primary: '#0F4C81'
+  primary-foreground: '#FFFFFF'
+  accent: '#F59E0B'
+  accent-foreground: '#1A1208'
+  primary-dark: '#5C8AC2'
+  primary-foreground-dark: '#0A1A2A'
+  accent-dark: '#FBC470'
+  accent-foreground-dark: '#1A1208'
+typography:
+  # Body, label, and muted inherit from shadcn (Geist Sans). Only display is overridden.
+  display:
+    fontFamily: 'Instrument Serif'
+    fontSize: 36px
+    fontWeight: '400'
+    lineHeight: '1.15'
+    letterSpacing: -0.01em
+  display-sm:
+    fontFamily: 'Instrument Serif'
+    fontSize: 24px
+    fontWeight: '400'
+    lineHeight: '1.2'
+rounded:
+  # Tighter than shadcn defaults — Drift reads sharper.
+  sm: 4px
+  md: 6px
+  lg: 8px
+spacing:
+  # shadcn / Tailwind defaults inherited; no overrides.
+components:
+  button-primary:
+    background: '{colors.primary}'
+    foreground: '{colors.primary-foreground}'
+    radius: '{rounded.md}'
+  focus-card:
+    background: '{colors.accent}'
+    foreground: '{colors.accent-foreground}'
+    radius: '{rounded.md}'
+    border: 'none'
+  command-palette-result-active:
+    background: '{colors.accent}'
+    foreground: '{colors.accent-foreground}'
+---
+
+## Brand & Style
+
+Drift is a focused task tracker for solo founders and small async teams. The product premise is that *work is a moving thing* — momentum matters more than perfectly groomed backlogs, and the right tool surfaces what you're working on *now* without making you administer a system to find it. The brand expression follows: a serif display moment in an otherwise sober sans-serif surface, a single warm accent that means *this is what's live*, and visual restraint everywhere else.
+
+Drift inherits shadcn/ui defaults wholesale. This DESIGN.md specifies only the brand-layer deltas — primary color, accent color, display typography, slightly tighter corners, and a handful of brand-specific components. The 80% of components that ship from shadcn (Button, Card, Dialog, Sheet, Command, Popover, Toast) inherit shadcn's visual specs as-is. Customizing those is *explicitly* against the brand discipline — shadcn's defaults are the contract.
+
+## Colors
+
+The Drift palette is two colors of brand-layer plus shadcn defaults for everything else.
+
+- **Primary Navy (`#0F4C81` light / `#5C8AC2` dark)** is the brand color. Used on primary buttons, active nav items, link underlines, and the "current week" indicator. Replaces shadcn's default `primary`.
+- **Focus Amber (`#F59E0B` light / `#FBC470` dark)** is the accent. Used exclusively to indicate the task or project currently in focus — the one you're working on *right now*. Never used for chrome, never used decoratively, never used for state badges. Amber means "live."
+- **All other tokens** (`background`, `foreground`, `muted`, `muted-foreground`, `border`, `input`, `ring`, `card`, `popover`, `destructive`) inherit from shadcn defaults. If the brand can't justify overriding a token, it doesn't override it.
+
+Avoid: chromatic flourishes, gradient surfaces, custom destructive colors (use shadcn's), more than two brand colors. The discipline is two-colors-and-stop.
+
+## Typography
+
+Body / label / caption inherit shadcn's Geist Sans ramp. Only the `display` role is brand-overridden, set in **Instrument Serif** at 36px (24px small variant). The serif moment appears in:
+
+- Empty-state hero text on Today and project surfaces
+- Project titles in the project detail header
+- The "Welcome back, {name}" greeting at first session of the day
+
+Everything else stays in Geist Sans. The serif is a punctuation mark, not a default voice.
+
+## Layout & Spacing
+
+shadcn / Tailwind spacing scale inherited as-is (the 4-based scale: 4, 8, 12, 16, 20, 24, 32, 40, 48, 64). Maximum content width: `max-w-3xl` (768px) — Drift is not a wide-table product, and forcing one-column reading keeps the surface focused.
+
+Single-column layout. Sidebar nav on `lg` (1024px+); on smaller viewports, the sidebar becomes a sheet triggered from the top bar.
+
+## Elevation & Depth
+
+Inherited from shadcn — subtle shadow on hover/active states, no elevation as a visual hierarchy device. Drift adds nothing on top of this; brand discipline is "shadcn's shadows are correct."
+
+## Shapes
+
+Tighter than shadcn defaults: `rounded/sm` (4px) for inputs, `rounded/md` (6px) for cards and buttons, `rounded/lg` (8px) for dialogs and the command palette. The crispness reads "tool" rather than "consumer app." Pill shapes (`rounded/full`) appear only on status badges.
+
+## Components
+
+Drift uses the following shadcn components as-is, unchanged: `Button`, `Card`, `Dialog`, `Sheet`, `Popover`, `DropdownMenu`, `Toast`, `Tabs`, `Avatar`, `Separator`. The contract: don't customize these.
+
+Brand-layer-overridden components:
+
+- **Button (primary variant)** — `{colors.primary}` fill, `{colors.primary-foreground}` text, `{rounded.md}` corner. Other variants (secondary, outline, ghost, destructive) inherit shadcn defaults.
+- **Focus card** — Custom Drift component. The "this is what you're working on now" card on Today and project detail. `{colors.accent}` fill, no border, slightly elevated. Appears at most once per surface.
+- **Command palette result (active)** — Override on shadcn's `Command` component: the highlighted/keyboard-selected result row uses `{colors.accent}` instead of shadcn's default `accent` token. Reinforces "this is what will fire if you hit Enter."
+
+## Do's and Don'ts
+
+| Do | Don't |
+|---|---|
+| Inherit shadcn defaults for everything not in the brand layer | Override shadcn's color tokens beyond `primary` and `accent` |
+| Use `{colors.accent}` only for "live / now / in-focus" | Use accent for state, chrome, or hover affordances |
+| `display` typography sparingly — empty states, hero greetings | Set body text in `display` to "make it pretty" |
+| Tighter corners than shadcn (4 / 6 / 8) | Use shadcn's default 6/8/12 (Drift reads sharper) |
+| Single-column layouts inside `max-w-3xl` | Wide multi-column tables (Drift is not a spreadsheet) |
diff --git a/.agents/skills/bmad-ux/assets/excalidraw-wireframe.md b/.agents/skills/bmad-ux/assets/excalidraw-wireframe.md
new file mode 100644
index 00000000..7f04c95e
--- /dev/null
+++ b/.agents/skills/bmad-ux/assets/excalidraw-wireframe.md
@@ -0,0 +1,19 @@
+# Excalidraw Wireframe Renderer
+
+Subagent prompt. Produce one `.excalidraw` file at `.working/ia-{date}.excalidraw` (IA diagram) or `.working/flow-{name}-{date}.excalidraw` (flow wireframe).
+
+## CRITICAL: two-character `index` fields only
+
+Every element's `index` field must be **exactly two characters** (`a0`, `aZ`, `b3`, ...). Three-character indices cause a silent *"Error: invalid file"* with no diagnostic output. Assign sequentially across all elements; advance the leading letter when the trailing alphanumeric exhausts (`a0..a9, aA..aZ`, then `b0..`). Verify before writing.
+
+## Shape
+
+Valid Excalidraw file: `{type: "excalidraw", version: 2, source: "https://excalidraw.com", elements: [...], appState: {gridSize: null, viewBackgroundColor: "#ffffff"}, files: {}}`. Each element needs the standard Excalidraw element fields (`id, type, x, y, width, height, angle, strokeColor, backgroundColor, fillStyle, strokeWidth, strokeStyle, roughness, opacity, groupIds, frameId, roundness, seed, version, versionNonce, isDeleted, boundElements, updated, link, locked, index`). Text elements add `text, fontSize, fontFamily, textAlign, verticalAlign, baseline, containerId, originalText, lineHeight`.
+
+## Content
+
+**IA diagram:** boxes-and-arrows of auth stack, main app surfaces, modal routes, settings stack, cross-cutting affordances. Color sparingly to distinguish category. Layout for human legibility, not graph correctness.
+
+**Flow wireframe:** screen-by-screen rectangles left-to-right, simple shapes inside (nav bar, CTA, content blocks) at low fidelity. Arrows labeled with the user action that causes transition. Annotations alongside for climax and edge-case beats.
+
+Return to the parent: file path, kind, one-line subject, element count, confirmation that all indices are two-character. Do not dump JSON into parent context. Tell the user to open in Excalidraw desktop or excalidraw.com.
diff --git a/.agents/skills/bmad-ux/assets/experience-example-mobile.md b/.agents/skills/bmad-ux/assets/experience-example-mobile.md
new file mode 100644
index 00000000..5bd05ee9
--- /dev/null
+++ b/.agents/skills/bmad-ux/assets/experience-example-mobile.md
@@ -0,0 +1,112 @@
+---
+name: Quill
+status: final
+sources:
+  - {planning_artifacts}/prds/quill-2025-08-15/prd.md
+updated: 2025-09-02
+---
+
+# Quill — Experience Spine
+
+> Illustrative example. Single-surface mobile (iOS + Android parity). Consumer posture, calm by default. Paired with `design-example-mobile.md` (Quill DESIGN.md). Demonstrates: microcopy as gating discipline, Inspiration & Anti-patterns earning its place, Responsive & Platform omitted (single-surface).
+
+## Foundation
+
+Single-surface mobile, iOS + Android with parity. No UI system named — inherits platform conventions for navigation, system gestures, dynamic type. `DESIGN.md` is the visual identity reference; this spine is the experience. Dark mode is the default surface; light is a setting.
+
+## Information Architecture
+
+| Surface | Reached from | Purpose |
+|---|---|---|
+| Today | App open (cold) | Today's prompt + entry composer |
+| Library | Tab bar | Past entries, searchable |
+| Entry detail | Library row tap | Read / edit one entry |
+| Settings | Today header gear | Account, export, theme |
+
+Bottom tab bar (Today / Library / Settings). No drawer. Modal stacks one level deep, never two.
+
+→ Composition reference: `mockups/today-cold.html`, `mockups/composer.html`. Spine wins on conflict.
+
+## Voice and Tone
+
+Microcopy. Brand voice and aesthetic posture live in `DESIGN.md`.
+
+| Do | Don't |
+|---|---|
+| "Today's prompt." | "Time to write!" |
+| "Saved." | "✓ Auto-saved successfully" |
+| "We couldn't reach the cloud — your work is on this device." | "Network error" |
+| Short, complete sentences. | Streak counters, encouragement, exclamation marks. |
+
+## Component Patterns
+
+Behavioral. Visual specs live in `DESIGN.md.Components`.
+
+| Component | Use | Behavioral rules |
+|---|---|---|
+| Prompt card | Today | One per day. Tap opens composer. |
+| Composer | Today + entry detail | No formatting toolbar in v1. Autosave on pause ≥ 600ms. |
+| Entry row | Library list | Tap → entry detail. Long-press reserved for system text selection. |
+| Save indicator | Composer header | Cycles `Editing…` → `Saved.` (≥ 800ms visible). |
+| Settings row | Settings list | Tap → detail or toggle. |
+
+## State Patterns
+
+| State | Surface | Treatment |
+|---|---|---|
+| Cold open | Today | Show today's prompt (cached). If no cache, `Today's prompt is loading.` with skeleton. |
+| Empty library | Library | `No entries yet — Today's prompt is your first.` Link to Today. |
+| Search empty | Library search | `No matches.` No suggestions. |
+| Offline write | Composer | Save locally. No banner. Sync on next foreground. |
+| Sync error | Settings → Account | Surfaced here only. Never block writing. |
+| Focus | Composer | Native cursor + keyboard. No custom focus chrome. |
+
+## Interaction Primitives
+
+- Tap to act. Long-press reserved for system text selection.
+- Swipe-to-delete on entry rows (native pattern, confirm sheet).
+- Pull-to-refresh on Library only.
+- **Banned:** carousels, hero animations on open, badge counts, streaks, push-notification re-engagement.
+
+## Accessibility Floor
+
+Behavioral. Visual contrast lives in `DESIGN.md`.
+
+- VoiceOver / TalkBack: every interactive element labeled with role + state. Save indicator announces `Saved` on transition.
+- Dynamic type honored through `DESIGN.md` typography tokens. UI must remain legible at largest setting — no truncated controls.
+- Reduce Motion: skip the save-indicator fade; show `Saved.` immediately.
+- Tap targets ≥ 44pt (iOS) / 48dp (Android).
+- Focus traversal follows reading order on every surface.
+
+## Inspiration & Anti-patterns
+
+- **Lifted from Day One:** the single daily entry framing — one prompt, one composer, no inbox.
+- **Lifted from iA Writer:** the no-toolbar composer; formatting is a settings-level decision, not a per-entry one.
+- **Rejected — Streaks (Duolingo, most habit apps):** streaks weaponize the user's calendar. Quill's value is showing up *today*, not punishing missed days.
+- **Rejected — AI prompt suggestions inside the composer:** the composer is for writing, not negotiating with a model. AI lives only in the daily prompt generation.
+
+## Key Flows
+
+### Flow 1 — Daily write (Mira, late evening, after work)
+
+1. Mira opens app.
+2. Today surface shows today's prompt (cached if offline).
+3. She taps the composer entry point.
+4. Composer opens, keyboard active.
+5. She writes; autosave fires on pause.
+6. She taps Back.
+7. **Climax:** Today surface shows `Saved.` and the entry's first line below the prompt — proof the day is captured.
+
+Failure: cold prompt fetch fails → composer still opens with cached generic prompt; banner on Today only after Mira returns.
+
+### Flow 2 — Recall past entry (Mira, three weeks later, looking for what she wrote about her mother)
+
+1. Mira taps Library.
+2. Scrolls or searches.
+3. Taps entry row.
+4. Entry detail opens in read mode.
+5. She taps anywhere to enter edit mode (cursor at tap point).
+6. Edits autosave.
+7. **Climax:** `Saved.` visible in entry header — the older self and the present self are in continuous conversation.
+
+Empty state: no entries → message routes back to Today.
diff --git a/.agents/skills/bmad-ux/assets/experience-example-shadcn.md b/.agents/skills/bmad-ux/assets/experience-example-shadcn.md
new file mode 100644
index 00000000..d46083c8
--- /dev/null
+++ b/.agents/skills/bmad-ux/assets/experience-example-shadcn.md
@@ -0,0 +1,133 @@
+---
+name: Drift
+status: final
+sources:
+  - {planning_artifacts}/prds/drift-2026-03-12/prd.md
+updated: 2026-04-02
+---
+
+# Drift — Experience Spine
+
+> Illustrative example. Single-surface responsive web. shadcn/ui on Next.js + Tailwind. Paired with `design-example-shadcn.md` (Drift DESIGN.md). Demonstrates: component-library inheritance, keyboard-first interaction primitives, the "shadcn + brand-layer" pattern that covers most modern web SaaS.
+
+## Foundation
+
+Single-surface responsive web. shadcn/ui on Next.js 15+ with Tailwind CSS. The component library does most of the work; brand discipline is "respect the defaults except where the brand layer overrides them." `DESIGN.md` is the visual identity reference and names the override surface; this spine is the experience. Single-tenant per project; users can belong to multiple projects but each project is a self-contained workspace.
+
+## Information Architecture
+
+| Surface | Reached from | Purpose |
+|---|---|---|
+| Today | App open / `g t` | Current focus, in-progress tasks pulled from all projects |
+| Projects | Sidebar / `g p` | List of active and archived projects |
+| Project detail | Projects row / `g 1`–`g 9` | Tasks in this project, organized by lane |
+| Search | `⌘K` / `Ctrl+K` | Command palette — surface, navigate, act |
+| Settings | Avatar menu | Account, theme, keyboard shortcuts, billing |
+
+Sidebar collapses to icons on `md`; becomes a `Sheet` on `sm`. Modal stacks one level deep (e.g., open `Dialog` on top of a surface, never on top of another dialog).
+
+→ Composition reference: `mockups/today.html`, `mockups/project-detail.html`, `mockups/command-palette.html`. Spine wins on conflict.
+
+## Voice and Tone
+
+Microcopy. Brand voice and aesthetic posture live in `DESIGN.md`.
+
+| Do | Don't |
+|---|---|
+| "What are you working on?" | "Let's get productive! 🚀" |
+| "3 tasks in motion" | "You have 3 active items." |
+| "Closed. Nice work." | "Task completed successfully ✓" |
+| "Nothing in motion. Pick something." | "No active tasks. Click below to get started!" |
+| Manager-facing: counts and verbs. Employee-facing: same. | Different tone per audience — Drift talks to everyone the same way. |
+
+## Component Patterns
+
+Behavioral. Visual specs live in `DESIGN.md.Components` (or in shadcn defaults, when inherited).
+
+| Component | Use | Behavioral rules |
+|---|---|---|
+| Task row | Projects, Today | Click anywhere on row opens edit dialog. Checkbox toggles done state with optimistic update. Hover reveals quick-actions (`focus`, `defer`, `archive`). |
+| Focus card | Today, Project detail | At most one focus card per surface — the task or project marked with `focus` state. `f` keyboard shortcut sets focus on the active row. |
+| Command palette | Global (⌘K) | Fuzzy search across all projects, tasks, and commands. `Enter` fires the highlighted result. `→` previews a result. Escape closes. |
+| Project header | Project detail | Inline-editable title (click to edit, blur to save). Status pill: active / archived / done. |
+| Empty state | Anywhere | shadcn's empty pattern + one Drift-specific sentence. `display-sm` for the headline, body text below, single primary action. |
+
+## State Patterns
+
+| State | Surface | Treatment |
+|---|---|---|
+| Cold app load | Today | shadcn `Skeleton` rows (4-6) match expected layout. Resolves on data. |
+| No focus | Today | `display-sm`: "Nothing in motion. Pick something." Below: list of in-progress tasks from all projects. |
+| Empty project | Project detail | `display-sm`: "{Project title} is empty." Body: "Add a first task to get going." Single primary button. |
+| Command palette no matches | ⌘K | "No matches. Start typing a task or project name, or pick an action below." Followed by 4-5 common commands. |
+| Offline | Global (status bar) | shadcn `Toast` once: "You're offline. Changes will sync when you reconnect." Local writes continue. |
+| Permission denied | Projects (others' private) | Surface hidden from sidebar. No "blocked" screen. |
+| Stale data | Project detail | If background refresh detects changes, shadcn `Toast`: "Updated by {user_name}. Refresh." Manual refresh, no auto. |
+
+## Interaction Primitives
+
+**Keyboard-first.** Drift's primary audience is developers and power users; the keyboard surface is the product, the mouse is fallback.
+
+- `⌘K` / `Ctrl+K` — Command palette (universal)
+- `g t` / `g p` — Go to Today / Projects (vim-style)
+- `g 1`–`g 9` — Go to project by sidebar position
+- `f` — Set focus on highlighted task/project
+- `c` — Create new task (context-aware: in the active project)
+- `Esc` — Close dialogs, exit edit mode, clear command palette
+- `/` — Focus search in current surface
+
+**Mouse:** click to act, drag deferred to v2. Hover reveals row actions on `md+` (touch users tap to reveal).
+
+**Banned everywhere:** infinite scroll (pagination only), drag-to-reorder in v1, hover-only affordances on `sm` viewports, modal stacks > 1 level deep.
+
+## Accessibility Floor
+
+Behavioral. Visual contrast lives in `DESIGN.md` (inherits shadcn's WCAG AA-compliant defaults; brand overrides verified to maintain ratios).
+
+- WCAG 2.2 AA across the responsive web surface.
+- Screen reader announces page surface on navigation: "Today, focus surface" / "Project: {name}, task list, {N} tasks."
+- Keyboard shortcuts available without modifier on most surfaces (vim-style `g t` etc.) — users with motor-control limitations get the same surface as power users.
+- `Tab` order matches reading order on every surface. `Esc` always closes the topmost modal/popover.
+- Command palette is fully keyboard-operable; results announce as they update via `aria-live`.
+- Focus rings inherit shadcn's `ring` token — visible at AA contrast against `background`.
+
+## Responsive & Platform
+
+| Breakpoint | Behavior |
+|---|---|
+| `≥ lg` (1024px+) | Sidebar visible. Today is a 2-column layout: focus + in-motion list. |
+| `md` (768–1023px) | Sidebar collapses to icons. Today stacks to single column. |
+| `< md` (`sm`) | Sidebar becomes a `Sheet` triggered from top bar. Command palette opens fullscreen. |
+
+Drift is responsive web, not a native mobile app. The product works on phones for read + simple-edit, but the primary surface is desktop / laptop.
+
+## Inspiration & Anti-patterns
+
+- **Lifted from Linear:** the keyboard-first discipline. `⌘K` is the command center; vim-style nav (`g t`); no drag for primary navigation; status pill vocabulary.
+- **Lifted from Notion:** inline-editable titles. Click-to-edit on project header, blur to save. No edit/view mode toggle.
+- **Lifted from shadcn:** the entire surface vocabulary. Drift's brand is *what we add to shadcn*, not a from-scratch design system. This is a deliberate posture, not a shortcut.
+- **Rejected — Streaks, badges, achievement notifications:** Drift is a tool, not a habit app. Task closure is its own reward; no celebratory animation, no "🎉 5-day streak!" toast.
+- **Rejected — AI-suggested next tasks:** Drift surfaces what's in motion, doesn't tell the user what to work on. The user picks focus; the tool surfaces consequences.
+- **Rejected — Multi-column kanban as default project view:** lists are linear; kanban hides progress behind columns. Optional v2; not the default.
+
+## Key Flows
+
+### Flow 1 — Morning focus (Sarah, solo founder, 8:45am Tuesday)
+
+1. Sarah opens Drift in a browser tab.
+2. App loads Today. `display-sm`: "Welcome back, Sarah." Focus card shows yesterday's marked task — "Finish landing page hero copy" — still in motion.
+3. She hits `⌘K`, types "ship hero", sees the matching task and presses Enter to open it.
+4. Inline edit: she updates the task description with two new bullets. Tab + Tab triggers save.
+5. **Climax:** Sarah closes the dialog. Today re-renders: focus card still shows the hero task, but now with the updated body text visible at a glance. She doesn't have to navigate anywhere — the surface that greeted her now reflects the work she just did. She picks up her coffee and starts writing.
+
+Failure: data save fails → shadcn `Toast` (destructive variant): "Couldn't save. Trying again." Inline edit retained; another `Enter` retries.
+
+### Flow 2 — Async handoff (Devon and Mara, small remote team, mid-afternoon)
+
+1. Devon finishes wiring up the auth flow and marks the task `done`.
+2. Mara, time-zoned three hours ahead and online during overlap, opens Drift.
+3. Today loads; her focus is on her own front-end work, but the in-motion list shows Devon's auth task now marked done and one task below it newly assigned to her — "Wire up post-auth redirect" — that Devon set during checkout.
+4. She hits `f` on the row to mark it as her focus, then `Enter` to open it.
+5. **Climax:** The focus card swaps. Her surface now shows the post-auth redirect task as the live thing she's working on; the projects sidebar shows the Auth project highlighted; the command palette `⌘K` defaults its first result to "Go to Auth project." The state of the team's progress is *embedded in her surface* — no Slack thread to scroll, no status doc to read.
+
+Failure: Devon hadn't actually assigned the follow-up — Mara mis-assigned to herself. She hits `Esc`, `f` again to unfocus, and reassigns to Devon. No "are you sure" dialog; Drift trusts the user.
diff --git a/.agents/skills/bmad-ux/assets/headless-schemas.md b/.agents/skills/bmad-ux/assets/headless-schemas.md
new file mode 100644
index 00000000..1211393c
--- /dev/null
+++ b/.agents/skills/bmad-ux/assets/headless-schemas.md
@@ -0,0 +1,84 @@
+# Headless Mode JSON Schemas
+
+Every headless run ends with one of these payloads. Omit keys for artifacts not produced.
+
+## Common fields
+
+- `status` — `"complete"`, `"blocked"`, or `"partial"`
+- `intent` — `"create"`, `"update"`, or `"validate"` (matches the detected intent)
+- `reason` — required when `status` is `"blocked"`; one-sentence explanation
+- `assumptions` — array of inferred values that were not directly confirmed by inputs
+- `open_questions` — array of items that need a human decision before the artifact can be considered final
+
+## Create
+
+```json
+{
+  "status": "complete",
+  "intent": "create",
+  "design": "{doc_workspace}/DESIGN.md",
+  "experience": "{doc_workspace}/EXPERIENCE.md",
+  "memlog": "{doc_workspace}/.memlog.md",
+  "working_artifacts": ["{doc_workspace}/.working/color-themes-1.html"],
+  "promoted_artifacts": {
+    "mockups": ["{doc_workspace}/mockups/direction-calm-sage.html"],
+    "wireframes": ["{doc_workspace}/wireframes/ia-2026-05-19.excalidraw"]
+  },
+  "open_questions": [],
+  "assumptions": [],
+  "external_handoffs": [
+    {"directive": "Confluence upload", "tool": "corp:confluence_upload", "url": "https://confluence.corp/DESIGN/123", "status": "ok"}
+  ]
+}
+```
+
+The `working_artifacts` and `promoted_artifacts` keys are optional and omitted entirely when empty. Headless Create runs default to not enabling creative tools — both keys are typically absent in headless output unless the caller enabled them.
+
+## Update
+
+```json
+{
+  "status": "complete",
+  "intent": "update",
+  "design": "{doc_workspace}/DESIGN.md",
+  "experience": "{doc_workspace}/EXPERIENCE.md",
+  "memlog": "{doc_workspace}/.memlog.md",
+  "changes_summary": "1-3 sentences describing what changed and why",
+  "conflicts_with_prior_decisions": [],
+  "open_questions": [],
+  "external_handoffs": [
+    {"directive": "Confluence upload", "tool": "corp:confluence_upload", "url": "https://confluence.corp/DESIGN/123", "status": "ok"}
+  ]
+}
+```
+
+## Validate
+
+```json
+{
+  "status": "complete",
+  "intent": "validate",
+  "validation_report": "{doc_workspace}/validation-report.md",
+  "findings_summary": {
+    "critical": 0,
+    "high": 0,
+    "medium": 0,
+    "low": 0
+  },
+  "offer_to_update": true
+}
+```
+
+`validation_report` is always written for Validate intent — the path here is required, not optional.
+
+## Blocked
+
+```json
+{
+  "status": "blocked",
+  "intent": "update",
+  "reason": "Change signal ambiguous — could be a brand refresh or an accessibility audit response; no inferred direction"
+}
+```
+
+Always include the intent (best-guess if not certain) and a one-sentence `reason`.
diff --git a/.agents/skills/bmad-ux/assets/key-screens.md b/.agents/skills/bmad-ux/assets/key-screens.md
new file mode 100644
index 00000000..62f057c9
--- /dev/null
+++ b/.agents/skills/bmad-ux/assets/key-screens.md
@@ -0,0 +1,29 @@
+# Key Screens Renderer
+
+Subagent prompt. Fired at Finalize (or during late Discovery once layout decisions firm up). Produces 1:1 HTML mocks of the load-bearing surfaces so the spine can link to them as visual reference. Spine remains the contract; mocks illustrate.
+
+## Inputs
+
+`.memlog.md`, the current drafts `DESIGN.md` and `EXPERIENCE.md`, `.working/` (especially the chosen color-theme and direction mocks), source PRD. The user names which surfaces to render — typically 2-4: the canonical entry surface, the most complex flow's hero screen, any load-bearing overlay/modal, and (when present) the Week / list / dashboard view.
+
+## What to render
+
+One HTML file per screen, at `.working/key-{slug}.html`. Each file: realistic device frame (phone or browser), real product content from the conversation (no lorem), every visible string voice-checked against `.memlog.md`, all decided tokens applied. Show one canonical state per screen; if a surface has a load-bearing alternate state (focus, error, crisis-card-present), render it as a second column or section in the same file.
+
+Inline CSS, system fonts, no JS, no network. The mock must render fully offline. Comment block at the top of the `<style>` notes which spine sections govern this screen so a future reader knows what to check.
+
+## What to return
+
+A compact summary to the parent:
+- file path per screen
+- one-line caption per screen ("Today picker at rest; accent on Thought record")
+- which spine sections each mock illustrates (Component Patterns rows, State Patterns rows, Flow steps)
+
+The parent, at Finalize "Promote working artifacts," uses this summary to insert inline `mockups/...` links into the relevant spine sections.
+
+## Anti-patterns
+
+- Do not invent layout — every composition decision must trace to a `.working/` artifact or a confirmation in `.memlog.md`. If a layout question is open, the mock is premature.
+- Do not show every screen of every flow — 2-4 load-bearing surfaces, not 14.
+- Do not stage marketing copy. Strings come from `.memlog.md` and voice rules.
+- Do not introduce a new pattern not in the spine's Component Patterns table. If you need one, log it and ask before rendering.
diff --git a/.agents/skills/bmad-ux/assets/validation-report-template.html b/.agents/skills/bmad-ux/assets/validation-report-template.html
new file mode 100644
index 00000000..a5d39be8
--- /dev/null
+++ b/.agents/skills/bmad-ux/assets/validation-report-template.html
@@ -0,0 +1,319 @@
+<!DOCTYPE html>
+<!--
+  UX Design Validation Report — skeleton template.
+
+  This file is a starter the synthesis pass fills in directly. There is no
+  substitution engine. The LLM:
+    1. Reads {doc_workspace}/review-rubric.md and every review-{slug}.md from
+       additional reviewers.
+    2. Copies this skeleton.
+    3. Replaces the placeholder content (everything between TEMPLATE markers)
+       with the consolidated review, preserving the structure and CSS.
+    4. Writes the result to {doc_workspace}/validation-report.html.
+    5. Writes a markdown twin to {doc_workspace}/validation-report.md.
+
+  Visual rules the LLM must preserve:
+    - The container width, the color tokens, the typography.
+    - One dimension = one collapsible <section class="dimension">.
+    - Verdict pill uses the verdict-* class matching its judgment.
+    - Severity badge uses the badge-sev-* class matching its level.
+    - Each extra reviewer (accessibility, adversarial, etc.) gets its own
+      collapsible section below the rubric dimensions.
+    - The footer always shows the artifact paths and timestamp.
+-->
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<title>UX Design Validation: TEMPLATE_UX_SPEC_NAME</title>
+<style>
+  :root {
+    --bg: #fafaf9;
+    --surface: #ffffff;
+    --border: #e7e5e4;
+    --text: #1c1917;
+    --muted: #78716c;
+
+    --verdict-strong: #16a34a;
+    --verdict-adequate: #65a30d;
+    --verdict-thin: #d97706;
+    --verdict-broken: #dc2626;
+
+    --sev-low: #64748b;
+    --sev-medium: #ca8a04;
+    --sev-high: #ea580c;
+    --sev-critical: #dc2626;
+  }
+  * { box-sizing: border-box; }
+  html, body { margin: 0; padding: 0; }
+  body {
+    font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Inter, system-ui, sans-serif;
+    background: var(--bg);
+    color: var(--text);
+    line-height: 1.6;
+    font-size: 15px;
+  }
+  .container { max-width: 960px; margin: 0 auto; padding: 32px 24px 64px; }
+
+  header.report-header {
+    display: flex;
+    align-items: flex-start;
+    justify-content: space-between;
+    gap: 24px;
+    padding-bottom: 16px;
+    border-bottom: 1px solid var(--border);
+    margin-bottom: 24px;
+  }
+  .title h1 { margin: 0; font-size: 22px; font-weight: 600; letter-spacing: -0.01em; }
+  .title .subtitle { color: var(--muted); font-size: 13px; margin-top: 4px; font-family: ui-monospace, "SF Mono", Menlo, monospace; }
+
+  .synthesis {
+    background: var(--surface);
+    border: 1px solid var(--border);
+    border-left: 3px solid var(--muted);
+    border-radius: 8px;
+    padding: 18px 22px;
+    margin-bottom: 24px;
+    font-size: 15.5px;
+  }
+  .synthesis p { margin: 0 0 10px; }
+  .synthesis p:last-child { margin-bottom: 0; }
+
+  .dimension-summary {
+    display: grid;
+    grid-template-columns: repeat(auto-fit, minmax(180px, 1fr));
+    gap: 10px;
+    margin-bottom: 24px;
+  }
+  .dim-card {
+    background: var(--surface);
+    border: 1px solid var(--border);
+    border-radius: 8px;
+    padding: 12px 14px;
+  }
+  .dim-card .dim-name { font-size: 13px; color: var(--muted); margin-bottom: 6px; }
+  .dim-card .dim-verdict { font-size: 14px; font-weight: 600; }
+
+  section.dimension, section.reviewer-section { margin-bottom: 14px; }
+  section.dimension details, section.reviewer-section details {
+    background: var(--surface);
+    border: 1px solid var(--border);
+    border-radius: 8px;
+    overflow: hidden;
+  }
+  section summary {
+    padding: 14px 20px;
+    cursor: pointer;
+    user-select: none;
+    list-style: none;
+    display: flex;
+    align-items: center;
+    gap: 12px;
+  }
+  section summary::-webkit-details-marker { display: none; }
+  section summary::before {
+    content: "▸";
+    display: inline-block;
+    color: var(--muted);
+    transition: transform 0.15s ease;
+  }
+  section details[open] summary::before { transform: rotate(90deg); }
+  section summary h2 {
+    display: inline;
+    margin: 0;
+    font-size: 16px;
+    font-weight: 600;
+    letter-spacing: -0.005em;
+    flex: 1;
+  }
+  .verdict-pill {
+    font-size: 11px;
+    padding: 3px 10px;
+    border-radius: 999px;
+    font-weight: 600;
+    text-transform: uppercase;
+    letter-spacing: 0.04em;
+    color: white;
+  }
+  .verdict-strong { background: var(--verdict-strong); }
+  .verdict-adequate { background: var(--verdict-adequate); }
+  .verdict-thin { background: var(--verdict-thin); }
+  .verdict-broken { background: var(--verdict-broken); }
+
+  .dim-body { padding: 4px 20px 18px; }
+  .dim-judgment { color: var(--text); font-size: 14.5px; }
+  .dim-judgment p { margin: 0 0 10px; }
+
+  .findings-list { padding: 0 20px 4px; }
+  article.finding {
+    padding: 14px 0;
+    border-top: 1px solid var(--border);
+  }
+  article.finding:first-child { border-top: none; }
+  article.finding header {
+    display: flex;
+    align-items: center;
+    gap: 10px;
+    flex-wrap: wrap;
+    margin-bottom: 6px;
+  }
+  .badge {
+    font-size: 10.5px;
+    padding: 3px 8px;
+    border-radius: 4px;
+    font-weight: 600;
+    text-transform: uppercase;
+    letter-spacing: 0.04em;
+    line-height: 1.4;
+  }
+  .badge-sev-low { background: rgba(100, 116, 139, 0.12); color: var(--sev-low); }
+  .badge-sev-medium { background: rgba(202, 138, 4, 0.14); color: var(--sev-medium); }
+  .badge-sev-high { background: rgba(234, 88, 12, 0.14); color: var(--sev-high); }
+  .badge-sev-critical { background: rgba(220, 38, 38, 0.14); color: var(--sev-critical); }
+  .finding-title { margin: 0; font-size: 15px; font-weight: 500; flex: 1; min-width: 200px; }
+  .finding-location { font-family: ui-monospace, "SF Mono", Menlo, monospace; font-size: 12.5px; color: var(--muted); }
+  .finding-note, .finding-fix { margin-top: 6px; font-size: 14px; }
+  .finding-fix strong { color: var(--muted); font-weight: 500; }
+
+  .reviewer-source {
+    font-size: 12px;
+    color: var(--muted);
+    font-family: ui-monospace, "SF Mono", Menlo, monospace;
+  }
+
+  .mechanical {
+    background: var(--surface);
+    border: 1px solid var(--border);
+    border-radius: 8px;
+    padding: 16px 20px;
+    margin-top: 24px;
+    font-size: 14px;
+  }
+  .mechanical h3 { margin: 0 0 10px; font-size: 14px; font-weight: 600; color: var(--muted); text-transform: uppercase; letter-spacing: 0.04em; }
+  .mechanical ul { margin: 0; padding-left: 20px; }
+  .mechanical li { margin-bottom: 4px; color: var(--text); }
+
+  footer.report-footer {
+    margin-top: 40px;
+    padding-top: 16px;
+    border-top: 1px solid var(--border);
+    font-size: 12px;
+    color: var(--muted);
+    font-family: ui-monospace, "SF Mono", Menlo, monospace;
+  }
+  footer .meta { display: flex; gap: 24px; flex-wrap: wrap; }
+</style>
+</head>
+<body>
+  <div class="container">
+
+    <!-- TEMPLATE: header. Fill ux spec name and ux spec path. No overall grade —
+         per-category verdicts and severity counts below carry the picture. -->
+    <header class="report-header">
+      <div class="title">
+        <h1>TEMPLATE_UX_SPEC_NAME — UX Design Validation Report</h1>
+        <div class="subtitle">TEMPLATE_UX_SPEC_PATH</div>
+      </div>
+    </header>
+
+    <!-- TEMPLATE: overall synthesis paragraphs. Lift directly from
+         review-rubric.md "Overall verdict" section; expand if extra reviewers
+         (accessibility, adversarial, etc.) materially shift the picture.
+         Wrap each paragraph in <p>. -->
+    <div class="synthesis">
+      <p>TEMPLATE_SYNTHESIS_PARAGRAPH</p>
+    </div>
+
+    <!-- TEMPLATE: dimension summary cards. One per rubric category, in this
+         canonical order (from references/validate.md):
+           1. Flow coverage
+           2. Token completeness
+           3. Component coverage
+           4. State coverage
+           5. Visual reference coverage
+           6. Bloat & overspecification
+           7. Inheritance discipline
+           8. Shape fit
+         The dim-verdict text uses one of: strong | adequate | thin | broken. -->
+    <div class="dimension-summary">
+      <div class="dim-card">
+        <div class="dim-name">TEMPLATE_CATEGORY_NAME</div>
+        <div class="dim-verdict" style="color: var(--verdict-TEMPLATE_VERDICT)">TEMPLATE_VERDICT_TEXT</div>
+      </div>
+      <!-- repeat for each of the eight rubric categories -->
+    </div>
+
+    <!-- TEMPLATE: one section per rubric category, same canonical order as the
+         summary cards above. Skip a category entirely if the rubric review
+         marked it n/a for this UX spec. Open the section by default if verdict
+         is thin or broken. -->
+    <section class="dimension">
+      <details open>
+        <summary>
+          <h2>TEMPLATE_CATEGORY_NAME</h2>
+          <span class="verdict-pill verdict-TEMPLATE_VERDICT">TEMPLATE_VERDICT_TEXT</span>
+        </summary>
+        <div class="dim-body">
+          <div class="dim-judgment">
+            <p>TEMPLATE_DIMENSION_JUDGMENT</p>
+          </div>
+        </div>
+        <div class="findings-list">
+          <!-- TEMPLATE: zero or more findings -->
+          <article class="finding">
+            <header>
+              <span class="badge badge-sev-TEMPLATE_SEVERITY">TEMPLATE_SEVERITY</span>
+              <h3 class="finding-title">TEMPLATE_FINDING_TITLE</h3>
+              <span class="finding-location">TEMPLATE_LOCATION</span>
+            </header>
+            <div class="finding-note">TEMPLATE_FINDING_NOTE</div>
+            <div class="finding-fix"><strong>Fix:</strong> TEMPLATE_SUGGESTED_FIX</div>
+          </article>
+        </div>
+      </details>
+    </section>
+
+    <!-- TEMPLATE: one section per extra reviewer that ran (accessibility,
+         adversarial, brand-conformance, etc.). Skip this block entirely if
+         only the rubric walker ran. -->
+    <section class="reviewer-section">
+      <details>
+        <summary>
+          <h2>Accessibility review</h2>
+          <span class="reviewer-source">TEMPLATE_REVIEWER_SOURCE_FILE</span>
+        </summary>
+        <div class="dim-body">
+          <div class="dim-judgment">
+            <p>TEMPLATE_REVIEWER_PREAMBLE</p>
+          </div>
+        </div>
+        <div class="findings-list">
+          <article class="finding">
+            <header>
+              <span class="badge badge-sev-TEMPLATE_SEVERITY">TEMPLATE_SEVERITY</span>
+              <h3 class="finding-title">TEMPLATE_FINDING_TITLE</h3>
+              <span class="finding-location">TEMPLATE_LOCATION</span>
+            </header>
+            <div class="finding-note">TEMPLATE_FINDING_NOTE</div>
+            <div class="finding-fix"><strong>Fix:</strong> TEMPLATE_SUGGESTED_FIX</div>
+          </article>
+        </div>
+      </details>
+    </section>
+
+    <!-- TEMPLATE: mechanical notes — short, bulleted. Skip if there are none. -->
+    <div class="mechanical">
+      <h3>Mechanical notes</h3>
+      <ul>
+        <li>TEMPLATE_MECHANICAL_NOTE</li>
+      </ul>
+    </div>
+
+    <footer class="report-footer">
+      <div class="meta">
+        <span>Rubric: TEMPLATE_RUBRIC_PATH</span>
+        <span>Generated: TEMPLATE_TIMESTAMP</span>
+      </div>
+    </footer>
+  </div>
+</body>
+</html>
diff --git a/.agents/skills/bmad-ux/customize.toml b/.agents/skills/bmad-ux/customize.toml
new file mode 100644
index 00000000..aad0829f
--- /dev/null
+++ b/.agents/skills/bmad-ux/customize.toml
@@ -0,0 +1,102 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-ux.
+# Overrides:
+#   {project-root}/_bmad/custom/bmad-ux.toml         (team)
+#   {project-root}/_bmad/custom/bmad-ux.user.toml    (personal)
+# Merge rules: scalars override, arrays append.
+
+[workflow]
+
+# Steps to run before/after standard activation. Append-only.
+activation_steps_prepend = []
+activation_steps_append = []
+
+# Persistent facts loaded at activation and kept in mind for the run.
+# Entries: literal sentence, `skill:NAME`, or `file:PATH` (glob ok).
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Runs at workflow completion. String or array of instructions.
+on_complete = ""
+
+# Reference DESIGN.md spines the distillation subagent reads to anchor shape
+# and editorial richness. Convention-compliant with the Google Labs DESIGN.md
+# spec (https://github.com/google-labs-code/design.md). Append entries via
+# override TOML to seed an org-specific canonical aesthetic.
+# Each entry: `file:PATH` (or bare relative path, resolved skill-relative).
+design_md_examples = [
+  "assets/design-example-mobile.md",
+  "assets/design-example-shadcn.md",
+  "assets/design-example-editorial.md",
+]
+
+# Reference EXPERIENCE.md spines for the behavioral/flow/IA layer. Each entry:
+# `file:PATH` (or bare relative path, resolved skill-relative).
+experience_md_examples = [
+  "assets/experience-example-mobile.md",
+  "assets/experience-example-shadcn.md",
+]
+
+# Design handoff targets — external tools that can take over the design /
+# visual identity work. The user runs the tool externally and saves outputs
+# (whatever the tool produces — DESIGN.md, Figma files, React components,
+# HTML mocks) to {doc_workspace}.
+# Each entry: `tool:NAME: <directive>`, `skill:NAME`, or plain-text descriptor.
+# Default: Google Stitch (emits DESIGN.md + per-screen HTML). Other producers:
+# Vercel v0, Figma, Galileo, Anima, internal generators.
+design_handoffs = [
+  "Google Stitch (https://stitch.withgoogle.com) — emits DESIGN.md + per-screen HTML. Paste assembled prompt; save outputs to {doc_workspace}.",
+]
+
+# HTML skeleton filled in by the validation synthesis pass.
+validation_report_template = "assets/validation-report-template.html"
+
+# Run folder. DESIGN.md, EXPERIENCE.md, .memlog.md, .working/
+# (creative-tool artifacts), imports/ (user-supplied screens / brand decks /
+# Figma exports / sketches), optional mockups/ and wireframes/ (promoted
+# artifacts), optional validation-report.* all land inside
+# {ux_output_path}/{run_folder_pattern}/.
+ux_output_path = "{planning_artifacts}/ux-designs"
+run_folder_pattern = "ux-{project_name}-{date}"
+
+# Creative tools registry. Collaborative renderers invoked on demand during
+# Discovery and at Finalize. Entry forms: `file:PATH`, `skill:NAME`,
+# `tool:MCP_TOOL: <directive>`, or plain text. Defaults ship for HTML color
+# themes, HTML design directions, Excalidraw wireframes (Discovery), and
+# 1:1 HTML key-screen mockups (Finalize). Working artifacts land in
+# {doc_workspace}/.working/; finalize promotes those with lasting reference
+# value to mockups/ or wireframes/. See references/creative-tools.md.
+creative_tools = [
+  "file:assets/color-themes.md",
+  "file:assets/design-directions.md",
+  "file:assets/excalidraw-wireframe.md",
+  "file:assets/key-screens.md",
+]
+
+# Polish passes applied to DESIGN.md and EXPERIENCE.md at finalize.
+# Entries: `skill:NAME`, `file:PATH`, or plain text directive.
+# Suggested order: structural → content/voice → prose mechanics.
+# The default entry runs bmad-review's two editorial lenses in order:
+# structure, then prose on top of the structure findings. The `lenses=` suffix
+# names them; drop it to let bmad-review pick what fits the content.
+doc_standards = [
+  "skill:bmad-review lenses=structure,prose",
+]
+
+# Information retrieval registry. Consulted on demand when the conversation
+# surfaces a matching need. Distinct from creative_tools (artifact production).
+# Example: "When researching component patterns, consult corp:design_system_search."
+external_sources = []
+
+# Routes outputs beyond local files at Finalize. Returned URLs/IDs surfaced
+# to the user. Unavailable tools skipped and flagged.
+# Example: "Upload DESIGN.md to Confluence via corp:confluence_upload (space_key='DESIGN')."
+external_handoffs = []
+
+# Reviewers spawned at Finalize step 4 and at Validate intent, alongside
+# the rubric walker. Entries: `skill:NAME`, `file:PATH`, or plain text.
+# Common ad-hoc add (judged by the skill): accessibility-focused reviewer
+# for consumer / regulated work.
+finalize_reviewers = []
diff --git a/.agents/skills/bmad-ux/references/creative-tools.md b/.agents/skills/bmad-ux/references/creative-tools.md
new file mode 100644
index 00000000..f42a971c
--- /dev/null
+++ b/.agents/skills/bmad-ux/references/creative-tools.md
@@ -0,0 +1,19 @@
+# Creative Tools
+
+`{workflow.creative_tools}` is a registry of collaborative renderers invoked on demand when seeing options helps the user decide. Entries follow the standard prefix convention: `skill:NAME`, `file:PATH`, `tool:MCP_TOOL_NAME: <directive>`, or plain-text directive.
+
+Defaults ship for HTML color themes, HTML design directions, Excalidraw wireframes (Discovery), and 1:1 HTML key-screen mocks (Finalize). Teams append more via override TOML — Figma MCP, custom skills, prompt-based mood boards.
+
+## When to invoke
+
+Decision moments where a visual beats more conversation: picking color tokens, picking a visual personality among directions, sketching IA, mocking a tricky flow. Fast-path users typically skip; coaching-path users typically lean in. Read the room.
+
+## Artifact handling
+
+Every renderer writes to `{doc_workspace}/.working/` with a descriptive filename. `.working/` is the audit trail and survives the run. At Finalize, the facilitator walks `.working/` with the user and promotes artifacts with lasting reference value to `{doc_workspace}/mockups/` (HTML anchoring a brand or layout decision) or `{doc_workspace}/wireframes/` (Excalidraw a dev would glance at). Bar for promotion: *would a future reader of `DESIGN.md` or `EXPERIENCE.md` open this?* Default is leave-in-`.working/`.
+
+## Renderer contract
+
+The parent passes the subagent: current `.memlog.md`, relevant prior `.working/` captures, the user's stated intent for this pass, the output path. The subagent writes its artifact under `.working/` and returns ONLY a compact summary (file path, one line per variant, mode coverage). Parent never holds the full payload.
+
+For HTML, open in the browser when interactive with the platform opener — `open "PATH"` on macOS, `xdg-open` on Linux, `start ""` on Windows, path always double-quoted. On failure, give the user the path instead. Skip in headless.
diff --git a/.agents/skills/bmad-ux/references/design-md-spec.md b/.agents/skills/bmad-ux/references/design-md-spec.md
new file mode 100644
index 00000000..f685b2ba
--- /dev/null
+++ b/.agents/skills/bmad-ux/references/design-md-spec.md
@@ -0,0 +1,50 @@
+# DESIGN.md Spec — Working Reference
+
+Source of truth: [google-labs-code/design.md](https://github.com/google-labs-code/design.md) (Apache 2.0, Google Labs, April 2026). This file is a working summary; the URL wins on conflict.
+
+## Structure
+
+YAML frontmatter (machine-readable tokens) + markdown body (human-readable rationale, prose sections).
+
+## Frontmatter tokens
+
+| Key | Type | Notes |
+|---|---|---|
+| `name` | string | Required. Brand or system name. |
+| `description` | string | One-line statement of what this system is. |
+| `colors` | flat object | Kebab-case keys. Values are hex strings (`'#FBF9F4'`). |
+| `typography` | nested object | Each value: an object with any subset of `fontFamily`, `fontSize`, `fontWeight`, `lineHeight`, `letterSpacing`. |
+| `rounded` | object | Scale names (`sm`, `md`, `lg`, `xl`, `full`, `DEFAULT`) → CSS dimensions. `full` is conventionally `9999px`. |
+| `spacing` | object | Scale levels (`'1'`, `'2'`, ...) or named tokens (`gutter`, `margin-mobile`, `editorial-gap`) → dimensions. |
+| `components` | object | Component-name → object of component tokens mapped to values or `{path.to.token}` references. |
+
+## Body sections (omittable, order-locked when present)
+
+1. **Brand & Style** — Aesthetic posture in prose. The editorial voice — what *kind* of thing this is.
+2. **Colors** — Per-color story. Why each exists, where it's used, what it's *not* used for.
+3. **Typography** — Type roles, ramp, and rules. Platform conventions noted semantically when inherited.
+4. **Layout & Spacing** — Spacing scale narrative, grid behavior, margins, gutters, breakpoint rules.
+5. **Elevation & Depth** — Shadow language and tonal layering rules.
+6. **Shapes** — Corner radii rules and the aesthetic logic behind them.
+7. **Components** — Per-component visual specs: anatomy, color usage, sizing, state appearance.
+8. **Do's and Don'ts** — Hard visual rules — what to do, what to avoid.
+
+Sections may be omitted when not relevant; order is locked when present.
+
+## Cross-reference syntax
+
+`{path.to.token}` used in prose and inside component objects to reference frontmatter tokens. Examples:
+
+- `{colors.primary}`
+- `{typography.body.fontSize}`
+- `{rounded.md}`
+- `{spacing.4}`
+
+The path follows the YAML structure.
+
+## Common patterns
+
+- **Light/dark mode.** Either separate kebab-case tokens (`surface-base` / `surface-base-dark`) or separate DESIGN.md files per mode. The spec allows either; pick the form that reads cleanest for the product.
+- **Platform conventions.** When inheriting from native platforms (iOS UIKit, Android Compose, Apple Human Interface Guidelines), use a `note` field instead of literal values: `{ note: 'iOS Title 1 · Android Headline Small' }`. The spec is the spec; the platform owns the rendered values.
+- **UI-system inheritance.** When inheriting from shadcn / MUI / Tailwind / internal design system, reference the system's tokens by name rather than restating values. DESIGN.md specifies only the deltas (brand color overrides, typography swaps, component customizations).
+- **Component tokens.** The `components` frontmatter entry maps each named component (e.g., `button-primary`) to its specific token values. Use `{path.to.token}` references freely; the resolver flattens at consumption time.
diff --git a/.agents/skills/bmad-ux/references/headless.md b/.agents/skills/bmad-ux/references/headless.md
new file mode 100644
index 00000000..7f1f8588
--- /dev/null
+++ b/.agents/skills/bmad-ux/references/headless.md
@@ -0,0 +1,37 @@
+# Headless Mode
+
+Load this file when invoked headless. Follow it for the whole run.
+
+## Detection
+
+Headless when any of: caller sets `headless: true` (or harness equivalent); invocation is from another skill or non-interactive runner; `{workflow.activation_steps_prepend}` declares it; first message is an automation context pre-supplying inputs. Ambiguous → default interactive.
+
+## Inputs
+
+Free-form structured payload in the first message:
+
+- `intent` — `"create"`, `"update"`, or `"validate"`. If absent, infer from the artifact set.
+- **Create**: any source spec (PRD, brief, requirements list, design-thinking output, prior UX — text, path, or URL) plus brand / platform / accessibility notes; `doc_workspace` if a specific run folder is required.
+- **Update**: existing workspace containing `DESIGN.md` + `EXPERIENCE.md` (or path to either) + change signal.
+- **Validate**: existing workspace containing `DESIGN.md` + `EXPERIENCE.md` (or path to either). Workspace defaults to the spines' containing directory.
+
+Inferences → `assumptions[]`. Gaps needing a human decision → `open_questions[]`. Do not invent persona, brand, accessibility, or scope detail.
+
+Creative tools default off in headless. Caller can override; artifacts land in `.working/` and are not promoted unless the caller signals.
+
+## Behavior
+
+Do not ask. Do not greet. Complete the intent from what's provided, what exists in `{doc_workspace}`, or what you can discover. If intent stays ambiguous after inference, halt with `status: "blocked"` and a one-sentence `reason`.
+
+`status`:
+- `"complete"` — stands on its own.
+- `"partial"` — artifact produced but `open_questions[]` non-empty or critical inputs inferred.
+- `"blocked"` — no artifact produced.
+
+End with JSON matching `assets/headless-schemas.md`. `intent` reflects detected intent. Omit keys for artifacts not produced.
+
+## Mode-specific overrides
+
+**Update.** Apply the change. Log it via `uv run {project-root}/_bmad/scripts/memlog.py append --workspace {doc_workspace} --type change --text "<change + rationale>"`. Surface conflicts in `conflicts_with_prior_decisions[]`.
+
+**Validate.** Always write both `validation-report.html` and `validation-report.md` regardless of finding count. Always include `"offer_to_update": true`. Skip the browser-open step.
diff --git a/.agents/skills/bmad-ux/references/validate.md b/.agents/skills/bmad-ux/references/validate.md
new file mode 100644
index 00000000..86516bc7
--- /dev/null
+++ b/.agents/skills/bmad-ux/references/validate.md
@@ -0,0 +1,115 @@
+# Validate
+
+Critique an existing spine pair (`DESIGN.md` + `EXPERIENCE.md`) or any format of UX the user provides, without changing it. The synthesis pipeline below is also used at the Reviewer Gate during Create / Update Finalize.
+
+## Orient
+
+Subagent-extract from `.memlog.md`, sources in frontmatter, `imports/`, `mockups/`, `wireframes/`, `DESIGN.md`, `EXPERIENCE.md`. Parent assembles from extracts.
+
+## Reviewer Gate
+
+**Opt-in.** Reviewers are costly. At Finalize, ask first if the user wants to run UX validation with multiple subagent lenses. Default offered, easy skip. At Validate intent, skip that question, the user already invoked it.
+
+**Lens menu.** UNLESS HEADLESS MODE: Always present the lens picks before dispatching. Build the menu from: rubric walker (this file) + `{workflow.finalize_reviewers}` + ad-hoc reviewers the skill judges relevant. The user picks all, a subset, or none. Only picked lenses dispatch.
+
+Rubric walker prompt:
+
+> Validate the spine pair (`DESIGN.md` + `EXPERIENCE.md`) as the contract for downstream consumers (architecture, story-dev — human or AI). Can a consumer source-extract cleanly, with every reference resolving and every load-bearing decision committed? Read `{workflow.design_md_examples}` and `{workflow.experience_md_examples}` first.
+>
+> **Pass 1 — mechanical coverage.** Per category: extract, then list misses with location citations. No misses = **strong**.
+>
+> 1. **Flow coverage** (EXPERIENCE.md). Sources frontmatter → extract every UJ / requirement name. Verify each has a Key Flow with named protagonist, numbered steps, a climax beat, and a failure path where applicable.
+>
+> 2. **Token completeness** (DESIGN.md). Extract every token in the YAML frontmatter and every `{path.to.token}` reference in the prose. Verify each defined (see `references/design-md-spec.md` for type rules). **Color tokens missing hex (or light/dark pairs where applicable) are critical** — downstream code mirrors the spine. Platform conventions (native dynamic type, 8pt grid) may stay semantic. Contrast targets stated for load-bearing combinations.
+>
+> 3. **Component coverage** (both spines). Extract every component name used anywhere. Verify each has a row in DESIGN.md.Components (visual spec) *and* EXPERIENCE.md.Component Patterns (behavioral spec) — real rules, not one-word descriptions.
+>
+> 4. **State coverage** (EXPERIENCE.md). Walk every IA surface. List states it should have (empty, cold-load, focus, error, offline, permission-denied — whichever apply). Verify each covered.
+>
+> 5. **Visual reference coverage.** List every file in `mockups/`, `wireframes/`, `imports/`. Spines link to each inline at the relevant section and name what it illustrates; spines-win-on-conflict stated once. List orphans and unspecific references.
+>
+> **Pass 2 — judgment.** Verdict per category (*strong / adequate / thin / broken*); findings only where they add information.
+>
+> 6. **Bloat & overspecification.** Pixel specs where tokens cover it; source restatement (personas, FRs, scope); prose where a table works; sections no downstream consumer would read; decorative narrative untied to a decision. DESIGN.md prose may carry editorial voice; EXPERIENCE.md prose should not.
+>
+> 7. **Inheritance discipline.** `sources` frontmatter resolves. UJ / requirement names verbatim from sources. Glossary identical across spines and sources. Component names identical across all sections in both files. EXPERIENCE.md token references resolve to DESIGN.md tokens by name.
+>
+> 8. **Shape fit.** DESIGN.md sections in canonical order (Brand & Style → Colors → Typography → Layout & Spacing → Elevation & Depth → Shapes → Components → Do's and Don'ts; omittable but order-locked when present). EXPERIENCE.md required defaults present (Foundation, IA, Voice and Tone, Component Patterns, State Patterns, Interaction Primitives, Accessibility Floor, Key Flows). Dropped defaults defensible. Required-when-applicable present where triggered (Inspiration when sources / memlog show reference products or rejects; Responsive when multi-surface or breakpoints). Invented sections earn their place.
+>
+> Severity = downstream impact, not fix difficulty.
+>
+> Write to `{doc_workspace}/review-rubric.md`:
+>
+> ```markdown
+> # Spine Pair Review — {project_name}
+>
+> ## Overall verdict
+> [2–3 sentences]
+>
+> ## 1. Flow coverage — [verdict]
+> [What was checked.]
+> ### Findings
+> - **[critical|high|medium|low]** [finding] (location). *Fix:* [suggestion].
+>
+> (repeat 2–8)
+>
+> ## Mechanical notes
+> [Name inconsistencies, broken cross-refs, frontmatter completeness, Mermaid syntax.]
+> ```
+>
+> Return ONLY a compact summary: overall verdict, per-section verdicts, finding counts by severity, file path.
+
+The gate may dispatch `{workflow.finalize_reviewers}` and ad-hoc reviewers (accessibility for consumer / regulated). Each writes `review-{slug}.md` and returns a compact summary. Parallel.
+
+## Synthesis pipeline
+
+Under Validate intent, after every reviewer returns, render one consolidated report. Don't skip.
+
+1. Read every `{doc_workspace}/review-*.md`.
+2. Fill `{workflow.validation_report_template}`. No overall grade — the per-category verdicts and severity counts already say what's true. Synthesis paragraph lifts the rubric's overall verdict; add a second if extra reviewers shift the picture. One section per rubric category (open if thin / broken), one per extra reviewer (closed, adversarial voice preserved).
+3. Write `{doc_workspace}/validation-report.html`.
+4. Write the Markdown twin `{doc_workspace}/validation-report.md` — same content grouped by severity.
+5. Open HTML with the platform opener — `open "{doc_workspace}/validation-report.html"` on macOS, `xdg-open` on Linux, `start ""` on Windows, path always double-quoted. On failure, give the user the path instead. Skip headless.
+
+Re-running overwrites the consolidated report; individual `review-*.md` files persist.
+
+## Markdown twin shape
+
+```markdown
+# Validation Report — {project_name}
+
+- **DESIGN.md:** `{design_path}`
+- **EXPERIENCE.md:** `{experience_path}`
+- **Run at:** {ISO timestamp}
+
+## Overall verdict
+{synthesis paragraphs}
+
+## Category verdicts
+- Flow coverage — {verdict}
+- Token completeness — {verdict}
+- Component coverage — {verdict}
+- State coverage — {verdict}
+- Visual reference coverage — {verdict}
+- Bloat & overspecification — {verdict}
+- Inheritance discipline — {verdict}
+- Shape fit — {verdict}
+
+## Findings by severity
+
+### Critical (n)
+**[Category or Reviewer]** — Title (§ location)
+{Note}
+Fix: {suggested fix}
+
+### High (n) / Medium (n) / Low (n)
+...
+
+## Reviewer files
+- `review-rubric.md`
+- ...
+```
+
+## Close
+
+Surface artifact paths. Always offer to roll findings into an Update.
diff --git a/.agents/skills/bmad-validate-prd/SKILL.md b/.agents/skills/bmad-validate-prd/SKILL.md
new file mode 100644
index 00000000..3000d028
--- /dev/null
+++ b/.agents/skills/bmad-validate-prd/SKILL.md
@@ -0,0 +1,30 @@
+---
+name: bmad-validate-prd
+description: 'Deprecated — forwards to bmad-prd (validate intent).'
+---
+
+# DEPRECATED — forwards to bmad-prd (validate intent)
+
+This skill was consolidated into `bmad-prd`. It is retained as a thin compatibility shim so existing invocations by name and `_bmad/custom/bmad-validate-prd.toml` override files keep working. New work should invoke `bmad-prd` directly — it detects create / update / validate intent from the conversation.
+
+## On Activation
+
+1. Resolve customization: `uv run {project-root}/_bmad/scripts/resolve_customization.py --skill {skill-root} --key workflow`. This picks up any `{project-root}/_bmad/custom/bmad-validate-prd.toml` and `bmad-validate-prd.user.toml` overrides for the legacy fields (`activation_steps_prepend`, `activation_steps_append`, `persistent_facts`, `on_complete`).
+
+2. Load `{project-root}/_bmad/bmm/config.yaml` (and `config.user.yaml` if present) to resolve `{user_name}` and `{communication_language}`.
+
+3. Emit a deprecation notice to the user in `{communication_language}`:
+
+   > Notice: `bmad-validate-prd` is deprecated and will be removed in a future release. It now forwards to `bmad-prd` with validate intent. To silence this notice and access the full new customization surface (`prd_template`, `validation_checklist_template`, `validation_report_template`, `doc_standards`, `finalize_reviewers`, `external_sources`, `external_handoffs`, `prd_output_path`, `run_folder_pattern`), migrate `_bmad/custom/bmad-validate-prd.toml` to `_bmad/custom/bmad-prd.toml` and invoke `bmad-prd` directly next time. Customization fields that were in this version still remain in the new version and will be respected if present in `_bmad/custom/bmad-prd.toml`, but the new version also supports additional fields that you can take advantage of by migrating.
+
+4. Invoke `bmad-prd` with the following context. Pass these as the activating context so `bmad-prd` honors them instead of resolving its own customization from scratch:
+
+   - **Intent:** `validate` — skip `bmad-prd`'s usual intent detection step.
+   - **Pre-resolved legacy customization** — use these in place of resolving from `bmad-prd`'s own `customize.toml` for the four legacy fields. For everything else (`prd_template`, `validation_checklist_template`, `validation_report_template`, `doc_standards`, `finalize_reviewers`, `prd_output_path`, `run_folder_pattern`, `external_sources`, `external_handoffs`), use `bmad-prd`'s own defaults and overrides as normal:
+     - `activation_steps_prepend` = the resolved value from step 1
+     - `activation_steps_append` = the resolved value from step 1
+     - `persistent_facts` = the resolved value from step 1
+     - `on_complete` = the resolved value from step 1
+   - **Original user input:** forward whatever the user said when invoking this skill verbatim (the target PRD path, etc.).
+
+   `bmad-prd` takes the workflow from here. Do not execute any further steps in this shim.
diff --git a/.agents/skills/bmad-validate-prd/customize.toml b/.agents/skills/bmad-validate-prd/customize.toml
new file mode 100644
index 00000000..15ec851a
--- /dev/null
+++ b/.agents/skills/bmad-validate-prd/customize.toml
@@ -0,0 +1,42 @@
+# DO NOT EDIT -- overwritten on every update.
+#
+# Workflow customization surface for bmad-validate-prd. Mirrors the
+# agent customization shape under the [workflow] namespace.
+
+[workflow]
+
+# --- Configurable below. Overrides merge per BMad structural rules: ---
+#   scalars: override wins • arrays (persistent_facts, activation_steps_*): append
+#   arrays-of-tables with `code`/`id`: replace matching items, append new ones.
+
+# Steps to run before the standard activation (config load, greet).
+# Overrides append. Use for pre-flight loads, compliance checks, etc.
+
+activation_steps_prepend = []
+
+# Steps to run after greet but before the workflow begins.
+# Overrides append. Use for context-heavy setup that should happen
+# once the user has been acknowledged.
+
+activation_steps_append = []
+
+# Persistent facts the workflow keeps in mind for the whole run
+# (standards, compliance constraints, stylistic guardrails).
+# Distinct from the runtime memory sidecar — these are static context
+# loaded on activation. Overrides append.
+#
+# Each entry is either:
+#   - a literal sentence, e.g. "All PRDs must include a regulatory-risk section."
+#   - a file reference prefixed with `file:`, e.g. "file:{project-root}/docs/standards.md"
+#     (glob patterns are supported; the file's contents are loaded and treated as facts).
+
+persistent_facts = [
+  "file:{project-root}/**/project-context.md",
+]
+
+# Scalar: executed when the workflow reaches Step 13 (Validation Report Complete) and
+# the user exits via [X] Exit — not on [E] Use Edit Workflow (which chains to
+# bmad-edit-prd), [R] Review (which loops within), or [F] Fix (which loops within).
+# Override wins. Leave empty for no custom post-completion behavior.
+
+on_complete = ""
diff --git a/.env.docker.example b/.env.docker.example
index 2162f948..9584bbcd 100644
--- a/.env.docker.example
+++ b/.env.docker.example
@@ -7,8 +7,10 @@ FRONTEND_PORT=3010
 API_PORT=8031
 PUBLIC_URL=http://localhost:8031
 
-# Required. The installers generate all four values automatically.
+# API transport token is required for non-localhost binds and Fleet/Agent/API
+# integrations. The installer generates it automatically.
 API_AUTH_TOKEN=
+# Optional legacy emergency/OIDC bootstrap token; local login does not use it.
 BOOTSTRAP_ADMIN_TOKEN=
 SECRET_KEY=
 CREDENTIAL_ENCRYPTION_KEY=
diff --git a/.gitignore b/.gitignore
index 69454af3..48861385 100644
--- a/.gitignore
+++ b/.gitignore
@@ -88,6 +88,5 @@ odp-rs/target/
 .repowise/
 .understand-anything/
 .codex/
-.sentrux/agent-sessions/
-.sentrux/baseline.json
+.sentrux/
 artifacts/
diff --git a/DESIGN.md b/DESIGN.md
index 1d85ab08..5daaff7d 100644
--- a/DESIGN.md
+++ b/DESIGN.md
@@ -1,5 +1,77 @@
+---
+name: OpenCLI Workflow Editor Design System
+schema: design-pipeline.design-foundation.v0.1
+activeChange: workflow-editor-design-pipeline-v090
+status: active
+---
+
 # Design
 
+## Product Context
+
+OpenCLI is a dense, local-first operations console for people and Agents that build, validate, publish, run, and audit demand-driven workflows. The primary design target is the Studio workflow editor: a React Flow canvas backed by a canonical WorkflowProject, Yjs collaboration, immutable versions, runtime Trace, Evidence, Lineage, and grounded gates. The product must remain useful during partial capability availability, slow networks, and collaboration outages.
+
+## Overview
+
+The selected direction is **Calm Live Operations**: refine the existing Dark Ops Console rather than redesign it. Local authoring feels immediate; collaboration, durable persistence, validation, publication, execution, and evidence are distinct visible states. Business intent leads, implementation detail appears on demand, and AI output never outranks the original demand or source evidence.
+
+## Colors
+
+- Page and canvas surfaces use the existing zinc-based dark hierarchy from `docs/DESIGN_SYSTEM.md` and `frontend/app/globals.css`.
+- Primary blue is reserved for selection, focus, and the current actionable path.
+- Success, warning, danger, info, and Agent colors are semantic signal tokens and always appear with text or an icon.
+- LIVE, SYNCING, OFFLINE, BLOCKED, grounded, and ungrounded never rely on color alone.
+- New work must reconcile existing token drift instead of introducing additional literal color families.
+
+## Typography
+
+- Chinese interface text uses the existing Noto Sans SC/system UI stack.
+- Identifiers, ports, hashes, revisions, timings, room names, and Trace data use IBM Plex Mono or `font-mono`.
+- Critical controls and status text are at least 11px; 10px is limited to non-critical telemetry.
+- Mixed Chinese and Latin labels use explicit line-height and fallback behavior; truncation must preserve an accessible full label.
+
+## Layout
+
+- Desktop remains canvas-first with a single right-side Workflow Dock for outline, configuration, last run, and Trace.
+- Project identity, workflow switching, lifecycle, and collaboration state remain visible without covering the graph.
+- Spacing follows a 4px rhythm. Controls use 2px radius, nodes and panels 6px, and overlays 8px.
+- Narrow screens convert the Dock into a mutually exclusive sheet without resetting the canvas viewport.
+- Large graphs progressively disclose internals; the parent canvas shows business-level nodes and stable contracts.
+
+## Components
+
+- Reuse project-owned `frontend/components/ui/*`, shell components, React Flow canvas primitives, WorkflowNode, CommandPalette, Inspector, Workflow Dock, RunTracePanel, and lifecycle controls.
+- Yjs/y-websocket is the collaboration execution layer; Studio Workflow Draft remains the durable authority.
+- Node cards expose business name, capability truth, one key summary, and typed ports. Parameters belong in the Dock.
+- Demand labels, generated search terms, grounded entities, missing evidence, and recovery actions use the same canonical graph and project context; no parallel UI-only store is allowed.
+- Every asynchronous action completes its empty, loading, success, blocked, failure, retry, offline, and reduced-motion states.
+
+## Do's and Don'ts
+
+### Do
+
+- Show local node mutation immediately and durable acknowledgement separately.
+- Preserve keyboard selection, visible focus, palette search, undo, and recovery.
+- Derive labels and proposals from the original demand and display their provenance.
+- Keep REAL, BLOCKED, LIVE, OFFLINE, grounded, and ungrounded tied to backend facts.
+- Use bounded motion for spatial entry, connection creation, selection, and acknowledgement.
+
+### Don't
+
+- Do not rebuild React Flow, Yjs, y-websocket, or existing UI primitives.
+- Do not upload the full graph for every connected edit.
+- Do not use the public Yjs demo server for project data.
+- Do not add decorative gradients, glass-heavy cards, oversized marketing layouts, anonymous AI sparkle styling, or continuous idle animation.
+- Do not present model-discovered entities as grounded without a readable exact-product URL.
+
+## Source Decisions
+
+- **Adopted:** the existing Dark Ops Console, project navigation, compact node language, typed ports, Workflow Dock, React Flow, Zustand, Yjs/y-websocket, first-party Draft/Version lifecycle, and Evidence/Lineage boundaries.
+- **Adopted:** the Design Pipeline v0.9.0 discipline of requirements-first foundations, explicit motion posture, component-first reuse, browser evidence, and hash-bound change artifacts.
+- **Rejected:** public-demo collaboration, parallel workflow state stores, per-edit full graph PUT, decorative AI visual defaults, unverified component adoption, and UI-only capability claims.
+- **Rejected:** copying external product branding or component source merely because it appears in a reference catalog.
+- Active synthesis change: `openspec/changes/workflow-editor-design-pipeline-v090/`.
+
 ## Source of truth
 
 - Status: Active
diff --git a/MOTION.md b/MOTION.md
new file mode 100644
index 00000000..1c9cbb7a
--- /dev/null
+++ b/MOTION.md
@@ -0,0 +1,68 @@
+---
+schema: design-pipeline.motion-foundation.v0.1
+name: OpenCLI Workflow Editor Motion Language
+posture: minimal
+primitiveRegistry: design-pipeline.motion-primitives.v1
+activeChange: workflow-editor-design-pipeline-v090
+---
+
+# Motion
+
+## Motion Thesis
+
+Motion communicates spatial causality, direct manipulation, asynchronous acknowledgement, and recovery. It never decorates idle operational state. Node authoring appears locally before network persistence; motion must not make a durable save look complete before its revision is acknowledged.
+
+## Motion Principles
+
+- Direct manipulation leads: press feedback begins immediately and node placement follows the pointer or picker decision without waiting for persistence.
+- State changes stay bounded: use 70ms press, 120ms response, 160ms control, 200ms panel, and 300ms spatial transitions.
+- One owner controls each property and clock. React Flow owns canvas transforms; CSS/WAAPI owns DOM state transitions; Yjs owns collaboration state, not animation.
+- LIVE, SYNCING, OFFLINE, BLOCKED, grounded, and ungrounded remain understandable with motion disabled.
+- Interruption is safe: opening another panel, changing selection, undoing, navigating scope, or losing connection cancels or retargets motion without replaying the old state.
+
+## Motion Vocabulary
+
+- `press`: 70ms scale or surface response for direct controls; never applied to text-only status.
+- `response`: 120ms opacity/surface response after a local edit.
+- `control`: 160ms tabs, badges, focus-linked indicators, and compact disclosure.
+- `panel`: 200ms Workflow Dock, Sheet, palette, and Inspector entry/exit.
+- `spatial`: 300ms bounded canvas fit, scope navigation, and node focus.
+- `durable-ack`: a non-looping status transition when a revision is persisted.
+- primitive: reveal.trim-line — reserved for a newly created or actively traced edge; a static complete edge is the reduced-motion substitute.
+
+## Procedural Motion
+
+No procedural generator is selected for the workflow editor. Canvas pan/zoom and React Flow edge geometry are interaction state, not decorative procedural motion. Persistent particles, noise fields, orbiting status, shader effects, and autonomous loops are prohibited on the authoring surface.
+
+## Runtime Policy
+
+- CSS transitions are supported for press, response, status, and compact disclosure.
+- WAAPI is supported for interruptible panel and bounded spatial choreography when CSS cannot express cancellation cleanly.
+- React Flow is supported for viewport and edge geometry; it is the only owner of canvas transform state.
+- Motion One remains a supported project dependency only where existing components already use it and cleanup/reduced-motion behavior is explicit.
+- Anime.js, GSAP, Canvas, PixiJS, WebGL, and WebGPU are unsupported for ordinary workflow editor transitions. Adoption requires a separate design change and runtime owner.
+- Collaboration updates must not replay entry motion for the whole graph. Only the changed node, edge, or acknowledgement may transition.
+- Performance budget: no continuous main-thread animation; direct manipulation must remain responsive at the existing graph-size targets.
+
+## Reduced Motion
+
+When `prefers-reduced-motion: reduce` is active:
+
+The reduced-motion fallback substitutes static state, immediate positioning, and textual acknowledgement for every moving treatment.
+
+
+- press feedback uses color, border, or immediate state with no scale;
+- panels appear without translation and with zero or near-zero opacity duration;
+- canvas focus changes viewport immediately or uses the shortest non-animated positioning path;
+- reveal.trim-line becomes a fully visible static edge;
+- durable acknowledgement changes icon and text without movement;
+- collaboration cursors may update position without interpolated trails;
+- no information, ordering, status, or recovery action is lost.
+
+## Source Decisions
+
+- Adopted: the existing OpenCLI timing tiers, React Flow spatial ownership, project CSS tokens, Yjs collaboration states, and Design Pipeline v0.9.0 motion-foundation contract.
+- Adopted: reveal.trim-line only for causal edge creation or active trace because it clarifies graph direction.
+- Rejected: continuous glow, pulsing idle badges, orbiting indicators, particle backgrounds, springy card motion, and animation used as the sole status signal.
+- Rejected: copying implementations from external motion catalogs; the selected primitive is semantic metadata and `codeCopied: false`.
+- Provenance: `openspec/changes/workflow-editor-design-pipeline-v090/decisions/grill.md` and the existing repository design/runtime evidence.
diff --git a/README.md b/README.md
index b93b5cfc..3c0589ed 100644
--- a/README.md
+++ b/README.md
@@ -55,12 +55,13 @@ Invoke-WebRequest https://raw.githubusercontent.com/2233admin/opencli-Razormind/
 | API 文档 | http://localhost:8031/docs | REST API 与集成调试 |
 | 内置浏览器 | http://localhost:6080 | 扫码或登录需要账号的平台 |
 
-安装完成后，终端会打印：
+安装完成后可以直接使用本地管理员账号登录：
 
-- `BOOTSTRAP_ADMIN_TOKEN`：首次进入管理界面使用；
-- `API_AUTH_TOKEN`：Fleet、Agent、API 和 MCP 访问使用。
+- 用户名：`admin`
+- 密码：`admin`
+- 登录后可在「账户设置」修改密码
 
-两者同时保存在安装目录的 `.env`。不要公开 noVNC、令牌或浏览器调试端口；远程部署建议使用 HTTPS、反向代理或 SSH 隧道。
+`API_AUTH_TOKEN` 仅由 Fleet、Agent、API 和 MCP 传输使用，自动保存在安装目录的 `.env`，不需要填入管理界面。不要公开 API 令牌、noVNC 或浏览器调试端口；远程部署建议使用 HTTPS、反向代理或 SSH 隧道。
 
 ## 正常的研究流程
 
diff --git a/_bmad-output/implementation-artifacts/spec-gpu-accelerated-current-pages.md b/_bmad-output/implementation-artifacts/spec-gpu-accelerated-current-pages.md
new file mode 100644
index 00000000..caf1f499
--- /dev/null
+++ b/_bmad-output/implementation-artifacts/spec-gpu-accelerated-current-pages.md
@@ -0,0 +1,117 @@
+---
+title: '在现有页面内支持可降级的 GPU 加速渲染'
+type: 'feature'
+created: '2026-08-28'
+status: 'done'
+review_loop_iteration: 0
+baseline_commit: '0f4b56edb3fd739bab1f0b3cbb9a7c185dab50ca'
+context:
+  - '{project-root}/_bmad-output/implementation-artifacts/tests/frontend-regression-plan.md'
+---
+
+<frozen-after-approval reason="human-owned intent — do not modify unless human renegotiates">
+
+## Intent
+
+**Problem:** 当前登录背景、项目 Galaxy 和记录关系图已经分别使用 OGL、Three.js、Sigma 等 GPU/WebGL 渲染器，但没有统一的能力探测、可观察后端标记和安全降级边界；渲染器不可用时可能留下空白画布或影响页面可用性。
+
+**Approach:** 在现有 Next.js 页面内增加共享 `GpuSurface` 客户端边界，统一探测 WebGPU/WebGL2、暴露实际渲染后端并提供 DOM/Canvas 降级；先接入登录背景和现有图谱页面，不新增页面、不替换 Base UI，也不引入 GPUIX。
+
+## Boundaries & Constraints
+
+**Always:** GPU 只负责呈现，业务数据、选择状态和权限继续由 React/API 持有；探测必须 SSR 安全；无 GPU、创建上下文失败、减少动态效果或上下文丢失时页面核心操作仍可用；延续 `next/dynamic({ ssr: false })`、可见性暂停和显式资源释放模式；现有用户改动保持不动。
+
+**Ask First:** 引入新的 GPU 依赖、在浏览器启用实验性 WebGPU renderer、把 Canvas2D 关系图整体迁移到另一渲染库，或改变现有路由/产品导航。
+
+**Never:** 新建独立 GPU 页面或桌面应用；把 GPUIX 注入 Next.js；让 GPU 成为登录、工作流、证据或交付的前置条件；以能力探测冒充实际 WebGPU 渲染；删除 DOM fallback、可访问性或 reduced-motion 行为。
+
+## I/O & Edge-Case Matrix
+
+| Scenario | Input / State | Expected Output / Behavior | Error Handling |
+|----------|--------------|---------------------------|----------------|
+| GPU 可用 | 浏览器可创建 WebGL2，上层 GPU renderer 正常挂载 | 原页面内渲染 GPU island，并以稳定属性暴露实际后端 | renderer 异常由边界切换到 fallback，不影响页面其余区域 |
+| GPU 不可用 | `getContext('webgl2')` 返回空或抛错 | 登录页显示静态背景；图谱页显示可操作的非 GPU 降级视图 | 不重试死循环，不显示空白画布，不阻止表单和导航 |
+| SSR/首次加载 | `window`、`navigator` 尚不可用 | 输出确定的 pending/loading shell，客户端探测后再挂载 renderer | 不在模块加载期访问浏览器全局 |
+| 生命周期变化 | 页面隐藏、组件卸载、reduced-motion 或 context lost | 停止无意义帧循环并释放资源；恢复时从 React 状态重建 | 保留业务状态，明确降级原因，避免泄漏 renderer/context |
+
+</frozen-after-approval>
+
+## Code Map
+
+- `frontend/components/canvasui/Ripple.tsx` -- 复用其 SSR 安全探测、回退、暂停与销毁模式，不修改波纹业务。
+- `frontend/lib/rendering/gpu-capabilities.ts` -- 新增能力快照，区分可探测与实际采用。
+- `frontend/components/gpu/gpu-surface.tsx` -- 新增客户端边界、`data-*` 状态、fallback 和错误隔离。
+- `frontend/app/login/page.tsx:63-128` -- 接入边界；无 GPU 时保留静态品牌背景和登录表单。
+- `frontend/components/records/project-graph-explorer.tsx:41-57,233-270` -- 保持动态加载；Galaxy 降级时保留搜索、选择与 Inspector。
+- `frontend/app/(app)/records/graph/page.tsx:36-45,260-265` -- 为 Sigma WebGL 图增加同一边界。
+- `frontend/e2e/login.spec.mjs` -- 覆盖 GPU、无 WebGL2、SSR、reduced-motion 与 context-loss 恢复。
+- `frontend/e2e/gpu-graph.spec.mjs` -- 覆盖记录图和 Galaxy 无 WebGL2 时的节点选择与 Inspector。
+
+## Tasks & Acceptance
+
+**Execution:**
+- [x] `frontend/lib/rendering/gpu-capabilities.ts`, `frontend/components/gpu/gpu-surface.tsx` -- 实现 SSR 安全探测、实际后端状态、错误边界和降级合同。
+- [x] `frontend/app/login/page.tsx` -- 接入 GPU surface；失败时渲染静态背景，不改变登录行为和主题选择。
+- [x] `frontend/components/records/project-graph-explorer.tsx`, `frontend/app/(app)/records/graph/page.tsx` -- 在原路由内接入边界，保留非 GPU 可操作路径。
+- [x] `frontend/e2e/login.spec.mjs`, `frontend/e2e/gpu-graph.spec.mjs` -- 覆盖 GPU 生命周期和图谱 fallback 的可执行交互。
+
+**Acceptance Criteria:**
+- Given 支持 WebGL2 的浏览器，when 打开现有登录或图谱页面，then GPU island 在原页面内挂载且页面暴露实际 renderer 状态。
+- Given 浏览器不能创建 WebGL2，when 打开同一页面，then 显示非空降级内容且登录、搜索、导航和 Inspector 等 DOM 操作保持可用。
+- Given SSR、页面隐藏或组件卸载，when 生命周期变化，then 不发生浏览器全局访问错误、持续后台帧循环或遗留 GPU context。
+- Given WebGPU 仅被探测但没有实际 renderer，when 查看状态，then 产品不得标记为 WebGPU 正在渲染。
+
+## Spec Change Log
+
+## Design Notes
+
+`GpuSurface` 只控制 renderer 是否挂载；业务状态仍在 React/API。`data-gpu-backend` 记录实际采用的 `webgl2`、fallback 或 pending，不把设备理论能力宣称为已启用。
+
+## Verification
+
+**Commands:**
+- `npm --prefix frontend run lint` -- 新增边界和页面集成无 ESLint 错误。
+- `npm --prefix frontend run build` -- Next.js SSR/客户端边界与动态 imports 可生产构建。
+- `npm --prefix frontend run check:login-themes` -- 三种现有主题与 reduced-motion 合同保持通过。
+- `npm --prefix frontend run check:record-relationships` -- 现有图谱结构和交互合同保持通过。
+- `npm --prefix frontend run test:smoke -- login.spec.mjs gpu-graph.spec.mjs` -- Chromium 中 9 条 GPU、SSR、恢复、正常图谱挂载和图谱 fallback 场景通过。
+
+## Suggested Review Order
+
+**渲染边界**
+
+- 统一实际 backend、降级、可见性和 context-loss 恢复。
+  [`gpu-surface.tsx:66`](../../frontend/components/gpu/gpu-surface.tsx#L66)
+
+- 缓存 SSR 安全探测，WebGPU 仅报告 API presence。
+  [`gpu-capabilities.ts:19`](../../frontend/lib/rendering/gpu-capabilities.ts#L19)
+
+**现有页面接入**
+
+- 登录页在原页面切换 GPU 背景与静态 fallback。
+  [`page.tsx:105`](../../frontend/app/login/page.tsx#L105)
+
+- Galaxy 使用 WebGL2 边界，关系图诚实标记 Canvas2D。
+  [`project-graph-explorer.tsx:251`](../../frontend/components/records/project-graph-explorer.tsx#L251)
+
+- Sigma 记录图降级后仍保留节点选择与详情。
+  [`graph/page.tsx:317`](../../frontend/app/(app)/records/graph/page.tsx#L317)
+
+**可用降级**
+
+- SSR 未水合时输出确定的 pending 静态背景。
+  [`login/page.tsx:86`](../../frontend/app/login/page.tsx#L86)
+
+- 无 GPU 的 Galaxy 以有界节点列表保留 Inspector。
+  [`project-graph-explorer.tsx:294`](../../frontend/components/records/project-graph-explorer.tsx#L294)
+
+- 无 WebGL2 的记录图保留双向邻居查看入口。
+  [`graph/page.tsx:62`](../../frontend/app/(app)/records/graph/page.tsx#L62)
+
+**行为验证**
+
+- 覆盖 GPU、无 WebGL2、SSR、reduced-motion 与 context-loss。
+  [`login.spec.mjs:65`](../../frontend/e2e/login.spec.mjs#L65)
+
+- 覆盖记录图和 Galaxy fallback 的真实点击链路。
+  [`gpu-graph.spec.mjs:167`](../../frontend/e2e/gpu-graph.spec.mjs#L167)
diff --git a/_bmad-output/implementation-artifacts/spec-paw-local-enrichment-runtime.md b/_bmad-output/implementation-artifacts/spec-paw-local-enrichment-runtime.md
new file mode 100644
index 00000000..69287518
--- /dev/null
+++ b/_bmad-output/implementation-artifacts/spec-paw-local-enrichment-runtime.md
@@ -0,0 +1,123 @@
+---
+title: '为采集流水线增加受治理的 PAW 本地富化运行时'
+type: 'feature'
+created: '2026-08-28'
+status: 'done'
+review_loop_iteration: 0
+baseline_commit: 'c5804809f2688b110eaac4d9aabe1e3312b5d1d1'
+context:
+  - '{project-root}/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/addendum.md'
+---
+
+<frozen-after-approval reason="human-owned intent — do not modify unless human renegotiates">
+
+## Intent
+
+**Problem:** 高频短文本分类/提取缺少低成本离线路径；直接使用公共 Hub 或在主 API 加载 llama.cpp 又有供应链、隐私和隔离风险。
+
+**Approach:** 增加 `processor_type="paw"`，让现有流水线调用只执行固定程序的本地 sidecar；短文本通过身份、契约和 JSON Schema 验证后才持久化。
+
+## Boundaries & Constraints
+
+**Always:** 复用处理器注册表和 `process_with_ai`；sidecar 固定 `program_id`、默认离线、串行、有界 JSON 输出；健康信息暴露身份和就绪状态；容器非 root、只读、内部网络；错误记录不标记 `ai_processed`。
+
+**Ask First:** 联网准备模型、切换程序、提高并发、传入原始浏览器数据/凭据/完整 HTML、设为默认处理器。
+
+**Never:** 提供编译、Hub 自动安装或任意程序入口；向 PAW 托管服务发送生产数据；在 API/Celery 进程加载 PAW；失败时伪造成功或切换模型。
+
+## I/O & Edge-Case Matrix
+
+| Scenario | Input / State | Expected Output / Behavior | Error Handling |
+|----------|--------------|---------------------------|----------------|
+| 正常 | 固定程序、短 prompt | 匹配身份的 JSON object 写入 `ai_enrichment` | N/A |
+| 身份/契约错配 | 响应 ID/version 不符 | 拒绝且记录保持原状态 | 稳定错误码，不计成功数 |
+| 非法输出 | 非 object、过大或 schema 不符 | 不写入富化 | 返回验证错误 |
+| 不可用 | 缓存缺失、超时、连接/加载失败 | 健康 not-ready；同批其他记录继续 | 失败记录保持未富化 |
+| 超长 | prompt 超限 | 推理前拒绝 | 413/422，不隐式截断 |
+
+</frozen-after-approval>
+
+## Code Map
+
+- `backend/processors/external_http_processor.py`, `backend/processors/registry.py` -- guarded HTTP、逐记录容错和注册模式。
+- `backend/processors/paw_processor.py`, `backend/pipeline/ai_processor.py:135-145` -- PAW 验证及真实成功状态。
+- `compat/kats_runtime/app.py`, `compat/paw_runtime/` -- 复用 sidecar 边界并新增 PAW 0.4.4 离线服务、镜像和合同测试。
+- `docker-compose.yml` -- opt-in profile、只读缓存、内部网络和健康检查。
+- `frontend/components/agents/agent-form-dialog.tsx`, `frontend/app/(app)/agents/page.tsx`, `frontend/lib/api/types.ts` -- PAW 类型和安全配置提示。
+
+## Tasks & Acceptance
+
+**Execution:**
+- [x] `compat/paw_runtime/` -- 实现 `/health`、`/v1/enrich`、离线固定程序、串行推理和有界 JSON 输出。
+- [x] `backend/processors/paw_processor.py`, `backend/processors/registry.py`, `backend/config.py` -- 实现 HTTP adapter、身份/schema 验证和默认 URL。
+- [x] `backend/pipeline/ai_processor.py` -- 错误结果保持未富化并返回准确成功数。
+- [x] `docker-compose.yml` -- 增加隔离、默认关闭的 PAW sidecar 和只读缓存挂载。
+- [x] Agent 前端文件 -- 增加 PAW 选择、标签和配置示例，不增加编译/Hub UI。
+- [x] 聚焦测试 -- 覆盖矩阵全部场景及配置往返。
+
+**Acceptance Criteria:**
+- Given 已固定的本地 PAW 程序，when 标准化记录进入 `paw` Agent，then 验证后的 object 持久化为 `ai_enrichment`。
+- Given 部分响应失败，when 批处理结束，then 失败记录保持未富化，其他记录成功且计数准确。
+- Given 未启用 `paw` profile，when 运行现有服务，then 启动和行为不变。
+- Given 操作者配置 PAW，when 查看 UI/健康信息，then 可识别程序和离线就绪状态，且无编译/Hub 安装入口。
+
+## Spec Change Log
+
+- 2026-08-28：sidecar 从通用子进程入口改为固定的 `programasweights==0.4.4` SDK；仅以 `paw.function(program_id, offline=True)` 加载环境固定程序，并以 `paw.is_offline_ready` 报告只读缓存就绪状态。移除了可执行文件路径约定，禁止编译、联网回退和任意入口。
+- 2026-08-28：收紧 sidecar 仅本地 URL、敏感占位符、流式响应和 PAW Agent 配置边界；SDK callable 缓存复用且超时后 fail-closed。无经批准只读 `.paw` 缓存时，真实运行时保持 `503 not_ready`，不尝试准备、下载、编译或远程推理。
+
+## Design Notes
+
+契约固定为 `opencli.paw.runtime.v1`：请求含配置的 `programId`、渲染后 `input`、有界 `maxTokens`；响应含同一 ID、契约版本和 `enrichment` object。sidecar 只加载环境固定程序，ID 不匹配直接拒绝。默认 `PAW_OFFLINE=1`，缓存由操作者只读挂载；联网准备属于独立运维动作。
+
+## Verification
+
+**Commands:**
+- `pytest -q compat/paw_runtime/tests` -- sidecar 合同和边界通过。
+- `pytest -q tests/unit/test_paw_processor.py tests/unit/pipeline/test_ai_processor.py tests/integration/test_agents_api.py` -- adapter、状态和配置合同通过。
+- `npm --prefix frontend run lint && npm --prefix frontend run build` -- Agent UI 通过检查。
+- `docker compose --profile paw config` -- 隔离服务可解析且默认 profile 不启用。
+
+**Executed:**
+- `uv run ruff check ...` — PAW 新增/修改的 Python 文件全部通过。
+- `uv run pytest --no-cov -q compat/paw_runtime/tests tests/unit/test_paw_processor.py tests/unit/pipeline/test_ai_processor.py tests/integration/test_agents_api.py` — 49 passed。
+- `npm --prefix frontend run lint` — 0 errors；2 条既有无关 warning。
+- `npm --prefix frontend run build` — Next.js 生产构建通过。
+- fresh 3100 端口运行 `paw-agent.spec.mjs` — 1 passed，覆盖 PAW 选择、说明、空提示词阻断和合法配置提交。
+- `docker compose --profile paw build paw-runtime` — 镜像构建通过；PAW SDK 与 llama wheel 均校验 SHA-256。
+- `API_AUTH_TOKEN=test-token docker compose --profile paw config` — profile、内部网络、只读缓存和健康检查可解析；默认 services 不含 sidecar。
+- 真实容器在未配置 `PAW_PROGRAM_ID`/缓存时 `/health` 返回 `503 not_ready`，不使用伪造程序 ID、不联网回退。
+
+## Suggested Review Order
+
+**受治理的后端入口**
+
+- 仅传递渲染后的标准化短文本，并逐条隔离失败。
+  [`paw_processor.py:135`](../../backend/processors/paw_processor.py#L135)
+
+- 仅持久化已验证成功的富化结果。
+  [`ai_processor.py:142`](../../backend/pipeline/ai_processor.py#L142)
+
+**固定离线运行时**
+
+- 首次离线加载后缓存官方 PAW callable，串行复用于后续记录。
+  [`engine.py:87`](../../compat/paw_runtime/engine.py#L87)
+
+- 健康返回真实就绪状态，推理不阻塞事件循环。
+  [`app.py:122`](../../compat/paw_runtime/app.py#L122)
+
+**部署隔离与操作界面**
+
+- 以 opt-in profile 隔离只读、非特权的内部 sidecar。
+  [`docker-compose.yml:146`](../../docker-compose.yml#L146)
+
+- PAW 选择说明固定离线边界并要求短提示词。
+  [`agent-form-dialog.tsx:201`](../../frontend/components/agents/agent-form-dialog.tsx#L201)
+
+**回归覆盖**
+
+- 验证 sidecar 身份、边界、并发与固定调用合同。
+  [`test_contract.py:61`](../../compat/paw_runtime/tests/test_contract.py#L61)
+
+- 验证 adapter 契约、拒绝路径及批次连续性。
+  [`test_paw_processor.py:78`](../../tests/unit/test_paw_processor.py#L78)
diff --git a/_bmad-output/implementation-artifacts/tests/frontend-regression-plan.md b/_bmad-output/implementation-artifacts/tests/frontend-regression-plan.md
new file mode 100644
index 00000000..803aadd0
--- /dev/null
+++ b/_bmad-output/implementation-artifacts/tests/frontend-regression-plan.md
@@ -0,0 +1,89 @@
+# Frontend Regression Test Plan
+
+## BMAD QA activation and scope
+
+- Workflow: `bmad-qa-generate-e2e-tests`, activated under OMP 5.6 Luna.
+- Discovery mode: Workflow Step 1 auto-discover features (no single component supplied).
+- Activation resolution: `resolve_customization.py --skill bmad-qa-generate-e2e-tests --key workflow` completed successfully. Prepend and append activation steps were empty. Persistent fact glob `**/project-context.md` was checked; no files were present.
+- Config: project `snailfish`; communication/output language English; implementation artifacts resolve to `_bmad-output/implementation-artifacts`.
+- This is preparation only: no product or test source is changed and no tests are run.
+- Explicit non-scope: do not invent tests for future geoXI/OODA contracts; targets below are observable in the current UI.
+
+## Framework and repository conventions
+
+The frontend is Next.js/React with Playwright available as `@playwright/test` (frontend/package.json). `frontend/playwright.config.mjs` uses Chromium, `http://127.0.0.1:3000`, and starts the production app with `pnpm start --hostname 127.0.0.1 --port 3000`. The configured E2E directory is `frontend/e2e`; currently it contains only `login.spec.mjs`.
+
+The frontend has no Jest/Vitest/Cypress dependency. Existing unit/contract-style regression coverage is Node's built-in `node:test`, invoked by scripts such as `check:workflow-regressions`, `check:node-capabilities`, `check:control-plane`, `check:opencli-business-workflows`, and related scripts. These tests mostly inspect/import TypeScript source and data contracts rather than driving a browser. CI (`.github/workflows/ci.yml`) runs lint, TypeScript, workflow regressions, build, then Playwright browser smoke (`pnpm test:smoke`) in the frontend job; the dedicated workflow-check job runs contract scripts. New browser tests should therefore use Playwright in `frontend/e2e`, while deterministic projection/validation logic can remain Node `node:test`.
+
+## Current user surfaces and observed coverage
+
+- `/plugins` exposes installed plugin, node capability, and marketplace tabs; provider state/readiness labels and detail sheets are implemented in `frontend/app/(app)/plugins/page.tsx`.
+- `/opencli` exposes a website adapter/capability directory, refresh action, and search by site/domain/capability (`frontend/app/(app)/opencli/page.tsx`).
+- `/studio/new` generates a draft from requirements and explicitly computes readiness, capability gaps, and publish/run blocking (`frontend/app/(app)/studio/new/page.tsx`).
+- `/studio/projects/[projectId]` renders project readiness cards and the draft → validation → publish steps (`frontend/app/(app)/studio/projects/[projectId]/page.tsx`); project data links to workflow and evidence (`.../data/page.tsx`).
+- `/skills` and `/skills/[id]` render skill inventory, evidence history, correction proposal state, dismiss correction, and rollback controls (`frontend/app/(app)/skills/page.tsx`, `frontend/app/(app)/skills/[id]/page.tsx`).
+- `/studio/projects/[projectId]/operations`, `/evidence`, and `/data` provide operations, evidence, records, and lineage views; `records/page.tsx` includes a LineagePanel and data page links to workflow/evidence.
+- `/operations-agents` is the current schedules/operations surface (the `/schedules` page redirects there). It supports workspace selection, automation/agent views, contract draft editing, conflict handling, publishing, and run/activity views (`frontend/app/(app)/operations-agents/page.tsx`).
+- Existing browser coverage: one login rendering test (`frontend/e2e/login.spec.mjs`).
+- Existing non-browser coverage: many Node regression scripts, including workflow structure, capability catalog, records/relationships/hygiene, control-plane/dashboard/inbox/navigation checks, and OpenCLI business workflow source contracts. These are valuable guardrails but do not verify user-visible navigation, loading/error states, or interactions.
+
+## Auto-discovered regression candidates (priority order)
+
+### P0-1 Capability catalog and readiness surface — E2E + component/unit
+- **Source anchors:** `frontend/app/(app)/plugins/page.tsx:159-200` (`PluginPageTabs`); provider state logic around `providerStateLabel/providerStateTone`; capability detail sheet around `ProviderCard` and capability metrics.
+- **Existing anchors:** `frontend/scripts/check-node-capability-catalog-regressions.mjs`, `check-tool-capability-catalog-regressions.mjs`.
+- **Plausible bug:** tab selection or readiness state mapping silently regresses, showing a runnable capability as unavailable or making the detail sheet inaccessible.
+- **Acceptance:** Navigate to `/plugins`, switch Installed → Node Capabilities → Marketplace using semantic labels; selected tab exposes `aria-current="page"`; a provider card opens its detail sheet and displays declared capabilities plus readiness/missing information. Mock or fixture API responses should cover ready, partial, configuration, and unavailable states.
+
+### P0-2 Studio capability-to-workflow entry and readiness blocking — E2E
+- **Source anchors:** `frontend/app/(app)/studio/page.tsx:262-265` (blank/template/import choices); `frontend/app/(app)/studio/new/page.tsx:200-220,328-350,384-387,456-625` (generation, gaps, readiness, save/navigation).
+- **Existing anchors:** `frontend/scripts/check-workflow-regressions.mjs`, `check-opencli-business-workflows.mjs`, `test:workflow-contracts`.
+- **Plausible bug:** a capability gap is displayed but Publish/Run remains enabled, or the Studio entry loses the project/workflow query parameters after save.
+- **Acceptance:** Start from Studio, choose an existing supported template/blank path, submit valid requirements, verify readiness items and generated draft. With a known missing capability, assert Draft can save but publish/run controls are disabled and the gap is visible; with complete requirements, assert navigation reaches the workflow editor with workspace/project/workflow parameters.
+
+### P0-3 Skill correction proposal, dismiss, and rollback — E2E + unit
+- **Source anchors:** `frontend/app/(app)/skills/[id]/page.tsx:64-96,159-197` (`openProposal`, `findRollbackTarget`, mutation handlers); evidence rendering `:99-138`.
+- **Existing anchors:** no browser test; backend correction tests exist under `tests/skills`/unit coverage.
+- **Plausible bug:** an old correction proposal is shown after a later corrected/dismissed boundary, or rollback is offered twice after a rollback event.
+- **Acceptance:** Fixture a skill with a current correction proposal; detail shows proposal and dismiss confirmation, dismiss calls the mutation and removes/updates the proposal. Fixture a corrected version; rollback confirmation is available once, success updates history and disables the stale action. Unit-test boundary helpers with corrected/dismissed and already-rolled-back sequences.
+
+### P1-4 Run, evidence, and lineage navigation — E2E
+- **Source anchors:** `frontend/app/(app)/studio/projects/[projectId]/operations/page.tsx`, `.../evidence/page.tsx`, `.../data/page.tsx:261-262`; `frontend/app/(app)/records/page.tsx:96-115,330-334`.
+- **Existing anchors:** `frontend/scripts/check-record-hygiene-regressions.mjs`, `check-record-relationship-regressions.mjs`; backend `tests/integration/test_workflow_opencli_hda_trace_api.py`.
+- **Plausible bug:** selecting a record loses workflow/run/source identifiers or links to evidence with malformed query parameters.
+- **Acceptance:** From a project run/record, select a record, verify visible lineage fields (workflow, run, source), open “定位业务编排” and “查看逻辑与证据”, and assert destination URL retains project/workflow/record context. Verify empty and loading states do not render broken links.
+
+### P1-5 Operations Agents contract editing and publish — E2E
+- **Source anchors:** `frontend/app/(app)/operations-agents/page.tsx:138-203,209-258,268-283`.
+- **Existing anchors:** backend API tests `tests/api/test_operations_agents.py`; no browser coverage.
+- **Plausible bug:** provider/model validation is bypassed, JSON contract fields are not persisted, or publish remains enabled without a reason.
+- **Acceptance:** Select a workspace and agent, edit role/required capabilities/evidence requirements, verify invalid JSON and provider/model mismatch produce visible errors without mutation; enter valid values, save draft, enter publish reason, publish, and verify success plus version history. A 409 refresh path should show the conflict message and latest revision.
+
+### P1-6 Schedules/automations entry and run/activity view — E2E
+- **Source anchors:** `frontend/app/(app)/schedules/page.tsx:1-5` redirect; `frontend/app/(app)/operations-agents/page.tsx:268-283` and automation/run sections later in same file.
+- **Existing anchors:** `frontend/scripts/check-dashboard-regressions.mjs`, `tests/api/test_automations.py`, `tests/api/test_automation_starters.py`.
+- **Plausible bug:** legacy `/schedules` navigation dead-ends, or starting a run reports success while activity does not refresh.
+- **Acceptance:** `/schedules` redirects to `/operations-agents`; automation view loads for a workspace, run action is disabled while pending, success appears in activity, and API error is rendered without duplicate submissions.
+
+### P1-7 OpenCLI website/capability directory search and refresh — E2E + component
+- **Source anchors:** `frontend/app/(app)/opencli/page.tsx:150-190,227-230`.
+- **Existing anchors:** `frontend/scripts/check-opencli-business-workflows.mjs`, catalog scripts; no browser coverage.
+- **Plausible bug:** search filters only labels but not domains/capabilities, or refresh/loading/error states leave stale results.
+- **Acceptance:** Directory renders loading then results, search by site/domain/capability narrows cards, refresh shows pending state and replaces results, and API failure shows “网站适配目录暂时不可用” with the error text.
+
+### P2-8 Project readiness and workflow/evidence deep links — E2E
+- **Source anchors:** `frontend/app/(app)/studio/projects/[projectId]/page.tsx:202-212,308-310,341-349`; `.../data/page.tsx:261-262`.
+- **Existing anchors:** `tests/integration/test_studio_lifecycle_api.py`, `test_workflow_capabilities_api.py`; no browser coverage.
+- **Plausible bug:** readiness cards report published status from stale data, or project navigation points at the wrong workflow/evidence route.
+- **Acceptance:** Fixture project with/without primary workflow and published version; assert card values and step completion reflect state. Deep links open the correct route and preserve workspace/project identifiers.
+
+## Recommended first batch
+
+Implement P0-1, P0-2, and P0-3 first. They exercise the highest-risk frontend regression seams—capability readiness, the primary Studio creation path, and stateful correction/rollback—and complement existing source-contract checks with actual user-visible behavior. Add shared Playwright API fixtures/mocking only if the current app test harness already supports it; otherwise use deterministic seeded backend data and semantic locators. Keep the tests linear, isolated, and focused on visible outcomes.
+
+## Suggested acceptance gate for the eventual implementation
+
+- New browser specs live under `frontend/e2e` and use `@playwright/test` semantic locators.
+- Existing login and Node regression scripts remain unchanged.
+- Each candidate has a happy path plus one meaningful error/boundary assertion.
+- Run targeted Playwright specs locally, then the repository’s existing `pnpm test:smoke`; this preparation phase intentionally does not run either command.
diff --git a/_bmad-output/implementation-artifacts/tests/test-summary.md b/_bmad-output/implementation-artifacts/tests/test-summary.md
new file mode 100644
index 00000000..cf428a1c
--- /dev/null
+++ b/_bmad-output/implementation-artifacts/tests/test-summary.md
@@ -0,0 +1,67 @@
+# Test Automation Summary
+
+## Generated Tests
+
+### API Tests
+- [x] Not applicable: this QA batch targets existing frontend user workflows; API behavior is exercised through deterministic Playwright route fixtures and existing repository contract tests.
+
+### E2E Tests
+- [x] `frontend/e2e/p0-regressions.spec.mjs` — P0-1 Plugin Center installed → node capabilities → marketplace tabs, `aria-current` selection, and generic provider capability detail/readiness surface.
+- [x] `frontend/e2e/p0-regressions.spec.mjs` — P0-3 Skill detail current correction proposal, dismiss confirmation, corrected-version rollback, and stale rollback prevention after `rolled_back`.
+- [x] P0-2 Studio readiness/gap behavior is covered by existing deterministic Node workflow regressions, not a newly passing browser test. The attempted Studio E2E was removed because React hydration/remount made the interaction flaky under the current production Next harness. This summary does not claim Studio E2E coverage.
+
+## Coverage
+
+| Discovered target | Priority | Coverage status | Evidence / notes |
+|---|---:|---|---|
+| Plugin capability catalog, readiness states, and tabs/detail | P0-1 | Covered — E2E | `frontend/e2e/p0-regressions.spec.mjs`; deterministic plugin/capability fixtures; semantic tab and detail assertions |
+| Studio capability-to-workflow entry and readiness blocking | P0-2 | Covered — deterministic Node contract; no E2E | Existing `check-workflow-regressions` Agent Builder readiness/gap tests prove `canSave=true`, `canPublish=false`, `canRun=false`, blocking actions, and complete-path readiness; browser attempt removed as flaky hydration/remount |
+| Skill correction proposal, dismiss, and rollback | P0-3 | Covered — E2E | `frontend/e2e/p0-regressions.spec.mjs`; deterministic skill evidence and POST mutation fixtures |
+| Run, evidence, and lineage navigation | P1-4 | Remaining candidate | Planned; no new test generated in this batch |
+| Operations Agents contract editing and publish | P1-5 | Remaining candidate | Planned; no new test generated in this batch |
+| Schedules/automations entry and run/activity | P1-6 | Remaining candidate | Planned; no new test generated in this batch |
+| OpenCLI website/capability directory search and refresh | P1-7 | Remaining candidate | Planned; no new test generated in this batch |
+| Project readiness and workflow/evidence deep links | P2-8 | Remaining candidate | Planned; no new test generated in this batch |
+
+## Verification Evidence
+
+- Targeted P0 Playwright with `--repeat-each=3`: **6/6 passed** (P0-1 and P0-3, each repeated three times).
+- Normal P0 Playwright run: **2/2 passed**.
+- Existing frontend smoke (`pnpm test:smoke`): **3/3 passed**.
+- Workflow regression contracts (`pnpm run check:workflow-regressions`, including inspector workflow checks): **60/60 passed**.
+- Node capability/catalog contracts (`pnpm run check:node-capabilities`, which includes the capability/tool catalog checks): **16/16 passed**.
+- Targeted E2E lint (`pnpm exec eslint e2e/p0-regressions.spec.mjs`): **passed**.
+
+## Infrastructure Notes
+
+- Existing warning: Playwright's configured `next start --hostname 127.0.0.1 --port 3000` reports that `next start` is incompatible with the app's `output: standalone` configuration.
+- `pnpm build` succeeded. A source-grounded attempt to run `.next/standalone/server.js` on Windows failed before serving with `EPERM` while stat'ing the bundled pnpm React path. This is a test-infrastructure follow-up, not a product regression.
+- The alias command `check:tool-capability-catalog-regressions` is absent from `frontend/package.json`; the repository's existing `check:node-capabilities` script includes the tool capability catalog test and passed (16/16).
+
+## Checklist Outcome
+
+Validated against `skill://bmad-qa-generate-e2e-tests/checklist.md`:
+
+- API tests: **N/A** for this frontend workflow batch.
+- E2E tests: **PASS** for the generated P0-1/P0-3 browser tests.
+- Standard framework APIs and semantic locators: **PASS** (Playwright and Node `node:test`; accessible roles/labels/text).
+- Happy paths and critical boundary/error behavior: **PASS** for generated P0 tests and existing P0-2 Node contract coverage.
+- All generated tests run successfully: **PASS** (6/6 repeated and 2/2 normal).
+- Test independence and clear descriptions: **PASS**; each Playwright test uses its own context and deterministic route fixtures.
+- No hardcoded waits/sleeps: **PASS** in the final generated tests; bounded Playwright expectations are used for asynchronous UI state.
+- Summary, test locations, and coverage metrics: **PASS**; this file and the prior discovery plan are saved under implementation artifacts.
+- Validation commands: **PASS** for the executed targeted P0, smoke, workflow, and capability contract commands listed above.
+
+Overall checklist result: **PASS with documented infrastructure follow-up and explicit Studio E2E gap**.
+
+## Scope and Changes
+
+- No production source changed.
+- Test-only change: `frontend/e2e/p0-regressions.spec.mjs`.
+- Documentation artifact: `_bmad-output/implementation-artifacts/tests/frontend-regression-plan.md` and this summary.
+
+## Next Steps
+
+- Keep the deterministic Node Studio readiness contract as the regression gate.
+- Revisit Studio browser interaction after the Next/React hydration/remount harness is made stable; do not treat the removed attempt as E2E coverage.
+- Prioritize the remaining P1/P2 candidates from the regression plan.
diff --git a/_bmad-output/planning-artifacts/architecture/architecture-snailfish-2026-08-27/.memlog.md b/_bmad-output/planning-artifacts/architecture/architecture-snailfish-2026-08-27/.memlog.md
new file mode 100644
index 00000000..373b7ae4
--- /dev/null
+++ b/_bmad-output/planning-artifacts/architecture/architecture-snailfish-2026-08-27/.memlog.md
@@ -0,0 +1,14 @@
+---
+scope: Gaojixing live business chain feature
+purpose: build-substrate
+altitude: feature
+updated: 2026-08-27T23:05
+---
+
+- (constraint) OpenSpec change gaojixing-live-business-chain remains the normative source; this BMAD output only organizes and references it.
+- (decision) AD-1: Govern the live slice as a compiled WorkflowProject source-to-sink flow using the existing runtime and adapters; prevents a parallel Gaojixing executor or backend rewrite.
+- (decision) AD-2: Require one immutable question package digest and one lineage tuple across capture, normalization, record, evidence, and delivery; prevents source mutation or cross-run evidence attachment.
+- (decision) AD-3: Model evidence, transport, and business outcome independently; only destination ACK/equivalent tied to the delivery identity may confirm business outcome.
+- (decision) AD-4: Keep fixture/mock paths explicit and non-live; missing live prerequisites fail closed without fallback.
+- (constraint) Current runtime observations (seven running runs, Doubao session, Chrome pool, and ODP NOGROUP) are operator-reported blockers until bound to a fresh run/evidence receipt; no live success is inferred.
+- (event) spine finalized; companion context, prerequisite/ACK matrix, and reordered existing-task plan produced
diff --git a/_bmad-output/planning-artifacts/architecture/architecture-snailfish-2026-08-27/ARCHITECTURE-SPINE.md b/_bmad-output/planning-artifacts/architecture/architecture-snailfish-2026-08-27/ARCHITECTURE-SPINE.md
new file mode 100644
index 00000000..0680ba37
--- /dev/null
+++ b/_bmad-output/planning-artifacts/architecture/architecture-snailfish-2026-08-27/ARCHITECTURE-SPINE.md
@@ -0,0 +1,123 @@
+---
+name: Gaojixing live business chain
+type: architecture-spine
+purpose: build-substrate
+altitude: feature
+paradigm: compiled workflow graph with evidence-bearing source-to-sink pipeline
+scope: Gaojixing live Doubao capture through normalized record, delivery, and acknowledged business outcome
+status: final
+created: 2026-08-27
+updated: 2026-08-27
+binds:
+  - gaojixing-live-business-chain
+sources:
+  - openspec/changes/gaojixing-live-business-chain/proposal.md
+  - openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md
+  - docs/workflow-hda-demand-runtime-PRD.md
+  - docs/workflow-runtime-conformance.md
+  - docs/adr/0009-plan-ir-free-graph-two-tier-attribution.md
+  - docs/adr/0021-delivery-separates-submission-from-business-outcome.md
+companions:
+  - ../../../bmad-kickoff/current-project-context.md
+  - ../../../bmad-kickoff/external-prerequisites-and-ack-matrix.md
+  - ../../../bmad-kickoff/live-business-chain-next-steps.md
+---
+
+# Architecture Spine — Gaojixing Live Business Chain
+
+## Design Paradigm
+
+A compiled `WorkflowProject` graph owns orchestration; an evidence-bearing source-to-sink pipeline owns the business chain. The Gaojixing source is a live-mode Doubao adapter behind the existing runtime boundary, then crosses `source → normalize → accept → sink`; it does not introduce a parallel executor or data path. This ratifies the runtime PRD’s Canvas/compiler/III/OpenCLI/ODP reuse decision and the backend-authoritative graph ADR. [Runtime PRD](../../../../docs/workflow-hda-demand-runtime-PRD.md#L72-L91) [ADR-0009](../../../../docs/adr/0009-plan-ir-free-graph-two-tier-attribution.md#L20-L29)
+
+```mermaid
+flowchart LR
+  WP[WorkflowProject] --> C[Backend compiler and runtime registry]
+  C --> S[Live Gaojixing source]
+  S --> N[Normalize and canonical dedupe]
+  N --> A[Accept: record and evidence projections]
+  A --> K[Sink and delivery attempt]
+  K --> O[Business outcome]
+  S -. resource gates .-> R[Capability, session, network]
+  O -. confirmed only with matching ACK .-> D[Destination]
+```
+
+## Invariants & Rules
+
+### AD-1 — One compiled workflow path [ADOPTED]
+
+- **Binds:** all Gaojixing execution, runtime bindings, and projections.
+- **Prevents:** a Gaojixing-specific executor, backend rewrite, or browser-side result-producing preview.
+- **Rule:** Author the run as a `WorkflowProject`; compile it through the existing runtime registry and route the live Doubao source through the existing OpenCLI/III/ODP-compatible chain. Use Canvas/Plan health for downstream stages while preserving source-keyed attribution at the source boundary. [Runtime PRD](../../../../docs/workflow-hda-demand-runtime-PRD.md#L74-L91) [ADR-0009](../../../../docs/adr/0009-plan-ir-free-graph-two-tier-attribution.md#L11-L24)
+
+### AD-2 — Immutable package and complete lineage [ADOPTED]
+
+- **Binds:** question package, raw answer, citations, conversation projection, normalized record, evidence, replay, delivery, and audit views.
+- **Prevents:** mutable source settings replacing the effective question, cross-run evidence attachment, and dedupe that loses traceability.
+- **Rule:** Persist the canonical effective question package and digest before dispatch. Every artifact and projection carries the package digest and originating run/execution plus established project, source/binding, worker/runtime, and artifact references. A mismatch is rejected or quarantined; a raw response is not a record or a deliverable business item. [OpenSpec spec](../../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L21-L32) [OpenSpec spec](../../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L47-L58) [OpenSpec spec](../../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L86-L97)
+
+### AD-3 — Independent evidence and outcome states [ADOPTED]
+
+- **Binds:** capture and delivery projections.
+- **Prevents:** URL extraction posing as verified citation, unknown conversation identity becoming a guessed URL, and HTTP `202`/accepted transport becoming business success.
+- **Rule:** Persist answer, citation, and conversation as separate evidence projections with capture status and provenance. Persist transport and business outcome separately. Only a documented destination ACK/equivalent that matches the delivery attempt and lineage may set business outcome to `confirmed`; otherwise retain `unconfirmed`, `unknown`, `partial`, `blocked`, or `failed` as applicable. [OpenSpec spec](../../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L34-L45) [OpenSpec spec](../../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L60-L71) [ADR-0021](../../../../docs/adr/0021-delivery-separates-submission-from-business-outcome.md#L3-L13)
+
+### AD-4 — Live gates never fall back [ADOPTED]
+
+- **Binds:** capability readiness, session, network, fixture/mock mode, and acceptance reporting.
+- **Prevents:** a configured catalog entry posing as readiness, fixture/mock evidence satisfying live gates, or a missing dependency becoming a simulated success.
+- **Rule:** Live execution requires observed published executable live capability, authenticated healthy persistent Doubao session, permitted network, valid answer/evidence/lineage, and matching destination ACK. Fixture/mock output must expose non-live mode and provenance and is excluded from live acceptance. Any missing or contradictory prerequisite ends in an explicit typed blocked/unknown/failed state, never implicit fallback. [OpenSpec spec](../../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L7-L19) [OpenSpec spec](../../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L73-L110)
+
+## Consistency Conventions
+
+| Concern | Convention |
+| --- | --- |
+| IDs and lineage | Package digest, run/execution, project, source/binding, worker/runtime, artifact, and delivery-attempt identity are retained across every projection. |
+| Evidence | Answer, citation, and conversation are separate; citation extraction remains unverified unless an independent verifier records otherwise; unavailable values remain `unknown`/`null`. |
+| State | Readiness, transport, and business outcome never collapse to one status. A blocked prerequisite has its precise reason; business confirmation requires ACK evidence. |
+| Fixtures | Every test/preview artifact carries `fixture`/`mock` mode plus provenance and cannot feed live-acceptance reporting. |
+
+## Structural Seed
+
+```mermaid
+sequenceDiagram
+  participant O as Operator
+  participant W as WorkflowProject/runtime
+  participant G as Capability/session/network gates
+  participant D as Doubao/OpenCLI
+  participant P as Normalize/record/evidence
+  participant T as Delivery destination
+
+  O->>W: Submit effective question and options
+  W->>W: Persist canonical package + digest
+  W->>G: Observe capability, session health, network
+  alt any prerequisite absent
+    G-->>W: typed blocked/unknown reason
+    W-->>O: no live-success claim
+  else prerequisites observed
+    W->>D: Execute immutable package
+    D-->>W: raw answer + citation/conversation capture
+    W->>P: source → normalize → accept → sink with lineage
+    P->>T: delivery attempt with idempotency identity
+    alt matching documented ACK
+      T-->>P: ACK evidence
+      P-->>O: confirmed live business outcome
+    else no matching ACK
+      P-->>O: transport status + unconfirmed/unknown outcome
+    end
+  end
+```
+
+## Capability → Architecture Map
+
+| Capability / area | Lives in | Governed by |
+| --- | --- | --- |
+| Readiness and live capture | Existing Gaojixing/Doubao runtime and session-bound OpenCLI channel | AD-1, AD-4 |
+| Package identity | Canonical question snapshot/digest before dispatch | AD-2 |
+| Evidence and lineage | Answer/citation/conversation capture; normalize, record, evidence, replay projections | AD-2, AD-3 |
+| Delivery and acceptance | Idempotent delivery attempt plus destination ACK evidence | AD-3, AD-4 |
+| Runtime evidence | Run events, SSE parity, runtime contracts, and ODP/Redis mirror where configured | AD-1; [Runtime conformance](../../../../docs/workflow-runtime-conformance.md#L10-L29) |
+
+## Deferred
+
+- Do not choose a new provider, credential mechanism, destination protocol, quota model, or citation-verification service here; the OpenSpec explicitly excludes inferring those facts. [OpenSpec proposal](../../../../openspec/changes/gaojixing-live-business-chain/proposal.md#L48-L53)
+- Do not claim current live acceptance from repository tests, historical fixtures, HTTP `202`, or the operator-reported seven-running/Doubao-session/Chrome-pool/ODP-`NOGROUP` snapshot. These require a fresh run-scoped evidence receipt against the matrix companion. [OpenSpec proposal](../../../../openspec/changes/gaojixing-live-business-chain/proposal.md#L36-L46)
diff --git a/_bmad-output/planning-artifacts/bmad-kickoff/current-project-context.md b/_bmad-output/planning-artifacts/bmad-kickoff/current-project-context.md
new file mode 100644
index 00000000..4b7a9e49
--- /dev/null
+++ b/_bmad-output/planning-artifacts/bmad-kickoff/current-project-context.md
@@ -0,0 +1,48 @@
+---
+name: snailfish-bmad-kickoff-context
+type: project-context
+status: current-snapshot
+created: 2026-08-27
+source-of-truth: openspec/changes/gaojixing-live-business-chain/
+---
+
+# BMAD Kickoff — Current Project Context
+
+## Authority and scope
+
+`openspec/changes/gaojixing-live-business-chain/` remains the normative change contract: its proposal, requirements, and task checkboxes are not superseded or edited by this document. This is a BMAD orientation artifact that organizes the present Gaojixing slice for execution. The existing runtime PRD and ADRs remain supporting product and architecture records.
+
+- The change defines an attributable live question-to-business-outcome chain, not a channel demo. [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L1-L3)
+- The product architecture keeps `WorkflowProject` as the authored execution input, compiles it in the backend, routes high-concurrency work to III/OpenCLI, and extends rather than rewrites the existing browser pool and ODP path. [Runtime PRD](../../../docs/workflow-hda-demand-runtime-PRD.md#L72-L91)
+- A workflow runtime is evidenced by registry declaration, executable fixture, and observed run-event transcript; a fixture alone is not a live-delivery proof. [Runtime conformance](../../../docs/workflow-runtime-conformance.md#L3-L6) [Runtime conformance](../../../docs/workflow-runtime-conformance.md#L39-L42)
+
+## OpenSpec-derived implementation position
+
+| Area | Recorded position | Basis |
+| --- | --- | --- |
+| Contract | Tasks 1.1–1.4 are checked: immutable package/digest, independently attributable evidence, lineage, explicit fixture separation, and fail-closed blockers are specified. | [OpenSpec tasks](../../../openspec/changes/gaojixing-live-business-chain/tasks.md#L1-L6) |
+| Capability and capture | Tasks 2.1–2.5 are checked. The repository has a `chat-ai.capture` live-mode Gaojixing runtime that hashes a canonical question package and rejects unavailable capability, denied network, or unhealthy Doubao session with typed errors. This is implementation evidence, not a live-acceptance receipt. | [OpenSpec tasks](../../../openspec/changes/gaojixing-live-business-chain/tasks.md#L8-L14) [Runtime gates](../../../backend/workflow/gaojixing_runtime.py#L46-L93) [Runtime gates](../../../backend/workflow/gaojixing_runtime.py#L96-L148) |
+| Normalize, record, delivery | Tasks 3.1–3.6 are checked. The contract requires `source → normalize → accept → sink`, retained lineage, and separate transport from business outcome. | [OpenSpec tasks](../../../openspec/changes/gaojixing-live-business-chain/tasks.md#L16-L23) [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L47-L71) |
+| Remaining scope | Tasks 4.1–4.3 and 5.1–5.7 are unchecked. Their execution and evidence—not the checked status of predecessor tasks—determine whether the change can close. | [OpenSpec tasks](../../../openspec/changes/gaojixing-live-business-chain/tasks.md#L25-L39) |
+
+## Live versus deterministic boundary
+
+- A live claim requires distinct observed capability publication, executable adapter/live mode, authenticated persistent session, passing session health, and permitted network; a catalog entry or configured channel is insufficient. [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L7-L19)
+- Every fixture/mock artifact must carry non-live mode and provenance, be excluded from live acceptance, and never be a fallback for a disappeared live dependency. [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L73-L84)
+- URL extraction is not citation verification; an unavailable conversation reference stays `unknown`. [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L34-L45)
+- HTTP `202`, enqueue acceptance, or no error is transport evidence only. Business success requires an ACK/equivalent tied to the delivery identity and lineage. [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L60-L71) [ADR-0021](../../../docs/adr/0021-delivery-separates-submission-from-business-outcome.md#L3-L13)
+
+## Known runtime blockers at kickoff
+
+The following are **operator-reported current observations supplied for this kickoff**, not facts proved by a repository document and not grounds for a success claim: seven runs are `running`; the Doubao session and Chrome pool require current health/capacity evidence; and ODP reports `NOGROUP`. Preserve each as `blocked` or `unknown` until a run-scoped receipt identifies the resource, time, and typed state.
+
+| Observation | Required treatment | Repository grounding for the treatment |
+| --- | --- | --- |
+| 7 running runs | Do not infer capacity, progress, or completion; obtain per-run event/status evidence and retain the run identity. | Run lifecycle state must be observable; runtime conformance uses the run events surface as evidence. [Runtime PRD](../../../docs/workflow-hda-demand-runtime-PRD.md#L87-L90) [Runtime conformance](../../../docs/workflow-runtime-conformance.md#L3-L6) |
+| Doubao session | Re-run the immediate pre-execution health/identity gate; `unavailable`/CAPTCHA is a typed fail-closed outcome, not a fixture substitute. | [OpenSpec tasks](../../../openspec/changes/gaojixing-live-business-chain/tasks.md#L10-L14) [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L7-L19) |
+| Chrome pool | Record actual pool/resource status before scheduling browser-bound work; do not treat a configured pool as available capacity. | Browser capacity is a resource pool. [Runtime PRD](../../../docs/workflow-hda-demand-runtime-PRD.md#L81-L85) |
+| ODP `NOGROUP` | Capture stream/group and whether the group is absent; it is an explicit no-group condition, distinct from a Redis outage, and must not be hidden by a generic healthy result. | [ODP metrics](../../../backend/control/collectors/odp_metrics.py#L157-L182) |
+
+## BMAD routing
+
+This kickoff has completed the prerequisites for a feature-level architecture spine and an execution-oriented task plan. The next required work is the unchecked OpenSpec task sequence in `live-business-chain-next-steps.md`; do not create a parallel PRD, spec, epic list, or replacement task ledger.
diff --git a/_bmad-output/planning-artifacts/bmad-kickoff/external-prerequisites-and-ack-matrix.md b/_bmad-output/planning-artifacts/bmad-kickoff/external-prerequisites-and-ack-matrix.md
new file mode 100644
index 00000000..9db2af20
--- /dev/null
+++ b/_bmad-output/planning-artifacts/bmad-kickoff/external-prerequisites-and-ack-matrix.md
@@ -0,0 +1,30 @@
+---
+name: gaojixing-live-prerequisites-and-ack-matrix
+type: acceptance-evidence-matrix
+status: kickoff-baseline
+created: 2026-08-27
+normative-source: openspec/changes/gaojixing-live-business-chain/
+---
+
+# Gaojixing External Prerequisites and ACK Evidence Matrix
+
+This matrix operationalizes the OpenSpec live-acceptance prerequisites; it does not replace them. A row is **observed** only when its listed run-scoped receipt exists. Absent evidence stays `blocked` or `unknown`, never `completed live`. [OpenSpec proposal](../../../openspec/changes/gaojixing-live-business-chain/proposal.md#L36-L46) [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L99-L110)
+
+| Gate / evidence | Required observation and durable reference | Valid state | Explicit non-evidence | Kickoff state |
+| --- | --- | --- | --- | --- |
+| Capability publication | Catalog entry identifies Gaojixing/Doubao, executable adapter, explicit `live` mode, and per-run readiness result. | published + executable + live | Catalog metadata alone, configured channel, fixture adapter. | Unknown; obtain fresh catalog/readiness receipt. [OpenSpec proposal](../../../openspec/changes/gaojixing-live-business-chain/proposal.md#L40-L46) |
+| Doubao session | Worker-bound authenticated persistent session, site/session binding, immediate pre-run health check, identity/scope (with `whoami` fallback), and CAPTCHA result. Do not persist secrets. | authenticated + healthy | Historic login, browser window, fixture session, assumed account. | **Operator-reported blocker:** session needs current receipt. [OpenSpec tasks](../../../openspec/changes/gaojixing-live-business-chain/tasks.md#L10-L14) |
+| Chrome capacity | Pool/resource status and assignment for the browser-bound run; identify endpoint/profile binding without equating configuration with available capacity. | observed available or typed blocked | Pool configuration, earlier pool status, unrelated desktop Chrome. | **Operator-reported blocker:** Chrome pool needs current status. Browser capacity is a resource pool. [Runtime PRD](../../../docs/workflow-hda-demand-runtime-PRD.md#L81-L85) |
+| Network policy | Permitted access to Doubao plus required citation/destination endpoints; capture domain-policy/SSRF decision. | permitted | Local fixture reachability, HTTP client creation, absence of an error. | Unknown; recheck immediately before live execution. [OpenSpec proposal](../../../openspec/changes/gaojixing-live-business-chain/proposal.md#L40-L46) |
+| Immutable input | Persisted effective question package and canonical digest before dispatch, referenced by prompt, evidence, lineage, replay, and audit views. | snapshot + digest observed | Mutable configured question, later-edited source settings, payload hash for a different object. | Verify under task 5.2/5.3. [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L21-L32) |
+| Capture evidence | Raw answer artifact, citation extraction projection/status, and conversation projection/status all share package digest and run/execution lineage. | captured or typed unknown/unavailable | URL extraction called verified citation; inferred conversation URL; empty response. | Verify under task 5.2/5.6. [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L34-L45) |
+| Normalize/record lineage | Normalized record, evidence links, project/source/binding/worker/runtime references, and canonical dedupe result are traceable to the same source artifact and package. | accepted with complete lineage | Raw provider response, a nearby record, or a dedupe result missing original linkage. | Verify under task 5.2/5.3. [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L47-L58) |
+| Delivery transport | Delivery attempt has idempotency identity, transport result, and retry lineage. | attempted/accepted/failed/unknown | HTTP `202`, enqueue success, or no client error presented as business completion. | Verify under task 5.5. [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L60-L66) |
+| Destination ACK | Documented ACK/equivalent is attached to the delivery attempt and record/package lineage. | `confirmed` business outcome | Accepted transport, a historic success, unmatched callback, or uncorrelated destination event. | Unknown until matching ACK receipt exists. [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L68-L71) |
+| Fixture/mock exclusion | Every resulting artifact/projection records mode and provenance; reporting excludes it from live acceptance. | deterministic fixture/mock verification only | A passing test, mocked adapter output, local fixture, or browser preview used as live proof. | Unfinished OpenSpec tasks 4.1–4.3. [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L73-L84) |
+| ODP group health | Record stream, group, timestamp, and detailed `NOGROUP` condition; distinguish absent stream/group from connection/auth/timeout outage. | typed availability/absence | A generic healthy status with no group evidence. | **Operator-reported blocker:** `NOGROUP`; code treats it as a no-group/no-key condition distinct from outage. [ODP metrics](../../../backend/control/collectors/odp_metrics.py#L157-L182) |
+| Existing running work | For each of the seven operator-reported `running` runs, retain run ID, stage, latest event, resource binding, age, and terminal/blocked reason before scheduling more live work. | individually observed | Aggregate count, presumed queue state, historical completion. | **Operator-reported blocker:** 7 `running`; no run receipts supplied to this kickoff. Runtime evidence is run-event based. [Runtime conformance](../../../docs/workflow-runtime-conformance.md#L3-L6) |
+
+## Terminal-state rule
+
+Report `completed live business success` only when every live prerequisite above is observed and the delivery row contains a matching ACK/equivalent. For all other combinations, report the precise `fixture/mock`, `blocked`, `failed`, `partial`, `unconfirmed`, or `unknown` state with retained evidence. [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L99-L110)
diff --git a/_bmad-output/planning-artifacts/bmad-kickoff/live-business-chain-next-steps.md b/_bmad-output/planning-artifacts/bmad-kickoff/live-business-chain-next-steps.md
new file mode 100644
index 00000000..eb5374b0
--- /dev/null
+++ b/_bmad-output/planning-artifacts/bmad-kickoff/live-business-chain-next-steps.md
@@ -0,0 +1,44 @@
+---
+name: gaojixing-live-business-chain-next-steps
+type: execution-plan
+status: derived-from-existing-openspec-tasks
+created: 2026-08-27
+normative-task-ledger: openspec/changes/gaojixing-live-business-chain/tasks.md
+---
+
+# Gaojixing Live Business Chain — Next Execution Plan
+
+This is a dependency order for the **existing unchecked OpenSpec task IDs**. It neither edits task wording/status nor creates a parallel backlog. The OpenSpec task ledger remains authoritative. [OpenSpec tasks](../../../openspec/changes/gaojixing-live-business-chain/tasks.md#L25-L39)
+
+## Stage 0 — Establish live-operability evidence (not a new OpenSpec task)
+
+Before any live acceptance, inspect the seven operator-reported `running` runs and bind current evidence for Doubao session, Chrome pool capacity, network policy, and ODP `NOGROUP` to the affected run/resource. Use `external-prerequisites-and-ack-matrix.md`; any absent observation stays blocked/unknown. This is a sequencing prerequisite derived from the OpenSpec’s external-gate rule, not a change to its ledger. [OpenSpec proposal](../../../openspec/changes/gaojixing-live-business-chain/proposal.md#L36-L46)
+
+## Stage 1 — Close the non-live boundary first
+
+1. **4.1** — Label every fixture/mock run and artifact with mode/provenance and prevent it satisfying live acceptance.
+2. **4.2** — Remove or reject implicit fallback when live capability, session, network, answer, or destination ACK is unavailable.
+3. **4.3** — Add explicit fixture-only contract coverage for digest, evidence shape, lineage, and split transport/business outcome.
+
+Why first: deterministic contracts establish the negative boundary that all live verification depends upon; the spec forbids fixture/mock fallback and demands explicit non-live reporting. [OpenSpec tasks](../../../openspec/changes/gaojixing-live-business-chain/tasks.md#L25-L29) [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L73-L84)
+
+## Stage 2 — Verify gates and safety properties without asserting live success
+
+4. **5.1** — Verify precise blockers for adapter, authentication, session health, and network permission.
+5. **5.4** — Verify fixture/mock output is visibly non-live and cannot produce live business acceptance.
+6. **5.6** — Verify timeout, CAPTCHA, unauthenticated session, network denial, empty/malformed evidence, lineage mismatch, and destination failure fail closed with typed state.
+7. **5.2** — Verify one immutable snapshot/digest flows through answer, citation, conversation, record, evidence, and delivery projections.
+8. **5.3** — Verify post-dispatch source changes cannot alter the captured snapshot/digest or mix run evidence.
+9. **5.5** — Verify accepted transport without matching ACK remains unconfirmed/unknown; verify matching ACK confirms business outcome.
+
+This order proves refusal/mode boundaries before happy-path lineage and then the ACK boundary. It implements the OpenSpec’s fail-closed and no-HTTP-202-as-success requirements. [OpenSpec tasks](../../../openspec/changes/gaojixing-live-business-chain/tasks.md#L31-L38) [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L60-L71) [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L86-L110)
+
+## Stage 3 — Focused checks, then conditional live acceptance
+
+10. **5.7** — Run focused contract/integration checks. Run a live acceptance only after all matrix prerequisites are actually observed and the destination ACK evidence is bound to the delivery attempt.
+
+Do not convert a historical run, a fixture/mock, local fixture network result, `HTTP 202`, accepted transport, or no-error response into current business success. The relevant PTT evidence explicitly limits local fixture execution to adapter/kernel path evidence rather than internet/public acceptance. [OpenSpec tasks](../../../openspec/changes/gaojixing-live-business-chain/tasks.md#L38-L39) [OpenSpec proposal](../../../openspec/changes/gaojixing-live-business-chain/proposal.md#L40-L46) [PTT acceptance](../../../docs/ptt-acceptance.md#L197-L211)
+
+## Completion rule
+
+Do not mark this change live-accepted merely because tasks 1–3 are already checked or fixture checks pass. The final acceptance record must retain observed prerequisites and the matching destination ACK; otherwise retain the exact blocked/failed/partial/unconfirmed/unknown/fixture state. [OpenSpec spec](../../../openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md#L99-L110)
diff --git a/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/.memlog.md b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/.memlog.md
new file mode 100644
index 00000000..ba2fe3e8
--- /dev/null
+++ b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/.memlog.md
@@ -0,0 +1,44 @@
+---
+topic: Gaojixing Live Business Chain
+updated: 2026-08-28T08:32
+---
+
+- (assumption) Internal-production stakes adopted because the user repeatedly requested continue; this is a corrigible assumption.
+- (decision) Coaching path was explicitly chosen by the user.
+- (assumption) Vision + Features entry selected based on the platform-product recommendation; this is a corrigible assumption.
+- (change) User provided 2026-04-05 OpenCLI: Artificial Intelligence Agents in Emacs as a new product-idea input, emphasizing programmable agent environments, unified plugins, interaction crystallization, token economics, self-healing, composable primitives, CLI Hub, and dynamic loading; whether to expand the PRD from the Gaojixing single chain into a platform vision plus first use case remains pending user confirmation.
+- (decision) PRD adopts the OpenCLI programmable Agent environment as the platform vision, with Gaojixing Live Business Chain as the first production-grade use case; this phase does not require crystallizing a real Doubao session into a permanent skill, while skill crystallization, self-healing, and a general plugin ecosystem remain future platform boundaries unless explicitly included later.
+- (change) User added https://github.com/jackwener/opencli-website as a source input and requested that selected functionality be included in the current platform PRD; the exact scope is pending repository-fact extraction and user confirmation.
+- (decision) This phase prioritizes P1: usable platform feature entry points with correct Gaojixing context, not a marketing-site clone. P2 third-party community plugin discovery/governance is deferred. P3 downloads, Windows/macOS installation, release/integrity, public documentation, Chinese/English blogs, and vision articles are deferred; current focus is optimizing existing platform functionality.
+- (change) P1 scope drafted as an internal operational platform entry: capability discovery, truthful readiness, capability-to-workflow use, and Gaojixing live evidence/lineage/matching-ACK context; P2 community plugin governance and P3 public distribution/content are deferred, with article mechanics retained in addendum only.
+- (decision) S1 is included in this phase: improve the trustworthiness, failure trace, versioning, rollback, and correction-state presentation of the existing skill record → distill → execute → correct flow; this does not commit this phase to automatic self-healing, general crystallization, or dynamic loading.
+- (change) New product input: Gaojixing results are ultimately consumed by the geoXI project; this phase adds real-time keyword collection, scheduled keyword collection, and an OODA closed-loop requirement.
+- (decision) Destination is geoXI, and the PRD must treat geoXI consumption as the business destination while adding real-time and scheduled keyword collection plus an OODA loop.
+- (decision) The snailfish system itself owns and completes the full OODA loop; geoXI is an Act-stage business destination/consumer and one feedback source. Adopt geoXI confirmation type A: persist results to the corresponding geoXI project for later query, analysis, and next-round use, and set business outcome confirmed only after a consumption receipt matching the delivery identity.
+- (decision) Scheduled keyword collection punctuality is measured by starting or triggering within ±1 minute of the planned time, with a 99% target.
+- (decision) geoXI is a downstream interface; snailfish dispatches results through its own delivery capability, transport accepted is not geoXI consumed, and only a matching consumption receipt for a persisted/queryable result confirms business outcome.
+- (decision) Real-time collection latency will be baseline-measured first; complete system-owned OODA cycle success target is 95%.
+- (override) User authorized one AUTO drafting round; difficult or high-impact uncertainties remain for user discussion and must not be decided unilaterally.
+- (assumption) GeoXI consumption latency will be baseline-measured and targeted only after real-interface integration; authenticity guardrails target zero because existing fail-closed contracts make fixture-as-live, unmatched-receipt confirmation, and retry-duplicate outcomes invariants.
+- (decision) geoXI is a downstream product, not a snailfish internal module. geoXI owns persistence and produces the minimum user-observable consumption receipt; snailfish owns association validation, receipt retention/display, business-outcome updates, and OODA feedback handling.
+- (decision) Overall readiness is unknown, blocked, or ready. Independent gates are capability publication, executable adapter, required configuration, authentication, session health, network permission, browser/Chrome capacity, and runtime binding. Execution status queued/running/completed/failed and mode live/fixture/mock remain separate dimensions.
+- (decision) S1 uses dual-role production approval: the skill maintainer views traces, initiates correction, and proposes rollback; the platform administrator approves production enablement and rollback, may pause, and audits the full process; workflow/operations users may only view and pause runs and cannot designate a production version.
+- (decision) OODA actions use risk tiers: low-risk actions auto-Act; medium-risk actions require OODA strategy owner approval; high-risk business actions require business reviewer or platform administrator approval. Evidence, lineage, receipt, mode, or duplicate-side-effect anomalies stop execution and require human handling.
+- (decision) Initial real baseline: two real geoXI projects (one with normal business volume and one newly created/low volume); at least three keywords per project covering normal high-hit, low-hit/empty, and Chinese composite/boundary cases; both real-time and scheduled triggers; normal/delayed receipts, rejection/failure, retries, missing/mismatched receipts; seven consecutive days; at least 100 real live OODA cycles. Report stratified by project × keyword × trigger × outcome × receipt. External blockers are reported separately and never silently removed from denominators or counted as success.
+- (event) ALL reviewer plan written, execution pending.
+- (decision) OODA completed semantics confirmed: completion requires a cycle and immutable package plus minimum observable Observe/Orient/Decide/Act records; when external feedback is required it must be matching, consumed by the cycle, and record its effect on the next round or close. Valid no-action records a reason and feedback-not-required. Terminal states include waiting_feedback, partial, blocked, failed, expired, and completed; missing/late/duplicate/mismatched feedback cannot complete a cycle.
+- (decision) Adopt minimum evidence/lineage pack with layered acceptance: immutable package/digest; project/workflow/run/execution/source/binding/worker/runtime/mode/provenance; nonempty raw answer artifact; citation projection status/method/items where extraction is separate from verification; conversation projection status/reference allowing unknown/unavailable without guessing; normalize/accept outcome and reason with accepted record reference; end-to-end lineage; delivery identity/transport; confirmed only with matching geoXI receipt. Optional empty citations or unknown conversation may continue with bounded claims; missing core identity/evidence/lineage or any package/run/project/artifact mismatch fails closed.
+- (decision) Readiness uses static/dynamic gate freshness, coherent run-scoped admission, post-admission invalidation, and full recovery reevaluation: static publication/executable/config/runtime-binding gates remain valid only until their configuration/version/binding changes; dynamic authentication/session/network/browser-capacity gates are re-observed before each run admission and before waits, retries, or dependent actions; each run has one coherent all-required-gate evaluation; invalidation blocks/pauses and recovery requires re-observing all gates for a new admission.
+- (change) Workflow contract audit completed: generic webhook only; transport is distinct from business outcome; matching delivery identity exists; geoXI contract is absent.
+- (decision) T-H4 safe derivations adopted: identity matching, attempt visibility, ambiguous outcomes remain unconfirmed, duplicate/replay receipts rejected, and project/package/run/delivery mismatches fail closed; consumed requires a queryable persisted result.
+- (assumption) Exact geoXI receipt validity duration and compensation policy are deferred to geoXI product owner with OODA strategy owner approval, condition: before geoXI live readiness/integration acceptance; without approved policy live readiness remains blocked/unknown.
+- (change) S1 workflow audit complete.
+- (decision) S1 safe derivation: execution, correction, approval, rollback, version, trace, and known-good states remain explicitly bound; proposal/distill never promotes a production version, environment errors remain distinct from skill failures, and rollback effectiveness requires restored-version attribution.
+- (assumption) Exact known-good sample count and risk strata are deferred to the skill maintainer and platform administrator, with OODA strategy owner included for OODA risk; condition: before production promotion for the target capability. Until decided, remain under-review and never known-good.
+- (change) Review medium safe batch applied.
+- (decision) OODA approvals bind to the exact cycle, action, keyword package, target project, risk tier, policy version, evidence snapshot, actor, and time; any scope or evidence change invalidates approval and generic prior approval cannot be reused.
+- (assumption) Deferred low-item owners: platform operations owns in-flight pause semantics before the production pause/recovery runbook; operational acceptance owns freshness/time-order semantics before live baseline target setting; operational acceptance owns exact per-stratum baseline minimum before target setting.
+- (change) Medium safe batch verification correction: repaired missing schedule occurrence accounting, approval binding, mixed/partial non-live rules, acceptance-target labeling, OODA reporting, and triage anchors.
+- (event) External handoffs skipped: none configured.
+- (event) PRD finalized
+- (event) PRD finalized
diff --git a/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/addendum.md b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/addendum.md
new file mode 100644
index 00000000..36e219ab
--- /dev/null
+++ b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/addendum.md
@@ -0,0 +1,58 @@
+# Addendum — 平台机制与来源范围判定
+
+本文件保留用户提供的技术深度、外部网站来源分析与后置决定；它们不构成本期 P1 的验收承诺。
+
+## 1. OpenCLI「AI Agent 的 Emacs」文章：长期机制，不纳入本期验收
+
+来源：用户提供的 OpenCLI Team 2026-04-05 文章《OpenCLI：人工智能代理的 Emacs》。
+
+文章提出的长期平台机制：
+
+- **Everything is CLI** 的可编程环境理念；
+- 从 operate 的交互探索、network/API 发现，经 init 脚手架和 verify，最终将即兴浏览 crystallize 为永久 CLI 技能的路径；
+- TypeScript / YAML 适配器、动态加载，以及保存后下次调用即时生效；
+- 透明 adapter 源码与结构化诊断，以支持 Agent 原地自修复；
+- CLI Hub 统一外部工具；
+- 可组合的 pipeline primitives；
+- 一次开发、重复低 token 调用的经济性；文章中的约 **92% token reduction** 仅是示例，不是本期 KPI、SLA 或性能承诺。
+
+这些内容说明长期平台方向，但本期不验收通用 CLI 化、API 发现、永久技能 crystallization、TS/YAML 动态加载、自动 adapter 修复、CLI Hub 或 pipeline primitive。P1 只采用其产品含义：能力应在产品内可发现、可判断真实 readiness、可进入既有工作流。
+
+## 2. `jackwener/opencli-website` 来源分析与范围决定
+
+来源：用户提供的 GitHub 仓库，默认分支 `master`：<https://github.com/jackwener/opencli-website>。
+
+### P1：本期采用的产品含义
+
+不复制其营销首页。仅采用与产品内 operational platform entry 一致的能力：
+
+- 平台/命令目录的能力发现与按类别理解；
+- 将 browser、desktop、public 等运行语境和命令能力清楚区分；
+- 从能力详情进入实际工作流使用；
+- 以 Doubao 作为高吉星首个生产用例的正确平台上下文，而非一个孤立页面。
+
+对应网站来源包括 `src/components/Platforms.tsx`、`src/components/data.ts` 与 `src/App.tsx`；P1 在 snailfish 中以现有的产品内插件、OpenCLI 适配和工作流入口为优化对象，不承诺公开站点。
+
+### P2：后置的社区插件生态
+
+网站已实现/表达的社区插件目录、插件详情、安装命令、GitHub metadata、贡献 PR 入口和 YAML/TS 分类，不属于本期。
+
+后置原因：第三方社区插件进入生产平台前必须先解决审核、安装安全、权限、兼容性和治理。该阶段不因网站存在展示页而被视为已可安全交付。
+
+相关来源：`src/pages/PluginsPage.tsx`、`src/pages/PluginDetailPage.tsx`、`src/data/plugins.json`、`CONTRIBUTING_PLUGINS.md`。
+
+### P3：后置的公共分发和内容面
+
+以下均明确后置：Download、Windows/macOS 安装、Release/完整性、公开文档、中英博客与理念文章。
+
+网站的相应实现位于 `src/pages/DownloadPage.tsx`、`src/pages/BlogPage.tsx`、`src/pages/BlogPostPage.tsx`；`/docs/` 在部署时由另一仓库的 VitePress 文档构建并合并（`.github/workflows/deploy.yml`）。这些是未来公共产品面/分发责任，不是当前优化产品功能的 P1 交付。
+
+## 3. 本期边界结论
+
+本期标题中的「OpenCLI 可编程 Agent 平台」是产品愿景与上下文；本期承诺仅限 P1 的内部/产品内 capability discovery、真实 readiness、从能力到工作流使用，以及高吉星的 live evidence / lineage / matching destination ACK 上下文。长期平台愿景不自动扩张为本期范围。
+
+## 4. geoXI 与 OODA 的当前事实边界
+
+仓库当前没有已定义的 geoXI/GEO-XI 专属消费契约、消费事件或 ACK 规范；本期产品决定将 geoXI 作为 Act 阶段业务 destination/消费者，并要求匹配消费回执后才确认业务结果，但不在此处发明技术合同。
+
+`docs/SYSTEM_ANALYSIS.md` 与 `docs/CONTROL_THEORY_ARCHITECTURE.md` 提供 OODA、反馈和控制动作的现状分析/架构参考，指出当前采集管线存在单向开环、质量反馈弱和 egress ACK 缺口。它们是产品抽取与范围判断的参考，不是 snailfish 已完成 OODA 交付或 geoXI ACK 能力的证明。
diff --git a/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/prd.md b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/prd.md
new file mode 100644
index 00000000..008362d8
--- /dev/null
+++ b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/prd.md
@@ -0,0 +1,324 @@
+---
+title: OpenCLI 可编程 Agent 平台：高吉星实时业务闭环（首个生产用例）
+status: final
+created: 2026-08-28
+updated: 2026-08-28
+---
+
+## Executive Summary
+
+本产品将 OpenCLI 定义为面向 Agent 的**可编程运行平台**，而非单一浏览器工具：使用者应能在产品内发现可用能力、看到其真实运行条件，并把合适的能力带入工作流。
+
+高吉星 Live Business Chain 是这一平台的首个生产用例。系统自身通过实时或定时 keyword collection 完成 Observe → Orient → Decide → Act 的 OODA 闭环：真实浏览器/OpenCLI 调用 Doubao，处理可归因问题包，保留答案、引用与会话证据的血缘，经 normalize → accept → sink 后交给 geoXI 对应项目持久化、查询和分析；仅当 geoXI 返回与 delivery identity 匹配的消费回执时，才判定该投递业务成功。P1 的目标是让这一用例出现在正确的平台操作上下文中；它不是营销首页复制，也不将长期平台愿景误写成本期承诺。
+
+## Product Vision
+
+让 Agent 能力从“存在于某个适配器或命令中”变成可发现、可判断是否真实可用、可投入工作流并可审计其业务结果的产品能力。平台以真实运行事实为准：浏览器、登录、配置、运行绑定和业务确认均应被清楚区分。
+
+平台自身应能围绕实时与定时 keyword collection 完成可观察、可判断、可决策、可行动并吸收反馈的 OODA 循环。高吉星证明该平台可以承载生产级闭环：能力的调用结果不仅可被看到，还能被追溯到问题包、证据、血缘，并由 geoXI 消费回执确认业务结果。
+## Problem / Opportunity
+
+- 已登记的网站适配、Provider、节点和技能需要一个连贯的操作入口；使用者不应只看到静态目录，却无法判断何种能力当前可运行、需要什么条件或应在何处使用。
+- 从能力发现到工作流使用之间若缺少连续路径，平台的现有能力难以被可靠地转化为实际自动化。
+- 浏览器或 Agent 给出答案并不构成业务成功。高吉星需要在平台中明确呈现真实调用、证据与血缘，以及 geoXI 消费回执的匹配状态，避免将 fixture/mock、执行完成、发送成功或 geoXI 尚未消费冒充 live 业务完成。
+- 平台必须自身完成 OODA，而非把闭环责任转交 geoXI；实时与定时 keyword collection、反馈和下一轮行动构成本期生产验证范围。
+- 平台愿景包含可复用的 Agent 环境，但本期优先解决产品内可操作性；将公共营销、分发或开放生态提前纳入会稀释生产闭环的验证重点。
+## Product Principles
+
+1. **运行事实优先于目录宣称。** 已登记、已安装、已配置、可运行与业务成功是不同状态，产品不得混淆。
+2. **能力发现必须通向实际使用。** 使用者应能从能力上下文理解其限制与 readiness，并进入工作流使用路径。
+3. **Live 必须可证明。** 只有真实浏览器/OpenCLI 与真实 Doubao 会话产生的结果可计入高吉星 live；fixture、mock、缓存或演示结果必须保持可区分。
+4. **系统自身完成 OODA。** Observe、Orient、Decide、Act 及反馈回流必须可观察；完成必须有 cycle、immutable package 和各阶段最小记录，所需 feedback 必须被消费并记录影响，合法 no-action 必须说明原因；geoXI 是 Act 阶段的业务 destination/消费者和反馈来源之一，不是 OODA 的唯一责任主体。
+5. **证据和归因分层。** 必需 evidence/lineage pack 完整且 identity 一致才可 accepted/交付；可选 citation 内容为空或 conversation unknown 可继续但不得过度宣称；核心缺失或 package/run/project/artifact mismatch 必须 blocked/failed 或 fail closed；无匹配 geoXI receipt 不得 consumed/confirmed。
+6. **状态维度不得混用。** 总体 readiness 仅为 unknown、blocked 或 ready；独立 gates、execution status（queued/running/completed/failed）和 mode（live/fixture/mock）分开呈现，completed 不等于 business success。static gates 仅在配置/版本/绑定未变化时有效，dynamic gates 在 admission 与等待、重试、依赖动作前重观。
+7. **OODA 按风险分层。** 低风险可自动 Act；中风险需 OODA 策略责任人批准；高风险需业务复核者或平台管理员批准。低置信度、冲突、证据/lineage/receipt/mode 异常或重复副作用风险必须停止并人工处理。
+## Current Release Scope
+
+本期为 **P1：内部/产品内的 operational platform entry**，优化已有平台功能，不交付公共营销站。
+- **能力发现：** 在产品内提供网站适配、Provider、节点与相关能力的可理解入口，使使用者能按能力上下文进行发现。
+- **真实 readiness：** 独立呈现 gates；static gates 在配置/版本/绑定变化后失效并须重观，dynamic gates 在每次 run admission 前及等待/重试/依赖动作前重观；每次 run 必须有同一有效窗口内的 coherent run-scoped all-required-gate evaluation，未观察为 unknown，明确失败为 blocked，全部通过才 ready。
+
+- **Acceptance boundary:** 本节范围与下列指标是 acceptance target, not current proof；checked tasks、fixture、catalog/configuration 或历史记录均不证明 live、geoXI consumed 或 OODA completed。
+- **从能力到工作流使用：** 让已发现且总体 readiness 为 ready 的能力能够进入既有工作流/Studio 使用路径；queued/running/completed/failed 仅描述执行状态，completed 不代表业务成功。
+- **高吉星的正确平台上下文：** 将高吉星表述并呈现为首个生产用例：真实 Doubao 调用、可归因 keyword package、答案/引用/会话证据、normalize → accept → sink 血缘；snailfish 通过自身交付能力向 geoXI downstream interface 派发，结果进入对应项目供持久化、查询和分析，只有 matching consumption receipt 才确认业务结果。
+
+
+## Users & Roles
+
+### 工作流设计者 / 运营人员
+
+在产品内寻找合适的能力，理解当前 readiness 与使用前提，并将能力带入工作流以完成自动化任务；只能查看或暂停运行，不能指定生产技能版本。
+
+### 高吉星运行操作者
+
+发起并观察实时或定时 keyword collection，观察真实 Doubao 问题包的生产运行，区分执行状态、证据状态、geoXI 消费状态和 OODA cycle 状态。
+
+### OODA 策略责任人
+
+定义或复核系统 Observe、Orient、Decide、Act 的业务判断依据与行动边界，检查反馈是否形成下一轮，并对需要人工处理的决策负责。
+
+### 业务结果复核者
+
+从答案回查引用、会话证据、问题包、血缘与匹配 geoXI 消费回执，判断结果是否可接受、是否已被 geoXI 持久化并可供后续使用。
+
+### geoXI 下游产品责任方
+
+geoXI 是独立的下游产品，负责将结果持久化到对应项目并产生消费回执；其消费状态、时间、结果引用和失败原因由 snailfish 接收并用于业务复核。
+
+### 技能维护者
+
+查看技能失败 trace，发起纠正并提出回滚建议；不能单独批准生产启用或生产回滚。
+
+### 平台管理员
+
+维护 readiness，审批技能生产启用与回滚，可暂停运行，并审计全程；不以审批替代证据或真实性护栏。
+## Capabilities & Functional Requirements
+
+### 1. 能力目录与上下文发现
+
+产品应让内部用户在正确的平台上下文中发现可用能力，并理解能力适用的运行语境与下一步用途。
+
+#### FR-001
+用户可以按能力类别浏览已登记的网站适配、Provider、节点、Agent 与技能，并看到每项能力的名称、用途和适用运行语境。
+
+#### FR-002
+能力详情应区分总体 readiness（unknown、blocked、ready）与独立 gates；static gates 在配置、版本或绑定变化后须重新观察，dynamic gates 须在每次 run admission 前及等待、重试或依赖动作前重新观察。
+
+#### FR-003
+能力详情应提供进入既有 Studio/工作流使用路径的明确入口，并保留当前能力上下文，避免用户重新寻找或误选能力。
+
+### 2. 真实 readiness
+
+产品应以可观察的运行事实呈现能力是否真正具备执行条件，并在条件不足时阻止不真实的成功预期。
+
+#### FR-004
+用户可以看到每个 gate 当前是否通过、未观察或明确失败及原因；每次 run 只有在同一有效窗口内完成 coherent run-scoped all-required-gate evaluation 才可 admission。必要 gate 未观察为 unknown，任一明确失败为 blocked，全部通过才 ready。
+
+#### FR-005
+关键 gate 在 admission 后失效时，产品应将运行转为 blocked 或 paused，保留已有 partial evidence，不继续执行、不 fixture fallback、不复用旧 ready；恢复须重观所有必要 gates、记录 blocker 与恢复原因，全部通过后才产生新 admission。
+
+#### FR-006
+产品应将总体 readiness、execution status（queued、running、completed、failed）和 mode（live、fixture、mock）作为独立维度呈现；partial coverage 或任何 live+fixture/mock 混合 artifacts 必须明确标记 partial/mixed、non-live、non-confirmed，保留每项 provenance，且不得满足 live success。
+
+#### FR-007
+当 live 前提不满足或结果为 partial/mixed 时，产品不得静默切换或宣称 live success；应显示 blocked、partial 或 mixed mode 及可纠正原因。
+
+### 3. 能力到 Studio/工作流连续入口
+
+产品应把能力发现连接到现有 Studio/Canvas 工作流，使用户能在理解约束后实际编排和运行能力。
+
+#### FR-008
+用户从能力详情进入 Studio/Canvas 时，产品应带入该能力的正确上下文，并允许用户将其作为工作流中的候选能力使用。
+
+#### FR-009
+工作流编辑与运行入口应呈现独立 gates、总体 readiness、execution status 与 mode；运行 admission 前须有同一有效窗口的 coherent gate evaluation，旧 ready 不能跨 run/时间复用，execution completed 也不得呈现为 business success。
+
+#### FR-010
+工作流运行后，用户应能从项目操作或运行视图进入该次运行的状态、事件和结果，而不必依赖目录页面推断执行是否发生。
+
+### 4. 高吉星生产闭环上下文
+
+产品应将高吉星呈现为平台首个生产级用例，并让业务结果可由问题包、证据、血缘和 destination 确认共同复核。
+
+#### FR-011
+高吉星运行应展示不可变 keyword package/digest，以及 project、workflow、run、execution、source、binding、worker、runtime、mode 和 provenance 归因，使结果可确认对应本次运行。
+
+#### FR-012
+产品应分别呈现 nonempty raw answer artifact、citation projection（status、method、items，可为空且 extraction 不等于 verified）和 conversation projection（status、reference，可 unknown/unavailable，不猜）；缺失可选内容不得被过度宣称为已验证证据。
+
+#### FR-013
+被接受的答案应展示 normalize/accept outcome 与 reason、accepted record reference，并保留从 package、raw answer、evidence 到 record、delivery 和 geoXI receipt 的 end-to-end lineage；核心 package、raw answer、identity 或 lineage 缺失时不得 accepted/交付。
+
+#### FR-014
+产品应分别呈现每个 delivery transport attempt、geoXI receipt 状态（pending、consumed、rejected、expired）与 OODA cycle completed；transport accepted 或 execution completed 均不等于业务成功。
+
+#### FR-015
+receipt 必须匹配 target project、package、run、execution、delivery identity 和 persisted result reference；只有目标项目可查询且匹配的 receipt 才能显示 consumed/confirmed。late receipt 仅归原 attempt，不确认新 attempt；duplicate、replay、ambiguous 或 mismatch receipt 必须 rejected/unconfirmed。
+
+#### FR-016
+当核心 evidence/lineage 缺失，或 package、digest、run、project、artifact 任一 identity mismatch、receipt 异常或 timeout-after-send/concurrent/unknown outcome 未完成 reconciliation 时，产品应 fail closed，显示可行动原因，不得创建或暗示 live business success。
+
+产品应提升现有技能录制→蒸馏→执行→纠正流程的可信度，让用户能判断版本、失败原因、回滚选择和纠正状态。
+
+#### FR-017
+技能用户应能区分录制、蒸馏、可执行、执行中、执行成功、执行失败、candidate、under-review、known-good 和 rolled-back 状态，并看到每个状态绑定的 exact skill version。
+
+#### FR-018
+每次技能执行或纠正失败时，用户应能查看与 exact skill version 绑定的失败 trace、阶段和可理解原因，并区分 environment-error 与 skill failure。
+
+#### FR-019
+用户应能查看 execution、correction、approval 和 rollback 与 exact skill version 及 linked trace 的关联；proposal 或 distill 不得 promotion，candidate/under-review/known-good/rolled-back 不得互换。
+
+#### FR-020
+技能维护者可以提出纠正或回滚，平台管理员审批生产启用与回滚；工作流/运营人员不能指定生产版本，只能查看或暂停运行。
+
+#### FR-021
+纠正记录应关联 from/to version、prior snapshot 和触发 trace，并显示待处理、进行中、成功、失败或需人工复核；未满足条件的版本不得自动成为 known-good。
+
+#### FR-022
+known-good 至少要求目标能力验证通过、至少一条目标范围真实 passing execution、完整 linked evidence/lineage、无未解决严重失败且无真实性/重复护栏违规；由技能维护者提出、平台管理员确认，未确认保持 under-review。批准可见且可审计。
+
+### 6. 实时与定时 Keyword Collection
+
+产品应支持以 keyword 驱动高吉星的实时和定时采集，并让每次采集的时效、状态、失败与归因可被复核。
+
+#### FR-023
+用户应能从正确的平台上下文发起一次实时 keyword collection，并在 admission 前看到 coherent run-scoped readiness evaluation；static 或 dynamic gate 失效时不得开始 live 运行。
+#### FR-024
+用户应能配置、查看、启用、停用和修改按 keyword 运行的定时采集，并能看到下一次计划时间与最近一次运行状态；每个 planned occurrence 具有可观察的唯一 identity 与 outcome，duplicate、skipped、coalesced、late 均明确记录，±1 分钟基于该 occurrence 的 planned time，duplicate/missed 不得消失。
+
+#### FR-025
+每次实时或定时采集都应保留不可变 keyword package/digest 及完整 lineage；采集启动后修改不得改变该次归因，gate 失效或 recovery 后旧 admission 不得复用。
+
+#### FR-026
+采集结果应呈现 freshness、运行阶段、结果状态和失败原因；空结果、过期结果、超时、认证/网络阻塞或来源变化不得被显示为成功采集。
+
+#### FR-027
+当实时或定时采集缺少必要 readiness、证据或合法运行条件时，产品应 fail closed，显示 blocked、unknown 或 failed；不得以 fixture/mock 替代 live 结果。
+
+### 7. 系统级 OODA 与 geoXI 反馈闭环
+
+产品应让 snailfish 自身完成可观察的 OODA 循环，使用 geoXI 作为 Act 阶段业务消费者和反馈来源之一，并推动下一轮行动。
+
+#### FR-028
+用户应能查看每个 OODA cycle 的 Observe、Orient、Decide、Act 阶段的最小可观察记录，并确认该 cycle 与 immutable keyword package、采集运行和反馈关联。仅有阶段标签不能构成 completed。
+
+#### FR-029
+Observe 阶段应呈现采集结果、freshness、质量/有效性状态及 evidence/lineage pack；nonempty raw answer 是核心条件，可选 citation 内容为空或 conversation unknown 时保持状态并限制声明。
+
+#### FR-030
+Orient 阶段应呈现 normalize/accept outcome、reason、record reference、证据关联和上下文判断；核心归因缺失或不一致时不得进入 accepted。
+
+#### FR-031
+Decide 阶段应呈现触发 Act、暂缓、重试、人工复核或结束本轮的判断及依据，并按风险分层。中/高风险 approval 必须绑定 exact OODA cycle、proposed action、keyword package、target project、risk tier、policy version、evidence snapshot、actor 和 time；任一 scope/evidence 改变即失效，generic prior approval 不可复用。
+
+#### FR-032
+Act 阶段应呈现每个 transport attempt 以及 snailfish 向 geoXI downstream interface 派发的结果；低置信度、lineage mismatch、缺关键证据、receipt 缺失/错配、fixture/mock/live 混淆或重复副作用风险必须停止并转人工处理。
+
+#### FR-033
+geoXI 是独立 downstream product，负责持久化并提供可查询结果。receipt 必须包含并匹配 target project、package、run、execution、delivery identity 和 persisted result reference；snailfish 负责验证关联、保存和展示。receipt 状态为 pending、consumed、rejected 或 expired；只有匹配且目标项目可查询时才可显示 consumed/confirmed。
+
+#### FR-034
+产品应分别呈现各 transport attempt、receipt 状态与 OODA cycle completed；accepted transport 不等于 consumed，缺 receipt 或未完成 reconciliation 不得 confirmed。late receipt 仅归原 attempt，不确认新 attempt。
+
+#### FR-035
+geoXI 消费反馈应回流对应 OODA cycle；duplicate、replay、ambiguous 或 mismatch receipt 必须 rejected/unconfirmed，异常时不得虚构闭环完成并须人工处理。
+
+#### FR-036
+重试保持同一 business identity，各 transport attempt 可见；timeout-after-send、concurrent retry 和 unknown outcome 必须 reconciliation，重复 geoXI 结果计入 guardrail incident 且不得被 dedupe 隐藏。
+
+## Success Metrics
+### Scheduled keyword collection punctuality
+
+- **Definition:** Percentage of scheduled keyword collection planned occurrences started or triggered within ±1 minute of each occurrence’s planned time.
+- **Target:** 99%.
+- **Denominator:** All enabled, user-not-paused planned occurrences, including occurrences with a declared external blocker.
+- **Reporting:** Eligibility and blocker classification are fixed at planned trigger time and cannot be rewritten later; blocked/not-started occurrences remain in the denominator and are reported separately, never silently counted as success.
+
+### Real-time collection latency
+
+- **Definition:** Baseline measurement of elapsed time from an accepted real-time keyword collection request to its recorded start/trigger event, using real live runs only.
+- **Baseline sample contract:** The initial baseline uses two real geoXI projects (one normal-volume and one new/low-volume), at least three keywords per project (normal high-hit, low-hit/empty, and Chinese composite/boundary), both real-time and scheduled triggers, and at least 100 real live OODA cycles over seven consecutive days. It includes normal/delayed receipts, rejection/failure, retries, and missing/mismatched receipts, reported by project × keyword × trigger × outcome × receipt.
+- **Baseline rule:** Report external blockers separately; never silently remove them from denominators or count them as success. This metric has no pass claim until a numeric target is set after reviewing this representative baseline.
+- **Target-setting trigger:** Set a numeric target after the baseline review establishes the normal operating range.
+
+### Complete system-owned OODA cycle success
+
+- **Definition:** Percentage of all triggered live real-time or scheduled cycles with completion evidence.
+- **Target:** 95%.
+- **Denominator:** All triggered cycles; report admitted, blocked, waiting, completed, and no-action totals. Blocker/eligibility classification is fixed at admission/trigger and cannot be shrunk post hoc; only cycles with immutable package, minimum stage records, and required feedback evidence count completed.
+
+### GeoXI consumption latency
+
+- **Definition:** Baseline measurement from snailfish delivery dispatch to the matching geoXI consumption receipt for a persisted, queryable project result.
+- **Baseline rule:** Collect only after real geoXI interface integration, including successful and delayed/failed receipts; no target or pass claim is made before baseline review.
+- **Target-setting trigger:** Set a numeric target after real-interface integration provides a representative baseline and the consumer responsibility boundary is confirmed.
+
+### Authenticity and outcome guardrails
+
+- **Definition:** Count of incidents where fixture/mock is presented as live, an unmatched receipt marks business outcome confirmed, retry creates a duplicate business result, or a missing core evidence/lineage pack is accepted.
+- **Target:** 0 incidents.
+- **Denominator:** All live and fixture/mock runs, delivery attempts, receipts, retries, and acceptance decisions subject to these guardrails; optional empty/unknown projections are reported separately.
+### Counter-metrics
+
+- Report total triggered, admitted, blocked, waiting, completed, and no-action OODA cycles beside the 95% rate; blocker/eligibility classification is fixed at trigger/admission and cannot be shrunk post hoc.
+- Report eligible-run volume and blocked-run volume beside every rate; do not improve rates by excluding blocked work or reducing runs.
+- Report evidence completeness and lineage coverage by required pack element; optional empty/unknown projections are reported separately and cannot inflate completeness.
+- Report live-versus-fixture/mock volumes separately; fixture output cannot substitute for live execution.
+- Report retry count and duplicate-result incidents; suppressing retries or hiding duplicates cannot improve the guardrail metric.
+- Report no-action, waiting_feedback, partial, blocked, failed, and expired cycle volumes separately; no-action cannot be used to inflate the 95% completion rate.
+
+## User Journeys
+
+### UJ-001 — 工作流/运营人员：keyword 到运行
+
+**角色：** 林然，工作流/运营人员。**起点：** 在平台能力上下文输入一个 keyword 并选择实时或已启用的定时运行。**步骤：** 确认 keyword package 与 readiness；启动或等待采集；查看 freshness、状态、失败与 OODA 阶段；在阻塞时修正前置条件或等待恢复。**成功终点：** 采集结果进入可追溯的 OODA cycle，并按规则进入 Act；未满足条件时以明确 blocked/failed 终止而非假成功。
+
+### UJ-002 — 系统：完成 OODA 并派发 geoXI
+
+**角色：** snailfish 系统。**起点：** 接收到实时或定时 keyword collection 的观察结果。**步骤：** 记录 Observe；整理并解释为 Orient；依据业务判断完成 Decide；执行采集调整/记录处理并通过自身交付能力派发 geoXI；接收消费反馈并安排下一轮。**失败/恢复：** 缺少 readiness、证据、lineage、匹配回执或动作失败时保持 fail-closed，记录状态并按策略重试或请求人工处理。**成功终点：** OODA cycle 记录完整阶段与反馈状态，且 geoXI 消费回执与 delivery identity 匹配时该投递业务结果 confirmed。
+
+### UJ-003 — 业务结果复核者：追踪 evidence/lineage/receipt
+**角色：** 许宁，业务结果复核者。**起点：** 打开一次高吉星运行或待复核结果。**步骤：** 回查 immutable keyword package/digest、project/workflow/run/execution/source/binding/worker/runtime/mode/provenance、nonempty raw answer、citation/conversation projections、normalize/accept outcome、record、delivery 和 geoXI receipt。**失败/恢复：** 核心证据/归因缺失或 mismatch 时保持 blocked/failed；可选 citation 为空或 conversation unknown 时仅限制声明并继续复核；仅 transport accepted 时保持 unknown/unconfirmed。**成功终点：** 仅在 matching receipt 证明 geoXI 已持久化且可查询时接受业务结果。
+
+
+### UJ-004 — 平台管理员：处理 readiness/blocked
+**角色：** 周衡，平台管理员。**起点：** 能力或计划显示 blocked/unknown。**步骤：** 检查各 static/dynamic gate 的最新观察，修复或明确声明外部阻塞；重新观察所有必要 gates 并记录 blocker 与恢复原因，产生新的 admission。**成功终点：** 条件恢复后运行可继续，或保留有原因的 blocked/paused 状态；管理员不能用 fixture/mock 掩盖阻塞。
+
+### UJ-005 — 技能维护者：失败到纠正/回滚
+**角色：** 沈岚，技能维护者。**起点：** 技能执行失败并进入待纠正。**步骤：** 查看与 exact skill version 绑定的 failure trace，区分 environment-error 与 skill failure；提交纠正并记录 from/to version、prior snapshot；提出回滚，等待平台管理员审批。**失败/恢复：** 纠正失败或无 known-good 时保持 under-review 并转人工。批准回滚后 detail/后续 run 显示 prior version active，corrected version history=rolled_back，下一 target-scope execution 归因 restored version；仅 mutation 完成不算成功，无法执行或结果 unknown 时为 blocked/unknown/failed。**成功终点：** 纠正/回滚证据完整且生产审批可审计。
+
+## Non-Functional Requirements
+
+### NFR-001 — 可靠性与幂等
+同一 keyword package、delivery identity 和 OODA cycle 的重试不得产生重复业务结果；失败恢复后状态可继续追溯。
+### NFR-002 — 可审计与证据保留
+运行、阶段、证据、lineage、投递、receipt、技能版本、纠正、审批和回滚状态应可回查；每次生产启用/暂停/回滚必须关联 exact version 与 trace 并可审计。
+
+### NFR-003 — 安全与权限
+技能维护者只能查看 trace、发起纠正和提出回滚；平台管理员审批生产启用/回滚并可暂停；工作流/运营人员仅查看/暂停运行且不能指定生产版本。认证信息和 secret 不得暴露。
+
+### NFR-004 — 项目隔离
+keyword package、运行、证据、lineage、交付、receipt 和 geoXI 项目结果不得跨项目混用或互相确认；project identity mismatch 必须 fail closed。
+
+### NFR-005 — 性能与容量基线
+实时延迟、geoXI 消费延迟和可承载运行量先采集真实基线；在基线与范围确认前不宣称目标达成。
+
+### NFR-006 — 可观测与故障状态
+每个阶段和关键失败应显示可行动状态；OODA 终态至少包括 waiting_feedback、partial、blocked、failed、expired、completed；blocked/unknown、failed/unconfirmed、partial 与 confirmed 不得合并，completed 不得由阶段标签单独产生。
+
+### NFR-007 — schedule 准点
+启用且未暂停、无声明外部阻塞的定时运行，至少 99% 应在计划时间 ±1 分钟内启动或触发；阻塞单独报告。
+
+## Dependencies & Risks
+
+- 真实可执行的 Doubao capability、认证健康会话、浏览器容量和网络许可是 live 运行前置。
+- 首轮 baseline 按 project × keyword × trigger × outcome × receipt 分层；稀疏 strata 必须单独报告，不得用总量掩盖。exact per-stratum minimum 由 operational acceptance 在 target-setting 前确定。
+- geoXI 是独立下游产品；其接口必须支持对应项目持久化/查询/分析，并产生包含 target project、delivery identity、keyword package/run、consumption status、consumed timestamp、persisted result reference、failure/rejection reason 和 match result 的最小可观察消费回执。geoXI 负责持久化与回执，snailfish 负责派发、关联验证、保存/展示及更新 business outcome/OODA。
+- OODA 决策责任与动作风险边界必须明确；低风险自动 Act，中风险由 OODA 策略责任人批准，高风险由业务复核者或平台管理员批准。证据、lineage、receipt、mode 或 duplicate-side-effect 风险异常会阻断闭环。
+- 采集来源、目标项目、权限与 OODA 决策责任需保持明确，否则可能产生跨项目污染或无人负责的 Act。
+- 调度准点、实时延迟和消费延迟受外部资源/网络影响；声明的 blocker 必须单独计量，不能美化指标。
+- fixture/mock 与 live 混用、证据缺失、重试重复和 receipt 错配会造成错误业务确认，必须 fail closed。
+
+- **Acceptance boundary:** All scope and metrics statements are acceptance targets, not current implementation/live proof; checked tasks, fixtures, catalog/configuration, and history do not satisfy live, geoXI, or OODA acceptance.
+## Deferred Items
+
+- **GeoXI receipt validity duration and compensation policy:** Owner: geoXI product owner with OODA strategy owner approval. Condition: before geoXI live readiness/integration acceptance; until approved, live readiness remains blocked/unknown.
+- **S1 broader known-good sample count and risk strata:** Owner: skill maintainer + platform administrator (include OODA strategy owner for OODA risk). Condition: before production promotion for the target capability; until decided, the version remains under-review and cannot become known-good.
+- **In-flight pause semantics:** Owner: platform operations. Condition: before production pause/recovery runbook.
+- **Freshness/time-order semantics:** Owner: operational acceptance. Condition: before live baseline target setting.
+## Open Questions
+No phase-blocking open questions; see Deferred Items.
+
+## Product Glossary
+
+- **readiness:** overall unknown/blocked/ready result.
+- **gate:** independent prerequisite observation.
+- **admission:** coherent run-scoped decision that all required gates passed.
+- **live / fixture / mock:** execution modes; fixture/mock is non-live.
+- **transport attempt / accepted:** one delivery try / destination accepted transport, not business confirmation.
+- **receipt pending / consumed / rejected / expired:** geoXI consumption states.
+- **confirmed:** business outcome supported by matching receipt and queryable persisted result.
+- **OODA cycle / completed / no-action:** one system loop / completion evidence satisfied / valid decision to take no action with reason.
+- **evidence/lineage pack:** required artifacts and identity links proving what ran, what it produced, and where it was delivered.
+- **candidate / under-review / known-good / rolled-back:** skill version lifecycle states; only administrator-confirmed known-good is production-trusted.
diff --git a/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/reconcile-gaojixing-normative.md b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/reconcile-gaojixing-normative.md
new file mode 100644
index 00000000..7d404360
--- /dev/null
+++ b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/reconcile-gaojixing-normative.md
@@ -0,0 +1,66 @@
+# Reconciliation — Gaojixing normative/repository inputs
+
+Against: `prd.md` + `addendum.md`
+Date: 2026-08-28
+Mode: extraction only; no PRD/addendum/code changes
+
+## `openspec/changes/gaojixing-live-business-chain/proposal.md`
+
+- **Gap — live acceptance prerequisites:** The proposal requires published executable live capability, authenticated healthy Doubao session, permitted network, reachable destination with documented acknowledgement, and an operator-owned immutable package. The PRD carries these as dependencies/readiness, but the destination acknowledgement contract remains intentionally product-level and unresolved in repository fact.
+- **Gap — non-goal boundary:** The proposal explicitly does not claim current capability publication, live session, public network access, or destination ACK. PRD language should continue to distinguish intended acceptance from current proof.
+- **Semantic preservation:** Immutable package/digest, independent answer/citation/conversation evidence, lineage, fail-closed behavior, fixture separation, and transport/business outcome separation are represented. No implementation fact should be promoted to completed live acceptance.
+
+## `openspec/changes/gaojixing-live-business-chain/tasks.md`
+
+- **Gap — verification remains incomplete:** Tasks 4.1–4.3 (fixture/mock labeling, no implicit fallback, fixture-only coverage) and 5.1–5.7 (readiness, immutable lineage, fixture exclusion, ACK semantics, fail-closed cases, focused/live acceptance) are unchecked. The PRD’s requirements and metrics must not imply those tasks are delivered.
+- **Gap — implementation status versus product commitment:** Tasks 1.1–3.6 are checked, but checked task definitions are contract/implementation ledger evidence, not proof of an observed live run or destination consumption.
+- **Semantic preservation:** Replay without another provider call, canonical dedupe, retry lineage, typed blockers, and no HTTP-202-as-business-success are retained in PRD requirements or boundaries.
+
+## `openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md`
+
+- **Gap — receipt/geoXI detail:** The spec requires a documented destination ACK/equivalent tied to delivery identity, but does not define geoXI-specific receipt fields or consumer event semantics. Current PRD intentionally records the agreed minimum observable receipt at product level without inventing a schema/API.
+- **Gap — evidence verification:** Citation extraction is not citation verification; unavailable conversation identity remains unknown/null. This distinction must remain visible in product language and acceptance review.
+- **Gap — state separation:** Readiness, execution mode/status, transport, business outcome, and OODA completion must not collapse into one success state; fixture/mock output is excluded from live acceptance.
+
+## `_bmad-output/planning-artifacts/bmad-kickoff/current-project-context.md`
+
+- **Gap — operator observations are not repository proof:** Seven running runs, Doubao session/Chrome pool health needs, and ODP `NOGROUP` are operator-reported kickoff observations. They must remain blocked/unknown until run-scoped receipts exist.
+- **Gap — current capability evidence:** Existing runtime/fixture evidence is implementation evidence, not a live-acceptance receipt; no claim of live success should be inferred.
+- **Semantic preservation:** WorkflowProject/backend compiler/III/OpenCLI reuse and source→normalize→accept→sink are architecture context, not a new parallel executor or backend rewrite.
+
+## `_bmad-output/planning-artifacts/bmad-kickoff/external-prerequisites-and-ack-matrix.md`
+
+- **Gap — gate observation:** Capability publication, session identity/health, Chrome capacity, network policy, immutable input, capture evidence, lineage, transport, destination ACK, and ODP group health each require fresh run-scoped observations; the PRD has requirements and baseline plans but no such receipt.
+- **Gap — explicit non-evidence:** Catalog metadata, configured channels/pools, historic login, HTTP 202/enqueue success, URL extraction, fixture output, and aggregate running counts cannot satisfy live acceptance.
+- **Semantic preservation:** Missing or contradictory observations remain precise blocked/unknown/failed/partial/unconfirmed states, never silently successful.
+
+## `_bmad-output/planning-artifacts/bmad-kickoff/live-business-chain-next-steps.md`
+
+- **Gap — execution sequence is not delivery:** The plan prioritizes fixture boundary (4.x), then gate/safety verification (5.1/5.4/5.6), lineage/immutability (5.2/5.3), ACK boundary (5.5), and focused/live acceptance (5.7). The PRD should not represent this dependency order as completed work.
+- **Gap — Stage 0 blockers:** Fresh evidence for the seven running runs, Doubao session, Chrome pool, network policy, and ODP `NOGROUP` precedes live acceptance; these are not replaced by deterministic tests.
+- **Semantic preservation:** Historical runs, local fixtures, accepted transport, or no-error responses cannot be promoted to current business success.
+
+## `docs/workflow-hda-demand-runtime-PRD.md`
+
+- **Gap — existing platform surface versus Gaojixing outcome:** This PRD establishes WorkflowProject as authoring/execution source of truth, backend compilation, III/OpenCLI execution, browser-pool reuse, and projection surfaces; it does not prove Gaojixing live capture or geoXI consumption.
+- **Gap — user-facing product intent:** Canvas/HDA/package nodes, AI structured patches, node-level progress, and evidence/cluster projections are platform affordances. The Gaojixing PRD should preserve the platform-entry intent without claiming a new UI or implementation.
+- **Semantic preservation:** No backend rewrite, no parallel executor, and no raw OpenCLI/III payload authoring by AI remain consistent with current scope.
+
+## `docs/adr/0009-plan-ir-free-graph-two-tier-attribution.md`
+
+- **Gap — attribution is architecture guidance:** Backend-authoritative graph compilation and source-keyed/two-tier attribution guide lineage, but ADR evidence does not establish geoXI receipt or business confirmation.
+- **Gap — product observable boundary:** The PRD must express attributable project/source/run/artifact outcomes without exposing ADR implementation mechanisms as user commitments.
+- **Semantic preservation:** One authored WorkflowProject graph and preserved source/run attribution are consistent with the PRD’s lineage requirements.
+
+## `docs/adr/0021-delivery-separates-submission-from-business-outcome.md`
+
+- **Gap — destination policy remains open:** The ADR establishes that transport/execution result and business outcome are distinct, with destination policy governing callbacks, status queries, retries, compensation, or human confirmation. It does not define geoXI’s concrete receipt contract.
+- **Gap — pending outcome behavior:** Unknown/pending outcomes need not universally block workflows; the product must distinguish safe continuation policy from business confirmation. Current PRD fail-closed rules apply to evidence/lineage/receipt anomalies and do not turn transport acceptance into confirmation.
+- **Semantic preservation:** Agent actions cannot declare business success without outcome evidence; human review remains a valid escalation path.
+
+## Cross-input reconciliation summary
+
+- No source contradicts the platform vision, P1 internal entry focus, S1 scope, system-owned OODA, keyword collection, or geoXI downstream-product decision.
+- The main semantic risk is treating checked tasks, fixture/runtime evidence, catalog/configuration, transport acceptance, or historical/operator observations as live business acceptance.
+- The principal unresolved product contract is the user-observable geoXI consumption receipt and responsibility boundary; current PRD contains the agreed product-level minimum, while repository sources do not define a technical schema/API.
+- The PRD’s current baseline contract (two projects, keyword strata, real-time + scheduled, seven days, at least 100 live OODA cycles, stratified reporting, blockers separately reported) is a later product decision and is not asserted by these normative sources.
diff --git a/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/reconcile-opencli-emacs-article.md b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/reconcile-opencli-emacs-article.md
new file mode 100644
index 00000000..3886a5d6
--- /dev/null
+++ b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/reconcile-opencli-emacs-article.md
@@ -0,0 +1,33 @@
+# Reconciliation — OpenCLI Team article
+
+Input: User-pasted 2026-04-05 OpenCLI Team, 《OpenCLI：人工智能代理的 Emacs》
+Against: `prd.md` + `addendum.md`
+Mode: extraction only; no PRD/addendum/code changes
+
+## Product-spirit check
+
+- The article’s central product spirit is preserved: OpenCLI is a programmable environment for agents, not merely a collection of browser commands; capabilities should be discoverable, composable, reusable, diagnosable, and immediately useful in an agent workflow.
+- The current PRD correctly translates that spirit into the confirmed P1: internal platform entry points, correct Gaojixing context, real readiness, and a path from capability discovery into existing Studio/workflows.
+- The PRD also preserves the user’s confirmed framing that Gaojixing Live Business Chain is the first production-grade use case, while avoiding a marketing-site clone.
+
+## Mechanism and scope check
+
+- **Plugin unification / capability discovery:** Represented at product level through capability categories, context discovery, readiness, and workflow entry. The PRD does not incorrectly promise a public plugin ecosystem.
+- **Operate exploration → init/verify → crystallization:** The learning loop is retained as long-term product direction in `addendum.md`; current S1 covers existing record → distill → execute → correct trust, traces, versioning, rollback, and correction visibility only. Permanent skill crystallization is explicitly out of scope.
+- **Token economics:** The article’s approximately 92% token-reduction figure is correctly treated as an illustrative example, not a KPI, SLA, or current performance claim. No quantitative token target appears in the PRD.
+- **Transparent adapters / diagnostics / self-repair:** Transparency and diagnosability are reflected in readiness, failure traces, evidence, and correction states. Automatic adapter self-repair is explicitly excluded from the current scope; S1 does not claim automatic self-healing.
+- **Composable pipeline primitives:** The idea is preserved as platform horizon/context, while current requirements use product-level source → normalize → accept → sink and OODA behavior. No generic primitive ecosystem is promised this phase.
+- **CLI Hub:** Retained as future platform horizon and not included in P1/P2/P3 delivery commitments.
+- **Dynamic loading / instant feedback:** Dynamic loading is explicitly excluded from current scope. Instant feedback is preserved as observable run, stage, failure, receipt, and OODA feedback states, without claiming the article’s dynamic runtime mechanism.
+
+## Gaps and risks
+
+1. **Product-spirit gap:** The PRD is deliberately use-case-led and does not yet articulate a standalone success measure for “programmable agent environment” breadth or composability. This is an intentional P1 focus decision, not an accidental omission.
+2. **Mechanism-to-requirement gap:** The article does not define the exact product-level acceptance semantics for plugin unification, crystallization, adapter transparency, CLI Hub, or dynamic loading; inventing those would exceed the supplied input and current scope.
+3. **S1 boundary risk:** “Instant feedback” could be misread as automatic self-repair or automatic crystallization. Current PRD wording limits it to visible failure trace, correction state, version/rollback governance, and human approval; this distinction must survive review.
+4. **P2/P3 boundary:** Community plugin discovery/governance and public distribution/content (download, OS installation, release/integrity, public docs, bilingual blogs/vision articles) are explicitly deferred and must not be reintroduced from the article.
+5. **Evidence gap:** The article’s token-economics example and mechanism claims are qualitative/product inspiration, not repository acceptance evidence; no live capability, geoXI receipt, or OODA completion may be inferred from them.
+
+## Reconciliation verdict
+
+No contradiction with the confirmed P1/S1 scope or P2/P3 deferrals. Product spirit is present as platform vision and P1 capability-entry behavior. Article mechanisms remain addendum/platform-horizon context unless separately authorized; none should be treated as delivered this phase.
diff --git a/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/reconcile-opencli-website.md b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/reconcile-opencli-website.md
new file mode 100644
index 00000000..d6504cc7
--- /dev/null
+++ b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/reconcile-opencli-website.md
@@ -0,0 +1,31 @@
+# Reconciliation — opencli-website
+
+Input: `https://github.com/jackwener/opencli-website` (default `master`), as previously source-checked by the user
+Against: `prd.md` + `addendum.md`
+Mode: extraction only; no PRD/addendum/code changes
+
+## Source-fact extraction
+
+The supplied source facts cover a public product entry/terminal demo/platform-command catalog; community plugin list/detail/search/install/contribute; bilingual markdown blog; GitHub Releases download with OS/system requirements/integrity; and a docs entry. These are public-site/distribution surfaces, not evidence of snailfish runtime execution, OODA completion, GEO/Gaojixing behavior, or geoXI consumption.
+
+## P1 reconciliation
+
+- The PRD correctly narrows P1 to usable internal platform feature entry points and correct Gaojixing context, rather than copying the marketing homepage, terminal demo, public command showcase, or public visual/brand experience.
+- Existing snailfish internal capability catalog, skills, Studio/workflow entry, run/evidence/lineage, and geoXI/OODA product context are used as the P1 basis. The website source is an inspiration/reference for discoverability and contextual entry, not a replacement surface or proof of those capabilities.
+
+## P2/P3 reconciliation
+
+- Community plugin list/detail/search/install/contribute and related ecosystem governance are correctly deferred to P2; they are not silently included as current platform commitments.
+- Download, Windows/macOS installation, GitHub Releases, system requirements/integrity, public docs entry, and bilingual blog/vision content are correctly deferred to P3; they are not part of current P1 acceptance.
+
+## Gaps and risks
+
+1. **P1 boundary risk:** A public product entry, terminal demo, or platform-command catalog could be mistaken for the internal operational entry required by P1. The current PRD/addendum explicitly reject that substitution.
+2. **Internal capability gap:** The source does not prove that snailfish’s existing catalog, skills, Studio, evidence/lineage views, or workflow handoffs are complete or consistently connected; those remain internal product work and must not be inferred from website pages.
+3. **Deferred-scope risk:** Plugin installation/contribution/governance and public distribution/content are source features but remain P2/P3; importing their labels, flows, or acceptance criteria would contradict the confirmed scope.
+4. **Implementation/visual overreach:** Website implementation details, page structure, terminal demo behavior, and visual design are not automatically product requirements for snailfish; no copy, visual system, or implementation mechanism is adopted as a requirement.
+5. **GEO/OODA/geoXI evidence gap:** The website source provides no evidence of Gaojixing/GEO-XI live collection, system-owned OODA, immutable keyword lineage, geoXI persistence, or matching consumption receipt. Those claims remain governed by snailfish’s normative contract and real acceptance evidence, not this website input.
+
+## Reconciliation verdict
+
+No contradiction with the confirmed P1 internal-entry focus, S1 scope, system-owned OODA, geoXI downstream boundary, or P2/P3 deferrals. The source contributes discoverability/context inspiration only; it does not add current public-site features, visual requirements, implementation commitments, or GEO/OODA/geoXI proof.
diff --git a/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/review-adversarial-live-boundary.md b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/review-adversarial-live-boundary.md
new file mode 100644
index 00000000..44b0120b
--- /dev/null
+++ b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/review-adversarial-live-boundary.md
@@ -0,0 +1,113 @@
+# PRD Reviewer Gate — adversarial-live-boundary
+
+## Overall verdict
+
+**thin** — The PRD states the intended trust boundary unusually clearly, but several acceptance semantics remain underspecified at the exact points where a production system can manufacture a successful-looking result. The missing controls are not implementation details; they determine whether a displayed state is admissible as live business success.
+
+## Critical findings
+
+### C-1 — Readiness aggregation can certify stale or mixed-time evidence
+- **Anchor:** FR-004–FR-005 (lines 93–97), Current Release Scope readiness (lines 39–41).
+- **Why false success is possible:** The rule says how to aggregate gate states but not evidence freshness, run identity, or whether all gates were observed from one coherent evaluation. A previously healthy session/browser gate can be combined with a current failed network or runtime-binding condition, yielding `ready` for a run that cannot actually execute live.
+- **Recommended disposition:** **autofix** — define freshness/expiry and atomic, run-scoped gate evaluation; stale or mixed evaluations must remain unknown/blocked.
+
+### C-2 — `ready` is not bound to a specific live admission decision
+- **Anchor:** FR-009 (lines 109–116), FR-023 (lines 166–168), FR-027 (lines 178–179).
+- **Why false success is possible:** A capability can be ready when the user opens the page but lose authentication, Chrome capacity, or network permission before admission. The PRD does not require a final gate check tied to the exact collection/run, so the UI can show a run as admitted/live based on an obsolete readiness snapshot.
+- **Recommended disposition:** **autofix** — require an explicit run-scoped admission observation and preserve its identity in the resulting state.
+
+### C-3 — Immutable keyword package is required, but package integrity is not an acceptance condition
+- **Anchor:** FR-011 (lines 122–123), FR-025 (lines 170–174), NFR-001/NFR-004 (lines 276–286).
+- **Why false success is possible:** “不可变” and “一致性标识” do not say what happens when the stored package, digest, expansion, or supplied package disagrees at execution, retry, delivery, or receipt time. A result may be attributed to the displayed package while the actual query set changed.
+- **Recommended disposition:** **autofix** — make package identity/integrity mismatch an explicit blocked/failed condition at every boundary, including retries.
+
+### C-4 — Citation capture and conversation evidence can pass without proving answer provenance
+- **Anchor:** FR-012–FR-013 (lines 125–129), FR-029–FR-030 (lines 185–193).
+- **Why false success is possible:** The PRD correctly distinguishes URL extraction from citation verification, but defines no minimum acceptance rule for verified citation, answer-to-citation linkage, or conversation/session continuity. An answer with unrelated URLs or an unavailable conversation can therefore progress through normalize/accept while still appearing evidence-backed.
+- **Recommended disposition:** **user decision** — decide the minimum evidence completeness required for acceptance versus merely displayable/partial output; then state that rule explicitly.
+
+## High findings
+
+### H-1 — Transport/consumption states lack a terminal timeout and stale-receipt policy
+- **Anchor:** FR-014–FR-016 (lines 131–138), FR-033–FR-035 (lines 200–206).
+- **Why false success is possible:** `accepted`, `consumed`, and `confirmed` are separated, but no deadline or terminal transition is defined for a missing, delayed, superseded, or never-arriving receipt. Operators may treat a permanently pending delivery as operational success, or a late receipt may confirm the wrong historical attempt.
+- **Recommended disposition:** **user decision** — define pending/expired semantics and receipt validity window before calling an outcome confirmed.
+
+### H-2 — Receipt matching does not state anti-replay or authenticity requirements
+- **Anchor:** FR-015 and FR-033 (lines 134–135, 199–200), NFR-002 (lines 279–280).
+- **Why false success is possible:** Matching fields can be copied, duplicated, replayed, or associated with a prior attempt unless uniqueness, provenance, and duplicate/replay handling are acceptance rules. A stale receipt for the same project/package can falsely confirm a new delivery.
+- **Recommended disposition:** **autofix** — require receipt uniqueness and explicit rejection of replay/duplicate/ambiguous receipts as unconfirmed.
+
+### H-3 — Project isolation is a principle, not a demonstrated boundary at every join
+- **Anchor:** NFR-004 (lines 285–286), FR-015/FR-033 (lines 134–135, 199–200), UJ-003 (lines 262–264).
+- **Why false success is possible:** The requirement says records “不得跨项目混用” but does not require project identity to be checked independently at package, run, delivery, receipt, and persisted-result joins. A valid receipt from project A could be accepted for project B if the UI or association layer chooses the wrong context.
+- **Recommended disposition:** **autofix** — make every association mismatch observable and fail closed; include cross-project negative acceptance cases.
+
+### H-4 — Retry/idempotency coverage omits concurrent and ambiguous outcome cases
+- **Anchor:** FR-036 (lines 208–209), NFR-001 (lines 276–277), Authenticity guardrail metric (lines 239–243).
+- **Why false success is possible:** “不得产生重复业务结果” does not specify behavior after timeout-after-send, consumer rejection after transport acceptance, concurrent retries, or unknown receipt status. A retry can create two geoXI records while only one is displayed, or a duplicate can be hidden by dedupe and counted as success.
+- **Recommended disposition:** **autofix** — require observable attempt-level reconciliation for ambiguous and concurrent retries; duplicates must remain incidents and cannot improve success rates.
+
+### H-5 — OODA risk-tier approvals are not tied to a concrete decision instance or policy version
+- **Anchor:** Product Principle 6 (lines 32–34), FR-031–FR-032 (lines 194–198), NFR-003 (lines 282–283).
+- **Why false success is possible:** “Low/medium/high” is named, but there is no required record of who approved which exact action, under which policy, with what evidence, expiry, or scope. A broad prior approval could be reused for a changed keyword range, target project, or skill version.
+- **Recommended disposition:** **user decision** — define approval scope, freshness, actor accountability, and what evidence is frozen for the decision; avoid accepting a generic approval as authorization.
+
+### H-6 — S1 known-good can be promoted after one success despite untested rollback and boundary cases
+- **Anchor:** FR-019–FR-022 (lines 150–160), UJ-005 (lines 270–272).
+- **Why false success is possible:** “At least once” real success plus complete trace does not require success under relevant boundary inputs, receipt confirmation, rollback rehearsal, or absence of latent duplicate effects. A candidate can become known-good after a lucky run and then be used as the trusted fallback.
+- **Recommended disposition:** **user decision** — set the minimum evidence breadth and require rollback effectiveness evidence before known-good promotion.
+
+### H-7 — Fail-closed rules do not prevent a partial result from being presented as a successful collection
+- **Anchor:** FR-016, FR-026–FR-027 (lines 137–138, 175–179), UJ-001/UJ-002 (lines 254–260).
+- **Why false success is possible:** The PRD lists empty, stale, timeout, source-change, evidence, and destination failures, but does not define whether partial keyword/page/session coverage may be accepted. A run can be “completed” for some work and still be interpreted as a complete live collection.
+- **Recommended disposition:** **user decision** — decide partial-coverage semantics and require explicit completeness/coverage state before any collection-success claim.
+
+### H-8 — Counter-metrics can still hide blocked work through the eligibility boundary
+- **Anchor:** Scheduled metric denominator (lines 215–218), OODA metric denominator (lines 227–231), Counter-metrics (lines 245–250).
+- **Why false success is possible:** “Eligible” and “explicitly declared external blocker” are not operationally defined. A system can classify authentication, capacity, receipt, or ODP failures as blockers and remove them from rates while reporting only a small blocked count, creating an inflated 95%/99% appearance.
+- **Recommended disposition:** **autofix** — require immutable admission/eligibility classification, blocker taxonomy, and independent reconciliation of all triggered/enabled runs.
+
+## Medium findings
+
+### M-1 — Repository and checked-task evidence can still leak into product status
+- **Anchor:** Addendum §4 (lines 54–58), reconcile-gaojixing-normative §1–§6 (especially lines 14–17, 27–35), PRD Dependencies & Risks (lines 297–300).
+- **Why false success is possible:** The reconciliations warn that checked tasks, fixtures, catalog metadata, operator observations, and historic runs are not live proof, but the PRD has no explicit user-facing evidence provenance taxonomy. A reviewer or operator can still promote implementation evidence into a live acceptance badge.
+- **Recommended disposition:** **autofix** — require every acceptance claim to label evidence class and prohibit repository/configuration evidence from satisfying live gates.
+
+### M-2 — Live/fixture/mock separation is stated but source-of-truth precedence is undefined
+- **Anchor:** FR-006–FR-007 (lines 99–103), Metrics lines 239–250.
+- **Why false success is possible:** A run may begin live and fall back to fixture/mock after a provider or browser failure, or combine live answer data with fixture evidence. The PRD forbids silent fallback but does not define the final mode when mixed artifacts exist.
+- **Recommended disposition:** **autofix** — define mixed-mode as non-live and non-acceptable for live outcomes; retain per-artifact provenance.
+
+### M-3 — OODA completion can be recorded without proving feedback was causally consumed
+- **Anchor:** FR-028, FR-034–FR-035 (lines 185–206).
+- **Why false success is possible:** A cycle can record all four stage labels and a feedback status while feedback is empty, duplicated, late, or not used to influence the next decision. “Recorded” completion can therefore masquerade as a closed control loop.
+- **Recommended disposition:** **user decision** — define the minimum valid feedback and causal linkage required for `cycle completed`; otherwise use a distinct partial/blocked state.
+
+### M-4 — Scheduled-run identity and missed-run recovery are underspecified
+- **Anchor:** FR-024–FR-026 (lines 169–176), NFR-007 (lines 294–295), scheduled metric (lines 213–218).
+- **Why false success is possible:** A scheduler can trigger twice, skip a run, or execute a late run under a new package while reporting punctuality against the nominal schedule. Without a unique scheduled occurrence and explicit missed/coalesced semantics, both duplicate effects and inflated punctuality are possible.
+- **Recommended disposition:** **autofix** — require occurrence identity and explicit accounting for skipped, coalesced, duplicate, and late triggers.
+
+## Low findings
+
+### L-1 — “Freshness” has no product-level meaning for answers, evidence, or receipts
+- **Anchor:** FR-026 and FR-029 (lines 175–176, 188–189), NFR-006 (lines 291–292).
+- **Why false success is possible:** Without a defined observation time/age boundary, stale but complete evidence can be displayed as current and support a new OODA decision.
+- **Recommended disposition:** **defer with owner/condition** — define freshness during operational acceptance before live claims.
+
+### L-2 — “可查询” persistence is asserted without specifying query visibility/consistency condition
+- **Anchor:** FR-033–FR-034 (lines 199–203), UJ-003 (lines 262–264).
+- **Why false success is possible:** A consumer receipt may arrive before the result is actually visible to the intended project users or query path; the UI can mark confirmed while the business record is not usable.
+- **Recommended disposition:** **user decision** — define what observable queryability is sufficient for confirmation.
+
+### L-3 — Pause semantics do not cover in-flight delivery or OODA actions
+- **Anchor:** Users & Roles lines 47–49, 71–73; NFR-003 (lines 282–283); UJ-004 (lines 266–268).
+- **Why false success is possible:** “Pause” may stop new runs while an already admitted Act continues across a risk or credential change, producing an unreviewed side effect that is later counted as successful.
+- **Recommended disposition:** **defer with owner/condition** — clarify pause scope and status treatment for in-flight actions during operational acceptance.
+
+### L-4 — Acceptance does not require clock/time-order integrity across run, delivery, and receipt
+- **Anchor:** FR-011, FR-015, FR-033–FR-035 (lines 122–135, 199–206).
+- **Why false success is possible:** A receipt with inconsistent timestamps or an event arriving out of order can be matched and shown as current despite belonging to an earlier run or future/replayed event.
+- **Recommended disposition:** **defer with owner/condition** — make event ordering and timestamp anomalies explicit blocked/unconfirmed cases.
diff --git a/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/review-findings-triage.md b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/review-findings-triage.md
new file mode 100644
index 00000000..ed7eb4ca
--- /dev/null
+++ b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/review-findings-triage.md
@@ -0,0 +1,141 @@
+# Review Findings Triage — Gaojixing Live Business Chain
+
+Inputs: `review-rubric.md`, `review-adversarial-live-boundary.md`, `review-operational-acceptance.md`, `prd.md`, and reconciliation files. No PRD/addendum/code changes; no disposition decisions recorded in memlog.
+
+## Critical
+
+None.
+
+## High
+
+### T-H1 — OODA completion and feedback semantics are not repeatable
+- **Problem:** FRs and the 95% metric require recorded Observe/Orient/Decide/Act plus feedback, but did not define minimum completion evidence, valid no-feedback/late-feedback terminal states, or causal feedback use.
+- **Consequence:** Operators could classify the same cycle differently; a stage-label record could look like a closed loop without feedback being consumed.
+- **Anchors:** `prd.md` FR-028, FR-031–FR-032, FR-034–FR-036; Success Metrics “Complete system-owned OODA cycle success” and Counter-metrics; NFR-006; UJ-002.
+- **Sources:** rubric-walker; operational-acceptance; adversarial-live-boundary.
+- **Disposition:** **resolved** by user decision and PRD update: completion now requires cycle + immutable package + minimum stage records; required feedback must be matching, consumed, and record next-round/close impact; valid no-action records reason and feedback-not-required; terminal states include waiting_feedback, partial, blocked, failed, expired, completed. Missing/late/duplicate/mismatched feedback cannot complete.
+
+### T-H2 — Evidence and lineage minimum is not one acceptance boundary
+- **Problem:** “完整 lineage,” “required evidence,” and citation/conversation status were used as gates without one consistent minimum across answer, normalized record, delivery, receipt, and OODA.
+- **Consequence:** Incomplete or unrelated evidence could progress as live/accepted, or reviewers could disagree on whether a run was safe to deliver.
+- **Anchors:** `prd.md` Product Principles 5; FR-011–FR-016, FR-025, FR-029–FR-030, FR-033; NFR-002/NFR-004; UJ-003; Authenticity and outcome guardrails.
+- **Sources:** rubric-walker; adversarial-live-boundary; operational-acceptance.
+- **Disposition:** **resolved** by user decision and PRD update: minimum pack, layered acceptance, extraction/verification separation, same package/run/project fail-closed, and no receipt without consumed/confirmed.
+
+### T-H3 — Readiness freshness, run admission, and recovery are underspecified
+- **Problem:** Overall aggregation was defined, but gate evidence freshness/coherence, a final run-scoped admission check, and recovery criteria from unknown/blocked to ready were not.
+- **Consequence:** A stale or mixed-time healthy snapshot could admit an unexecutable live run, while blocked operators could retry unsafely or remain blocked indefinitely.
+- **Anchors:** `prd.md` Product Principles 6; Current Release Scope readiness; FR-002, FR-004–FR-006, FR-009, FR-023–FR-027; UJ-001/UJ-004; NFR-006; Dependencies & Risks; Success Metrics/Counter-metrics.
+- **Sources:** adversarial-live-boundary; operational-acceptance.
+- **Disposition:** **resolved** by user decision and PRD update: static/dynamic freshness, coherent run-scoped all-gate admission, post-admission invalidation to blocked/paused with retained partial evidence, no fallback/old-ready reuse, and full recovery reevaluation for a new admission.
+
+### T-H4 — GeoXI receipt, persistence/queryability, timeout, replay, and project isolation need one boundary
+- **Problem:** Receipt fields and matching were stated, but expiry/late receipt handling, replay/duplicate receipt rejection, proof of persisted/queryable result, and checks at every project/package/run/delivery join were not unified.
+- **Consequence:** A stale or replayed receipt, wrong-project result, or receipt-before-queryable persistence could mark a delivery confirmed incorrectly; permanently pending delivery could look successful.
+- **Anchors:** `prd.md` FR-014–FR-016, FR-033–FR-036, NFR-001/NFR-002/NFR-004/NFR-006, UJ-003, Dependencies & Risks, Success Metrics/Counter-metrics; `addendum.md` §4.
+- **Sources:** adversarial-live-boundary; operational-acceptance; rubric-walker.
+- **Disposition:** **resolved by safe derivation** and explicit deferment: identity matching, attempt visibility, ambiguous outcomes unconfirmed, duplicate/replay rejection, project/package/run/delivery mismatch fail closed, and consumed requires queryable persisted result. Exact receipt validity duration and compensation are **deferred** to `geoXI product owner + OODA strategy owner`, condition: before geoXI live readiness/integration acceptance; without approved policy readiness remains blocked/unknown.
+
+### T-H5 — S1 known-good evidence breadth is insufficiently operationalized
+- **Problem:** FR-022 required one target-scope real success and safety conditions but did not define evidence breadth, rollback effectiveness, or repeatable review criteria.
+- **Consequence:** A lucky success could promote a candidate to trusted fallback, or maintainer/admin could disagree on production approval.
+- **Anchors:** `prd.md` FR-017–FR-022, UJ-005, NFR-002/003, Deferred Items; `review-adversarial-live-boundary.md` H-6; `review-operational-acceptance.md` medium S1.
+- **Sources:** adversarial-live-boundary; operational-acceptance.
+- **Disposition:** **resolved by safe derivation** and explicit deferment: exact execution sample count/risk strata are owned by skill maintainer + platform administrator (OODA strategy owner for OODA risk) before production promotion; until decided, version remains under-review and cannot become known-good. Safe product invariants now bind exact version/trace, separate environment-error, require real target-scope passing execution and complete linked evidence, preserve version states, and require rollback effectiveness evidence.
+
+## Medium
+
+### T-M1 — Open Questions says none while review-critical decisions remain
+- **Problem:** `prd.md` Open Questions states `None at draft completion`, while the review identifies unresolved acceptance semantics.
+- **Consequence:** Downstream teams may treat unresolved decisions as closed and finalize prematurely.
+- **Anchors:** `prd.md` Open Questions; `reconcile-gaojixing-normative.md` cross-input summary; all three review files’ high findings.
+- **Sources:** rubric-walker; operational-acceptance; reconciliation.
+- **Disposition:** user-decision — reopen only the genuinely phase-blocking items; do not silently close them.
+
+### T-M2 — Retry/idempotency and ambiguous concurrent outcomes lack acceptance coverage
+- **Problem:** “no duplicate business result” is not bounded for timeout-after-send, concurrent retries, consumer rejection after acceptance, or unknown receipt.
+- **Consequence:** Two geoXI effects may occur while one is displayed, or dedupe may hide a duplicate and inflate success.
+- **Anchors:** `prd.md` FR-036, NFR-001, Authenticity guardrail metric; `review-adversarial-live-boundary.md` H-4.
+- **Sources:** adversarial-live-boundary.
+- **Disposition:** safe-autofix — add conservative observable reconciliation/fail-closed language only, without implementation prescription.
+
+### T-M3 — OODA risk approvals are not tied to a specific decision instance
+- **Problem:** Risk tiers and approver roles exist, but exact action scope, policy freshness, actor accountability, and frozen evidence are not stated.
+- **Consequence:** A broad or stale approval could be reused for changed keywords, target project, or production skill version.
+- **Anchors:** `prd.md` Product Principles 6, FR-031–FR-032, NFR-003; `review-adversarial-live-boundary.md` H-5.
+- **Sources:** adversarial-live-boundary.
+- **Disposition:** user-decision — define product-level approval scope and accountability.
+
+### T-M4 — Partial coverage and mixed-mode outcomes are not fully classified
+- **Problem:** Empty/stale/timeout/source-change cases are listed, but partial keyword/page/session coverage and live+fixture mixed artifacts lack a single final outcome rule.
+- **Consequence:** A partially collected or mixed-mode run may appear to be complete live success.
+- **Anchors:** `prd.md` FR-006–FR-007, FR-016, FR-026–FR-027; `review-adversarial-live-boundary.md` H-7/M-2.
+- **Sources:** adversarial-live-boundary.
+- **Disposition:** safe-autofix — state that partial or mixed-mode outcomes cannot satisfy live success and remain explicitly non-confirmed.
+
+### T-M5 — Metric eligibility and blocker accounting can still be gamed
+- **Problem:** Eligible runs and declared external blockers are not operationally bounded; blocked authentication/capacity/receipt work could be removed from denominators.
+- **Consequence:** 95%/99% rates can improve by excluding difficult runs or reducing triggered volume.
+- **Anchors:** `prd.md` Success Metrics denominators/counter-metrics; NFR-007; `review-adversarial-live-boundary.md` H-8; `review-operational-acceptance.md` medium schedule edge.
+- **Sources:** adversarial-live-boundary; operational-acceptance.
+- **Disposition:** safe-autofix — require independent accounting of all enabled/triggered runs and immutable eligibility/blocker classification, in product terms.
+
+### T-M6 — Scheduled occurrence identity and missed/coalesced handling are underspecified
+- **Problem:** ±1 minute punctuality does not state how duplicate, skipped, coalesced, or late occurrences are counted.
+- **Consequence:** Duplicate triggers or missed runs can be hidden while punctuality remains inflated.
+- **Anchors:** `prd.md` FR-024–FR-026, NFR-007, schedule metric; `review-adversarial-live-boundary.md` M-4.
+- **Sources:** adversarial-live-boundary.
+- **Disposition:** safe-autofix — clarify product accounting for each scheduled occurrence and late/duplicate/missed outcomes.
+
+### T-M7 — Current proof versus acceptance target needs explicit labeling
+- **Problem:** Baseline contracts and targets can be read as delivered capability despite reconciliations warning that checked tasks, fixtures, catalog state, and operator observations are not live proof.
+- **Consequence:** Reviewers or operators may issue a live acceptance claim without a live receipt.
+- **Anchors:** `prd.md` Current Release Scope, Success Metrics, Dependencies; `reconcile-gaojixing-normative.md` §§1–6; `review-adversarial-live-boundary.md` M-1; `review-rubric.md` Scope honesty medium.
+- **Sources:** adversarial-live-boundary; rubric-walker; reconciliation.
+- **Disposition:** safe-autofix — add product-level “acceptance target, not current proof” labeling.
+
+### T-M8 — Glossary/terminology consistency is missing
+- **Problem:** readiness, accepted, consumed, confirmed, completed, evidence, lineage, known-good, and mode have no canonical product glossary.
+- **Consequence:** UX, operations, and story teams may interpret status labels differently.
+- **Anchors:** `prd.md` FR-002, FR-006, FR-014–FR-016, FR-033–FR-036, NFR-006; `review-rubric.md` Downstream usability medium; `review-operational-acceptance.md` low terminology.
+- **Sources:** rubric-walker; operational-acceptance.
+- **Disposition:** safe-autofix — add a concise product glossary during polish; no technical contract.
+
+## Low
+
+### T-L1 — Pause scope for in-flight actions is unclear
+- **Problem:** Pause permissions are stated, but in-flight delivery/OODA behavior is not.
+- **Consequence:** An already admitted action may continue after a safety or credential change.
+- **Anchors:** `prd.md` Users & Roles, NFR-003, UJ-004; `review-adversarial-live-boundary.md` L-3.
+- **Sources:** adversarial-live-boundary.
+- **Disposition:** defer(owner=platform operations; condition=before production pause/recovery runbook).
+
+### T-L2 — Freshness and event time-order semantics are not explicit
+- **Problem:** Freshness and timestamp ordering are used but not bounded as product states.
+- **Consequence:** Stale or out-of-order evidence may support a current decision.
+- **Anchors:** `prd.md` FR-026, FR-029, FR-033–FR-035; `review-adversarial-live-boundary.md` L-1/L-4.
+- **Sources:** adversarial-live-boundary.
+- **Disposition:** defer(owner=operational acceptance; condition=before live baseline target setting).
+
+## Resolution order
+
+2. No high findings remain unresolved: T-H1–T-H5 are resolved or explicitly deferred. No polish/finalize while any critical finding exists or any high finding lacks disposition.
+## Medium/low batch resolution register
+
+- **T-M1 — resolved:** `prd.md` Open Questions now states “No phase-blocking open questions; see Deferred Items.”
+- **T-M2 — resolved:** `prd.md` FR-016, FR-034–FR-036; per-attempt visibility, same business identity, reconciliation before confirmation, duplicate guardrail.
+- **T-M3 — resolved:** `prd.md` FR-031 and NFR-003; approval binds exact cycle/action/package/project/risk/policy/evidence/actor/time and invalidates on change.
+- **T-M4 — resolved:** `prd.md` FR-006, FR-007, FR-016, FR-026–FR-027; partial/mixed is explicit non-live/non-confirmed with provenance.
+- **T-M5 — resolved:** `prd.md` Success Metrics scheduled denominator/counter-metrics; all planned occurrences remain counted and OODA totals are reported.
+- **T-M6 — resolved:** `prd.md` FR-024 and schedule metric; occurrence identity and duplicate/skipped/coalesced/late outcomes are visible.
+- **T-M7 — resolved:** `prd.md` Current Release Scope acceptance boundary and Dependencies acceptance boundary; targets are not current proof.
+- **T-M8 — resolved:** `prd.md` Product Glossary.
+- **T-L1 — deferred:** owner=platform operations; condition=before production pause/recovery runbook.
+
+## Final resolution summary
+
+- Critical findings: 0.
+- T-H1–T-H5: resolved or explicitly deferred with owner and condition.
+- T-M1–T-M8: resolved; T-L1/T-L2 remain deferred with owner and condition.
+- No finding remains without a disposition; polish/finalization may proceed subject to deferred-item conditions.
+- **T-L2 — deferred:** owner=operational acceptance; condition=before live baseline target setting.
diff --git a/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/review-operational-acceptance.md b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/review-operational-acceptance.md
new file mode 100644
index 00000000..4f155dfb
--- /dev/null
+++ b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/review-operational-acceptance.md
@@ -0,0 +1,39 @@
+# Operational Acceptance Review — Gaojixing Live Business Chain
+
+## Overall verdict
+
+**Adequate.** The draft gives operators concrete journeys, role boundaries, schedule and OODA targets, a representative live baseline contract, and clear transport/consumption distinctions. Acceptance is still blocked at the product-definition level by missing observable minimums for OODA completion, readiness reasons, and the boundary between a geoXI receipt being present and a result being queryable; these should be resolved before downstream story/architecture handoff.
+
+## Critical findings
+
+- None.
+
+## High findings
+
+- **[high] OODA completion lacks a repeatable acceptance test** — **Anchor:** `prd.md` FR-028, FR-034, FR-035 and Success Metrics “Complete system-owned OODA cycle success”. **Operational consequence:** Two operators could classify the same cycle differently when feedback is delayed, absent, rejected, or not applicable; the 95% metric would not be reproducible. **Disposition:** User decision: define the minimum observable evidence for each completed cycle and the permitted terminal states for missing/late feedback, without prescribing implementation.
+
+- **[high] GeoXI consumption acceptance still has an ambiguity at the handoff boundary** — **Anchor:** `prd.md` FR-033–FR-035; `Dependencies & Risks`; `addendum.md` §4. **Operational consequence:** A receipt may be present but not prove that the result is persisted and queryable, or may identify a project/result without a sufficiently clear match; operators cannot consistently set `consumed`/`confirmed`. **Disposition:** User decision: confirm the minimum user-observable proof that “persisted and queryable” is satisfied and who owns exception handling; do not invent a technical contract.
+
+- **[high] Readiness blocker recovery is not operationally bounded** — **Anchor:** `prd.md` FR-004, FR-005, FR-009 and UJ-004. **Operational consequence:** Administrators know a run is blocked but may not know what evidence is sufficient to move a gate from unknown/blocked to ready, causing unsafe retries or indefinite stoppage. **Disposition:** User decision or defer with named owner: define product-facing gate evidence and recovery criteria for the eight independent gates.
+
+## Medium findings
+
+- **[medium] Schedule punctuality denominator has an unresolved timing edge** — **Anchor:** Success Metrics “Scheduled keyword collection punctuality” and NFR-007. **Operational consequence:** Runs paused or externally blocked near the planned time may be classified inconsistently, affecting the 99% rate and counter-metrics. **Disposition:** Safe clarification/autofix: state that eligibility is evaluated at planned trigger time and that a declared blocker must be recorded before/at that time; otherwise defer to operations owner.
+
+- **[medium] Baseline coverage is defined but sampling balance is not** — **Anchor:** Success Metrics “Real-time collection latency” baseline sample contract. **Operational consequence:** The minimum 100 cycles could be dominated by one project, trigger type, or normal outcome while still satisfying the count, producing a misleading target-setting baseline. **Disposition:** Defer with acceptance owner: set minimum per-stratum coverage or explicitly report sparse strata before target setting; do not add an arbitrary distribution now.
+
+- **[medium] S1 known-good evidence is named but not operationally reviewable** — **Anchor:** FR-022 and UJ-005. **Operational consequence:** Maintainers and administrators may disagree on “target ability verification”, “target-scope success”, or “serious unresolved failure”, delaying production approval or enabling an unsafe rollback. **Disposition:** User decision: define product-level review checklist/owner for each criterion, preserving dual-role approval.
+
+## Low findings
+
+- **[low] Role journey coverage does not give geoXI responsibility party a named operational journey** — **Anchor:** Users & Roles “geoXI 下游产品责任方” and UJ-002/UJ-003. **Operational consequence:** Receipt rejection or delayed consumption may lack a clearly rehearsed cross-product escalation path. **Disposition:** Defer with owner: add a downstream-incident journey only if cross-product operations require it; current scope can proceed with UJ-003.
+
+- **[low] Baseline and acceptance terminology mixes English labels without a compact product glossary** — **Anchor:** FR-002, FR-006, FR-014–FR-016, FR-033–FR-036 and NFR-006. **Operational consequence:** Different operators may interpret “accepted”, “consumed”, “confirmed”, “blocked”, and “completed” inconsistently. **Disposition:** Safe autofix or defer to polish: add a short product glossary, not technical schema/API definitions.
+
+## Mechanical and scope checks
+
+- FR-001–FR-036 and NFR-001–NFR-007 are contiguous and unique.
+- The five UJs cover workflow/operations, system OODA and geoXI dispatch, business review, readiness administration, and S1 correction/rollback; protagonists are named where a human role is involved.
+- P1 internal entry, S1 governance, system-owned OODA, and geoXI downstream consumption remain in scope; P2 community plugin governance and P3 public distribution/content remain deferred.
+- The seven-day/two-project/keyword-strata/100-live-cycle baseline is an acceptance sampling contract, not proof that live prerequisites or geoXI integration currently exist.
+- Recommendations are product/acceptance decisions only; no API, schema, database, or implementation design is proposed.
diff --git a/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/review-plan.md b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/review-plan.md
new file mode 100644
index 00000000..1c21a716
--- /dev/null
+++ b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/review-plan.md
@@ -0,0 +1,92 @@
+# Reviewer Gate Plan — Gaojixing Live Business Chain
+
+Status: plan written; execution pending
+Model/executor: OMP 5.6 Luna via Orca
+Workspace: `_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/`
+
+## Review inputs
+
+Reviewers read and source-extract:
+
+- `prd.md`
+- `addendum.md`
+- `reconcile-gaojixing-normative.md`
+- `reconcile-opencli-emacs-article.md`
+- `reconcile-opencli-website.md`
+- `.memlog.md` — decision consistency only; not a product-requirement source and not a substitute for PRD/addendum evidence
+
+No reviewer may modify `prd.md`, `addendum.md`, source inputs, or code.
+
+## Reviewers
+
+### `rubric-walker`
+
+**Goal:** Judge the PRD against the complete seven-dimension quality rubric at high rigor for an internal production system.
+
+**Checks:** decision-readiness and honest trade-offs; substance over persona/innovation/NFR/vision theater; strategic coherence of P1/S1/OODA/geoXI scope and metrics; testable done-ness of FRs/NFRs/UJs/metrics; scope honesty and deferred P2/P3 boundaries; downstream usability, glossary and cross-reference consistency; shape fit for a brownfield internal platform; mechanical ID continuity, assumptions, named protagonists, and required sections.
+
+**Output:** `review-rubric.md`
+
+### `adversarial-live-boundary`
+
+**Goal:** Attack the production trust boundary and find plausible paths to false success.
+
+**Checks:** readiness gate aggregation and unknown/blocked/ready semantics; live versus fixture/mock separation; execution status versus business outcome; immutable keyword package and lineage; evidence completeness and citation/conversation provenance; transport accepted versus geoXI consumed versus OODA cycle completed; matching receipt and project isolation; retry/idempotency and duplicate side effects; OODA risk-tier approvals; S1 candidate/under-review/known-good and rollback controls; fail-closed behavior and counter-metrics; whether repository evidence is incorrectly treated as live proof.
+
+**Output:** `review-adversarial-live-boundary.md`
+
+### `operational-acceptance`
+
+**Goal:** Determine whether the PRD can guide observable acceptance across roles and real operations without inventing implementation contracts.
+
+**Checks:** real-time and scheduled keyword journeys; schedule ±1 minute/99% definition, denominator, blocker treatment; real-time and geoXI latency baseline rules; seven-day/two-project/keyword-strata/100-live-cycle baseline sample; OODA 95% denominator and completion semantics; geoXI downstream responsibility and minimum receipt observability; role permissions and risk-tier intervention; evidence/lineage/receipt visibility; S1 trace/correction/version/approval journey; NFR measurability and counter-metrics; remaining phase-blocking ambiguity.
+
+**Output:** `review-operational-acceptance.md`
+
+## Output contract
+
+Each reviewer writes only to its assigned `review-<slug>.md` and returns a compact summary. The file format MUST include:
+
+- Overall verdict (`strong`, `adequate`, `thin`, or `broken`, as applicable)
+- Findings grouped by severity: `critical`, `high`, `medium`, `low`
+- Specific evidence anchors (PRD section/FR/NFR/UJ/metric and, where relevant, reconcile or addendum path)
+- Recommended disposition for every finding: autofix, user decision, defer with owner/condition, or ignore with rationale
+
+Findings must be about product quality and acceptance; implementation code/API/schema suggestions are out of scope.
+
+## Execution orchestration
+
+Because Orca worker count is constrained:
+
+1. **Batch 1, parallel:** `rubric-walker` and `adversarial-live-boundary`.
+2. **Batch 2, after Batch 1 settles:** `operational-acceptance`.
+
+Each reviewer receives an isolated context containing the review inputs and this output contract. They do not coordinate through mutable PRD files, do not edit PRD/addendum/code, and do not run project-wide validation.
+
+## Acceptance gate
+
+Before finding resolution:
+
+- Confirm all three review files exist at the exact output paths.
+- Every critical/high finding has a concrete PRD anchor and a repair recommendation.
+- Reject or reframe any recommendation that pollutes the review with implementation code/API/schema design.
+- Collect all three compact summaries before opening finding resolution.
+
+## Finding resolution plan
+
+1. Deduplicate overlapping findings while preserving the strongest evidence anchor.
+2. Resolve critical/high findings one at a time: obtain a user decision or apply only a safe, scope-preserving autofix.
+3. Record medium/low findings with disposition, owner, and revisit condition where applicable.
+4. Append every accepted change, decision, deferral, or override to `.memlog.md`.
+5. An unresolved critical finding blocks polish and finalization; high findings require explicit disposition before polish.
+
+## Subsequent Finalize sequence
+
+After all findings are resolved or explicitly dispositioned:
+
+1. Open-item triage.
+2. Apply `skill:bmad-review lenses=structure,prose` in the declared document-standards order.
+3. Set `prd.md` status to `final` and update the final date.
+4. Append the finalized event to `.memlog.md`.
+
+This file records planning only. Reviewer execution is pending and is not performed by this plan-writing step.
diff --git a/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/review-rubric.md b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/review-rubric.md
new file mode 100644
index 00000000..fc38169b
--- /dev/null
+++ b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/review-rubric.md
@@ -0,0 +1,64 @@
+# PRD Quality Review — Gaojixing Live Business Chain
+
+## Overall verdict
+
+**Fair.** The draft has a coherent platform-entry thesis, explicit P1/S1/OODA/geoXI boundaries, strong state-separation principles, and unusually clear anti-false-success guardrails. It is not yet green-light ready for an internal production build because several acceptance-critical product semantics remain qualitative (especially OODA completion, evidence completeness, and baseline governance), while the draft’s Open Questions section says `None at draft completion` despite unresolved questions documented in the reconciliation material.
+
+## Decision-readiness — adequate
+
+The scope and major trade-offs are explicit: internal P1 over marketing/public distribution, S1 governance over automatic self-healing, and system-owned OODA with geoXI as downstream consumer. Roles and risk-tier approvals are stated. However, the remaining hard decisions are not consistently surfaced in the PRD itself; the reconciliation files identify unresolved baseline/receipt evidence boundaries that should not be treated as settled merely because a product-level choice exists.
+
+### Findings
+- **[high] Open-item state is overstated (§ Open Questions, line 308) —** `None at draft completion` conflicts with unresolved acceptance questions visible in `reconcile-gaojixing-normative.md` and the earlier discovery decisions, including what constitutes sufficient live evidence and how baseline observations become targets. *Fix:* user decision: either resolve and record those items before review closure, or list them as explicit open items with owner and revisit condition.
+
+## Substance over theater — strong
+
+The PRD avoids public-site theater and keeps the article mechanisms in the addendum as platform horizon. Personas are tied to decisions and operations, and metrics include counter-metrics rather than activity-only measures. The 92% token example is correctly excluded from KPI claims.
+
+### Findings
+- No substantive finding.
+
+## Strategic coherence — adequate
+
+The thesis is consistent: make existing agent capabilities discoverable and trustworthy, then prove the platform with a Gaojixing live chain and system-owned OODA. P1, S1, geoXI, and P2/P3 boundaries align. The breadth of seven metrics and a 100-cycle baseline may still compete with the stated P1 optimization focus unless the baseline is explicitly framed as acceptance evidence rather than an additional product surface.
+
+### Findings
+- **[medium] Baseline purpose is ambiguous (§ Success Metrics, Real-time collection latency and GeoXI consumption latency) —** the draft specifies a seven-day, two-project, 100-cycle baseline but does not state which decisions the baseline is authorized to support beyond later target-setting. *Fix:* disposition as user decision: name the target-setting and go/no-go decisions the baseline may inform, without inventing latency targets.
+
+## Done-ness clarity — thin
+
+FRs generally state observable outcomes and preserve important failure distinctions. Several requirements rely on unbounded terms such as “完整 lineage,” “可行动原因,” “质量/有效性状态,” “required feedback,” and “目标范围真实成功,” making consistent acceptance difficult. NFRs improve the bounds for schedule punctuality and OODA success, but evidence completeness and cycle completion still lack an observable minimum.
+
+### Findings
+- **[high] OODA completion is under-specified (FR-028, FR-034, FR-035; Success Metrics OODA definition) —** a cycle must record stages and “required feedback status,” but the draft does not define the minimum observable condition for a completed cycle or the distinction between a valid no-feedback terminal state and an incomplete cycle. *Fix:* user decision: define the product-level completion evidence and terminal states before story decomposition.
+- **[high] Evidence/lineage completeness is unbounded (FR-013, FR-016, FR-025, FR-036; NFR-002) —** “complete lineage,” “key evidence,” and “required evidence” are used as acceptance gates without a single product-level minimum list across live answer, normalized record, delivery, and geoXI receipt. *Fix:* safe autofix only if limited to the already agreed observable receipt/evidence lists; otherwise user decision, not an implementation proposal.
+- **[medium] Readiness and failure observability remain qualitative (FR-004/005/009/026) —** “对应原因,” “可纠正,” and “可行动” do not establish the minimum information a reviewer needs to accept a gate or recover a run. *Fix:* user decision or defer with an owner to define the product-facing reason vocabulary; do not prescribe API/schema behavior.
+
+## Scope honesty — adequate
+
+Out-of-scope P2/P3 items and article mechanisms are explicit. Live acceptance is repeatedly separated from fixture evidence and transport acceptance. The draft does not promise automatic crystallization or dynamic loading. The main concern is not hidden scope, but the possibility that the extensive baseline and OODA requirements are read as already available rather than acceptance commitments.
+
+### Findings
+- **[medium] Baseline and live acceptance status may be mistaken for delivered capability (§ Current Release Scope, Success Metrics, Dependencies) —** the draft states the required baseline contract and 95%/99% targets but does not consistently label them as future acceptance evidence rather than current repository capability. *Fix:* safe autofix: add an explicit “target/acceptance requirement, not current proof” qualifier to the metric and dependency framing.
+
+## Downstream usability — adequate
+
+The FR-001–FR-036 and NFR-001–NFR-007 identifiers are contiguous, UJs have named protagonists, and the document is extractable by functional group. Reconciliations provide useful provenance. A glossary is absent from this draft, so domain terms such as readiness, OODA cycle, consumed, confirmed, known-good, and evidence may drift during UX/architecture/story work.
+
+### Findings
+- **[medium] No PRD glossary (§ entire draft) —** the document uses readiness, mode, execution status, transport accepted, consumed, confirmed, evidence, lineage, OODA cycle, candidate, under-review, and known-good with high consequence but no canonical definitions in the PRD. *Fix:* safe autofix or defer: add a concise product glossary before downstream story extraction; do not add technical schemas.
+
+## Shape fit — strong
+
+For a brownfield internal production platform, the combination of capability requirements, named multi-role journeys, operational metrics, NFRs, dependencies, and explicit boundaries is appropriate. The draft is more rigorous than a single-operator capability spec, but the multiple roles, downstream geoXI, and OODA governance justify that shape.
+
+### Findings
+- No substantive finding.
+
+## Mechanical notes
+
+- ID continuity: FR-001–FR-036 and NFR-001–NFR-007 are contiguous and unique.
+- UJ protagonist naming: UJ-001, UJ-003, UJ-004, and UJ-005 have named human protagonists; UJ-002 intentionally names the system as the actor and is not floating.
+- Scope labels: P2/P3 and article mechanisms are explicitly deferred; S1 is explicitly bounded.
+- Assumptions: the draft contains no inline `[ASSUMPTION]` tags even though earlier memlog entries recorded corrigible assumptions; confirm whether the project requires an inline assumptions index before finalization.
+- Open Questions: currently states `None at draft completion`; this needs reconciliation with the high/medium findings above before finalization.
diff --git a/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/s1-workflow-audit.md b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/s1-workflow-audit.md
new file mode 100644
index 00000000..2523008c
--- /dev/null
+++ b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/s1-workflow-audit.md
@@ -0,0 +1,76 @@
+# S1 Skill Workflow Contract Audit
+
+只读审计；未修改 PRD/code，未运行测试。范围聚焦现有 skill API/UI/correction/distillation/evidence 合同及相关 OpenSpec/tests/docs。
+
+## 1. 实际 lifecycle map
+
+```text
+record (Skill row/elements/source trace)
+  -> distill (initial skill fields)
+  -> execute (skill-channel self-eval appends executed evidence)
+  -> failure trace (trace_id/loop outcome/passed)
+  -> 3 eligible consecutive failures
+  -> correction_proposed (flag only; never auto-redistill)
+  -> human redistill from most recent supplied trace
+  -> corrected, version n -> n+1 (body replaced; prior body stashed)
+  -> execute new version
+  -> rollback latest corrected version
+  -> rolled_back, restore prior body/version
+```
+
+Correction proposal is deliberately not execution: `maybe_propose_correction` appends a marker; dock/endpoint triggers `re_distill`. Rollback is only for the latest corrected entry and rejects double rollback.
+
+## 2. Current observable evidence, roles, actions
+
+| Stage | Observable evidence / status | Role/action | Anchors | Contract status |
+|---|---|---|---|---|
+| Record/list/detail | Skill brief/detail includes version, model/source fields, evidence log; UI has `/skills` and `/skills/:id`. | Users view skills. | `backend/api/v1/skills.py:38-117`; `frontend/app/(app)/skills/page.tsx`; `frontend/app/(app)/skills/[id]/page.tsx` | Implemented read surface |
+| Distill | `re_distill` accepts one/list trace, uses latest list entry, resolves provider, calls distill kernel, replaces body/elements/model/source trace, increments version. | Human correction trigger; endpoint/dock. | `backend/skills/correction.py:67-156`; `backend/api/v1/skills.py:154-201` | Implemented |
+| Execute | Skill channel self-eval records `executed`, `passed`, trace and loop outcome; failure streak ignores `loop_outcome=error`. | Runtime executes; system appends evidence. | `backend/skills/correction.py:244-306`; `backend/skills/*`; tests `tests/integration/test_skills_api.py`, `tests/skills/test_skill_channel.py` | Implemented evidence contract |
+| Failure/proposal | Three consecutive eligible failures since last corrected/dismissed boundary produce one `correction_proposed`; no duplicate open proposals. | System proposes; human reviews. | `correction.py:227-306`; UI detail openProposal mirror lines 64-78 | Implemented conservative trigger |
+| Correction | `corrected` records from/to versions, trace id, timestamp, prior body/model/source snapshot. | Maintainer/human invokes redistill; no auto self-healing. | `correction.py:111-150`; PRD FR-017–022 | Implemented, but production authorization is not evidenced by this function |
+| Rollback | Restores stashed previous fields, decrements to prior version, appends `rolled_back`; refuses absent/latest-already-rolled-back. | UI rollback action; API authoritative on stale read. | `correction.py:159-212`; `backend/api/v1/skills.py:139-151`; UI lines 81-97 | Implemented mechanical rollback |
+| Dismiss | `correction_dismissed` resets streak boundary; UI/API exposes dismiss. | Human operator says failures not actionable. | `skills.py:120-136`; correction boundary lines 230-241 | Implemented |
+| Production/known-good | UI/API vocabulary and PRD distinguish candidate/under-review/known-good, but audited correction code has no demonstrated promotion, approval, production enablement, or rollback-effectiveness state. | PRD assigns maintainer propose, admin approve. | `prd.md` FR-019–022; `correction.py` | Mostly declared, not evidenced |
+
+## 3. Existing known-good / production / rollback semantics
+
+**Existing facts:** version is monotonic for redistill and restored on rollback; evidence is append-only; prior body is stashed before overwrite; latest-only and no-double-rollback guards exist; correction proposal is human-triggered rather than automatic. The UI explicitly treats backend as authoritative for stale reads.
+
+**Not evidenced:** a durable `known-good` promotion workflow, production enablement approval, role enforcement in the endpoint itself, representative real-success requirement, rollback approval, post-rollback execution proof, or a state proving the restored version is actually serving production runs. The endpoint signatures shown are DB-dependent and do not establish actor authorization.
+
+Thus PRD FR-019–022 are new product commitments beyond the demonstrated lifecycle. OpenSpec/reconciliation also says checked task/implementation evidence is not live acceptance evidence.
+
+## 4. T-H5 minimum evidence breadth safely derivable
+
+A conservative minimum can be derived without asking the user:
+
+1. The exact skill version and source trace must be identified for every execution.
+2. The evidence must include the failed execution trace(s), failure stage/loop outcome, and whether the failure was environment noise or a skill failure (the existing proposal logic already excludes `loop_outcome=error`).
+3. A correction must record from-version, to-version, trace id, timestamp, and the prior body/source snapshot (already implemented).
+4. A proposed correction must be distinguishable from an executed correction; no proposal may silently promote or enable a version.
+5. Before calling a version known-good, evidence must show at least one target-scope real execution that passed, with its trace/evidence linked to that exact version; fixture/mock or mere distillation cannot qualify (directly follows PRD authenticity and current evidence separation).
+6. Rollback must identify the corrected version, restore the prior version, append an auditable event, and make double rollback impossible (existing invariant).
+7. Candidate/under-review/known-good/rolled-back states must remain distinct and visible.
+
+This is the minimum evidence breadth, not an invented sample-size policy.
+
+## 5. Rollback effectiveness as observable product behavior
+
+Express effectiveness as a user-verifiable outcome, not an internal operation: after an approved rollback, the skill detail and subsequent run trace show the prior version as the active version; the corrected version remains in history as rolled back; the next target-scope execution is attributable to the restored version; its outcome and failure trace (if any) are visible; and the system does not label the rollback successful merely because the database mutation completed. If the restored version cannot execute or its outcome is unknown, status stays blocked/unknown/failed and production trust is not restored.
+
+## 6. Exact sample count / risk strata
+
+The repository provides no basis for an exact number of known-good runs, domains, boundary cases, or risk strata. The PRD’s “at least one real success” is a product commitment, not an existing implementation fact. **Conservative defer:** assign the exact breadth to the platform administrator + skill maintainer / OODA strategy owner, conditioned on target capability risk and real target-scope evidence being available; until decided, do not claim known-good beyond the minimum one linked real pass and keep the version under review. No need to ask for a number during this audit.
+
+## 7. Recommended T-H5 resolution patch plan (no implementation design)
+
+1. Preserve current safe mechanics: proposal-only trigger, environment-error exclusion, append-only evidence, exact version transitions, prior-body rollback snapshot, latest-only rollback.
+2. Add product language that every execution/approval/correction/rollback is tied to an explicit skill version and linked trace evidence.
+3. State that `candidate`, `under-review`, `known-good`, and `rolled-back` are mutually non-interchangeable; proposal or distillation never promotes a version.
+4. Adopt the derived minimum breadth above: one target-scope real passing execution with complete linked evidence is the provisional floor; fixtures, mocks, catalog state, and repository/task completion cannot qualify.
+5. Define rollback success by observed post-rollback active version plus attributable follow-up execution outcome, not by mutation completion.
+6. Defer exact sample count/risk strata to named operational owners and record the condition for revisiting; do not invent a number.
+7. Require production enablement/rollback approval to be visible and auditable, while leaving actor/permission implementation out of this product-semantic patch.
+
+**Unique question that genuinely remains:** What minimum real-success breadth (sample count and risk strata) should the business require before a skill version is promoted to `known-good` for production? Everything else above can be resolved conservatively from existing contracts and observed invariants.
diff --git a/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/workflow-contract-audit.md b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/workflow-contract-audit.md
new file mode 100644
index 00000000..7153b085
--- /dev/null
+++ b/_bmad-output/planning-artifacts/prds/prd-snailfish-2026-08-28/workflow-contract-audit.md
@@ -0,0 +1,98 @@
+# Workflow Contract Audit
+
+只读盘点；未修改 PRD/addendum/code，未运行测试。结论基于仓库当前代码、OpenSpec 与已指定文档；geoXI 专属消费契约在仓库中不存在。
+
+## 1. End-to-end workflow map
+
+```text
+Capability catalog/context
+  -> readiness (declared gates; runtime has partial preflight)
+  -> realtime or scheduled collection
+  -> Observe (capture artifact)
+  -> Orient (normalize/lineage/evidence; product-declared)
+  -> Decide (risk policy/control contracts; product-declared)
+  -> Act (webhook delivery)
+  -> geoXI delivery/receipt (repo has generic destination ACK parsing only; geoXI contract absent)
+  -> feedback/next cycle (OODA product commitment; no demonstrated geoXI loop)
+
+S1: record -> distill -> execute -> failure trace/correction -> maintainer proposes rollback
+   -> platform administrator approves production rollback/enablement -> known-good governance
+```
+
+## 2. Stage inventory
+
+| Stage | Current implemented / declared behavior | Observable evidence / vocabulary | Anchors | Status |
+|---|---|---|---|---|
+| Capability/readiness | PRD defines eight independent gates and unknown/blocked/ready aggregation. Runtime checks capability id/availability, configured blockers, network permission, channel health; it does not implement the full aggregate gate model. | `GaojixingReadinessError` with typed codes; PRD gate states. | `prd.md` FR-002/004–007; `backend/workflow/gaojixing_runtime.py:107-161` | Partial implementation; declared broader product contract |
+| Package | Runtime resolves question/options once, canonicalizes JSON, computes SHA-256 digest, freezes options. | `GaojixingQuestionPackage`, `digest`, `questionPackage`. | `gaojixing_runtime.py:38-104` | Implemented for one question package; PRD expands immutable run/package lineage |
+| Live capture | Preflight then real `DoubaoResearchChannel`; explicitly never uses fixtures. | live/fixture/mock constants; channel health/readiness; answer item. | `gaojixing_runtime.py:16-25,107-161` | Implemented source boundary |
+| Evidence | Capture mapping separates answer, citations, conversation. Citation `verified` is always false; conversation missing is `unknown`; URL extraction is not verification. | evidence schema; statuses `captured`, `unavailable`, `empty`, `unknown`; mode/provenance. | `gaojixing_runtime.py:164-232` | Implemented conservative evidence mapping |
+| Normalize/lineage | Delivery context validates package/evidence mode/provenance and source lineage; mixed batch contexts fail. | `WorkflowWebhookDeliveryError` codes; lineage mismatch/context mismatch. | `webhook_delivery.py:152-263,291-343` | Implemented boundary checks; not full OODA |
+| Transport Act | Generic webhook sends payload and reports `transportStatus=accepted` only when notifier returns delivered. Network/notifier failures are explicit. | `delivered`, `transportStatus`, `businessOutcome`. | `webhook_delivery.py:26-150` | Implemented transport, not geoXI consumption |
+| Receipt/business outcome | Generic response parser recognizes `businessAck=true`, `business_ack=true`, `acknowledged=true`, or status confirmed/acknowledged; requires matching delivery id for live confirmation. | `ackEvidence`, `liveAccepted`, `confirmed/unconfirmed`. | `webhook_delivery.py:118-150,266-288` | Generic placeholder; not geoXI contract |
+| OODA/feedback | PRD requires system-owned Observe/Orient/Decide/Act and feedback-linked next cycle. Control package provides policy/cycle/ledger/kill-switch concepts, but current evidence does not demonstrate Gaojixing end-to-end closure or geoXI feedback consumption. | PRD statuses; control module contracts; no geoXI receipt/event. | `prd.md` FR-028–036; `docs/SYSTEM_ANALYSIS.md`; `docs/CONTROL_THEORY_ARCHITECTURE.md`; `backend/control/*` | Declared product commitment, not proven live workflow |
+| Schedule | Scheduling/automation surfaces exist and PRD defines enabled/unpaused ±1 minute, 99% metric. Audit evidence does not establish Gaojixing package/lineage/receipt semantics for scheduled occurrences. | schedule enabled/paused/next run/recent status are product needs. | `prd.md` FR-024–026/NFR-007 | Existing platform surface; Gaojixing contract gap |
+| S1 correction/rollback | PRD declares trace, stages, candidate/under-review/known-good, maintainer proposal and administrator approval. Reconciled OpenSpec says this is trust/governance scope, not automatic self-healing/crystallization. | candidate, under-review, known-good, correction/rollback states. | `prd.md` FR-017–022; `openspec/.../spec.md`, `tasks.md` | Product contract; acceptance tasks remain incomplete |
+
+## 3. Invariants directly adoptable (no user question)
+
+1. Live capture must call the real Doubao channel and must never silently fall back to fixture/mock (`gaojixing_runtime.py:107-161`; PRD FR-006/007).
+2. Missing effective question is a typed readiness blocker, not a successful empty run (`gaojixing_runtime.py:62-77`).
+3. Package digest is canonicalized and frozen for the runtime package (`gaojixing_runtime.py:95-104`).
+4. Answer, citation capture, and conversation evidence are independent; citation extraction is not verification and missing conversation remains unknown (`gaojixing_runtime.py:182-207`).
+5. Delivery rejects incomplete, contradictory, or mixed Gaojixing source contexts and lineage (`webhook_delivery.py:179-237,291-343`).
+6. Transport delivered/accepted is not business confirmation; live confirmation requires a response acknowledgement whose delivery id matches (`webhook_delivery.py:118-150,266-288`).
+7. Fixture/mock cannot satisfy live acceptance; mode and provenance remain visible.
+8. Repository checked tasks, fixtures, catalog/configuration, historical runs, and operator observations are not live proof (`reconcile-gaojixing-normative.md:13-17,25-35`; `addendum.md:54-58`).
+9. geoXI is a downstream product, not a snailfish internal module; no geoXI technical contract may be invented (`addendum.md:54-58`).
+10. S1 does not include automatic self-healing, general crystallization, dynamic loading, or a public plugin ecosystem (`addendum.md:5-19`).
+
+## 4. PRD commitments beyond current facts
+
+- Full eight-gate product readiness aggregation, freshness, and user-facing gate reasons.
+- Realtime and scheduled immutable package lineage across complete runs.
+- Product-level Orient/Decide/Act and feedback-linked next-cycle behavior owned by snailfish.
+- Matching, persisted/queryable geoXI receipt as business confirmation, with project isolation.
+- Risk-tier approval behavior for OODA actions.
+- 100 live cycles over seven days across two real geoXI projects and keyword strata as baseline evidence.
+- S1 known-good and rollback acceptance governance.
+- 99% schedule punctuality and 95% complete system-owned OODA targets.
+
+These are commitments, not implementation evidence. OpenSpec explicitly says acceptance tasks 4.x/5.x remain unchecked (`reconcile-gaojixing-normative.md:13-17`).
+
+## 5. Reviewer T-H4/T-H5 and medium findings
+
+### T-H4: retry/idempotency and duplicate side effects
+**Safe derivation:** Existing delivery id is deterministic from workflow/run/node/package digest (`webhook_delivery.py:248-255`); lineage and mixed-context checks already fail closed; PRD NFR-001/FR-036 already prohibit duplicate outcomes. Therefore the PRD can be patched without user input to require attempt-level visibility, preserve the same identity on retry, and classify ambiguous timeout-after-send/concurrent retry as unconfirmed until reconciled. The current code does not prove downstream idempotency or duplicate prevention.
+
+**Genuine business decision:** Whether an ambiguous transport may safely continue, how long to wait before expiry, and whether compensation/manual review is required are destination/business policies—not derivable from repo facts.
+
+### T-H5: OODA risk approvals
+**Safe derivation:** Existing control package has policy, gate, cycle, ledger, kill-switch, actuator, and operations-agent policy concepts; PRD already names role tiers. Patch can require every decision to retain its risk tier, decision evidence, actor, and approval state, and stop on evidence/lineage/receipt anomalies. This does not invent an API/schema.
+
+**Genuine business decision:** Which specific actions are low versus medium/high risk, and approval expiry/scope for frequency, project, capability, or production skill changes, require the OODA/business owners.
+
+### Medium findings
+- **Stale/mixed readiness:** safe derivation from runtime preflight plus PRD gate semantics: require a run-scoped, fresh admission result; no user decision needed.
+- **Repository evidence mistaken for live proof:** safe derivation from addendum/reconciliation; require evidence-class labeling and prohibit checked tasks/fixtures/configuration from satisfying live acceptance.
+- **Mixed live/fixture artifacts:** safe derivation from explicit no-fallback invariant; any mixed-mode run is non-live and cannot confirm.
+- **OODA completion without causal feedback:** genuine business decision on minimum valid feedback and whether a cycle may be complete without a next action.
+- **Partial collection semantics:** genuine business decision on whether partial keyword coverage is usable, and what outcome label it receives.
+- **Freshness and queryability:** business/operations decision for acceptable age and when a persisted result is truly queryable; geoXI contract is absent.
+
+## 6. geoXI contract boundary
+
+**Repo absent.** The repository contains no defined geoXI/GEO-XI-specific consumption contract, consumer event, ACK specification, persistence visibility guarantee, receipt freshness/expiry rule, or anti-replay rule. `webhook_delivery.py` only parses generic response fields and a matching delivery id. No geoXI behavior, receipt fields, or project persistence may be claimed from this code.
+
+## 7. Concrete next PRD patch plan (product semantics only)
+
+1. Add run-scoped readiness admission and evidence freshness; define stale/mixed gate behavior.
+2. State immutable package mismatch, mixed mode, and lineage mismatch as explicit non-live terminal outcomes.
+3. Add evidence-class/provenance labels separating repository, fixture, execution, and live business proof.
+4. Clarify citation/conversation acceptance versus displayable partial evidence.
+5. Add attempt-level retry semantics and ambiguous-send handling while retaining deterministic attribution.
+6. Require risk decision records and approvals tied to the exact action/evidence; leave risk classification matrix to owner decision.
+7. Define partial collection, OODA feedback sufficiency, and pending/expired destination outcome semantics.
+8. Keep geoXI requirements at product boundary until geoXI owner supplies the real receipt/persistence contract; then add acceptance conditions without inventing fields.
+9. Make baseline denominators auditable from all enabled/admitted occurrences, with blocker taxonomy and counter-metrics.
+10. Strengthen S1 known-good/rollback acceptance with representative real evidence and explicit production approval.
diff --git a/agent/Dockerfile b/agent/Dockerfile
index e842b6b9..250633b3 100644
--- a/agent/Dockerfile
+++ b/agent/Dockerfile
@@ -6,6 +6,9 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
     curl ca-certificates git procps \
     && curl -fsSL https://deb.nodesource.com/setup_22.x | bash - \
     && apt-get install -y --no-install-recommends nodejs \
+    && if ! command -v npm >/dev/null 2>&1; then \
+        apt-get install -y --no-install-recommends npm; \
+    fi \
     && rm -rf /var/lib/apt/lists/*
 
 # ── Optional: headless Chrome + virtual display ───────────────────────────────
@@ -48,6 +51,7 @@ WORKDIR /app
 COPY backend/agent_server.py ./backend/agent_server.py
 COPY backend/agent_runtimes/ ./backend/agent_runtimes/
 COPY backend/miniflow/ ./backend/miniflow/
+COPY backend/security/ ./backend/security/
 RUN touch ./backend/__init__.py
 
 COPY agent/entrypoint.sh /entrypoint.sh
@@ -61,6 +65,8 @@ ARG INSTALL_CHROME=false
 ENV PYTHONPATH=/app \
     PYTHONDONTWRITEBYTECODE=1 \
     PYTHONUNBUFFERED=1 \
+    # MiniFlow executes only workflows packaged under this allowlisted root.
+    MINIFLOW_WORKFLOW_ROOT=/app/workflows \
     # Bridge daemon port (used when Chrome is embedded OR daemon runs on host)
     OPENCLI_DAEMON_PORT=19825 \
     # Baked at build time so agent_server knows whether Chrome is bundled.
diff --git a/backend/agent_runtimes/base.py b/backend/agent_runtimes/base.py
index 5bc4f1c6..8080bf43 100644
--- a/backend/agent_runtimes/base.py
+++ b/backend/agent_runtimes/base.py
@@ -16,44 +16,102 @@
 that is what prevents typos in ``type`` strings and missing ``task_id`` fields
 from ever reaching a caller.
 """
-
 from abc import ABC, abstractmethod
 from collections.abc import AsyncIterator
 from dataclasses import dataclass, field
-from typing import Any
+from typing import Any, Literal
+
+RUNTIME_CAPABILITY_STREAMING = "streaming"
+RUNTIME_CAPABILITY_TOOL_EVENTS = "tool_events"
+RUNTIME_CAPABILITY_PERSISTENT_SESSION = "persistent_session"
+RUNTIME_CAPABILITY_RESUMABLE = "resumable"
+RUNTIME_CAPABILITY_BROWSER = "browser"
+RUNTIME_CAPABILITY_SUBAGENTS = "subagents"
+RUNTIME_CAPABILITY_WORKSPACE_READ = "workspace_read"
+RUNTIME_CAPABILITY_WORKSPACE_WRITE = "workspace_write"
+RUNTIME_CAPABILITY_SCHEDULES = "schedules"
+RUNTIME_CAPABILITY_HEARTBEATS = "heartbeats"
+RUNTIME_CAPABILITY_AGENT_MESSAGING = "agent_messaging"
+RUNTIME_CAPABILITY_MODEL_SELECTION = "model_selection"
+
+
+@dataclass(frozen=True)
+class RuntimeReadiness:
+    """Safe, typed evidence used before dispatching a runtime task.
+
+    Readiness is intentionally separate from ``RuntimeCapabilities``: a
+    registered adapter may be known to the node while its executable or
+    working directory is unavailable.  Implementations MUST keep secrets out
+    of this structure; it is suitable for Fleet diagnostics and wire output.
+    """
+
+    runtime: str
+    status: Literal["ready", "blocked"]
+    binary_present: bool
+    version: str | None = None
+    permitted_project_root: str | None = None
+    working_directory: str | None = None
+    reason_code: str | None = None
+    reason: str | None = None
 
 #: Closed tagged-union of runtime event types. Adapters MUST NOT emit any
-#: `type` outside this set — an unrecognized native event from the underlying
-#: framework is either mapped onto one of these or dropped (with a debug log),
-#: never passed through verbatim.
+#: ``type`` outside this set. Artifact, evidence, and audit events use the
+#: same envelope across every runtime so callers never parse native shapes.
 EVENT_TYPES: frozenset[str] = frozenset(
-    {"started", "text", "tool_call", "tool_result", "state", "done", "error"}
+    {
+        "started",
+        "text",
+        "tool_call",
+        "tool_result",
+        "state",
+        "artifact",
+        "evidence",
+        "audit",
+        "done",
+        "error",
+    }
 )
 
 
 @dataclass(frozen=True)
 class RuntimeCapabilities:
-    """What an agent runtime adapter can do; callers branch on this
-    declaration, exactly like ``channels.base.Capabilities``."""
+    """Stable capability advertisement used by Fleet selection."""
 
     transport: str  # "stdio" | "http" | "inprocess"
     streaming: bool = True
-    resume_by_id: bool = False  # can reopen a session by launcher-assigned id
+    resume_by_id: bool = False
     checkpoint: str = "none"  # none | memory | sqlite | postgres
     concurrent_sessions: bool = True
-
+    features: frozenset[str] = frozenset()
+
+    def names(self) -> frozenset[str]:
+        names = set(self.features)
+        if self.streaming:
+            names.add(RUNTIME_CAPABILITY_STREAMING)
+        if self.resume_by_id:
+            names.add(RUNTIME_CAPABILITY_RESUMABLE)
+        if self.checkpoint != "none":
+            names.add(RUNTIME_CAPABILITY_PERSISTENT_SESSION)
+        return frozenset(names)
 
 @dataclass
 class AgentTask:
-    """One agent run request handed to a ``RuntimeAdapter.invoke()``."""
+    """Runtime-neutral task handed to one selected ``RuntimeAdapter``."""
 
     task_id: str
-    workflow: str  # runtime-native workflow/agent identifier
+    workflow: str
     instructions: str = ""
     input: dict[str, Any] = field(default_factory=dict)
-    # Runtime-specific settings such as model, tools, cwd, or sidecar config.
+    # Only task-scoped policy belongs here. Binary paths, provider credentials,
+    # launch environment, and other Fleet-owned settings stay on the edge.
     config: dict[str, Any] = field(default_factory=dict)
-    session_id: str | None = None  # resume handle
+    session_id: str | None = None
+    provider: str | None = None
+    model: str | None = None
+    required_capabilities: tuple[str, ...] = ()
+    permissions: dict[str, Any] = field(default_factory=dict)
+    budget: dict[str, Any] = field(default_factory=dict)
+    evidence_requirements: tuple[str, ...] = ()
 
 
 # ── Event constructors ───────────────────────────────────────────────────────
@@ -104,6 +162,18 @@ def event_state(task_id: str, state: dict[str, Any]) -> dict[str, Any]:
     return {"type": "state", "task_id": task_id, "state": state}
 
 
+def event_artifact(task_id: str, artifact: dict[str, Any]) -> dict[str, Any]:
+    return {"type": "artifact", "task_id": task_id, "artifact": artifact}
+
+
+def event_evidence(task_id: str, evidence: dict[str, Any]) -> dict[str, Any]:
+    return {"type": "evidence", "task_id": task_id, "evidence": evidence}
+
+
+def event_audit(task_id: str, audit: dict[str, Any]) -> dict[str, Any]:
+    return {"type": "audit", "task_id": task_id, "audit": audit}
+
+
 def event_done(task_id: str, result: dict[str, Any] | None = None) -> dict[str, Any]:
     return {"type": "done", "task_id": task_id, "result": result or {}}
 
@@ -151,6 +221,23 @@ async def health(self) -> bool:
         """Cheap liveness check for this runtime (binary present, sidecar
         reachable, ...). Does not run a task."""
 
+    async def readiness(self, config: dict[str, Any] | None = None) -> RuntimeReadiness:
+        """Return safe pre-dispatch evidence for this runtime.
+
+        Adapters with richer checks override this method.  The default keeps
+        existing adapters source-compatible while giving callers a typed
+        readiness shape.
+        """
+        ready = await self.health()
+        return RuntimeReadiness(
+            runtime=self.runtime_type,
+            status="ready" if ready else "blocked",
+            binary_present=ready,
+            reason_code=None if ready else "unavailable",
+            reason=None if ready else f"runtime {self.runtime_type!r} is unavailable",
+        )
+
+
     @abstractmethod
     def validate_config(self, config: dict[str, Any]) -> list[str]:
         """Validate an AgentTask.config dict; return list of error strings
diff --git a/backend/agent_runtimes/bbx_adapter.py b/backend/agent_runtimes/bbx_adapter.py
index 61b534ba..c2f633b6 100644
--- a/backend/agent_runtimes/bbx_adapter.py
+++ b/backend/agent_runtimes/bbx_adapter.py
@@ -39,6 +39,7 @@ class BbxRuntimeAdapter(RuntimeAdapter):
         resume_by_id=False,
         checkpoint="none",
         concurrent_sessions=True,
+        features=frozenset({"browser", "tool_events"}),
     )
 
     def validate_config(self, config: dict[str, Any]) -> list[str]:
diff --git a/backend/agent_runtimes/codex_adapter.py b/backend/agent_runtimes/codex_adapter.py
new file mode 100644
index 00000000..9f00674b
--- /dev/null
+++ b/backend/agent_runtimes/codex_adapter.py
@@ -0,0 +1,446 @@
+"""Registered-Agent adapter for the local Codex CLI.
+
+The adapter is deliberately an edge-side subprocess adapter.  The control
+plane sends an ``agent_task`` over the authenticated Agent transport; only the
+registered Agent process imports this module and starts ``codex``.  No shell is
+used and no provider credential is copied into readiness or runtime events.
+
+Codex ``exec --json`` emits JSONL.  The native protocol has changed names a few
+times, so translation accepts the stable ``thread.*``, ``turn.*`` and
+``item.*`` envelopes while keeping our event envelope closed.
+"""
+
+from __future__ import annotations
+
+import asyncio
+import json
+import logging
+import re
+import shutil
+from collections.abc import AsyncIterator
+from pathlib import Path
+from typing import Any
+
+from backend.agent_runtimes.base import (
+    AgentTask,
+    RuntimeAdapter,
+    RuntimeCapabilities,
+    RuntimeReadiness,
+    event_done,
+    event_error,
+    event_started,
+    event_state,
+    event_text,
+    event_tool_call,
+    event_tool_result,
+)
+from backend.agent_runtimes.registry import register_runtime
+
+logger = logging.getLogger(__name__)
+
+_DEFAULT_TIMEOUT_SECONDS = 1800
+_MAX_TIMEOUT_SECONDS = 3600
+_VERSION_TIMEOUT_SECONDS = 5
+_KILL_GRACE_SECONDS = 10
+_STDERR_TAIL_BYTES = 2048
+_PERMISSION_MODES = frozenset({"approval_required", "full_auto", "read_only", "suggest_changes"})
+_SANDBOX_MODES = frozenset({"read-only", "workspace-write", "danger-full-access"})
+_VERSION_RE = re.compile(r"\bcodex(?:[- ]cli)?(?:\s+version)?\s+([0-9][0-9A-Za-z.+-]*)\b", re.I)
+_BARE_VERSION_RE = re.compile(r"\b([0-9]+\.[0-9]+(?:\.[0-9]+)?(?:[-+][0-9A-Za-z.-]+)?)\b")
+
+
+@register_runtime
+class CodexRuntimeAdapter(RuntimeAdapter):
+    """Run ``codex exec --json`` on a registered local Agent node."""
+
+    runtime_type = "codex"
+    capabilities = RuntimeCapabilities(
+        transport="stdio",
+        streaming=True,
+        resume_by_id=False,
+        checkpoint="none",
+        concurrent_sessions=True,
+        features=frozenset(
+            {"tool_events", "model_selection", "workspace_read", "workspace_write"}
+        ),
+    )
+
+    def validate_config(self, config: dict[str, Any]) -> list[str]:
+        errors: list[str] = []
+        binary = config.get("binary", "codex")
+        if not isinstance(binary, str) or not binary.strip():
+            errors.append("'binary' must be a non-empty string")
+        elif "\x00" in binary:
+            errors.append("'binary' must not contain NUL bytes")
+
+        for key in ("cwd", "project_root"):
+            if key in config and config[key] is not None:
+                value = config[key]
+                if not isinstance(value, str) or not value.strip():
+                    errors.append(f"'{key}' must be a non-empty string when provided")
+                elif "\x00" in value:
+                    errors.append(f"'{key}' must not contain NUL bytes")
+
+        if "args" in config and config["args"] is not None:
+            args = config["args"]
+            if not isinstance(args, list) or not all(isinstance(arg, str) for arg in args):
+                errors.append("'args' must be a list of strings when provided")
+
+        permission_mode = config.get("permission_mode")
+        if permission_mode is not None and permission_mode not in _PERMISSION_MODES:
+            errors.append(
+                "'permission_mode' must be one of " + ", ".join(sorted(_PERMISSION_MODES))
+            )
+        sandbox_mode = config.get("sandbox_mode")
+        if sandbox_mode is not None and sandbox_mode not in _SANDBOX_MODES:
+            errors.append("'sandbox_mode' must be one of " + ", ".join(sorted(_SANDBOX_MODES)))
+
+        if "timeout_seconds" in config and config["timeout_seconds"] is not None:
+            timeout = config["timeout_seconds"]
+            if (
+                not isinstance(timeout, (int, float))
+                or isinstance(timeout, bool)
+                or not 0 < timeout <= _MAX_TIMEOUT_SECONDS
+            ):
+                errors.append(
+                    f"'timeout_seconds' must be between 0 and {_MAX_TIMEOUT_SECONDS} when provided"
+                )
+        return errors
+
+    async def health(self) -> bool:
+        return self.is_available()
+
+    @classmethod
+    def is_available(cls, binary: str = "codex") -> bool:
+        """Cheap check used by the Agent registration handshake."""
+        if not isinstance(binary, str) or not binary or "\x00" in binary:
+            return False
+        return shutil.which(binary) is not None
+
+    async def readiness(self, config: dict[str, Any] | None = None) -> RuntimeReadiness:
+        config = config or {}
+        errors = self.validate_config(config)
+        if errors:
+            return RuntimeReadiness(
+                runtime=self.runtime_type,
+                status="blocked",
+                binary_present=False,
+                reason_code="invalid_config",
+                reason="; ".join(errors),
+            )
+
+        binary = config.get("binary") or "codex"
+        resolved_binary = shutil.which(binary)
+        if resolved_binary is None:
+            return RuntimeReadiness(
+                runtime=self.runtime_type,
+                status="blocked",
+                binary_present=False,
+                reason_code="missing_binary",
+                reason=f"codex binary not found: {binary!r}",
+            )
+
+        try:
+            project_root, cwd = self._resolve_paths(config)
+        except ValueError as exc:
+            return RuntimeReadiness(
+                runtime=self.runtime_type,
+                status="blocked",
+                binary_present=True,
+                permitted_project_root=self._display_path(config.get("project_root")),
+                working_directory=self._display_path(config.get("cwd")),
+                reason_code="invalid_path",
+                reason=str(exc),
+            )
+
+        version = await self._detect_version(
+            resolved_binary, config.get("args") or [], timeout_seconds=_VERSION_TIMEOUT_SECONDS
+        )
+        return RuntimeReadiness(
+            runtime=self.runtime_type,
+            status="ready",
+            binary_present=True,
+            version=version,
+            permitted_project_root=str(project_root),
+            working_directory=str(cwd),
+        )
+
+    def _compose_argv(
+        self,
+        config: dict[str, Any],
+        prompt: str = "",
+        *,
+        model: str | None = None,
+    ) -> list[str]:
+        binary = config.get("binary") or "codex"
+        argv = [binary, *(config.get("args") or []), "exec", "--json", "--color", "never"]
+        permission_mode = config.get("permission_mode")
+        if permission_mode == "full_auto":
+            # Codex exposes automatic review, not the old generic approval flag.
+            # Keep the sandbox bounded; the dangerous bypass flag is never
+            # selected by the Agent runtime.
+            argv.extend(("--approve-for-me", "--sandbox", "workspace-write"))
+        elif permission_mode == "read_only":
+            argv.extend(("--sandbox", "read-only"))
+        elif permission_mode in {"suggest_changes", "approval_required"}:
+            # Default Codex approval flow is the governed on-request mode.
+            pass
+
+        sandbox_mode = config.get("sandbox_mode")
+        if sandbox_mode is not None and permission_mode not in {"read_only", "full_auto"}:
+            argv.extend(("--sandbox", sandbox_mode))
+        if model:
+            argv.extend(("--model", model))
+        argv.append(prompt)
+        return argv
+
+    def _compose_prompt(self, task: AgentTask) -> str:
+        payload = task.input if isinstance(task.input, dict) else {}
+        message = payload.get("message") or payload.get("prompt") or ""
+        if not isinstance(message, str):
+            message = str(message)
+        if task.instructions:
+            return f"{task.instructions}\n\n{message}".strip()
+        return message
+
+    async def invoke(self, task: AgentTask) -> AsyncIterator[dict[str, Any]]:
+        config = task.config or {}
+        config_errors = self.validate_config(config)
+        if config_errors:
+            yield event_error(task.task_id, "; ".join(config_errors), error_type="ConfigError")
+            return
+        if task.provider not in {None, "openai", "openai-codex"}:
+            yield event_error(
+                task.task_id,
+                f"codex runtime does not support provider {task.provider!r}",
+                error_type="ConfigError",
+            )
+            return
+
+        binary = config.get("binary") or "codex"
+        resolved_binary = shutil.which(binary)
+        if resolved_binary is None:
+            yield event_error(
+                task.task_id,
+                f"codex binary not found: {binary!r}",
+                error_type="FileNotFoundError",
+            )
+            return
+        try:
+            _project_root, cwd = self._resolve_paths(config)
+        except ValueError as exc:
+            yield event_error(task.task_id, str(exc), error_type="PathError")
+            return
+
+        timeout_seconds = config.get("timeout_seconds") or _DEFAULT_TIMEOUT_SECONDS
+        version = await self._detect_version(
+            resolved_binary,
+            config.get("args") or [],
+            timeout_seconds=min(timeout_seconds, _VERSION_TIMEOUT_SECONDS),
+        )
+        argv = self._compose_argv(
+            config,
+            self._compose_prompt(task),
+            model=task.model,
+        )
+
+        try:
+            proc = await asyncio.create_subprocess_exec(
+                *argv,
+                stdin=asyncio.subprocess.DEVNULL,
+                stdout=asyncio.subprocess.PIPE,
+                stderr=asyncio.subprocess.PIPE,
+                cwd=str(cwd),
+            )
+        except FileNotFoundError as exc:
+            yield event_error(task.task_id, f"codex binary not found: {binary!r}", type(exc).__name__)
+            return
+        except OSError as exc:
+            yield event_error(task.task_id, f"failed to spawn codex: {exc}", type(exc).__name__)
+            return
+
+        yield event_started(task.task_id)
+        yield event_state(
+            task.task_id,
+            {"runtime": self.runtime_type, "codex_version": version, "working_directory": str(cwd)},
+        )
+
+        accumulated_text: list[str] = []
+        native_error: str | None = None
+
+        async def _read_events() -> AsyncIterator[dict[str, Any]]:
+            assert proc.stdout is not None
+            while True:
+                line = await proc.stdout.readline()
+                if not line:
+                    break
+                stripped = line.decode(errors="replace").strip("\r\n")
+                if not stripped:
+                    continue
+                try:
+                    native = json.loads(stripped)
+                except json.JSONDecodeError:
+                    logger.debug("codex_adapter: skipping non-JSON stdout line: %r", stripped[:200])
+                    continue
+                if not isinstance(native, dict):
+                    continue
+                translated = self._translate_event(task.task_id, native)
+                if translated is not None:
+                    yield translated
+
+        try:
+            async with asyncio.timeout(timeout_seconds):
+                async for event in _read_events():
+                    if event["type"] == "text":
+                        accumulated_text.append(event.get("text", ""))
+                    elif event["type"] == "error":
+                        native_error = event.get("message") or "Codex reported an error"
+                        break
+                    yield event
+        except (TimeoutError, asyncio.CancelledError) as exc:
+            await self._stop_process(proc)
+            if isinstance(exc, asyncio.CancelledError):
+                raise
+            yield event_error(
+                task.task_id,
+                f"codex run timed out after {timeout_seconds}s",
+                error_type="TimeoutError",
+            )
+            return
+
+        returncode = await proc.wait()
+        if native_error is not None:
+            yield event_error(task.task_id, native_error, error_type="RuntimeInvocationError")
+            return
+        if returncode != 0:
+            stderr_tail = b""
+            if proc.stderr is not None:
+                stderr_tail = await proc.stderr.read()
+            tail = stderr_tail[-_STDERR_TAIL_BYTES:].decode(errors="replace")
+            detail = f": {tail}" if tail else ""
+            yield event_error(
+                task.task_id,
+                f"codex exited with code {returncode}{detail}",
+                error_type="ProcessExitError",
+            )
+            return
+
+        yield event_done(
+            task.task_id,
+            result={
+                "runtime": self.runtime_type,
+                "codex_version": version,
+                "exit_code": returncode,
+                "text": "".join(accumulated_text),
+            },
+        )
+
+    async def _detect_version(
+        self,
+        binary: str,
+        args: list[str] | None = None,
+        timeout_seconds: float = _VERSION_TIMEOUT_SECONDS,
+    ) -> str | None:
+        proc: asyncio.subprocess.Process | None = None
+        try:
+            proc = await asyncio.create_subprocess_exec(
+                binary,
+                *(args or []),
+                "--version",
+                stdin=asyncio.subprocess.DEVNULL,
+                stdout=asyncio.subprocess.PIPE,
+                stderr=asyncio.subprocess.PIPE,
+            )
+            stdout, _stderr = await asyncio.wait_for(
+                proc.communicate(), timeout=timeout_seconds
+            )
+        except TimeoutError:
+            if proc is not None and proc.returncode is None:
+                proc.kill()
+                await proc.wait()
+            return None
+        except OSError:
+            return None
+        except asyncio.CancelledError:
+            if proc is not None and proc.returncode is None:
+                proc.kill()
+                await proc.wait()
+            raise
+        line = stdout.decode(errors="replace").splitlines()[0].strip() if stdout else ""
+        match = _VERSION_RE.search(line) or _BARE_VERSION_RE.search(line)
+        return match.group(0) if match else None
+
+    async def _stop_process(self, proc: asyncio.subprocess.Process) -> None:
+        if proc.returncode is not None:
+            return
+        proc.terminate()
+        try:
+            await asyncio.wait_for(proc.wait(), timeout=_KILL_GRACE_SECONDS)
+        except TimeoutError:
+            proc.kill()
+            await proc.wait()
+
+    def _resolve_paths(self, config: dict[str, Any]) -> tuple[Path, Path]:
+        project_root_raw = config.get("project_root") or config.get("cwd") or str(Path.cwd())
+        cwd_raw = config.get("cwd") or project_root_raw
+        project_root = Path(project_root_raw).expanduser().resolve()
+        cwd = Path(cwd_raw).expanduser().resolve()
+        if not project_root.is_dir():
+            raise ValueError(f"permitted project root is not a directory: {project_root}")
+        if not cwd.is_dir():
+            raise ValueError(f"working directory is not a directory: {cwd}")
+        try:
+            cwd.relative_to(project_root)
+        except ValueError as exc:
+            raise ValueError(
+                f"working directory {cwd} is outside permitted project root {project_root}"
+            ) from exc
+        return project_root, cwd
+
+    @staticmethod
+    def _display_path(value: object) -> str | None:
+        if not isinstance(value, str) or not value.strip():
+            return None
+        return str(Path(value).expanduser().resolve())
+
+    def _translate_event(self, task_id: str, native: dict[str, Any]) -> dict[str, Any] | None:
+        native_type = native.get("type")
+        if native_type in {"error", "turn.error"}:
+            return event_error(task_id, str(native.get("message") or native.get("error") or "Codex error"))
+        if native_type == "thread.started":
+            return event_state(task_id, {"thread_id": native.get("thread_id")})
+        if native_type == "turn.started":
+            return event_state(task_id, {"turn": "started"})
+        if native_type == "turn.completed":
+            state: dict[str, Any] = {"turn": "completed"}
+            if isinstance(native.get("usage"), dict):
+                state["usage"] = native["usage"]
+            return event_state(task_id, state)
+
+        item = native.get("item") if isinstance(native.get("item"), dict) else native
+        item_type = item.get("type")
+        if item_type in {"agent_message", "assistant_message", "text"}:
+            text = item.get("text") or item.get("content") or native.get("text")
+            return event_text(task_id, text) if isinstance(text, str) and text else None
+        if item_type in {"command_execution", "tool_call", "function_call"}:
+            if native_type in {"item.completed", "tool_result", "function_result"}:
+                output = item.get("aggregated_output", item.get("output", item.get("result")))
+                exit_code = item.get("exit_code")
+                return event_tool_result(
+                    task_id,
+                    name=str(item.get("command") or item.get("name") or "codex_tool"),
+                    result=output,
+                    call_id=item.get("id") or item.get("call_id"),
+                    is_error=bool(item.get("is_error")) or exit_code not in (None, 0),
+                )
+            return event_tool_call(
+                task_id,
+                name=str(item.get("command") or item.get("name") or "codex_tool"),
+                args=item.get("arguments") if isinstance(item.get("arguments"), dict) else {"command": item.get("command", "")},
+                call_id=item.get("id") or item.get("call_id"),
+            )
+        if native_type in {"text", "output_text.delta", "response.output_text.delta"}:
+            text = native.get("text") or native.get("delta")
+            return event_text(task_id, text) if isinstance(text, str) and text else None
+        logger.debug("codex_adapter: skipping unmapped native event type %r", native_type)
+        return None
diff --git a/backend/agent_runtimes/miniflow_adapter.py b/backend/agent_runtimes/miniflow_adapter.py
index 4d12a86f..029f004a 100644
--- a/backend/agent_runtimes/miniflow_adapter.py
+++ b/backend/agent_runtimes/miniflow_adapter.py
@@ -23,7 +23,7 @@
 from backend.agent_runtimes.registry import register_runtime
 from backend.miniflow.audit import AuditLog
 from backend.miniflow.loader import confine_to_workflow_root, load_workflow_file
-from backend.miniflow.model import Outcome
+from backend.miniflow.model import Outcome, Workflow, get_builtin_workflow
 from backend.miniflow.runner import RunResult, run_workflow
 
 _RUN_WORKFLOWS = {"run", "workflow.run", "miniflow.run"}
@@ -43,6 +43,7 @@ class MiniFlowRuntimeAdapter(RuntimeAdapter):
         resume_by_id=False,
         checkpoint="none",
         concurrent_sessions=True,
+        features=frozenset({"tool_events"}),
     )
 
     def validate_config(self, config: dict[str, Any]) -> list[str]:
@@ -133,7 +134,11 @@ def _run_sync(
         queue: asyncio.Queue[dict[str, Any]],
         loop: asyncio.AbstractEventLoop,
     ) -> RunResult:
-        workflow = load_workflow_file(request.workflow_path)
+        workflow = (
+            request.builtin_workflow
+            if request.builtin_workflow is not None
+            else load_workflow_file(request.workflow_path)
+        )
 
         def emit(entry: dict[str, Any]) -> None:
             loop.call_soon_threadsafe(queue.put_nowait, entry)
@@ -154,7 +159,8 @@ class _RunRequest:
     def __init__(
         self,
         *,
-        workflow_path: Path,
+        workflow_path: Path | None,
+        builtin_workflow: Workflow | None,
         audit_log: Path | None,
         max_attempts: int,
         breaker_threshold: int,
@@ -163,6 +169,7 @@ def __init__(
         sleep: Callable[[float], Any],
     ) -> None:
         self.workflow_path = workflow_path
+        self.builtin_workflow = builtin_workflow
         self.audit_log = audit_log
         self.max_attempts = max_attempts
         self.breaker_threshold = breaker_threshold
@@ -187,12 +194,15 @@ def _build_run_request(task: AgentTask) -> _RunRequest:
     audit_ref = _string(payload.get("audit_log")) or _string(config.get("audit_log"))
     clock = config.get("_clock") if callable(config.get("_clock")) else time.monotonic
     sleep = config.get("_sleep") if callable(config.get("_sleep")) else time.sleep
-    # Confine both to the MiniFlow allowlist root: workflow_path is executed
-    # by loader.load_workflow_file (RCE surface) and audit_log is opened for
-    # append with parents auto-created (arbitrary-file-write surface). Both
-    # are attacker-influenced via task.input/task.config.
-    workflow_path = confine_to_workflow_root(
-        _resolve_path(workflow_ref, cwd), label="workflow_path"
+    # Stable built-ins ship inside backend.miniflow on Docker and shell Agents;
+    # caller-provided files retain the existing allowlisted-root confinement.
+    builtin_workflow = get_builtin_workflow(workflow_ref)
+    workflow_path = (
+        None
+        if builtin_workflow is not None
+        else confine_to_workflow_root(
+            _resolve_path(workflow_ref, cwd), label="workflow_path"
+        )
     )
     audit_log = (
         confine_to_workflow_root(_resolve_path(audit_ref, cwd), label="audit_log")
@@ -201,6 +211,7 @@ def _build_run_request(task: AgentTask) -> _RunRequest:
     )
     return _RunRequest(
         workflow_path=workflow_path,
+        builtin_workflow=builtin_workflow,
         audit_log=audit_log,
         max_attempts=_positive_int(payload, config, "max_attempts", _DEFAULT_MAX_ATTEMPTS),
         breaker_threshold=_positive_int(
diff --git a/backend/agent_runtimes/opentabs_adapter.py b/backend/agent_runtimes/opentabs_adapter.py
index 1ebae5e7..e1cff947 100644
--- a/backend/agent_runtimes/opentabs_adapter.py
+++ b/backend/agent_runtimes/opentabs_adapter.py
@@ -57,6 +57,7 @@ class OpenTabsRuntimeAdapter(RuntimeAdapter):
         resume_by_id=False,
         checkpoint="none",
         concurrent_sessions=True,
+        features=frozenset({"browser", "tool_events"}),
     )
 
     def validate_config(self, config: dict[str, Any]) -> list[str]:
diff --git a/backend/agent_runtimes/pi_adapter.py b/backend/agent_runtimes/pi_adapter.py
index 92972ddc..c939ad4b 100644
--- a/backend/agent_runtimes/pi_adapter.py
+++ b/backend/agent_runtimes/pi_adapter.py
@@ -88,7 +88,7 @@
 
 logger = logging.getLogger(__name__)
 
-_DEFAULT_TIMEOUT_SECONDS = 300
+_DEFAULT_TIMEOUT_SECONDS = 1800
 _KILL_GRACE_SECONDS = 10
 _STDERR_TAIL_BYTES = 2048
 _READ_ONLY_PROFILE_MODES = frozenset({"observe_only", "suggest_changes"})
@@ -106,7 +106,12 @@ class PiRuntimeAdapter(RuntimeAdapter):
         resume_by_id=False,  # see module docstring: no --session-id in pi's RPC protocol
         checkpoint="none",
         concurrent_sessions=True,
+        features=frozenset(
+            {"tool_events", "model_selection", "workspace_read", "workspace_write"}
+        ),
     )
+    binary_name = "pi"
+    session_dir_env = "PI_CODING_AGENT_SESSION_DIR"
 
     def validate_config(self, config: dict[str, Any]) -> list[str]:
         errors: list[str] = []
@@ -118,7 +123,7 @@ def validate_config(self, config: dict[str, Any]) -> list[str]:
                     "read-only permission modes cannot override Fleet launch config: "
                     + ", ".join(unsupported)
                 )
-        binary = config.get("binary", "pi")
+        binary = config.get("binary", self.binary_name)
         if not isinstance(binary, str) or not binary:
             errors.append("'binary' must be a non-empty string")
         if "cwd" in config and config["cwd"] is not None and not isinstance(config["cwd"], str):
@@ -145,17 +150,23 @@ async def health(self) -> bool:
         return self.is_available()
 
     @classmethod
-    def is_available(cls, binary: str = "pi") -> bool:
+    def is_available(cls, binary: str | None = None) -> bool:
         """Cheap sync check used by ``registry.available_runtimes()``."""
-        return shutil.which(binary) is not None
+        return shutil.which(binary or cls.binary_name) is not None
 
     # ── argv / env composition ───────────────────────────────────────────────
     # Split into small methods (OpenAlice CliAdapter pattern cited in the GOAL
     # doc) rather than one monolithic spawn(): argv, env, and the request
     # payload are each independently testable and independently overridable.
 
-    def _compose_argv(self, config: dict[str, Any]) -> list[str]:
-        binary = config.get("binary") or "pi"
+    def _compose_argv(
+        self,
+        config: dict[str, Any],
+        *,
+        provider: str | None = None,
+        model: str | None = None,
+    ) -> list[str]:
+        binary = config.get("binary") or self.binary_name
         extra_args = config.get("args") or []
         # `args` is inserted BEFORE `--mode rpc` so tests can point `binary`
         # at a bare interpreter (e.g. sys.executable) and supply the script
@@ -176,6 +187,10 @@ def _compose_argv(self, config: dict[str, Any]) -> list[str]:
                     _READ_ONLY_TOOLS,
                 ]
             )
+        if provider:
+            argv.extend(("--provider", provider))
+        if model:
+            argv.extend(("--model", model))
         return [*argv, "--mode", "rpc"]
 
     def _compose_env(self, config: dict[str, Any]) -> dict[str, str] | None:
@@ -186,7 +201,7 @@ def _compose_env(self, config: dict[str, Any]) -> dict[str, str] | None:
         if provider_dir:
             # See module docstring: pi has no PI_CODING_AGENT_DIR; the real
             # override is PI_CODING_AGENT_SESSION_DIR.
-            extra_env.setdefault("PI_CODING_AGENT_SESSION_DIR", provider_dir)
+            extra_env.setdefault(self.session_dir_env, provider_dir)
         if not extra_env:
             return None
         return {**os.environ, **extra_env}
@@ -209,7 +224,11 @@ async def invoke(self, task: AgentTask) -> AsyncIterator[dict[str, Any]]:
         if config_errors:
             yield event_error(task.task_id, "; ".join(config_errors), error_type="ConfigError")
             return
-        argv = self._compose_argv(config)
+        argv = self._compose_argv(
+            config,
+            provider=task.provider,
+            model=task.model,
+        )
         env = self._compose_env(config)
         cwd = config.get("cwd")
         timeout_seconds = config.get("timeout_seconds") or _DEFAULT_TIMEOUT_SECONDS
@@ -224,10 +243,18 @@ async def invoke(self, task: AgentTask) -> AsyncIterator[dict[str, Any]]:
                 env=env,
             )
         except FileNotFoundError as exc:
-            yield event_error(task.task_id, f"pi binary not found: {argv[0]!r}", error_type=type(exc).__name__)
+            yield event_error(
+                task.task_id,
+                f"{self.runtime_type} binary not found: {argv[0]!r}",
+                error_type=type(exc).__name__,
+            )
             return
         except OSError as exc:
-            yield event_error(task.task_id, f"failed to spawn pi: {exc}", error_type=type(exc).__name__)
+            yield event_error(
+                task.task_id,
+                f"failed to spawn {self.runtime_type}: {exc}",
+                error_type=type(exc).__name__,
+            )
             return
 
         yield event_started(task.task_id)
@@ -288,7 +315,7 @@ async def _read_events() -> AsyncIterator[dict[str, Any]]:
                 raise
             yield event_error(
                 task.task_id,
-                f"pi run timed out after {timeout_seconds}s",
+                f"{self.runtime_type} run timed out after {timeout_seconds}s",
                 error_type="TimeoutError",
             )
             return
@@ -306,7 +333,7 @@ async def _read_events() -> AsyncIterator[dict[str, Any]]:
             tail = stderr_tail[-_STDERR_TAIL_BYTES:].decode(errors="replace")
             yield event_error(
                 task.task_id,
-                f"pi exited with code {returncode}: {tail}",
+                f"{self.runtime_type} exited with code {returncode}: {tail}",
                 error_type="ProcessExitError",
             )
             return
@@ -323,7 +350,10 @@ def _translate_event(self, task_id: str, native: dict[str, Any]) -> dict[str, An
 
         if native_type == "response":
             if native.get("success") is False:
-                return event_error(task_id, native.get("error") or "pi command failed")
+                return event_error(
+                    task_id,
+                    native.get("error") or f"{self.runtime_type} command failed",
+                )
             return None
 
         if native_type == "message_update":
diff --git a/backend/agent_runtimes/prime_agent_adapter.py b/backend/agent_runtimes/prime_agent_adapter.py
new file mode 100644
index 00000000..9bfd4dd1
--- /dev/null
+++ b/backend/agent_runtimes/prime_agent_adapter.py
@@ -0,0 +1,33 @@
+"""Prime Agent adapter using its documented pi-compatible JSONL RPC mode."""
+
+from backend.agent_runtimes.base import RuntimeCapabilities
+from backend.agent_runtimes.pi_adapter import PiRuntimeAdapter
+from backend.agent_runtimes.registry import register_runtime
+
+
+@register_runtime
+class PrimeAgentRuntimeAdapter(PiRuntimeAdapter):
+    """Run ``prime-agent --mode rpc`` without moving provider secrets off-node."""
+
+    runtime_type = "prime-agent"
+    binary_name = "prime-agent"
+    session_dir_env = "PRIME_AGENT_SESSION_DIR"
+    capabilities = RuntimeCapabilities(
+        transport="stdio",
+        streaming=True,
+        resume_by_id=False,
+        checkpoint="none",
+        concurrent_sessions=True,
+        features=frozenset(
+            {
+                "agent_messaging",
+                "heartbeats",
+                "model_selection",
+                "schedules",
+                "subagents",
+                "tool_events",
+                "workspace_read",
+                "workspace_write",
+            }
+        ),
+    )
diff --git a/backend/agent_runtimes/registry.py b/backend/agent_runtimes/registry.py
index dc74a438..90196ff0 100644
--- a/backend/agent_runtimes/registry.py
+++ b/backend/agent_runtimes/registry.py
@@ -43,13 +43,26 @@ def available_runtimes() -> list[str]:
     return available
 
 
+def available_runtime_capabilities() -> dict[str, list[str]]:
+    """Capability manifests for runtimes that are usable on this node."""
+
+    capabilities: dict[str, list[str]] = {}
+    for runtime_type, instance in _REGISTRY.items():
+        is_available = getattr(type(instance), "is_available", None)
+        if is_available is not None and is_available():
+            capabilities[runtime_type] = sorted(instance.capabilities.names())
+    return capabilities
+
+
 def _load_all_runtimes() -> None:
     """Import all agent-runtime adapter modules to trigger registration."""
     from backend.agent_runtimes import (  # noqa: F401
         bbx_adapter,
+        codex_adapter,
         miniflow_adapter,
         opentabs_adapter,
         pi_adapter,
+        prime_agent_adapter,
     )
 
 
diff --git a/backend/agent_server.py b/backend/agent_server.py
index d8d886bf..7cdaa0ee 100644
--- a/backend/agent_server.py
+++ b/backend/agent_server.py
@@ -57,6 +57,7 @@
 from contextlib import asynccontextmanager
 from typing import Any
 from urllib.parse import urlparse
+from urllib.request import proxy_bypass
 
 import yaml
 from fastapi import FastAPI, Header, HTTPException
@@ -69,7 +70,11 @@
 # dependency set (see module docstring); this avoids depending on the package
 # __init__ staying lightweight as more adapters are added later.
 from backend.agent_runtimes.base import AgentTask, RuntimeInvocationError
-from backend.agent_runtimes.registry import available_runtimes, get_runtime
+from backend.agent_runtimes.registry import (
+    available_runtime_capabilities,
+    available_runtimes,
+    get_runtime,
+)
 
 logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s [%(name)s] %(message)s")
 logger = logging.getLogger("agent_server")
@@ -207,14 +212,18 @@ def _detect_advertise_url() -> str:
     return f"http://{ip}:{_AGENT_PORT}"
 
 
+def _center_proxy() -> str | None:
+    """Return the configured outbound proxy unless the center host is bypassed."""
+    center_host = urlparse(_CENTRAL_API_URL).hostname
+    if center_host and proxy_bypass(center_host):
+        return None
+    return _HTTPS_PROXY or _HTTP_PROXY or None
+
+
 def _build_proxies() -> dict:
-    """Build httpx proxy dict from environment variables."""
-    proxies: dict = {}
-    if _HTTPS_PROXY:
-        proxies["https://"] = _HTTPS_PROXY
-    if _HTTP_PROXY:
-        proxies["http://"] = _HTTP_PROXY
-    return proxies
+    """Build the legacy httpx proxy map for center registration."""
+    proxy = _center_proxy()
+    return {"https://": proxy, "http://": proxy} if proxy else {}
 
 
 async def _register_with_center(advertise_url: str) -> None:
@@ -229,6 +238,7 @@ async def _register_with_center(advertise_url: str) -> None:
         "label": _AGENT_LABEL,
         "agent_protocol": "http",
         "runtimes": available_runtimes(),
+        "runtime_capabilities": available_runtime_capabilities(),
         "profile_kind": _BROWSER_PROFILE_KIND,
     }
     proxies = _build_proxies()
@@ -345,10 +355,22 @@ async def _send_result(result: dict) -> None:
             input=msg.get("input") or {},
             config=msg.get("config") or {},
             session_id=msg.get("session_id"),
+            provider=msg.get("provider"),
+            model=msg.get("model"),
+            required_capabilities=tuple(msg.get("required_capabilities") or ()),
+            permissions=msg.get("permissions") or {},
+            budget=msg.get("budget") or {},
+            evidence_requirements=tuple(msg.get("evidence_requirements") or ()),
         )
         config_errors = adapter.validate_config(task.config)
         if config_errors:
             raise RuntimeInvocationError("; ".join(config_errors), error_type="ConfigError")
+        logger.info(
+            "WS agent_task started request_id=%s runtime=%s workflow=%s",
+            request_id,
+            runtime_type,
+            task.workflow,
+        )
 
         terminal_event: dict | None = None
         async for event in adapter.invoke(task):
@@ -375,6 +397,12 @@ async def _send_result(result: dict) -> None:
                 "error_type": "RuntimeInvocationError",
             }
         await _send_result(terminal_event)
+        logger.info(
+            "WS agent_task finished request_id=%s runtime=%s terminal=%s",
+            request_id,
+            runtime_type,
+            terminal_event.get("type"),
+        )
     except RuntimeInvocationError as exc:
         logger.exception(
             "WS agent_task request_id=%s: adapter invocation error: %s",
@@ -427,7 +455,7 @@ async def _register_via_ws(advertise_url: str) -> None:
         .rstrip("/")
         + "/api/v1/nodes/ws"
     )
-    _proxy = _HTTPS_PROXY or _HTTP_PROXY or None
+    _proxy = _center_proxy()
     # Computed once (not per reconnect attempt): available_runtimes() does a
     # handful of cheap shutil.which() checks, not worth repeating on every
     # reconnect. A node's installed runtimes don't change without a restart.
@@ -438,6 +466,7 @@ async def _register_via_ws(advertise_url: str) -> None:
         "node_type": _AGENT_DEPLOY_TYPE,
         "label": _AGENT_LABEL,
         "runtimes": available_runtimes(),
+        "runtime_capabilities": available_runtime_capabilities(),
         "profile_kind": _BROWSER_PROFILE_KIND,
     })
 
diff --git a/backend/api/v1/__init__.py b/backend/api/v1/__init__.py
index e42907ec..5da3b07d 100644
--- a/backend/api/v1/__init__.py
+++ b/backend/api/v1/__init__.py
@@ -16,6 +16,9 @@
     geo_acquisition,
     identity,
     image_studio,
+    internal_agent_runs,
+    internal_automations,
+    internal_collaboration,
     model_defaults,
     nodes,
     notifications,
@@ -48,6 +51,9 @@
 v1_router.include_router(agents.router)
 v1_router.include_router(geo_acquisition.router)
 v1_router.include_router(automations.router)
+v1_router.include_router(internal_automations.router)
+v1_router.include_router(internal_agent_runs.router)
+v1_router.include_router(internal_collaboration.router)
 v1_router.include_router(image_studio.router)
 v1_router.include_router(browser_act.router)
 v1_router.include_router(browsers.router)
diff --git a/backend/api/v1/agents.py b/backend/api/v1/agents.py
index d8f21dcc..d85c8127 100644
--- a/backend/api/v1/agents.py
+++ b/backend/api/v1/agents.py
@@ -8,7 +8,12 @@
 
 from backend.database import get_db
 from backend.models.agent import AIAgent
-from backend.schemas.agent import AIAgentCreate, AIAgentRead, AIAgentUpdate
+from backend.schemas.agent import (
+    AIAgentCreate,
+    AIAgentRead,
+    AIAgentUpdate,
+    validate_paw_agent_config,
+)
 from backend.schemas.common import ApiResponse
 
 router = APIRouter(prefix="/agents", tags=["agents"])
@@ -29,6 +34,10 @@ async def list_agents(
 
 @router.post("", response_model=ApiResponse[AIAgentRead], status_code=201)
 async def create_agent(body: AIAgentCreate, db: AsyncSession = Depends(get_db)):
+    try:
+        validate_paw_agent_config(body.processor_type, body.prompt_template, body.processor_config)
+    except ValueError as exc:
+        raise HTTPException(status_code=422, detail=str(exc)) from exc
     agent = AIAgent(**body.model_dump())
     db.add(agent)
     await db.commit()
@@ -53,7 +62,17 @@ async def update_agent(
     agent = result.scalar_one_or_none()
     if not agent:
         raise HTTPException(status_code=404, detail="Agent not found")
-    for field, value in body.model_dump(exclude_unset=True).items():
+    changes = body.model_dump(exclude_unset=True)
+    candidate = {
+        "processor_type": changes.get("processor_type", agent.processor_type),
+        "prompt_template": changes.get("prompt_template", agent.prompt_template),
+        "processor_config": changes.get("processor_config", agent.processor_config),
+    }
+    try:
+        validate_paw_agent_config(**candidate)
+    except ValueError as exc:
+        raise HTTPException(status_code=422, detail=str(exc)) from exc
+    for field, value in changes.items():
         setattr(agent, field, value)
     await db.commit()
     await db.refresh(agent)
diff --git a/backend/api/v1/automations.py b/backend/api/v1/automations.py
index 93da16eb..50c2f5e0 100644
--- a/backend/api/v1/automations.py
+++ b/backend/api/v1/automations.py
@@ -4,13 +4,61 @@
 
 from backend.database import get_db
 from backend.models.automation import Automation
-from backend.schemas.automation import AutomationCreate, AutomationRead, AutomationUpdate
+from backend.schemas.automation import (
+    AutomationCreate,
+    AutomationRead,
+    AutomationUpdate,
+    StarterInstallationPreview,
+    StarterInstallationResult,
+)
 from backend.schemas.common import ApiResponse
+from backend.schemas.operations_agent import OperationsAgentRunRead
 from backend.security.identity import RequestIdentity, get_request_identity
 from backend.security.workspace_rbac import WorkspacePermission, get_workspace_access, require_permission
-
+from backend.services.automation_schedule_service import (
+    AutomationBindingError,
+    create_bound_automation_run,
+    validate_automation_binding,
+)
+from backend.services.automation_starter_service import (
+    install_starters,
+    preview_starter_installation,
+)
 router = APIRouter(prefix="/workspaces/{workspace_id}/automations", tags=["automations"])
 
+@router.get(
+    "/starters/preview",
+    response_model=ApiResponse[StarterInstallationPreview],
+)
+async def preview_automation_starters(
+    workspace_id: str,
+    identity: RequestIdentity = Depends(get_request_identity),
+    db: AsyncSession = Depends(get_db),
+) -> ApiResponse:
+    access = await get_workspace_access(db, workspace_id, identity)
+    require_permission(access, WorkspacePermission.READ)
+    preview = await preview_starter_installation(db, workspace_id=workspace_id)
+    return ApiResponse.ok(preview)
+
+
+@router.post(
+    "/starters/install",
+    response_model=ApiResponse[StarterInstallationResult],
+)
+async def install_automation_starters(
+    workspace_id: str,
+    identity: RequestIdentity = Depends(get_request_identity),
+    db: AsyncSession = Depends(get_db),
+) -> ApiResponse:
+    access = await get_workspace_access(db, workspace_id, identity)
+    require_permission(access, WorkspacePermission.MANAGE_AGENT_IDENTITIES)
+    result = await install_starters(
+        db,
+        workspace_id=workspace_id,
+        created_by_user_id=access.user_id,
+    )
+    return ApiResponse.ok(result)
+
 
 @router.get("", response_model=ApiResponse[list[AutomationRead]])
 async def list_automations(
@@ -38,6 +86,11 @@ async def create_automation(
     row = Automation(workspace_id=workspace_id, created_by_user_id=access.user_id, **body.model_dump())
     db.add(row)
     await db.flush()
+    if row.enabled:
+        try:
+            await validate_automation_binding(db, row)
+        except AutomationBindingError as exc:
+            raise HTTPException(status.HTTP_422_UNPROCESSABLE_CONTENT, str(exc)) from exc
     return ApiResponse.ok(AutomationRead.model_validate(row))
 
 
@@ -58,5 +111,49 @@ async def update_automation(
         raise HTTPException(status.HTTP_404_NOT_FOUND, "Automation not found")
     for field, value in body.model_dump(exclude_unset=True).items():
         setattr(row, field, value)
+    row.revision += 1
+    if row.enabled:
+        try:
+            await validate_automation_binding(db, row)
+        except AutomationBindingError as exc:
+            raise HTTPException(status.HTTP_422_UNPROCESSABLE_CONTENT, str(exc)) from exc
     await db.flush()
     return ApiResponse.ok(AutomationRead.model_validate(row))
+
+
+@router.post(
+    "/{automation_id}/runs",
+    response_model=ApiResponse[OperationsAgentRunRead],
+    status_code=201,
+)
+async def start_automation_run(
+    workspace_id: str,
+    automation_id: str,
+    identity: RequestIdentity = Depends(get_request_identity),
+    db: AsyncSession = Depends(get_db),
+) -> ApiResponse:
+    access = await get_workspace_access(db, workspace_id, identity)
+    require_permission(access, WorkspacePermission.RUN_OPERATIONS_AGENTS)
+    row = await db.scalar(
+        select(Automation)
+        .where(Automation.workspace_id == workspace_id, Automation.id == automation_id)
+        .with_for_update()
+    )
+    if row is None:
+        raise HTTPException(status.HTTP_404_NOT_FOUND, "Automation not found")
+    if not row.enabled:
+        raise HTTPException(status.HTTP_409_CONFLICT, "Paused Automation cannot run")
+    try:
+        run, _ = await create_bound_automation_run(
+            db,
+            row,
+            trigger_type="manual",
+            started_by_user_id=access.user_id,
+        )
+    except AutomationBindingError as exc:
+        raise HTTPException(status.HTTP_409_CONFLICT, str(exc)) from exc
+    return ApiResponse.ok(OperationsAgentRunRead.model_validate(run))
+
+
+
+
diff --git a/backend/api/v1/collaboration_schemas.py b/backend/api/v1/collaboration_schemas.py
new file mode 100644
index 00000000..3d23267a
--- /dev/null
+++ b/backend/api/v1/collaboration_schemas.py
@@ -0,0 +1,33 @@
+"""Request and response schemas for Studio collaboration service callbacks."""
+
+from typing import Any, Literal
+
+from pydantic import BaseModel, Field
+
+
+class CollaborationRoomRequest(BaseModel):
+    room: str = Field(min_length=1, max_length=512)
+
+
+class CollaborationRoomRead(BaseModel):
+    room: str
+
+
+class YMapJSON(BaseModel):
+    """JSON representation emitted by ``Y.Map.toJSON()``."""
+
+    type: Literal["Map"]
+    content: dict[str, dict[str, Any]] = Field(default_factory=dict)
+
+
+class CollaborationSnapshotData(BaseModel):
+    nodes: YMapJSON
+    edges: YMapJSON
+
+
+class CollaborationSnapshotRequest(CollaborationRoomRequest):
+    data: CollaborationSnapshotData
+
+
+class CollaborationSnapshotRead(BaseModel):
+    revision: int
diff --git a/backend/api/v1/identity.py b/backend/api/v1/identity.py
index 362ad6fe..613385bc 100644
--- a/backend/api/v1/identity.py
+++ b/backend/api/v1/identity.py
@@ -1,15 +1,70 @@
-"""Request identity endpoint."""
+"""Local administrator and request identity endpoints."""
 
 from typing import Annotated
 
-from fastapi import APIRouter, Depends
+from fastapi import APIRouter, Depends, HTTPException, status
+from pydantic import BaseModel, Field
 
+from backend.config import get_settings
 from backend.schemas.common import ApiResponse
 from backend.security.identity import RequestIdentity, get_request_identity
+from backend.security.local_auth import (
+    hash_password,
+    issue_local_token,
+    persist_password_hash,
+    verify_password,
+)
+
+
+class LocalLoginRequest(BaseModel):
+    username: str = Field(min_length=1, max_length=255)
+    password: str = Field(min_length=1, max_length=255)
+
+
+class ChangePasswordRequest(BaseModel):
+    current_password: str = Field(min_length=1, max_length=255)
+    new_password: str = Field(min_length=6, max_length=255)
+
 
 router = APIRouter(prefix="/auth", tags=["auth"])
 
 
+@router.post("/login", response_model=ApiResponse[dict])
+async def local_login(body: LocalLoginRequest) -> ApiResponse:
+    settings = get_settings()
+    if body.username != settings.local_admin_username or not verify_password(
+        body.password, settings.local_admin_password_hash
+    ):
+        raise HTTPException(status.HTTP_401_UNAUTHORIZED, "用户名或密码错误")
+
+    return ApiResponse.ok(
+        {
+            "access_token": issue_local_token(settings.local_admin_username, settings.secret_key),
+            "token_type": "bearer",
+            "using_default_password": verify_password(
+                "admin", settings.local_admin_password_hash
+            ),
+        }
+    )
+
+
+@router.post("/password", response_model=ApiResponse[dict])
+async def change_local_password(
+    body: ChangePasswordRequest,
+    identity: Annotated[RequestIdentity, Depends(get_request_identity)],
+) -> ApiResponse:
+    if identity.auth_method != "local":
+        raise HTTPException(status.HTTP_403_FORBIDDEN, "仅本地管理员可以修改本地密码")
+    settings = get_settings()
+    if not verify_password(body.current_password, settings.local_admin_password_hash):
+        raise HTTPException(status.HTTP_400_BAD_REQUEST, "当前密码错误")
+    if body.new_password == body.current_password:
+        raise HTTPException(status.HTTP_400_BAD_REQUEST, "新密码不能与当前密码相同")
+    persist_password_hash(hash_password(body.new_password))
+    get_settings.cache_clear()
+    return ApiResponse.ok({"message": "密码已更新"})
+
+
 @router.get("/me", response_model=ApiResponse[dict])
 async def read_identity(
     identity: Annotated[RequestIdentity, Depends(get_request_identity)],
diff --git a/backend/api/v1/internal_agent_runs.py b/backend/api/v1/internal_agent_runs.py
new file mode 100644
index 00000000..7307f252
--- /dev/null
+++ b/backend/api/v1/internal_agent_runs.py
@@ -0,0 +1,90 @@
+"""Token-authenticated API-owned WS dispatch for durable scheduled Agent runs."""
+
+from fastapi import APIRouter, Depends, HTTPException, status
+from pydantic import ValidationError
+from sqlalchemy import select
+
+from backend.database import AsyncSessionLocal
+from backend.models.operations_agent import (
+    OperationsAgentRun,
+    PublishedOperationsAgentVersion,
+)
+from backend.schemas.common import ApiResponse
+from backend.schemas.operations_agent import (
+    OperationsAgentRunRead,
+    agent_contract_from_model_configuration,
+    agent_runtime_binding_from_model_configuration,
+)
+from backend.security.internal_service import require_internal_service_token
+from backend.services.agent_runtime_selection import (
+    RuntimeSelectionError,
+    select_agent_runtime,
+)
+from backend.services.operations_agent_runtime_service import schedule_operations_agent_run
+
+router = APIRouter(
+    prefix="/internal/operations-agent-runs",
+    tags=["internal-operations-agent-runs"],
+)
+
+
+@router.post(
+    "/{run_id}/dispatch",
+    response_model=ApiResponse[OperationsAgentRunRead],
+    dependencies=[Depends(require_internal_service_token)],
+)
+async def dispatch_scheduled_agent_run(run_id: str) -> ApiResponse:
+    async with AsyncSessionLocal() as session:
+        run = await session.get(OperationsAgentRun, run_id)
+        if run is None:
+            raise HTTPException(status.HTTP_404_NOT_FOUND, "Operations Agent Run not found")
+        if run.trigger_type != "scheduled":
+            raise HTTPException(
+                status.HTTP_409_CONFLICT,
+                "Internal durable dispatch accepts scheduled runs only",
+            )
+        if run.status != "queued":
+            return ApiResponse.ok(OperationsAgentRunRead.model_validate(run))
+        version = await session.scalar(
+            select(PublishedOperationsAgentVersion).where(
+                PublishedOperationsAgentVersion.operations_agent_id
+                == run.operations_agent_id,
+                PublishedOperationsAgentVersion.version == run.published_version,
+            )
+        )
+        if version is None:
+            raise HTTPException(
+                status.HTTP_409_CONFLICT,
+                "Published Operations Agent version is missing",
+            )
+        try:
+            binding = agent_runtime_binding_from_model_configuration(
+                version.model_configuration
+            )
+            contract = agent_contract_from_model_configuration(version.model_configuration)
+        except ValidationError as exc:
+            raise HTTPException(
+                status.HTTP_409_CONFLICT,
+                "Published Operations Agent contract or runtime binding is invalid",
+            ) from exc
+        if binding is None or contract is None:
+            raise HTTPException(
+                status.HTTP_409_CONFLICT,
+                "Published Operations Agent requires contract and runtime binding",
+            )
+        try:
+            run.execution_binding = await select_agent_runtime(
+                session,
+                contract=contract,
+                binding=binding,
+            )
+        except RuntimeSelectionError as exc:
+            raise HTTPException(
+                status.HTTP_503_SERVICE_UNAVAILABLE,
+                str(exc),
+            ) from exc
+        await session.commit()
+        queued_response = OperationsAgentRunRead.model_validate(run)
+
+    schedule_operations_agent_run(run_id)
+    return ApiResponse.ok(queued_response)
diff --git a/backend/api/v1/internal_automations.py b/backend/api/v1/internal_automations.py
new file mode 100644
index 00000000..e2fe740b
--- /dev/null
+++ b/backend/api/v1/internal_automations.py
@@ -0,0 +1,35 @@
+"""Token-authenticated Automation scheduler boundary for Celery workers."""
+
+from datetime import timedelta
+
+from fastapi import APIRouter, Depends
+
+from backend.schemas.automation import (
+    AutomationSchedulerTickRequest,
+    AutomationSchedulerTickResult,
+)
+from backend.schemas.common import ApiResponse
+from backend.security.internal_service import require_internal_service_token
+from backend.services.automation_schedule_service import dispatch_due_automations
+from backend.services.scheduled_run_recovery import list_queued_scheduled_run_ids
+
+router = APIRouter(prefix="/internal/automations", tags=["internal-automations"])
+
+
+@router.post(
+    "/scheduler/tick",
+    response_model=ApiResponse[AutomationSchedulerTickResult],
+    dependencies=[Depends(require_internal_service_token)],
+)
+async def automation_scheduler_tick(body: AutomationSchedulerTickRequest) -> ApiResponse:
+    runs = await dispatch_due_automations(
+        body.fired_at - timedelta(seconds=90),
+        body.fired_at,
+    )
+    return ApiResponse.ok(
+        AutomationSchedulerTickResult(
+            run_ids=[run.id for run in runs],
+            occurrence_references=[run.trigger_reference or "" for run in runs],
+            queued_run_ids=await list_queued_scheduled_run_ids(),
+        )
+    )
diff --git a/backend/api/v1/internal_collaboration.py b/backend/api/v1/internal_collaboration.py
new file mode 100644
index 00000000..a278ac9b
--- /dev/null
+++ b/backend/api/v1/internal_collaboration.py
@@ -0,0 +1,150 @@
+"""Authenticated Studio Draft collaboration boundary for the Yjs sidecar."""
+
+from __future__ import annotations
+
+from collections.abc import Mapping
+
+from fastapi import APIRouter, Depends, HTTPException, status
+from pydantic import ValidationError
+from sqlalchemy import select
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from backend.api.v1.collaboration_schemas import (
+    CollaborationRoomRead,
+    CollaborationRoomRequest,
+    CollaborationSnapshotRead,
+    CollaborationSnapshotRequest,
+)
+from backend.api.v1.studio_helpers import canonicalize_studio_graph, get_workflow
+from backend.database import get_db
+from backend.models.studio import StudioWorkflowDraft
+from backend.schemas import workflow as workflow_schemas
+from backend.schemas.common import ApiResponse
+
+router = APIRouter(prefix="/internal/collaboration", tags=["internal-collaboration"])
+
+# Snapshot calls originate in the self-hosted collaboration sidecar. The API's
+# fleet-token middleware authenticates that service before this router runs.
+COLLABORATION_ACTOR_ID = "collaboration-service"
+_ROOM_LABELS = ("workspace", "project", "workflow")
+
+
+def collaboration_room(*, workspace_id: str, project_id: str, workflow_id: str) -> str:
+    """Return the sole permitted room name for a Studio workflow."""
+
+    return f"workspace:{workspace_id}:project:{project_id}:workflow:{workflow_id}"
+
+
+def parse_collaboration_room(room: str) -> tuple[str, str, str]:
+    """Parse a canonical Studio workflow room without accepting aliases."""
+
+    parts = room.split(":")
+    if (
+        len(parts) != 6
+        or tuple(parts[::2]) != _ROOM_LABELS
+        or any(not part or part != part.strip() for part in parts[1::2])
+    ):
+        raise ValueError("Invalid collaboration room")
+    workspace_id, project_id, workflow_id = parts[1::2]
+    if collaboration_room(
+        workspace_id=workspace_id,
+        project_id=project_id,
+        workflow_id=workflow_id,
+    ) != room:
+        raise ValueError("Invalid collaboration room")
+    return workspace_id, project_id, workflow_id
+
+
+async def _owned_workflow_from_room(
+    db: AsyncSession, room: str
+) -> tuple[str, str, str]:
+    try:
+        workspace_id, project_id, workflow_id = parse_collaboration_room(room)
+    except ValueError as exc:
+        raise HTTPException(status.HTTP_400_BAD_REQUEST, str(exc)) from exc
+    await get_workflow(db, workspace_id, project_id, workflow_id)
+    return workspace_id, project_id, workflow_id
+
+
+def _map_values(content: Mapping[str, dict], kind: str) -> list[dict]:
+    """Convert a Y.Map JSON object to graph entries while rejecting duplicate IDs."""
+
+    entries = list(content.values())
+    if any(entry.get("id") != map_key for map_key, entry in content.items()):
+        raise HTTPException(
+            status.HTTP_422_UNPROCESSABLE_ENTITY,
+            f"Collaboration {kind} map key does not match entry id",
+        )
+    ids = [entry.get("id") for entry in entries]
+    if any(not isinstance(entry_id, str) for entry_id in ids):
+        raise HTTPException(
+            status.HTTP_422_UNPROCESSABLE_ENTITY,
+            f"Collaboration {kind} map has an entry without a string id",
+        )
+    if len(ids) != len(set(ids)):
+        raise HTTPException(
+            status.HTTP_422_UNPROCESSABLE_ENTITY,
+            f"Collaboration {kind} map contains duplicate ids",
+        )
+    return entries
+
+
+@router.post(
+    "/authorize",
+    response_model=ApiResponse[CollaborationRoomRead],
+)
+async def authorize_collaboration_room(
+    body: CollaborationRoomRequest,
+    db: AsyncSession = Depends(get_db),
+) -> ApiResponse:
+    """Authorize one websocket room through the workflow ownership hierarchy."""
+
+    workspace_id, project_id, workflow_id = await _owned_workflow_from_room(db, body.room)
+    return ApiResponse.ok(
+        CollaborationRoomRead(
+            room=collaboration_room(
+                workspace_id=workspace_id,
+                project_id=project_id,
+                workflow_id=workflow_id,
+            )
+        )
+    )
+
+
+@router.post(
+    "/snapshot",
+    response_model=ApiResponse[CollaborationSnapshotRead],
+)
+async def snapshot_collaboration_draft(
+    body: CollaborationSnapshotRequest,
+    db: AsyncSession = Depends(get_db),
+) -> ApiResponse:
+    """Durably apply a Yjs graph snapshot without replacing Draft metadata."""
+
+    _, _, workflow_id = await _owned_workflow_from_room(db, body.room)
+    draft = await db.scalar(
+        select(StudioWorkflowDraft)
+        .where(StudioWorkflowDraft.workflow_id == workflow_id)
+        .with_for_update()
+    )
+    if draft is None:
+        raise HTTPException(status.HTTP_404_NOT_FOUND, "Workflow draft not found")
+
+    graph = {
+        **draft.graph,
+        "nodes": _map_values(body.data.nodes.content, "nodes"),
+        "edges": _map_values(body.data.edges.content, "edges"),
+    }
+    try:
+        workflow_schemas.WorkflowProject.model_validate({**graph, "id": workflow_id})
+    except ValidationError as exc:
+        raise HTTPException(
+            status.HTTP_422_UNPROCESSABLE_ENTITY,
+            "Invalid collaboration graph",
+        ) from exc
+
+    draft.graph = canonicalize_studio_graph(graph, workflow_id=workflow_id)
+    draft.revision += 1
+    draft.updated_by_user_id = COLLABORATION_ACTOR_ID
+    await db.flush()
+    return ApiResponse.ok(CollaborationSnapshotRead(revision=draft.revision))
diff --git a/backend/api/v1/nodes.py b/backend/api/v1/nodes.py
index b225d863..5b33c9d8 100644
--- a/backend/api/v1/nodes.py
+++ b/backend/api/v1/nodes.py
@@ -4,10 +4,11 @@
 Both HTTP-mode agents (center calls agent) and WS-mode agents (agent initiates
 reverse channel) register here and have their online/offline history tracked.
 """
-
+import io
 import logging
 import re
 import shlex
+import tarfile
 from datetime import UTC, datetime
 from pathlib import Path
 from typing import TYPE_CHECKING
@@ -21,7 +22,7 @@
     WebSocket,
     WebSocketDisconnect,
 )
-from fastapi.responses import PlainTextResponse
+from fastapi.responses import PlainTextResponse, Response
 from pydantic import BaseModel
 from sqlalchemy import func, select
 from sqlalchemy.ext.asyncio import AsyncSession
@@ -53,6 +54,7 @@ async def _upsert_node(
     ip: str | None = None,
     node_type: str = "chrome",
     runtimes: list[str] | None = None,
+    runtime_capabilities: dict[str, list[str]] | None = None,
 ) -> "EdgeNode":
     from backend.models.edge_node import EdgeNode
 
@@ -71,6 +73,8 @@ async def _upsert_node(
             node.ip = ip
         if runtimes is not None:
             node.runtimes = runtimes
+        if runtime_capabilities is not None:
+            node.runtime_capabilities = runtime_capabilities
     else:
         node = EdgeNode(
             url=url,
@@ -82,6 +86,7 @@ async def _upsert_node(
             last_seen_at=now,
             ip=ip,
             runtimes=runtimes,
+            runtime_capabilities=runtime_capabilities,
         )
         db.add(node)
     await db.flush()
@@ -157,6 +162,7 @@ class NodeRegisterRequest(BaseModel):
     label: str = ""
     agent_protocol: str = "http"
     runtimes: list[str] | None = None
+    runtime_capabilities: dict[str, list[str]] | None = None
     profile_kind: str = "authenticated"
 
 
@@ -198,6 +204,7 @@ async def register_node(
         ip,
         body.node_type,
         runtimes=body.runtimes,
+        runtime_capabilities=body.runtime_capabilities,
     )
     await _write_event(
         db,
@@ -209,6 +216,7 @@ async def register_node(
             "node_type": body.node_type,
             "protocol": body.agent_protocol,
             "runtimes": body.runtimes,
+            "runtime_capabilities": body.runtime_capabilities,
             "profile_kind": body.profile_kind,
         },
     )
@@ -394,6 +402,52 @@ async def get_opencli_runtime_patch() -> PlainTextResponse:
             )
     raise HTTPException(status_code=404, detail="OpenCLI runtime patch not packaged")
 
+@router.get("/install/agent-runtime.tar.gz")
+async def get_agent_runtime_bundle() -> Response:
+    """Serve the Python packages required by non-Docker Agents."""
+    source_roots = [
+        Path(__file__).parent.parent.parent.parent,
+        Path("/app"),
+    ]
+    source_root = next(
+        (
+            root
+            for root in source_roots
+            if (root / "backend" / "agent_runtimes").is_dir()
+        ),
+        None,
+    )
+    if source_root is None:
+        raise HTTPException(status_code=404, detail="Agent runtime package not packaged")
+
+    package_dirs = (
+        Path("backend/agent_runtimes"),
+        Path("backend/miniflow"),
+        Path("backend/security"),
+    )
+    archive = io.BytesIO()
+    with tarfile.open(fileobj=archive, mode="w:gz") as tar:
+        for package_dir in package_dirs:
+            package_root = source_root / package_dir
+            for path in sorted(package_root.rglob("*.py")):
+                info = tar.gettarinfo(
+                    str(path),
+                    arcname=path.relative_to(source_root).as_posix(),
+                )
+                info.mtime = 0
+                info.uid = 0
+                info.gid = 0
+                info.uname = ""
+                info.gname = ""
+                with path.open("rb") as stream:
+                    tar.addfile(info, stream)
+
+    return Response(
+        content=archive.getvalue(),
+        media_type="application/gzip",
+        headers={"Content-Disposition": 'attachment; filename="opencli-agent-runtime.tar.gz"'},
+    )
+
 
 @router.get("/install/agent.sh", response_class=PlainTextResponse)
 async def get_install_script(request: Request) -> PlainTextResponse:
@@ -750,6 +804,7 @@ async def node_ws_endpoint(ws: WebSocket) -> None:
         node_type = data.get("node_type", "chrome")
         label = data.get("label", "")
         runtimes = data.get("runtimes")
+        runtime_capabilities = data.get("runtime_capabilities")
         profile_kind = data.get("profile_kind", "authenticated")
 
         if not agent_url.startswith("http"):
@@ -766,6 +821,21 @@ async def node_ws_endpoint(ws: WebSocket) -> None:
         ):
             await ws.close(code=1008, reason="runtimes must be a list of strings")
             return
+        if runtime_capabilities is not None and (
+            not isinstance(runtime_capabilities, dict)
+            or any(
+                not isinstance(runtime_name, str)
+                or not isinstance(capabilities, list)
+                or not all(isinstance(capability, str) for capability in capabilities)
+                for runtime_name, capabilities in runtime_capabilities.items()
+            )
+            or set(runtime_capabilities) != set(runtimes or [])
+        ):
+            await ws.close(
+                code=1008,
+                reason="runtime_capabilities must map every advertised runtime to string names",
+            )
+            return
         if profile_kind not in ("anonymous", "authenticated"):
             await ws.close(
                 code=1008,
@@ -784,6 +854,7 @@ async def node_ws_endpoint(ws: WebSocket) -> None:
                     mode,
                     node_type=node_type,
                     runtimes=runtimes,
+                    runtime_capabilities=runtime_capabilities,
                 )
                 await _write_event(
                     db,
@@ -794,6 +865,7 @@ async def node_ws_endpoint(ws: WebSocket) -> None:
                         "node_type": node_type,
                         "protocol": "ws",
                         "profile_kind": profile_kind,
+                        "runtime_capabilities": runtime_capabilities,
                     },
                 )
                 # BrowserInstance compat
diff --git a/backend/api/v1/operations_agents.py b/backend/api/v1/operations_agents.py
index c58ddc0c..32b21900 100644
--- a/backend/api/v1/operations_agents.py
+++ b/backend/api/v1/operations_agents.py
@@ -5,9 +5,7 @@
 from sqlalchemy import select, update
 from sqlalchemy.ext.asyncio import AsyncSession
 
-from backend import ws_agent_manager
 from backend.database import get_db, queue_after_commit
-from backend.models.edge_node import EdgeNode
 from backend.models.identity import Team, WorkspaceRole
 from backend.models.operations_agent import (
     AgentPermissionProfile,
@@ -29,6 +27,7 @@
     OperationsAgentRead,
     OperationsAgentRunCreate,
     OperationsAgentRunRead,
+    OperationsAgentTeamRead,
     PublishedOperationsAgentVersionRead,
     agent_contract_from_model_configuration,
     agent_runtime_binding_from_model_configuration,
@@ -41,6 +40,10 @@
     get_workspace_access,
     require_permission,
 )
+from backend.services.agent_runtime_selection import (
+    RuntimeSelectionError,
+    select_agent_runtime,
+)
 from backend.services.operations_agent_runtime_service import (
     cancel_operations_agent_run,
     schedule_operations_agent_run,
@@ -120,6 +123,24 @@ def _read_agent(
     )
 
 
+@router.get("/teams", response_model=ApiResponse[list[OperationsAgentTeamRead]])
+async def list_operations_agent_teams(
+    workspace_id: str,
+    identity: RequestIdentity = Depends(get_request_identity),
+    db: AsyncSession = Depends(get_db),
+) -> ApiResponse:
+    access = await get_workspace_access(db, workspace_id, identity)
+    require_permission(access, WorkspacePermission.READ)
+    teams = (
+        await db.execute(
+            select(Team).where(Team.workspace_id == workspace_id).order_by(Team.name)
+        )
+    ).scalars().all()
+    return ApiResponse.ok(
+        [OperationsAgentTeamRead.model_validate(team) for team in teams]
+    )
+
+
 @router.get("", response_model=ApiResponse[list[OperationsAgentRead]])
 async def list_operations_agents(
     workspace_id: str,
@@ -151,14 +172,20 @@ async def create_operations_agent(
 ) -> ApiResponse:
     access = await get_workspace_access(db, workspace_id, identity)
     require_permission(access, WorkspacePermission.MANAGE_AGENT_IDENTITIES)
-    team = await db.scalar(
-        select(Team).where(Team.id == body.owning_team_id).where(Team.workspace_id == workspace_id)
-    )
+    team_query = select(Team).where(Team.workspace_id == workspace_id)
+    if body.owning_team_id is not None:
+        team_query = team_query.where(Team.id == body.owning_team_id)
+        team = await db.scalar(team_query)
+    else:
+        teams = (await db.execute(team_query.limit(2))).scalars().all()
+        team = teams[0] if len(teams) == 1 else None
     if team is None:
-        raise HTTPException(
-            status.HTTP_422_UNPROCESSABLE_CONTENT,
-            "Owning Team must belong to Workspace",
+        detail = (
+            "Owning Team must belong to Workspace"
+            if body.owning_team_id is not None
+            else "owning_team_id is required unless Workspace has exactly one Team"
         )
+        raise HTTPException(status.HTTP_422_UNPROCESSABLE_CONTENT, detail)
 
     agent = OperationsAgentIdentity(
         workspace_id=workspace_id,
@@ -445,36 +472,37 @@ async def start_agent_run(
     except ValidationError as exc:
         raise HTTPException(
             status.HTTP_409_CONFLICT,
-            "Published Agent Version contains an invalid AgentContractV1",
+            "Published Agent Version contains an invalid AgentContractV2",
         ) from exc
-    if runtime_binding is None:
+    if contract is None:
         raise HTTPException(
             status.HTTP_409_CONFLICT,
-            "Published Agent Version requires an AgentRuntimeBindingV1",
+            "Published Agent Version requires an AgentContractV2",
         )
-    runtime_node = await db.scalar(
-        select(EdgeNode).where(EdgeNode.url == runtime_binding.agent_url)
-    )
-    if (
-        runtime_node is None
-        or runtime_node.status != "online"
-        or runtime_node.protocol != "ws"
-        or runtime_binding.runtime not in (runtime_node.runtimes or [])
-        or not ws_agent_manager.is_connected(runtime_binding.agent_url)
-    ):
+    if runtime_binding is None:
         raise HTTPException(
             status.HTTP_409_CONFLICT,
-            "Agent Runtime Fleet precheck failed",
+            "Published Agent Version requires an AgentRuntimeBindingV2",
+        )
+    try:
+        execution_binding = await select_agent_runtime(
+            db,
+            contract=contract,
+            binding=runtime_binding,
         )
-    if contract is not None:
-        try:
-            validate_agent_contract_payload(contract, "input_schema", body.input_payload)
-            validate_agent_contract_payload(contract, "state_schema", body.state_payload)
-        except ValueError as exc:
-            raise HTTPException(
-                status.HTTP_422_UNPROCESSABLE_CONTENT,
-                f"Operations Agent run payload violates AgentContractV1: {exc}",
-            ) from exc
+    except RuntimeSelectionError as exc:
+        raise HTTPException(
+            status.HTTP_409_CONFLICT,
+            f"Agent Runtime Fleet precheck failed: {exc}",
+        ) from exc
+    try:
+        validate_agent_contract_payload(contract, "input_schema", body.input_payload)
+        validate_agent_contract_payload(contract, "state_schema", body.state_payload)
+    except ValueError as exc:
+        raise HTTPException(
+            status.HTTP_422_UNPROCESSABLE_CONTENT,
+            f"Operations Agent run payload violates AgentContractV2: {exc}",
+        ) from exc
     profile = await _get_profile(db, agent)
     if profile.mode == AgentProfileMode.LOW_RISK_AUTOMATIC:
         raise HTTPException(
@@ -491,6 +519,8 @@ async def start_agent_run(
         target_resource_id=body.target_resource_id,
         input_payload=body.input_payload,
         state_payload=body.state_payload,
+        execution_binding=execution_binding,
+        evidence_payload=None,
         status="queued",
         started_by_user_id=access.user_id,
     )
@@ -575,3 +605,4 @@ async def assign_agent_profile(
     agent.current_profile_version = profile.version
     await db.flush()
     return ApiResponse.ok(AgentProfileRead.model_validate(profile))
+
diff --git a/backend/api/v1/studio_schemas.py b/backend/api/v1/studio_schemas.py
index 1e8fb2b0..76af8453 100644
--- a/backend/api/v1/studio_schemas.py
+++ b/backend/api/v1/studio_schemas.py
@@ -3,7 +3,7 @@
 from datetime import datetime
 from typing import Literal
 
-from pydantic import BaseModel, ConfigDict, Field, field_serializer
+from pydantic import AliasChoices, BaseModel, ConfigDict, Field, field_serializer
 
 from backend.schemas import workflow as workflow_schemas
 from backend.schemas.common import UTCModel
@@ -84,9 +84,7 @@ class DraftRead(UTCModel):
     updated_at: datetime
 
     @field_serializer("graph")
-    def serialize_graph(
-        self, graph: workflow_schemas.WorkflowProject
-    ) -> dict[str, object]:
+    def serialize_graph(self, graph: workflow_schemas.WorkflowProject) -> dict[str, object]:
         return graph.model_dump(mode="json", exclude_none=True)
 
 
@@ -162,6 +160,10 @@ class PublishedWorkflowRunStart(BaseModel):
 
     inputs: dict = Field(default_factory=dict)
     response_mode: Literal["async"] = "async"
-    user: str = Field(min_length=1, max_length=255)
+    user: str = Field(
+        min_length=1,
+        max_length=255,
+        validation_alias=AliasChoices("user", "initiated_by"),
+    )
     request_id: str | None = Field(default=None, max_length=255)
     idempotency_key: str | None = Field(default=None, max_length=255)
diff --git a/backend/api/v1/studio_workflows.py b/backend/api/v1/studio_workflows.py
index 269e9cd6..4ad8bf12 100644
--- a/backend/api/v1/studio_workflows.py
+++ b/backend/api/v1/studio_workflows.py
@@ -23,7 +23,13 @@
     WorkflowCreate,
     WorkflowRead,
 )
-from backend.api.v1.workflows import dispatch_materialized_image_jobs
+from backend.api.v1.workflows import (
+    build_evidence_projection,
+    dispatch_materialized_image_jobs,
+    get_evidence_batch,
+    list_evidence_batches,
+    parse_projection_includes,
+)
 from backend.database import get_db
 from backend.models.studio import (
     StudioProject,
@@ -38,6 +44,7 @@
     get_workflow_run_checkpoint,
     get_workflow_run_projection,
     list_workflow_run_events,
+    replay_downstream_from_persisted_gaojixing_source,
     start_workflow_run,
 )
 
@@ -107,6 +114,35 @@ async def _project_runtime_scope(
     return workflow_names, {version.id: version.version for version in versions}
 
 
+async def _get_project_workflow_run(
+    db: AsyncSession,
+    *,
+    workspace_id: str,
+    project_id: str,
+    workflow_id: str,
+    run_id: str,
+) -> tuple[WorkflowRun, int]:
+    """Load one published run only through its owning project and version."""
+
+    await get_workflow(db, workspace_id, project_id, workflow_id)
+    result = await db.execute(
+        select(WorkflowRun, StudioWorkflowVersion.version)
+        .join(
+            StudioWorkflowVersion,
+            WorkflowRun.studio_workflow_version_id == StudioWorkflowVersion.id,
+        )
+        .where(
+            WorkflowRun.id == run_id,
+            WorkflowRun.workflow_id == workflow_id,
+            StudioWorkflowVersion.workflow_id == workflow_id,
+        )
+    )
+    scoped_run = result.one_or_none()
+    if scoped_run is None:
+        raise HTTPException(status.HTTP_404_NOT_FOUND, "Workflow run not found")
+    return scoped_run
+
+
 @router.get(
     "/workspaces/{workspace_id}/projects/{project_id}/workflows",
     response_model=ApiResponse[list[WorkflowRead]],
@@ -197,8 +233,7 @@ async def get_project_runtime_summary(
             blocked_runs=sum(1 for row in aggregate_rows if row.status in blocked),
             running_runs=sum(1 for row in aggregate_rows if row.status in running),
             total_events=sum(
-                int((row.projection or {}).get("eventCount", 0))
-                for row in aggregate_rows
+                int((row.projection or {}).get("eventCount", 0)) for row in aggregate_rows
             ),
             recent_logs=[
                 _runtime_log(
@@ -254,12 +289,7 @@ async def list_project_runtime_logs(
             )
         )
 
-    total = int(
-        await db.scalar(
-            select(func.count()).select_from(WorkflowRun).where(*filters)
-        )
-        or 0
-    )
+    total = int(await db.scalar(select(func.count()).select_from(WorkflowRun).where(*filters)) or 0)
     rows = list(
         (
             await db.execute(
@@ -292,10 +322,7 @@ async def list_project_runtime_logs(
 
 
 @router.post(
-    (
-        "/workspaces/{workspace_id}/projects/{project_id}/workflows/{workflow_id}"
-        "/runs"
-    ),
+    ("/workspaces/{workspace_id}/projects/{project_id}/workflows/{workflow_id}/runs"),
     response_model=ApiResponse[workflow_schemas.WorkflowRunProjection],
     status_code=202,
 )
@@ -379,6 +406,55 @@ async def start_published_workflow_run(
     return ApiResponse.ok(projection)
 
 
+@router.post(
+    (
+        "/workspaces/{workspace_id}/projects/{project_id}/workflows/{workflow_id}"
+        "/runs/{run_id}/downstream-replay"
+    ),
+    response_model=ApiResponse[workflow_schemas.WorkflowRunProjection],
+    status_code=202,
+)
+async def replay_persisted_gaojixing_source_downstream(
+    workspace_id: str,
+    project_id: str,
+    workflow_id: str,
+    run_id: str,
+    db: AsyncSession = Depends(get_db),
+) -> ApiResponse:
+    """Replay normalization through sink from completed persisted Gaojixing evidence."""
+
+    workflow = await get_workflow(db, workspace_id, project_id, workflow_id)
+    if workflow.current_published_version is None:
+        raise HTTPException(
+            status.HTTP_409_CONFLICT, "Workflow must be published before downstream replay"
+        )
+    version = await db.scalar(
+        select(StudioWorkflowVersion).where(
+            StudioWorkflowVersion.workflow_id == workflow_id,
+            StudioWorkflowVersion.version == workflow.current_published_version,
+        )
+    )
+    if version is None:
+        raise HTTPException(status.HTTP_409_CONFLICT, "Published workflow version is unavailable")
+    await _get_project_workflow_run(
+        db,
+        workspace_id=workspace_id,
+        project_id=project_id,
+        workflow_id=workflow_id,
+        run_id=run_id,
+    )
+    try:
+        projection = await replay_downstream_from_persisted_gaojixing_source(
+            run_id,
+            expected_workflow_id=workflow_id,
+            expected_studio_workflow_version_id=version.id,
+            session=db,
+        )
+    except ValueError as exc:
+        raise HTTPException(status.HTTP_409_CONFLICT, str(exc)) from exc
+    return ApiResponse.ok(projection)
+
+
 @router.get(
     (
         "/workspaces/{workspace_id}/projects/{project_id}/workflows/{workflow_id}"
@@ -397,10 +473,13 @@ async def get_project_runtime_trace(
 ) -> ApiResponse:
     """Return one project-owned run trace without opening generic run access."""
 
-    await get_workflow(db, workspace_id, project_id, workflow_id)
-    row = await db.get(WorkflowRun, run_id)
-    if row is None or row.workflow_id != workflow_id:
-        raise HTTPException(status.HTTP_404_NOT_FOUND, "Workflow run not found")
+    row, workflow_version = await _get_project_workflow_run(
+        db,
+        workspace_id=workspace_id,
+        project_id=project_id,
+        workflow_id=workflow_id,
+        run_id=run_id,
+    )
     projection = await get_workflow_run_projection(run_id, session=db)
     checkpoint = await get_workflow_run_checkpoint(run_id, session=db)
     events = await list_workflow_run_events(
@@ -413,13 +492,6 @@ async def get_project_runtime_trace(
         raise HTTPException(status.HTTP_404_NOT_FOUND, "Workflow run not found")
     request = row.request or {}
     input_payload = request.get("input") or {}
-    workflow_version = None
-    if row.studio_workflow_version_id:
-        workflow_version = await db.scalar(
-            select(StudioWorkflowVersion.version).where(
-                StudioWorkflowVersion.id == row.studio_workflow_version_id
-            )
-        )
     return ApiResponse.ok(
         ProjectRuntimeTraceRead(
             workflow_version=workflow_version,
@@ -440,6 +512,187 @@ async def get_project_runtime_trace(
     )
 
 
+@router.get(
+    ("/workspaces/{workspace_id}/projects/{project_id}/workflows/{workflow_id}/runs/{run_id}"),
+    response_model=ApiResponse[workflow_schemas.WorkflowRunProjection],
+)
+async def get_project_workflow_run_projection(
+    workspace_id: str,
+    project_id: str,
+    workflow_id: str,
+    run_id: str,
+    db: AsyncSession = Depends(get_db),
+) -> ApiResponse:
+    """Return a projection through the owning published workflow."""
+
+    await _get_project_workflow_run(
+        db,
+        workspace_id=workspace_id,
+        project_id=project_id,
+        workflow_id=workflow_id,
+        run_id=run_id,
+    )
+    projection = await get_workflow_run_projection(run_id, session=db)
+    if projection is None:
+        raise HTTPException(status.HTTP_404_NOT_FOUND, "Workflow run not found")
+    return ApiResponse.ok(projection)
+
+
+@router.get(
+    (
+        "/workspaces/{workspace_id}/projects/{project_id}/workflows/{workflow_id}"
+        "/runs/{run_id}/events"
+    ),
+    response_model=ApiResponse[list[workflow_schemas.WorkflowNodeRunEvent]],
+)
+async def list_project_workflow_run_events(
+    workspace_id: str,
+    project_id: str,
+    workflow_id: str,
+    run_id: str,
+    after_sequence: int | None = Query(default=None, ge=0, alias="afterSequence"),
+    limit: int | None = Query(default=None, ge=1, le=500),
+    db: AsyncSession = Depends(get_db),
+) -> ApiResponse:
+    """Replay persisted events through the owning published workflow."""
+
+    await _get_project_workflow_run(
+        db,
+        workspace_id=workspace_id,
+        project_id=project_id,
+        workflow_id=workflow_id,
+        run_id=run_id,
+    )
+    events = await list_workflow_run_events(
+        run_id,
+        session=db,
+        after_sequence=after_sequence,
+        limit=limit,
+    )
+    if events is None:
+        raise HTTPException(status.HTTP_404_NOT_FOUND, "Workflow run not found")
+    return ApiResponse.ok(events)
+
+
+@router.get(
+    (
+        "/workspaces/{workspace_id}/projects/{project_id}/workflows/{workflow_id}"
+        "/runs/{run_id}/evidence-batches"
+    ),
+    response_model=ApiResponse[workflow_schemas.WorkflowEvidenceBatchListResponse],
+)
+async def list_project_workflow_evidence_batches(
+    workspace_id: str,
+    project_id: str,
+    workflow_id: str,
+    run_id: str,
+    node_id: str | None = Query(default=None),
+    source_group: str | None = Query(default=None),
+    cursor: str | None = Query(default=None),
+    limit: int = Query(default=50, ge=1, le=200),
+    db: AsyncSession = Depends(get_db),
+) -> ApiResponse:
+    """List compact evidence metadata through the owning published workflow."""
+
+    await _get_project_workflow_run(
+        db,
+        workspace_id=workspace_id,
+        project_id=project_id,
+        workflow_id=workflow_id,
+        run_id=run_id,
+    )
+    projection = await get_workflow_run_projection(run_id, session=db)
+    if projection is None:
+        raise HTTPException(status.HTTP_404_NOT_FOUND, "Workflow run not found")
+    try:
+        batches = list_evidence_batches(
+            projection,
+            node_id=node_id,
+            source_group=source_group,
+            cursor=cursor,
+            limit=limit,
+        )
+    except ValueError as exc:
+        raise HTTPException(status.HTTP_400_BAD_REQUEST, str(exc)) from exc
+    return ApiResponse.ok(batches)
+
+
+@router.get(
+    (
+        "/workspaces/{workspace_id}/projects/{project_id}/workflows/{workflow_id}"
+        "/runs/{run_id}/evidence-batches/{batch_id}"
+    ),
+    response_model=ApiResponse[workflow_schemas.WorkflowEvidenceBatchDetail],
+)
+async def get_project_workflow_evidence_batch(
+    workspace_id: str,
+    project_id: str,
+    workflow_id: str,
+    run_id: str,
+    batch_id: str,
+    db: AsyncSession = Depends(get_db),
+) -> ApiResponse:
+    """Return one evidence manifest through the owning published workflow."""
+
+    await _get_project_workflow_run(
+        db,
+        workspace_id=workspace_id,
+        project_id=project_id,
+        workflow_id=workflow_id,
+        run_id=run_id,
+    )
+    projection = await get_workflow_run_projection(run_id, session=db)
+    if projection is None:
+        raise HTTPException(status.HTTP_404_NOT_FOUND, "Workflow run not found")
+    batch = get_evidence_batch(projection, batch_id)
+    if batch is None:
+        raise HTTPException(status.HTTP_404_NOT_FOUND, "Evidence batch not found")
+    return ApiResponse.ok(batch)
+
+
+@router.get(
+    (
+        "/workspaces/{workspace_id}/projects/{project_id}/workflows/{workflow_id}"
+        "/runs/{run_id}/projection"
+    ),
+    response_model=ApiResponse[workflow_schemas.WorkflowEvidenceProjection],
+)
+async def get_project_workflow_evidence_projection(
+    workspace_id: str,
+    project_id: str,
+    workflow_id: str,
+    run_id: str,
+    node_id: str | None = Query(default=None),
+    source_group: str | None = Query(default=None),
+    include: str | None = Query(default=None),
+    db: AsyncSession = Depends(get_db),
+) -> ApiResponse:
+    """Project evidence through the owning published workflow."""
+
+    await _get_project_workflow_run(
+        db,
+        workspace_id=workspace_id,
+        project_id=project_id,
+        workflow_id=workflow_id,
+        run_id=run_id,
+    )
+    projection = await get_workflow_run_projection(run_id, session=db)
+    if projection is None:
+        raise HTTPException(status.HTTP_404_NOT_FOUND, "Workflow run not found")
+    try:
+        includes = parse_projection_includes(include)
+    except ValueError as exc:
+        raise HTTPException(status.HTTP_400_BAD_REQUEST, str(exc)) from exc
+    return ApiResponse.ok(
+        build_evidence_projection(
+            projection,
+            node_id=node_id,
+            source_group=source_group,
+            includes=includes,
+        )
+    )
+
+
 @router.post(
     "/workspaces/{workspace_id}/projects/{project_id}/workflows",
     response_model=ApiResponse[WorkflowRead],
diff --git a/backend/api/v1/system.py b/backend/api/v1/system.py
index 06d0d0f4..719966a9 100644
--- a/backend/api/v1/system.py
+++ b/backend/api/v1/system.py
@@ -1,16 +1,20 @@
-"""System configuration endpoint."""
+"""System-level configuration and deployment status endpoints."""
+
+from __future__ import annotations
 
 import os
 import re
+from typing import Literal
 
-from fastapi import APIRouter, HTTPException
-from pydantic import BaseModel
+from fastapi import APIRouter
+from pydantic import BaseModel, Field
 
 from backend.config import get_settings
 from backend.schemas.common import ApiResponse
 
 router = APIRouter(prefix="/system", tags=["system"])
 
+
 def _resolve_env_path() -> str:
     if explicit := os.environ.get("ENV_FILE_PATH"):
         return explicit
@@ -20,15 +24,14 @@ def _resolve_env_path() -> str:
     ]:
         if os.path.exists(candidate):
             return candidate
-    # Fallback: project root .env (will be created if missing)
     return os.path.join(os.path.dirname(__file__), "..", "..", "..", ".env")
 
 
 def _update_env_file(key: str, value: str) -> None:
     path = _resolve_env_path()
     try:
-        with open(path) as f:
-            content = f.read()
+        with open(path, encoding="utf-8") as env_file:
+            content = env_file.read()
     except FileNotFoundError:
         content = ""
     new_line = f"{key}={value}"
@@ -37,44 +40,77 @@ def _update_env_file(key: str, value: str) -> None:
         content = re.sub(pattern, new_line, content, flags=re.MULTILINE)
     else:
         content = content.rstrip("\n") + f"\n{new_line}\n"
-    with open(path, "w") as f:
-        f.write(content)
+    with open(path, "w", encoding="utf-8") as env_file:
+        env_file.write(content)
 
 
 class ConfigPatch(BaseModel):
-    collection_mode: str | None = None
+    collection_mode: Literal["local", "agent"] | None = None
+    collection_orchestrator: Literal["admin", "iii"] | None = None
+    local_max_concurrent_pipelines: int | None = Field(default=None, ge=1, le=64)
+    opencli_timeout: int | None = Field(default=None, ge=1, le=3600)
+    default_timezone: str | None = Field(default=None, min_length=1, max_length=64)
+    public_url: str | None = Field(default=None, max_length=2048)
+    fleet_network_provider: Literal["lan", "netbird", "wireguard", "ssh", "custom"] | None = None
+    netbird_mode: Literal["off", "host", "docker"] | None = None
+    opencli_cdp_endpoint: str | None = Field(default=None, min_length=1, max_length=2048)
+    agent_pool_endpoints: str | None = Field(default=None, max_length=8192)
+    llm_request_timeout_seconds: int | None = Field(default=None, ge=1, le=3600)
+    llm_max_concurrency: int | None = Field(default=None, ge=1, le=64)
+    control_mode: Literal["advisory", "automatic"] | None = None
+    control_kill_switch: bool | None = None
+
+
+def _system_payload() -> dict:
+    settings = get_settings()
+    return {
+        "app_name": settings.app_name,
+        "app_env": settings.app_env,
+        "debug": settings.debug,
+        "collection_mode": settings.collection_mode,
+        "collection_orchestrator": settings.collection_orchestrator,
+        "task_executor": settings.task_executor,
+        "local_max_concurrent_pipelines": settings.local_max_concurrent_pipelines,
+        "opencli_timeout": settings.opencli_timeout,
+        "default_timezone": settings.default_timezone,
+        "public_url": settings.public_url,
+        "fleet_network_provider": settings.fleet_network_provider,
+        "netbird_mode": settings.netbird_mode,
+        "opencli_cdp_endpoint": settings.opencli_cdp_endpoint,
+        "agent_pool_endpoints": [
+            endpoint.strip()
+            for endpoint in settings.agent_pool_endpoints.split(",")
+            if endpoint.strip()
+        ],
+        "effective_cdp_endpoints": settings.cdp_endpoints,
+        "llm_request_timeout_seconds": settings.llm_request_timeout_seconds,
+        "llm_max_concurrency": settings.llm_max_concurrency,
+        "control_mode": settings.control_mode,
+        "control_kill_switch": settings.control_kill_switch,
+        "image_tag": settings.image_tag,
+        "database_kind": "sqlite" if settings.is_sqlite else "postgresql",
+        "api_auth_configured": bool(settings.api_auth_token),
+        "oidc_configured": bool(os.getenv("OIDC_ISSUER") and os.getenv("OIDC_AUDIENCE")),
+        "smtp_configured": bool(settings.smtp_host and settings.smtp_from),
+        "credential_encryption_configured": bool(settings.credential_encryption_key),
+    }
 
 
 @router.get("/config", response_model=ApiResponse[dict])
 async def get_config() -> ApiResponse:
-    s = get_settings()
-    return ApiResponse.ok(
-        {
-            "collection_mode": s.collection_mode,
-            "task_executor": s.task_executor,
-            "image_tag": s.image_tag,
-        }
-    )
+    return ApiResponse.ok(_system_payload())
 
 
 @router.patch("/config", response_model=ApiResponse[dict])
 async def update_config(body: ConfigPatch) -> ApiResponse:
-    if body.collection_mode is not None:
-        if body.collection_mode not in ("local", "agent"):
-            raise HTTPException(
-                status_code=400, detail="collection_mode must be 'local' or 'agent'"
-            )
-        _update_env_file("COLLECTION_MODE", body.collection_mode)
-        # Also update the process env var so pydantic-settings picks up the new value
-        # (env vars take priority over .env file in pydantic-settings v2)
-        os.environ["COLLECTION_MODE"] = body.collection_mode
-        get_settings.cache_clear()
-
-    s = get_settings()
-    return ApiResponse.ok(
-        {
-            "collection_mode": s.collection_mode,
-            "task_executor": s.task_executor,
-            "image_tag": s.image_tag,
-        }
-    )
+    updates = body.model_dump(exclude_none=True)
+    if not updates:
+        return ApiResponse.ok(_system_payload())
+
+    for key, value in updates.items():
+        env_key = key.upper()
+        env_value = str(value).lower() if isinstance(value, bool) else str(value)
+        _update_env_file(env_key, env_value)
+        os.environ[env_key] = env_value
+    get_settings.cache_clear()
+    return ApiResponse.ok(_system_payload())
diff --git a/backend/api/v1/workspaces.py b/backend/api/v1/workspaces.py
index 17969488..79a0bded 100644
--- a/backend/api/v1/workspaces.py
+++ b/backend/api/v1/workspaces.py
@@ -3,7 +3,7 @@
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from backend.database import get_db
-from backend.models.identity import User, Workspace, WorkspaceMembership, WorkspaceRole
+from backend.models.identity import Team, User, Workspace, WorkspaceMembership, WorkspaceRole
 from backend.models.workflow import Project
 from backend.schemas.common import ApiResponse
 from backend.schemas.workflow_asset import ProjectRead
@@ -54,6 +54,51 @@ async def _get_or_create_user(
         raise HTTPException(status.HTTP_409_CONFLICT, "Disabled user cannot join a Workspace")
     return user
 
+async def _ensure_local_admin_workspace(
+    db: AsyncSession,
+    identity: RequestIdentity,
+) -> None:
+    if identity.auth_method != "local":
+        return
+
+    user = await db.scalar(select(User).where(User.subject == identity.subject))
+    if user is None:
+        user = User(
+            subject=identity.subject,
+            display_name=identity.name or "本地管理员",
+        )
+        db.add(user)
+        await db.flush()
+
+    workspace = await db.scalar(select(Workspace).where(Workspace.slug == "opencli-default"))
+    if workspace is None:
+        workspace = Workspace(name="OpenCLI 工作区", slug="opencli-default")
+        db.add(workspace)
+        await db.flush()
+
+    membership = await db.scalar(
+        select(WorkspaceMembership)
+        .where(WorkspaceMembership.workspace_id == workspace.id)
+        .where(WorkspaceMembership.user_id == user.id)
+    )
+    if membership is None:
+        db.add(
+            WorkspaceMembership(
+                workspace_id=workspace.id,
+                user_id=user.id,
+                role=WorkspaceRole.ADMIN,
+            )
+        )
+
+    team = await db.scalar(
+        select(Team)
+        .where(Team.workspace_id == workspace.id)
+        .where(Team.slug == "default")
+    )
+    if team is None:
+        db.add(Team(workspace_id=workspace.id, name="默认团队", slug="default"))
+        await db.flush()
+
 
 @router.get(
     "/governance/workspaces",
@@ -63,6 +108,7 @@ async def list_accessible_workspaces(
     identity: RequestIdentity = Depends(get_request_identity),
     db: AsyncSession = Depends(get_db),
 ) -> ApiResponse:
+    await _ensure_local_admin_workspace(db, identity)
     rows = (
         (
             await db.execute(
diff --git a/backend/automation_schedule.py b/backend/automation_schedule.py
new file mode 100644
index 00000000..a53a743e
--- /dev/null
+++ b/backend/automation_schedule.py
@@ -0,0 +1,74 @@
+"""Timezone-aware occurrence matching for the Automation schedule contract."""
+
+from dataclasses import dataclass
+from datetime import UTC, datetime, timedelta
+import re
+from zoneinfo import ZoneInfo, ZoneInfoNotFoundError
+
+_SCHEDULE_RE = re.compile(r"^(daily|weekdays|weekly)@(\d{2}):(\d{2})$")
+
+
+@dataclass(frozen=True)
+class AutomationScheduleSpec:
+    kind: str
+    hour: int | None = None
+    minute: int = 0
+
+
+def parse_automation_schedule(value: str) -> AutomationScheduleSpec:
+    if value == "hourly":
+        return AutomationScheduleSpec(kind="hourly")
+    match = _SCHEDULE_RE.fullmatch(value)
+    if match is None:
+        raise ValueError(
+            "schedule must be hourly, daily@HH:MM, weekdays@HH:MM, or weekly@HH:MM"
+        )
+    kind, raw_hour, raw_minute = match.groups()
+    hour = int(raw_hour)
+    minute = int(raw_minute)
+    if hour > 23 or minute > 59:
+        raise ValueError("schedule time must be a valid 24-hour HH:MM value")
+    return AutomationScheduleSpec(kind=kind, hour=hour, minute=minute)
+
+
+def parse_automation_timezone(value: str) -> ZoneInfo:
+    try:
+        return ZoneInfo(value)
+    except ZoneInfoNotFoundError as exc:
+        raise ValueError(f"unknown automation timezone: {value}") from exc
+
+
+def automation_fire_times(
+    schedule: str,
+    timezone_name: str,
+    window_start: datetime,
+    window_end: datetime,
+) -> list[datetime]:
+    """Match UTC minutes after local conversion for deterministic DST behavior."""
+    if window_start.tzinfo is None or window_end.tzinfo is None:
+        raise ValueError("automation scheduler windows must be timezone-aware")
+    if window_end <= window_start:
+        return []
+    spec = parse_automation_schedule(schedule)
+    timezone = parse_automation_timezone(timezone_name)
+    cursor = window_start.astimezone(UTC).replace(second=0, microsecond=0)
+    if cursor <= window_start.astimezone(UTC):
+        cursor += timedelta(minutes=1)
+    end_utc = window_end.astimezone(UTC)
+    fires: list[datetime] = []
+    while cursor <= end_utc:
+        local = cursor.astimezone(timezone)
+        matches_time = (
+            local.minute == 0
+            if spec.kind == "hourly"
+            else local.hour == spec.hour and local.minute == spec.minute
+        )
+        matches_day = (
+            spec.kind not in {"weekdays", "weekly"}
+            or (spec.kind == "weekdays" and local.weekday() < 5)
+            or (spec.kind == "weekly" and local.weekday() == 0)
+        )
+        if matches_time and matches_day:
+            fires.append(cursor)
+        cursor += timedelta(minutes=1)
+    return fires
diff --git a/backend/channels/doubao_research_channel.py b/backend/channels/doubao_research_channel.py
index 6a9f430a..e5c40036 100644
--- a/backend/channels/doubao_research_channel.py
+++ b/backend/channels/doubao_research_channel.py
@@ -1,6 +1,6 @@
 """Collect a cited Doubao research answer through the installed OpenCLI adapter."""
 
-import os
+import asyncio
 import re
 from typing import Any
 
@@ -18,6 +18,34 @@
     "验证码",
 )
 
+_AUTHENTICATED_LOGIN_VALUES = {"true", "yes", "logged_in", "authenticated"}
+_LOGGED_OUT_LOGIN_VALUES = {"false", "no", "logged_out", "unauthenticated"}
+_DOUBAO_WORKSPACE_URL_RE = re.compile(r"^https://(?:www\.)?doubao\.com/chat(?:[/?#]|$)", re.I)
+_ACCOUNT_IDENTITY_KEYS = ("id", "uid", "user_id", "email", "phone", "name", "nickname", "account")
+
+
+def _row_value(row: dict[str, Any], *keys: str) -> str:
+    """Read an OpenCLI table key without depending on its display casing."""
+    values = {str(key).lower(): value for key, value in row.items()}
+    return next((str(values[key]).strip() for key in keys if values.get(key) is not None), "")
+
+
+def _is_authenticated_doubao_workspace(row: dict[str, Any]) -> bool:
+    """Require an unambiguous Doubao chat workspace, never just a login redirect."""
+    url = _row_value(row, "url")
+    title = _row_value(row, "title")
+    return bool(_DOUBAO_WORKSPACE_URL_RE.match(url) and "豆包" in title)
+
+
+def _has_authenticated_account(rows: list[dict[str, Any]]) -> bool:
+    """Accept exactly one non-empty identity from the provider's read-only whoami output."""
+    identities = {
+        _row_value(row, *_ACCOUNT_IDENTITY_KEYS)
+        for row in rows
+        if _row_value(row, *_ACCOUNT_IDENTITY_KEYS)
+    }
+    return len(identities) == 1
+
 
 def _citations(text: str) -> list[dict[str, str]]:
     """Extract and de-duplicate URLs while preserving the answer's order."""
@@ -70,7 +98,7 @@ async def _run_doubao_command(command: list[str]) -> tuple[int, str, str]:
     """Late import avoids the channel registry's legacy OpenCLI import cycle."""
     from backend.channels.opencli_channel import _run_opencli
 
-    return await _run_opencli(command, os.environ.copy())
+    return await _run_opencli(command)
 
 
 def _opencli_binary() -> str:
@@ -105,6 +133,13 @@ async def collect(self, config: dict[str, Any], parameters: dict[str, Any]) -> C
             return ChannelResult.fail("'question' is required for doubao_research channel")
 
         extract_citations = bool(config.get("extract_citations", True))
+        try:
+            settle_seconds = float(config.get("settle_seconds", 0))
+        except (TypeError, ValueError):
+            return ChannelResult.fail("'settle_seconds' must be a non-negative number")
+        if settle_seconds < 0:
+            return ChannelResult.fail("'settle_seconds' must be a non-negative number")
+        site_session = str(config.get("site_session", "ephemeral"))
         # Prompt wording belongs to the research brief.  Appending a fixed
         # instruction made the browser adapter lose its active conversation;
         # extract URLs from the returned answer without altering the query.
@@ -117,7 +152,7 @@ async def collect(self, config: dict[str, Any], parameters: dict[str, Any]) -> C
             "-f",
             "json",
             "--site-session",
-            str(config.get("site_session", "ephemeral")),
+            site_session,
         ]
         try:
             returncode, stdout, stderr = await _run_doubao_command(command)
@@ -147,6 +182,31 @@ async def collect(self, config: dict[str, Any], parameters: dict[str, Any]) -> C
         if not answer:
             return ChannelResult.fail("Doubao returned no assistant text")
 
+        # OpenCLI 1.8.5 can return after Doubao creates its first progress
+        # message while deep research continues in the same conversation.
+        # The Gaojixing capability opts into one delayed, read-only snapshot so
+        # evidence stores the completed answer instead of that progress text.
+        if settle_seconds:
+            await asyncio.sleep(settle_seconds)
+            read_command = [
+                _opencli_binary(),
+                "doubao",
+                "read",
+                "-f",
+                "json",
+                "--site-session",
+                site_session,
+            ]
+            try:
+                read_rc, read_stdout, _ = await _run_doubao_command(read_command)
+                settled_answer = (
+                    _answer(_parse_opencli_rows(read_stdout)) if read_rc == 0 else ""
+                )
+                if len(settled_answer) > len(answer):
+                    answer = settled_answer
+            except Exception:
+                pass
+
         # Best-effort conversation URL: `doubao status -f json` exposes the
         # active chat id (https://www.doubao.com/chat/<id>).  This is a
         # read-only query against the same browser session; a failure here
@@ -160,7 +220,7 @@ async def collect(self, config: dict[str, Any], parameters: dict[str, Any]) -> C
                 "-f",
                 "json",
                 "--site-session",
-                str(config.get("site_session", "ephemeral")),
+                site_session,
             ]
             try:
                 rc, so, se = await _run_doubao_command(status_command)
@@ -189,6 +249,94 @@ async def collect(self, config: dict[str, Any], parameters: dict[str, Any]) -> C
             citation_capture="answer_url_extraction" if extract_citations else "disabled",
         )
 
+    async def health_check(
+        self,
+        config: dict[str, Any] | None = None,
+        source_id: str | None = None,
+    ) -> bool:
+        """Return whether the selected session satisfies live capture readiness."""
+        del source_id
+        return await self.readiness_code(config) is None
+
+    async def readiness_code(self, config: dict[str, Any] | None = None) -> str | None:
+        """Classify a failed live-session probe without exposing provider secrets."""
+        session = str((config or {}).get("site_session", "persistent"))
+        status_command = [
+            _opencli_binary(),
+            "doubao",
+            "status",
+            "-f",
+            "json",
+            "--site-session",
+            session,
+        ]
+        try:
+            returncode, stdout, stderr = await _run_doubao_command(status_command)
+        except FileNotFoundError:
+            return "adapter_missing"
+        except (TimeoutError, OSError):
+            return "session_unavailable"
+        if _is_captcha_block(stderr, stdout):
+            return "captcha_challenge"
+        if returncode:
+            return "session_unavailable"
+        try:
+            rows = _parse_opencli_rows(stdout)
+        except Exception:
+            return "session_unavailable"
+
+        connected_rows = [
+            row
+            for row in rows
+            if _row_value(row, "status").lower() in {"connected", "ready", "available"}
+        ]
+        if not connected_rows:
+            return "session_unavailable"
+        logins = [_row_value(row, "login").lower() for row in connected_rows]
+        if any(login in _LOGGED_OUT_LOGIN_VALUES for login in logins):
+            return "authentication_required"
+        for row, login in zip(connected_rows, logins, strict=True):
+            if login not in _AUTHENTICATED_LOGIN_VALUES:
+                continue
+            if _row_value(row, "url", "title") and not _is_authenticated_doubao_workspace(row):
+                return "session_unavailable"
+            return None
+
+        workspace_rows = [
+            row
+            for row, login in zip(connected_rows, logins, strict=True)
+            if login in {"", "unknown"} and _is_authenticated_doubao_workspace(row)
+        ]
+        if len(workspace_rows) != 1:
+            return "session_unavailable"
+        whoami_command = [
+            _opencli_binary(),
+            "doubao",
+            "whoami",
+            "-f",
+            "json",
+            "--site-session",
+            session,
+        ]
+        try:
+            returncode, stdout, stderr = await _run_doubao_command(whoami_command)
+        except FileNotFoundError:
+            return "adapter_missing"
+        except (TimeoutError, OSError):
+            return "session_unavailable"
+        if _is_captcha_block(stderr, stdout):
+            return "captcha_challenge"
+        if returncode:
+            return "authentication_required"
+        try:
+            return (
+                None
+                if _has_authenticated_account(_parse_opencli_rows(stdout))
+                else "authentication_required"
+            )
+        except Exception:
+            return "authentication_required"
+
     async def validate_config(self, config: dict[str, Any]) -> list[str]:
         return (
             []
diff --git a/backend/channels/opencli_channel.py b/backend/channels/opencli_channel.py
index c86c9f8b..282e81a4 100644
--- a/backend/channels/opencli_channel.py
+++ b/backend/channels/opencli_channel.py
@@ -522,7 +522,9 @@ async def _cleanup_cdp_tabs(cdp_endpoint: str, pre_existing_ids: set[str]) -> No
         logger.warning("cleanup: could not close CDP tabs at %s: %s", cdp_endpoint, exc)
 
 
-async def _run_opencli(cmd: list[str], env: dict) -> tuple[int, str, str]:
+async def _run_opencli(
+    cmd: list[str], env: dict[str, str] | None = None
+) -> tuple[int, str, str]:
     """Run opencli subprocess, return (returncode, stdout, stderr).
 
     Kills the process on timeout before re-raising so it doesn't linger.
@@ -533,7 +535,7 @@ async def _run_opencli(cmd: list[str], env: dict) -> tuple[int, str, str]:
             *cmd,
             stdout=asyncio.subprocess.PIPE,
             stderr=asyncio.subprocess.PIPE,
-            env=env,
+            env=env if env is not None else os.environ.copy(),
             **_process_group_kwargs(),
         )
         from backend.config import get_settings
diff --git a/backend/config.py b/backend/config.py
index 36b32378..093b2754 100644
--- a/backend/config.py
+++ b/backend/config.py
@@ -3,6 +3,8 @@
 
 from pydantic_settings import BaseSettings, SettingsConfigDict
 
+from backend.security.local_auth import DEFAULT_LOCAL_ADMIN_PASSWORD_HASH
+
 
 class Settings(BaseSettings):
     model_config = SettingsConfigDict(
@@ -46,6 +48,8 @@ class Settings(BaseSettings):
     redis_url: str = "redis://localhost:6379/0"
     celery_broker_url: str = "redis://localhost:6379/0"
     celery_result_backend: str = "redis://localhost:6379/1"
+    # Internal API URL used by Celery workers for API-owned WS Agent dispatch.
+    control_plane_url: str = "http://api:8000"
 
     # API Security
     # (api_key_enabled/api_key predate fleet auth and were never enforced by
@@ -59,18 +63,10 @@ class Settings(BaseSettings):
     # guard only allows on a localhost bind. Env: API_AUTH_TOKEN.
     api_auth_token: str = ""
 
-    # OIDC identity verification + emergency bootstrap admin token
-    # (backend/security/identity.py). Read via Settings — not raw
-    # os.getenv() — so these are correctly populated under plain
-    # `uv run uvicorn ...` even when uv does not inject .env into the
-    # process environment (uv only does that for `uv run --env-file .env`;
-    # BaseSettings' own env_file=".env" parsing is what actually reads
-    # these today). Empty (default) = OIDC not configured / bootstrap
-    # token disabled.
-    oidc_issuer: str = ""
-    oidc_audience: str = ""
-    oidc_jwks_url: str = ""
-    bootstrap_admin_token: str = ""
+    # Local-first account used by the NAS/server deployment. The password hash
+    # is persisted in .env after the user changes the default password.
+    local_admin_username: str = "admin"
+    local_admin_password_hash: str = DEFAULT_LOCAL_ADMIN_PASSWORD_HASH
 
     # CLI channel binary allowlist (ADR-0005, audit P0-4). The cli channel is
     # an arbitrary-binary-execution surface, so it only runs binaries the
@@ -153,6 +149,12 @@ def cdp_endpoints(self) -> list[str]:
     # pinned Python 3.10 sidecar instead of the Python 3.13 API process.
     kats_runtime_url: str = "http://localhost:8096"
     kats_runtime_timeout_seconds: float = 120.0
+    # PAW runs only in its own offline sidecar. The URL and program identity are
+    # deployment-owned settings rather than agent-editable processor config.
+    paw_runtime_url: str = "http://localhost:8097"
+    paw_runtime_timeout_seconds: float = 30.0
+    paw_program_id: str = ""
+    paw_max_tokens: int = 512
     # Managed acquisition runtime. The commit/version are code-owned pins;
     # this path merely locates the installed checkout on every platform.
     ohmyopencli_root: str = "/opt/ohmyopencli"
diff --git a/backend/main.py b/backend/main.py
index a09b027d..6431abec 100644
--- a/backend/main.py
+++ b/backend/main.py
@@ -138,7 +138,6 @@ async def lifespan(app: FastAPI):
     # Mark stale pending/running tasks as failed (lost on previous restart)
     from sqlalchemy import update
 
-    from backend.models.operations_agent import OperationsAgentRun
     from backend.models.task import CollectionTask
     async with AsyncSessionLocal() as session:
         await session.execute(
@@ -146,15 +145,12 @@ async def lifespan(app: FastAPI):
             .where(CollectionTask.status.in_(["pending", "running", "ai_processing"]))
             .values(status="failed", error_message="Task lost on server restart")
         )
-        await session.execute(
-            update(OperationsAgentRun)
-            .where(OperationsAgentRun.status.in_(["queued", "running"]))
-            .values(
-                status="failed",
-                error_message="Operations Agent run interrupted by server restart",
-            )
-        )
         await session.commit()
+    from backend.services.scheduled_run_recovery import (
+        recover_operations_agent_runs_on_startup,
+    )
+
+    await recover_operations_agent_runs_on_startup()
     logger.info("Recovered stale tasks on startup")
 
     # Managed acquisitions are durable submit-and-observe work. Unlike legacy
diff --git a/backend/migrations/versions/aa1b2c3d4e5f_add_automation_starter_key.py b/backend/migrations/versions/aa1b2c3d4e5f_add_automation_starter_key.py
new file mode 100644
index 00000000..cb3c53d1
--- /dev/null
+++ b/backend/migrations/versions/aa1b2c3d4e5f_add_automation_starter_key.py
@@ -0,0 +1,24 @@
+"""add stable first-party starter identity to automations"""
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "aa1b2c3d4e5f"
+down_revision = "k8l9m0n1o2p3"
+depends_on = None
+
+def upgrade() -> None:
+    op.add_column("automations", sa.Column("starter_key", sa.String(64), nullable=True))
+    # A unique index is portable to SQLite (where ALTER TABLE cannot add a
+    # table-level unique constraint) and has the same uniqueness semantics.
+    op.create_index(
+        "uq_automations_workspace_starter_key",
+        "automations",
+        ["workspace_id", "starter_key"],
+        unique=True,
+    )
+
+
+def downgrade() -> None:
+    op.drop_index("uq_automations_workspace_starter_key", table_name="automations")
+    op.drop_column("automations", "starter_key")
diff --git a/backend/migrations/versions/ab2c3d4e5f6a_add_collection_lineage_envelope.py b/backend/migrations/versions/ab2c3d4e5f6a_add_collection_lineage_envelope.py
new file mode 100644
index 00000000..1590e28e
--- /dev/null
+++ b/backend/migrations/versions/ab2c3d4e5f6a_add_collection_lineage_envelope.py
@@ -0,0 +1,29 @@
+"""add collection lineage envelopes to records and notification logs
+
+Revision ID: ab2c3d4e5f6a
+Revises: aa1b2c3d4e5f, f5a6b7c8d9e0, w3c4d5e6f7g8
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "ab2c3d4e5f6a"
+down_revision = ("aa1b2c3d4e5f", "f5a6b7c8d9e0", "w3c4d5e6f7g8")
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column(
+        "collected_records",
+        sa.Column("lineage", sa.JSON(), nullable=True),
+    )
+    op.add_column(
+        "notification_logs",
+        sa.Column("lineage", sa.JSON(), nullable=True),
+    )
+
+
+def downgrade() -> None:
+    op.drop_column("notification_logs", "lineage")
+    op.drop_column("collected_records", "lineage")
diff --git a/backend/migrations/versions/bc3d4e5f6a7b_schedule_bound_automations.py b/backend/migrations/versions/bc3d4e5f6a7b_schedule_bound_automations.py
new file mode 100644
index 00000000..704e56f6
--- /dev/null
+++ b/backend/migrations/versions/bc3d4e5f6a7b_schedule_bound_automations.py
@@ -0,0 +1,138 @@
+"""bind scheduled automations to pinned agent versions and durable occurrences
+
+Revision ID: bc3d4e5f6a7b
+Revises: ab2c3d4e5f6a
+"""
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "bc3d4e5f6a7b"
+down_revision = "ab2c3d4e5f6a"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    bind = op.get_bind()
+    is_sqlite = bind.dialect.name == "sqlite"
+
+    def has_column(table: str, column: str) -> bool:
+        return column in {
+            item["name"] for item in sa.inspect(bind).get_columns(table)
+        }
+
+    def has_index(table: str, index: str) -> bool:
+        return index in {
+            item["name"] for item in sa.inspect(bind).get_indexes(table)
+        }
+
+    if not has_column("automations", "revision"):
+        op.add_column(
+            "automations",
+            sa.Column("revision", sa.Integer(), nullable=False, server_default="1"),
+        )
+    if not has_column("automations", "operations_agent_id"):
+        op.add_column(
+            "automations",
+            sa.Column(
+                "operations_agent_id",
+                sa.String(36),
+                *(
+                    ()
+                    if is_sqlite
+                    else (
+                        sa.ForeignKey(
+                            "operations_agent_identities.id",
+                            ondelete="RESTRICT",
+                        ),
+                    )
+                ),
+                nullable=True,
+            ),
+        )
+    if not has_column("automations", "operations_agent_version"):
+        op.add_column(
+            "automations",
+            sa.Column("operations_agent_version", sa.Integer(), nullable=True),
+        )
+    if not has_index("automations", "ix_automations_operations_agent_id"):
+        op.create_index(
+            "ix_automations_operations_agent_id",
+            "automations",
+            ["operations_agent_id"],
+        )
+    # Previously enabled rows had no executable binding. Fail closed until an
+    # operator explicitly pins a compatible published Agent version.
+    op.execute("UPDATE automations SET enabled = 0 WHERE operations_agent_id IS NULL")
+
+    run_columns = (
+        (
+            "automation_id",
+            sa.Column(
+                "automation_id",
+                sa.String(36),
+                *(
+                    ()
+                    if is_sqlite
+                    else (
+                        sa.ForeignKey("automations.id", ondelete="RESTRICT"),
+                    )
+                ),
+                nullable=True,
+            ),
+        ),
+        ("automation_revision", sa.Column("automation_revision", sa.Integer(), nullable=True)),
+        ("automation_snapshot", sa.Column("automation_snapshot", sa.JSON(), nullable=True)),
+        (
+            "scheduled_for",
+            sa.Column("scheduled_for", sa.DateTime(timezone=True), nullable=True),
+        ),
+        (
+            "schedule_timezone",
+            sa.Column("schedule_timezone", sa.String(64), nullable=True),
+        ),
+    )
+    for name, column in run_columns:
+        if not has_column("operations_agent_runs", name):
+            op.add_column("operations_agent_runs", column)
+    if not has_index(
+        "operations_agent_runs",
+        "ix_operations_agent_runs_automation_id",
+    ):
+        op.create_index(
+            "ix_operations_agent_runs_automation_id",
+            "operations_agent_runs",
+            ["automation_id"],
+        )
+    if not has_index(
+        "operations_agent_runs",
+        "uq_operations_agent_runs_automation_occurrence",
+    ):
+        op.create_index(
+            "uq_operations_agent_runs_automation_occurrence",
+            "operations_agent_runs",
+            ["automation_id", "scheduled_for"],
+            unique=True,
+        )
+
+
+def downgrade() -> None:
+    op.drop_index(
+        "uq_operations_agent_runs_automation_occurrence",
+        table_name="operations_agent_runs",
+    )
+    op.drop_index(
+        "ix_operations_agent_runs_automation_id",
+        table_name="operations_agent_runs",
+    )
+    op.drop_column("operations_agent_runs", "schedule_timezone")
+    op.drop_column("operations_agent_runs", "scheduled_for")
+    op.drop_column("operations_agent_runs", "automation_snapshot")
+    op.drop_column("operations_agent_runs", "automation_revision")
+    op.drop_column("operations_agent_runs", "automation_id")
+
+    op.drop_index("ix_automations_operations_agent_id", table_name="automations")
+    op.drop_column("automations", "operations_agent_version")
+    op.drop_column("automations", "operations_agent_id")
+    op.drop_column("automations", "revision")
diff --git a/backend/migrations/versions/cd4e5f6a7b8c_universal_agent_runtime.py b/backend/migrations/versions/cd4e5f6a7b8c_universal_agent_runtime.py
new file mode 100644
index 00000000..270a0bef
--- /dev/null
+++ b/backend/migrations/versions/cd4e5f6a7b8c_universal_agent_runtime.py
@@ -0,0 +1,137 @@
+"""add capability-selected universal agent runtime envelopes
+
+Revision ID: cd4e5f6a7b8c
+Revises: bc3d4e5f6a7b
+"""
+
+from copy import deepcopy
+
+import sqlalchemy as sa
+from alembic import op
+
+revision = "cd4e5f6a7b8c"
+down_revision = "bc3d4e5f6a7b"
+branch_labels = None
+depends_on = None
+
+
+def _has_table(bind, table: str) -> bool:
+    return table in sa.inspect(bind).get_table_names()
+
+
+def _has_column(bind, table: str, column: str) -> bool:
+    return column in {item["name"] for item in sa.inspect(bind).get_columns(table)}
+
+
+def _migrate_configuration(configuration: dict) -> dict:
+    migrated = deepcopy(configuration)
+    contract = migrated.get("agent_contract")
+    if isinstance(contract, dict) and contract.get("schema_version") == "agent.contract.v1":
+        contract["schema_version"] = "agent.contract.v2"
+        contract.setdefault("role", "operations_agent")
+        contract.setdefault("required_capabilities", ["streaming"])
+        contract.setdefault("tool_policy", {})
+        contract.setdefault("budget", {})
+        contract.setdefault("quality_gates", [])
+        contract.setdefault("evidence_requirements", [])
+
+    binding = migrated.get("runtime_binding")
+    if isinstance(binding, dict) and binding.get("schema_version") == "agent.runtime-binding.v1":
+        agent_url = binding.get("agent_url")
+        runtime = binding.get("runtime")
+        migrated["runtime_binding"] = {
+            "schema_version": "agent.runtime-binding.v2",
+            "workflow": binding.get("workflow") or "default",
+            "preferred_agent_urls": [agent_url] if isinstance(agent_url, str) and agent_url else [],
+            "preferred_runtimes": [runtime] if isinstance(runtime, str) and runtime else [],
+            "model_binding": None,
+            "config": binding.get("config") or {},
+            "dispatch_timeout_seconds": binding.get("dispatch_timeout_seconds", 1800),
+        }
+    return migrated
+
+
+def _downgrade_configuration(configuration: dict) -> dict:
+    downgraded = deepcopy(configuration)
+    contract = downgraded.get("agent_contract")
+    if isinstance(contract, dict) and contract.get("schema_version") == "agent.contract.v2":
+        downgraded["agent_contract"] = {
+            "schema_version": "agent.contract.v1",
+            "input_schema": contract.get("input_schema") or {},
+            "output_schema": contract.get("output_schema") or {},
+            "state_schema": contract.get("state_schema") or {},
+        }
+
+    binding = downgraded.get("runtime_binding")
+    if isinstance(binding, dict) and binding.get("schema_version") == "agent.runtime-binding.v2":
+        urls = binding.get("preferred_agent_urls") or []
+        runtimes = binding.get("preferred_runtimes") or []
+        runtime = (
+            runtimes[0]
+            if runtimes and runtimes[0] in {"miniflow", "pi", "codex"}
+            else "miniflow"
+        )
+        downgraded["runtime_binding"] = {
+            "schema_version": "agent.runtime-binding.v1",
+            "agent_url": urls[0] if urls else "http://unbound.invalid",
+            "runtime": runtime,
+            "workflow": binding.get("workflow") or "default",
+            "config": binding.get("config") or {},
+            "dispatch_timeout_seconds": binding.get("dispatch_timeout_seconds", 1800),
+        }
+    return downgraded
+
+
+def _rewrite_agent_configurations(bind, transform) -> None:
+    for table_name in ("operations_agent_drafts", "published_operations_agent_versions"):
+        if not _has_table(bind, table_name):
+            continue
+        table = sa.table(
+            table_name,
+            sa.column("id", sa.String()),
+            sa.column("model_configuration", sa.JSON()),
+        )
+        rows = bind.execute(sa.select(table.c.id, table.c.model_configuration)).mappings()
+        for row in rows:
+            configuration = row["model_configuration"]
+            if isinstance(configuration, dict):
+                bind.execute(
+                    table.update()
+                    .where(table.c.id == row["id"])
+                    .values(model_configuration=transform(configuration))
+                )
+
+
+def upgrade() -> None:
+    bind = op.get_bind()
+    if _has_table(bind, "edge_nodes") and not _has_column(
+        bind, "edge_nodes", "runtime_capabilities"
+    ):
+        op.add_column(
+            "edge_nodes",
+            sa.Column("runtime_capabilities", sa.JSON(), nullable=True),
+        )
+    if _has_table(bind, "operations_agent_runs"):
+        if not _has_column(bind, "operations_agent_runs", "execution_binding"):
+            op.add_column(
+                "operations_agent_runs",
+                sa.Column("execution_binding", sa.JSON(), nullable=True),
+            )
+        if not _has_column(bind, "operations_agent_runs", "evidence_payload"):
+            op.add_column(
+                "operations_agent_runs",
+                sa.Column("evidence_payload", sa.JSON(), nullable=True),
+            )
+    _rewrite_agent_configurations(bind, _migrate_configuration)
+
+
+def downgrade() -> None:
+    bind = op.get_bind()
+    _rewrite_agent_configurations(bind, _downgrade_configuration)
+    if _has_table(bind, "operations_agent_runs"):
+        if _has_column(bind, "operations_agent_runs", "evidence_payload"):
+            op.drop_column("operations_agent_runs", "evidence_payload")
+        if _has_column(bind, "operations_agent_runs", "execution_binding"):
+            op.drop_column("operations_agent_runs", "execution_binding")
+    if _has_table(bind, "edge_nodes") and _has_column(bind, "edge_nodes", "runtime_capabilities"):
+        op.drop_column("edge_nodes", "runtime_capabilities")
diff --git a/backend/miniflow/model.py b/backend/miniflow/model.py
index fc690cc7..4e14d677 100644
--- a/backend/miniflow/model.py
+++ b/backend/miniflow/model.py
@@ -6,6 +6,7 @@
 import graphlib
 from collections.abc import Callable
 from dataclasses import dataclass, field
+from pathlib import Path
 from typing import Any
 
 
@@ -67,3 +68,40 @@ def ordered(self) -> list[Step]:
                 ordered.append(by_name[name])
             sorter.done(*ready)
         return ordered
+
+
+READ_ONLY_READINESS = "builtin.read_only_readiness"
+
+
+def _inspect_runtime_root() -> str:
+    root = Path.cwd().resolve()
+    if not root.is_dir():
+        raise RuntimeError("Agent runtime working directory is unavailable")
+    return str(root)
+
+
+def _verify_runtime_bundle() -> str:
+    module_path = Path(__file__).resolve()
+    if not module_path.is_file():
+        raise RuntimeError("Packaged MiniFlow runtime bundle is unavailable")
+    return module_path.name
+
+
+_READ_ONLY_READINESS_WORKFLOW = Workflow(
+    name="operations-read-only-readiness",
+    steps=[
+        Step(name="inspect-runtime-root", run=_inspect_runtime_root),
+        Step(
+            name="verify-runtime-bundle",
+            run=_verify_runtime_bundle,
+            depends_on=["inspect-runtime-root"],
+        ),
+    ],
+)
+
+
+def get_builtin_workflow(name: str) -> Workflow | None:
+    """Resolve a stable built-in workflow alias without filesystem assumptions."""
+    if name == READ_ONLY_READINESS:
+        return _READ_ONLY_READINESS_WORKFLOW
+    return None
diff --git a/backend/models/automation.py b/backend/models/automation.py
index a481fdba..0ec12119 100644
--- a/backend/models/automation.py
+++ b/backend/models/automation.py
@@ -1,4 +1,4 @@
-from sqlalchemy import JSON, Boolean, ForeignKey, String, Text
+from sqlalchemy import JSON, Boolean, ForeignKey, Integer, String, Text, UniqueConstraint
 from sqlalchemy.orm import Mapped, mapped_column
 
 from backend.models.base import TimestampMixin
@@ -8,10 +8,23 @@ class Automation(TimestampMixin):
     """Provider-neutral scheduled agent task, configurable by UI or API."""
 
     __tablename__ = "automations"
+    __table_args__ = (
+        UniqueConstraint("workspace_id", "starter_key", name="uq_automations_workspace_starter_key"),
+    )
 
     workspace_id: Mapped[str] = mapped_column(
         ForeignKey("workspaces.id", ondelete="CASCADE"), nullable=False, index=True
     )
+    # First-party starter identity.  Null keeps existing user-created automations
+    # compatible while making starter installation concurrency-safe.
+    starter_key: Mapped[str | None] = mapped_column(String(64), nullable=True)
+    revision: Mapped[int] = mapped_column(Integer, nullable=False, default=1)
+    operations_agent_id: Mapped[str | None] = mapped_column(
+        ForeignKey("operations_agent_identities.id", ondelete="RESTRICT"),
+        nullable=True,
+        index=True,
+    )
+    operations_agent_version: Mapped[int | None] = mapped_column(Integer, nullable=True)
     name: Mapped[str] = mapped_column(String(255), nullable=False)
     prompt: Mapped[str] = mapped_column(Text, nullable=False)
     precheck: Mapped[str | None] = mapped_column(Text, nullable=True)
diff --git a/backend/models/edge_node.py b/backend/models/edge_node.py
index 5fcc784d..654086e6 100644
--- a/backend/models/edge_node.py
+++ b/backend/models/edge_node.py
@@ -39,6 +39,8 @@ class EdgeNode(TimestampMixin):
     # node hasn't registered since this field was added, or registered over the
     # HTTP (non-WS) path, which doesn't carry runtime advertisement.
     runtimes: Mapped[Optional[list]] = mapped_column(JSON, nullable=True)
+    # Per-runtime capability names measured by the edge adapter registry.
+    runtime_capabilities: Mapped[Optional[dict]] = mapped_column(JSON, nullable=True)
 
 
 class EdgeNodeEvent(TimestampMixin):
diff --git a/backend/models/notification.py b/backend/models/notification.py
index f0135f41..4d28e959 100644
--- a/backend/models/notification.py
+++ b/backend/models/notification.py
@@ -39,6 +39,9 @@ class NotificationLog(TimestampMixin):
         String(36), ForeignKey("notification_rules.id", ondelete="CASCADE"), nullable=False
     )
     record_id: Mapped[str | None] = mapped_column(String(36), nullable=True)
+    # Collection provenance copied from the originating record. Nullable for
+    # historical/manual notifications and records created before the envelope.
+    lineage: Mapped[dict | None] = mapped_column(JSON, nullable=True)
     # sent | failed
     status: Mapped[str] = mapped_column(String(50), nullable=False)
     response_data: Mapped[dict | None] = mapped_column(JSON, nullable=True)
diff --git a/backend/models/operations_agent.py b/backend/models/operations_agent.py
index 2013e9db..27eabd37 100644
--- a/backend/models/operations_agent.py
+++ b/backend/models/operations_agent.py
@@ -1,8 +1,10 @@
+from datetime import datetime
 from enum import StrEnum
 
 from sqlalchemy import (
     JSON,
     Boolean,
+    DateTime,
     CheckConstraint,
     ForeignKey,
     Integer,
@@ -121,6 +123,11 @@ class OperationsAgentRun(TimestampMixin):
             "trigger_type IN ('manual', 'scheduled', 'event')",
             name="ck_operations_agent_runs_trigger_type",
         ),
+        UniqueConstraint(
+            "automation_id",
+            "scheduled_for",
+            name="uq_operations_agent_runs_automation_occurrence",
+        ),
     )
 
     workspace_id: Mapped[str] = mapped_column(
@@ -135,11 +142,22 @@ class OperationsAgentRun(TimestampMixin):
     profile_version: Mapped[int] = mapped_column(Integer, nullable=False)
     trigger_type: Mapped[str] = mapped_column(String(16), nullable=False)
     trigger_reference: Mapped[str | None] = mapped_column(String(255), nullable=True)
+    automation_id: Mapped[str | None] = mapped_column(
+        ForeignKey("automations.id", ondelete="RESTRICT"),
+        nullable=True,
+        index=True,
+    )
+    automation_revision: Mapped[int | None] = mapped_column(Integer, nullable=True)
+    automation_snapshot: Mapped[dict | None] = mapped_column(JSON, nullable=True)
+    scheduled_for: Mapped[datetime | None] = mapped_column(DateTime(timezone=True), nullable=True)
+    schedule_timezone: Mapped[str | None] = mapped_column(String(64), nullable=True)
     target_resource_type: Mapped[str] = mapped_column(String(100), nullable=False)
     target_resource_id: Mapped[str] = mapped_column(String(255), nullable=False)
     input_payload: Mapped[dict] = mapped_column(JSON, nullable=False, default=dict)
     state_payload: Mapped[dict] = mapped_column(JSON, nullable=False, default=dict)
     output_payload: Mapped[dict | None] = mapped_column(JSON, nullable=True)
+    execution_binding: Mapped[dict | None] = mapped_column(JSON, nullable=True)
+    evidence_payload: Mapped[dict | None] = mapped_column(JSON, nullable=True)
     error_message: Mapped[str | None] = mapped_column(Text, nullable=True)
     status: Mapped[str] = mapped_column(String(16), nullable=False, default="queued")
     started_by_user_id: Mapped[str] = mapped_column(
diff --git a/backend/models/record.py b/backend/models/record.py
index c8f420b5..1623ac98 100644
--- a/backend/models/record.py
+++ b/backend/models/record.py
@@ -37,6 +37,10 @@ class CollectedRecord(TimestampMixin):
         String(36), nullable=True, index=True
     )
 
+    # Immutable collection provenance envelope. Nullable for pre-envelope rows
+    # and for paths that did not establish optional references.
+    lineage: Mapped[dict | None] = mapped_column(JSON, nullable=True)
+
     # Raw data as returned by the channel
     raw_data: Mapped[dict] = mapped_column(JSON, nullable=False, default=dict)
     # Normalized standard fields: title, url, content, author, published_at, ...
diff --git a/backend/notifiers/base.py b/backend/notifiers/base.py
index 8c276f78..c77dd27f 100644
--- a/backend/notifiers/base.py
+++ b/backend/notifiers/base.py
@@ -11,6 +11,7 @@ class NotificationPayload:
     record_id: str | None = None
     data: dict[str, Any] = field(default_factory=dict)
     ai_enrichment: dict[str, Any] | None = None
+    lineage: dict[str, Any] | None = None
 
 
 @dataclass
diff --git a/backend/pipeline/ai_processor.py b/backend/pipeline/ai_processor.py
index 071d7936..a8576c80 100644
--- a/backend/pipeline/ai_processor.py
+++ b/backend/pipeline/ai_processor.py
@@ -83,7 +83,7 @@ async def process_with_ai(
     records actually enriched.
 
     ai_config keys:
-        processor_type: claude | openai | local
+        processor_type: claude | openai | local | paw
         model: model name
         prompt_template: Jinja2 template
         provider_id: model-provider runtime PR-F — governed ModelProvider reference; wins
@@ -138,8 +138,24 @@ async def process_with_ai(
         config=resolved_config,
     )
 
+    if len(result.enrichments) != len(records) or any(
+        not isinstance(index, int) or index < 0 or index >= len(records)
+        for index in result.failed_indices
+    ):
+        logger.error(
+            "processor.contract_invalid | processor_type=%s records=%d "
+            "enrichments=%d failed_indices=%r",
+            processor_type,
+            len(records),
+            len(result.enrichments),
+            result.failed_indices,
+        )
+        return 0
+
     enriched = 0
-    for record, enrichment in zip(records, result.enrichments):
+    for index, (record, enrichment) in enumerate(zip(records, result.enrichments)):
+        if index in result.failed_indices:
+            continue
         record.ai_enrichment = enrichment
         record.status = "ai_processed"
         enriched += 1
diff --git a/backend/pipeline/notifier_dispatch.py b/backend/pipeline/notifier_dispatch.py
index 748380f5..7df0ade8 100644
--- a/backend/pipeline/notifier_dispatch.py
+++ b/backend/pipeline/notifier_dispatch.py
@@ -39,6 +39,11 @@ def _normalize_send_result(result: bool | NotificationSendResult) -> tuple[bool,
         return result.success, result.response_data
     return bool(result), None
 
+def _record_lineage(record: CollectedRecord) -> dict[str, Any] | None:
+    """Return only a JSON object; compatibility callers may use lightweight mocks."""
+    value = getattr(record, "lineage", None)
+    return dict(value) if isinstance(value, dict) else None
+
 
 @dataclass
 class _PendingSend:
@@ -123,14 +128,15 @@ async def dispatch_notifications(
                 rule.id, rule.notifier_type,
             )
             continue
-
         ack_required = bool(_ack_secret(rule.notifier_config))
         for record in records:
             log_id = str(uuid.uuid4())
+            lineage = _record_lineage(record)
             session.add(NotificationLog(
                 id=log_id,
                 rule_id=rule.id,
                 record_id=record.id,
+                lineage=lineage,
                 status="pending",
                 ack_status="not_required",
             ))
@@ -143,6 +149,7 @@ async def dispatch_notifications(
                     source_id=source_id,
                     delivery_id=log_id,
                     record_id=record.id,
+                    lineage=lineage,
                     data=record.normalized_data,
                     ai_enrichment=record.ai_enrichment,
                 ),
diff --git a/backend/pipeline/pipeline.py b/backend/pipeline/pipeline.py
index f151db00..d481cfd6 100644
--- a/backend/pipeline/pipeline.py
+++ b/backend/pipeline/pipeline.py
@@ -122,6 +122,7 @@ async def run_pipeline(
     agent_config: dict[str, Any] | None = None,
     run_id: str | None = None,
     sink=None,  # ItemSink | None — write destination; defaults to LegacyDbSink
+    collection_lineage=None,  # CollectionLineage | None
 ) -> PipelineResult:
     """Execute the full collection pipeline. Each write step uses its own
     short-lived session so no write lock is held during long-running I/O."""
@@ -291,6 +292,7 @@ async def run_pipeline(
         source_id=source.id,
         provider=source.channel_type,
         run_id=run_id,
+        lineage=collection_lineage,
     )
     logger.info("[task:%s] step2-3/sink start | sink=%s items=%d",
                 task_id, type(active_sink).__name__, channel_result.count)
diff --git a/backend/pipeline/runner.py b/backend/pipeline/runner.py
index 08c0172c..6da6e674 100644
--- a/backend/pipeline/runner.py
+++ b/backend/pipeline/runner.py
@@ -166,11 +166,18 @@ async def run_collection_pipeline(
         {k: v for k, v in (agent_config or {}).items() if k != "prompt_template"},
     )
 
-    # ── Phase 3: run pipeline (no session held during collection) ─────────────
-    # Hold a per-domain slot for the run so the fleet stays polite to a site even
-    # when many sources target it (in-process cap; cross-worker would need Redis).
-    from backend.pipeline.domain_limiter import domain_slot
-
+    # The legacy task runner knows the durable task/run and worker identities.
+    # Source/binding revisions, acquisition execution, runtime, trace, and
+    # artifacts remain null unless an upstream handoff establishes them; this
+    # path deliberately does not infer an acquisition-to-record bridge.
+    from backend.pipeline.sinks.base import CollectionLineage
+
+    collection_lineage = CollectionLineage(
+        task_id=task_id,
+        source_id=source.id,
+        collection_run_id=run_id,
+        worker_id=worker_id,
+    )
     async with domain_slot(source):
         try:
             pipeline_result = await run_pipeline(
@@ -179,6 +186,7 @@ async def run_collection_pipeline(
                 parameters=merged_params,
                 agent_config=agent_config,
                 run_id=run_id,
+                collection_lineage=collection_lineage,
             )
         except Exception as exc:
             # run_pipeline only re-raises errors its taxonomy classified as
diff --git a/backend/pipeline/sinks/__init__.py b/backend/pipeline/sinks/__init__.py
index bfccdc32..80fe3b49 100644
--- a/backend/pipeline/sinks/__init__.py
+++ b/backend/pipeline/sinks/__init__.py
@@ -5,13 +5,14 @@
     from backend.pipeline.sinks import LegacyDbSink, OdpSink, DualSink, ItemSink, RunContext, SinkResult
 """
 
-from backend.pipeline.sinks.base import ItemSink, RunContext, SinkResult
+from backend.pipeline.sinks.base import CollectionLineage, ItemSink, RunContext, SinkResult
 from backend.pipeline.sinks.legacy_db_sink import LegacyDbSink
 from backend.pipeline.sinks.odp_sink import OdpSink
 from backend.pipeline.sinks.dual_sink import DualSink
 from backend.pipeline.sinks.strategy import select_sink
 
 __all__ = [
+    "CollectionLineage",
     "ItemSink",
     "RunContext",
     "SinkResult",
diff --git a/backend/pipeline/sinks/base.py b/backend/pipeline/sinks/base.py
index 6829615a..017360f7 100644
--- a/backend/pipeline/sinks/base.py
+++ b/backend/pipeline/sinks/base.py
@@ -18,6 +18,88 @@
 from typing import Any, Protocol, Sequence
 
 
+@dataclass(frozen=True, slots=True)
+class CollectionLineage:
+    """Immutable provenance envelope shared by collection projections.
+
+    Every value is an identifier or reference owned by an existing subsystem.
+    ``None`` means that subsystem did not establish the value at this boundary;
+    callers must not synthesize replacements.
+    """
+
+    task_id: str | None = None
+    source_id: str | None = None
+    provider: str | None = None
+    ingest_mode: str | None = None
+    collection_run_id: str | None = None
+    acquisition_execution_id: str | None = None
+    source_revision_id: str | None = None
+    source_binding_revision_id: str | None = None
+    account_revision_id: str | None = None
+    credential_revision_id: str | None = None
+    project_id: str | None = None
+    scope_ref: str | None = None
+    worker_id: str | None = None
+    runtime_id: str | None = None
+    trace_id: str | None = None
+    trace_ref: str | None = None
+    artifact_refs: tuple[Any, ...] | None = None
+
+    def __post_init__(self) -> None:
+        if self.artifact_refs is not None and not isinstance(self.artifact_refs, tuple):
+            object.__setattr__(self, "artifact_refs", tuple(self.artifact_refs))
+
+    def to_dict(self) -> dict[str, Any]:
+        """Return the stable JSON representation persisted on projections."""
+        values = {
+            "task_id": self.task_id,
+            "source_id": self.source_id,
+            "provider": self.provider,
+            "ingest_mode": self.ingest_mode,
+            "collection_run_id": self.collection_run_id,
+            "acquisition_execution_id": self.acquisition_execution_id,
+            "source_revision_id": self.source_revision_id,
+            "source_binding_revision_id": self.source_binding_revision_id,
+            "account_revision_id": self.account_revision_id,
+            "credential_revision_id": self.credential_revision_id,
+            "project_id": self.project_id,
+            "scope_ref": self.scope_ref,
+            "worker_id": self.worker_id,
+            "runtime_id": self.runtime_id,
+            "trace_id": self.trace_id,
+            "trace_ref": self.trace_ref,
+            "artifact_refs": (
+                list(self.artifact_refs) if self.artifact_refs is not None else None
+            ),
+        }
+        return values
+
+    @classmethod
+    def from_dict(cls, values: dict[str, Any] | None) -> "CollectionLineage | None":
+        if values is None:
+            return None
+        fields = {
+            "task_id",
+            "source_id",
+            "provider",
+            "ingest_mode",
+            "collection_run_id",
+            "acquisition_execution_id",
+            "source_revision_id",
+            "source_binding_revision_id",
+            "account_revision_id",
+            "credential_revision_id",
+            "project_id",
+            "scope_ref",
+            "worker_id",
+            "runtime_id",
+            "trace_id",
+            "trace_ref",
+            "artifact_refs",
+        }
+        return cls(**{key: values[key] for key in fields if key in values})
+
+
 @dataclass
 class RunContext:
     """Identity of one collection run, threaded to whichever sink handles it.
@@ -25,7 +107,8 @@ class RunContext:
     ``provider`` is the channel_type (e.g. ``"rss"``); it becomes the ODP
     ``provider`` and the legacy ``channel_type``. ``ingest_mode`` is
     ``snapshot`` (full re-list) or ``stream`` (incremental), mirroring the ODP
-    contract.
+    contract. Optional references remain absent when the owning subsystem has
+    not established them.
     """
 
     task_id: str
@@ -34,7 +117,45 @@ class RunContext:
     ingest_mode: str = "snapshot"
     run_id: str | None = None
     trace_id: str | None = None
-
+    acquisition_execution_id: str | None = None
+    source_revision_id: str | None = None
+    source_binding_revision_id: str | None = None
+    account_revision_id: str | None = None
+    credential_revision_id: str | None = None
+    project_id: str | None = None
+    scope_ref: str | None = None
+    worker_id: str | None = None
+    runtime_id: str | None = None
+    trace_ref: str | None = None
+    artifact_refs: list[Any] | tuple[Any, ...] | None = None
+    lineage: CollectionLineage | None = None
+
+    def lineage_envelope(self) -> CollectionLineage:
+        """Build one immutable envelope without inventing missing references."""
+        values = self.lineage.to_dict() if self.lineage is not None else {}
+        values.update(
+            task_id=self.task_id,
+            source_id=self.source_id,
+            provider=self.provider,
+            ingest_mode=self.ingest_mode,
+        )
+        optional = {
+            "collection_run_id": self.run_id,
+            "trace_id": self.trace_id,
+            "acquisition_execution_id": self.acquisition_execution_id,
+            "source_revision_id": self.source_revision_id,
+            "source_binding_revision_id": self.source_binding_revision_id,
+            "account_revision_id": self.account_revision_id,
+            "credential_revision_id": self.credential_revision_id,
+            "project_id": self.project_id,
+            "scope_ref": self.scope_ref,
+            "worker_id": self.worker_id,
+            "runtime_id": self.runtime_id,
+            "trace_ref": self.trace_ref,
+            "artifact_refs": self.artifact_refs,
+        }
+        values.update({key: value for key, value in optional.items() if value is not None})
+        return CollectionLineage.from_dict(values)  # type: ignore[return-value]
 
 @dataclass
 class SinkResult:
diff --git a/backend/pipeline/sinks/legacy_db_sink.py b/backend/pipeline/sinks/legacy_db_sink.py
index 733acd9d..cee5c11d 100644
--- a/backend/pipeline/sinks/legacy_db_sink.py
+++ b/backend/pipeline/sinks/legacy_db_sink.py
@@ -86,6 +86,7 @@ async def write_batch(self, ctx: RunContext, items: Sequence[dict]) -> SinkResul
                 session, ctx.task_id, ctx.source_id, triples,
                 channel_type=ctx.provider, forward_to_odp=self.forward_to_odp,
                 identities=identities,
+                lineage=ctx.lineage_envelope(),
             )
             await session.commit()
 
diff --git a/backend/pipeline/storer.py b/backend/pipeline/storer.py
index d216ea91..5c859e34 100644
--- a/backend/pipeline/storer.py
+++ b/backend/pipeline/storer.py
@@ -2,6 +2,7 @@
 
 import logging
 import os
+from typing import Any
 
 from sqlalchemy import select
 from sqlalchemy.exc import IntegrityError
@@ -9,7 +10,7 @@
 
 from backend.models.record import CollectedRecord
 from backend.pipeline import odp_client
-
+from backend.pipeline.sinks.base import CollectionLineage
 logger = logging.getLogger(__name__)
 
 # SQLite's default SQLITE_MAX_VARIABLE_NUMBER is 999 on older builds (32766 on
@@ -62,7 +63,6 @@ async def _existing_by_identity(
             existing[record.identity_key] = record
     return existing
 
-
 async def store_records(
     session: AsyncSession,
     task_id: str,
@@ -74,6 +74,7 @@ async def store_records(
     workflow_id: str | None = None,
     workflow_run_id: str | None = None,
     identities: list[str | None] | None = None,
+    lineage: CollectionLineage | dict[str, Any] | None = None,
 ) -> tuple[list[CollectedRecord], int]:
     """Insert new records; skip existing ones by content_hash.
 
@@ -108,8 +109,18 @@ async def store_records(
         completely unchanged — content_hash-only dedup, exactly as before
         C7. This is the only path channels without identity() ever take.
 
+
     Returns (new_records, skipped_count).
     """
+    if isinstance(lineage, CollectionLineage):
+        lineage_payload = lineage.to_dict()
+    elif lineage is None:
+        lineage_payload = None
+    else:
+        # Accept already-serialized envelopes for callers crossing a process
+        # boundary, while keeping one canonical shape at this persistence seam.
+        lineage_payload = CollectionLineage.from_dict(lineage)
+        lineage_payload = lineage_payload.to_dict() if lineage_payload else None
     if not normalized_triples:
         return [], 0
 
@@ -174,8 +185,8 @@ async def store_records(
                 continue
             if identity in seen_identities_in_batch:
                 # Two triples in this same batch share an identity (e.g. a
-                # feed listed the same entry twice) — keep the first, skip
-                # the rest rather than fight over which one "wins".
+                # feed listed the same entry twice) — keep the first, skip the
+                # rest rather than fight over which one "wins".
                 skipped += 1
                 continue
             seen_identities_in_batch.add(identity)
@@ -190,6 +201,7 @@ async def store_records(
             source_id=source_id,
             workflow_id=workflow_id,
             workflow_run_id=workflow_run_id,
+            lineage=lineage_payload,
             raw_data=raw,
             normalized_data=normalized,
             content_hash=content_hash,
diff --git a/backend/processors/base.py b/backend/processors/base.py
index 8966d5e7..c1779598 100644
--- a/backend/processors/base.py
+++ b/backend/processors/base.py
@@ -11,6 +11,7 @@ class ProcessingResult:
     success: bool
     enrichments: list[dict[str, Any]] = field(default_factory=list)
     error: str | None = None
+    failed_indices: set[int] = field(default_factory=set)
 
 
 class AbstractProcessor(ABC):
diff --git a/backend/processors/paw_processor.py b/backend/processors/paw_processor.py
new file mode 100644
index 00000000..f53aa29e
--- /dev/null
+++ b/backend/processors/paw_processor.py
@@ -0,0 +1,251 @@
+"""Governed adapter for the fixed local PAW enrichment sidecar."""
+import ipaddress
+import json
+import logging
+import re
+from typing import TYPE_CHECKING, Any
+from urllib.parse import urlparse
+
+from jsonschema import Draft202012Validator
+from jsonschema.exceptions import SchemaError, ValidationError
+
+from backend.config import get_settings
+from backend.processors.base import AbstractProcessor, ProcessingResult
+from backend.processors.registry import register_processor
+from backend.security.url_guard import SSRFValidationError, guarded_async_client
+
+if TYPE_CHECKING:
+    from backend.models.record import CollectedRecord
+
+logger = logging.getLogger(__name__)
+
+CONTRACT_VERSION = "opencli.paw.runtime.v1"
+MAX_INPUT_CHARS = 8_192
+MAX_TOKENS = 512
+MAX_OUTPUT_BYTES = 65_536
+_PLACEHOLDER_RE = re.compile(r"\{\{(\w+)\}\}")
+_PROGRAM_ID_RE = re.compile(r"^[a-f0-9]{16,64}$")
+_RESPONSE_SCHEMA = {
+    "type": "object",
+    "additionalProperties": False,
+    "required": ["contractVersion", "programId", "enrichment"],
+    "properties": {
+        "contractVersion": {"const": CONTRACT_VERSION},
+        "programId": {"type": "string", "minLength": 1},
+        "enrichment": {"type": "object"},
+    },
+}
+_SENSITIVE_PLACEHOLDER_TERMS = (
+    "cookie",
+    "token",
+    "secret",
+    "password",
+    "auth",
+    "headers",
+    "raw_html",
+    "html",
+)
+
+
+def _is_local_paw_url(url: str) -> bool:
+    parsed = urlparse(url)
+    if parsed.scheme != "http" or parsed.username or parsed.password:
+        return False
+    hostname = parsed.hostname
+    if hostname in {"paw-runtime", "localhost"}:
+        return True
+    if hostname is None:
+        return False
+    try:
+        return ipaddress.ip_address(hostname).is_loopback
+    except ValueError:
+        return False
+
+
+def _has_sensitive_placeholder(template: str) -> bool:
+    return any(
+        any(term in placeholder.lower() for term in _SENSITIVE_PLACEHOLDER_TERMS)
+        for placeholder in _PLACEHOLDER_RE.findall(template)
+    )
+_RESPONSE_VALIDATOR = Draft202012Validator(_RESPONSE_SCHEMA)
+
+
+def _merged_config(config: dict[str, Any]) -> dict[str, Any]:
+    nested = config.get("config")
+    if isinstance(nested, dict):
+        return {**nested, **{key: value for key, value in config.items() if key != "config"}}
+    return config
+
+
+def _render(template: str, data: dict[str, Any]) -> str:
+    return _PLACEHOLDER_RE.sub(lambda match: str(data.get(match.group(1), "")), template)
+
+
+def _validate_response(
+    data: Any,
+    program_id: str,
+    output_validator: Draft202012Validator | None,
+) -> dict[str, Any]:
+    try:
+        _RESPONSE_VALIDATOR.validate(data)
+    except ValidationError as exc:
+        raise ValueError("response.schema_invalid") from exc
+    if data["programId"] != program_id:
+        raise ValueError("response.program_mismatch")
+    enrichment = data["enrichment"]
+    if output_validator is not None:
+        try:
+            output_validator.validate(enrichment)
+        except ValidationError as exc:
+            raise ValueError("response.output_schema_invalid") from exc
+    try:
+        encoded = json.dumps(
+            enrichment, ensure_ascii=False, separators=(",", ":"), allow_nan=False
+        ).encode("utf-8")
+    except (TypeError, ValueError) as exc:
+        raise ValueError("response.enrichment_invalid") from exc
+    if len(encoded) > MAX_OUTPUT_BYTES:
+        raise ValueError("response.too_large")
+    return enrichment
+
+
+async def _read_bounded_json(response: Any) -> Any:
+    body = bytearray()
+    try:
+        async for chunk in response.aiter_bytes():
+            if len(body) + len(chunk) > MAX_OUTPUT_BYTES:
+                raise ValueError("response.too_large")
+            body.extend(chunk)
+    finally:
+        await response.aclose()
+    try:
+        return json.loads(
+            body,
+            parse_constant=lambda value: (_ for _ in ()).throw(ValueError(value)),
+        )
+    except (json.JSONDecodeError, ValueError) as exc:
+        raise ValueError("response.invalid_json") from exc
+
+
+@register_processor
+class PawProcessor(AbstractProcessor):
+    """Send only rendered short text to the code-owned PAW sidecar."""
+
+    processor_type = "paw"
+
+    async def process(
+        self,
+        records: list["CollectedRecord"],
+        prompt_template: str,
+        config: dict[str, Any],
+    ) -> ProcessingResult:
+        cfg = _merged_config(config)
+        settings = get_settings()
+        if not _PROGRAM_ID_RE.fullmatch(settings.paw_program_id):
+            return ProcessingResult(success=False, error="runtime.program_not_configured")
+        if not _is_local_paw_url(settings.paw_runtime_url):
+            return ProcessingResult(
+                success=False,
+                error="runtime.endpoint_rejected: paw.local_only",
+            )
+        if _has_sensitive_placeholder(prompt_template):
+            return ProcessingResult(success=False, error="request.sensitive_placeholder")
+        try:
+            max_tokens = int(cfg.get("max_tokens", settings.paw_max_tokens))
+        except (TypeError, ValueError):
+            return ProcessingResult(success=False, error="request.max_tokens_invalid")
+        if not 1 <= max_tokens <= min(MAX_TOKENS, settings.paw_max_tokens):
+            return ProcessingResult(success=False, error="request.max_tokens_invalid")
+        output_schema = cfg.get("output_schema")
+        if output_schema is None:
+            output_validator = None
+        elif not isinstance(output_schema, dict):
+            return ProcessingResult(success=False, error="request.output_schema_invalid")
+        else:
+            try:
+                Draft202012Validator.check_schema(output_schema)
+                output_validator = Draft202012Validator(output_schema)
+            except SchemaError:
+                return ProcessingResult(success=False, error="request.output_schema_invalid")
+
+        enrichments: list[dict[str, Any]] = [{} for _ in records]
+        failed_indices: set[int] = set()
+        inputs: list[str | None] = []
+        for index, record in enumerate(records):
+            context = getattr(record, "normalized_data", {})
+            if not isinstance(context, dict):
+                context = {}
+            input_text = _render(prompt_template, context)
+            if not input_text.strip():
+                failed_indices.add(index)
+                inputs.append(None)
+                logger.warning(
+                    "PAW input rejected [%d/%d]: request.input_empty", index + 1, len(records)
+                )
+                continue
+            if len(input_text) > MAX_INPUT_CHARS:
+                failed_indices.add(index)
+                inputs.append(None)
+                logger.warning(
+                    "PAW input rejected [%d/%d]: request.input_too_large", index + 1, len(records)
+                )
+                continue
+            inputs.append(input_text)
+
+        if len(failed_indices) == len(records):
+            return ProcessingResult(
+                success=False,
+                enrichments=enrichments,
+                error="paw.enrichment_failed",
+                failed_indices=failed_indices,
+            )
+
+        try:
+            client, endpoint = await guarded_async_client(
+                settings.paw_runtime_url,
+                allow_private=True,
+                timeout=settings.paw_runtime_timeout_seconds,
+            )
+        except SSRFValidationError as exc:
+            return ProcessingResult(success=False, error=f"runtime.endpoint_rejected: {exc}")
+
+        async with client as opened_client:
+            for index, input_text in enumerate(inputs):
+                if input_text is None:
+                    continue
+                response = None
+                try:
+                    request = opened_client.build_request(
+                        "POST",
+                        f"{endpoint.rstrip('/')}/v1/enrich",
+                        json={
+                            "programId": settings.paw_program_id,
+                            "input": input_text,
+                            "maxTokens": max_tokens,
+                        },
+                    )
+                    response = await opened_client.send(request, stream=True)
+                    response.raise_for_status()
+                    enrichments[index] = _validate_response(
+                        await _read_bounded_json(response),
+                        settings.paw_program_id,
+                        output_validator,
+                    )
+                except Exception as exc:
+                    failed_indices.add(index)
+                    logger.warning(
+                        "PAW enrichment rejected [%d/%d]: %s",
+                        index + 1,
+                        len(records),
+                        exc,
+                    )
+                finally:
+                    if response is not None:
+                        await response.aclose()
+
+        return ProcessingResult(
+            success=not failed_indices,
+            enrichments=enrichments,
+            error="paw.enrichment_failed" if failed_indices else None,
+            failed_indices=failed_indices,
+        )
diff --git a/backend/processors/registry.py b/backend/processors/registry.py
index 88a43c6d..ac2fe274 100644
--- a/backend/processors/registry.py
+++ b/backend/processors/registry.py
@@ -28,6 +28,7 @@ def _load_all_processors() -> None:
         external_http_processor,
         local_processor,
         openai_processor,
+        paw_processor,
     )
 
 
diff --git a/backend/scheduler.py b/backend/scheduler.py
index edc4af7a..ae7c263e 100644
--- a/backend/scheduler.py
+++ b/backend/scheduler.py
@@ -144,6 +144,36 @@ async def _scheduler_loop() -> None:
                         "schedule %s dispatch failed: %s", sched["schedule_id"], exc,
                     )
 
+            try:
+                from backend.services.automation_schedule_service import (
+                    dispatch_due_automations,
+                )
+
+                automation_runs = await dispatch_due_automations(last_tick, now)
+                if automation_runs:
+                    logger.info(
+                        "Dispatched %d scheduled Automation occurrence(s): %s",
+                        len(automation_runs),
+                        [run.id for run in automation_runs],
+                    )
+            except Exception as exc:
+                logger.warning("Automation scheduler tick failed: %s", exc)
+
+            try:
+                from backend.services.scheduled_run_recovery import (
+                    recover_queued_scheduled_runs_local,
+                )
+
+                queued_run_ids = await recover_queued_scheduled_runs_local()
+                if queued_run_ids:
+                    logger.info(
+                        "Recovered %d queued scheduled Operations Agent run(s): %s",
+                        len(queued_run_ids),
+                        queued_run_ids,
+                    )
+            except Exception as exc:
+                logger.warning("Scheduled Operations Agent recovery failed: %s", exc)
+
             last_tick = now
         except asyncio.CancelledError:
             break
diff --git a/backend/schemas/agent.py b/backend/schemas/agent.py
index 82056c0c..f5134420 100644
--- a/backend/schemas/agent.py
+++ b/backend/schemas/agent.py
@@ -1,10 +1,44 @@
 from datetime import datetime
 from typing import Any, Optional
 
+from jsonschema import Draft202012Validator
+from jsonschema.exceptions import SchemaError
 from pydantic import BaseModel, Field
 
 from backend.schemas.common import UTCModel
 
+PAW_MAX_PROMPT_CHARS = 8_192
+PAW_MAX_TOKENS = 512
+PAW_CONFIG_KEYS = frozenset({"max_tokens", "output_schema"})
+
+
+def validate_paw_agent_config(
+    processor_type: str, prompt_template: str, processor_config: dict[str, Any]
+) -> None:
+    if processor_type != "paw":
+        return
+    if not isinstance(prompt_template, str) or not prompt_template.strip():
+        raise ValueError("paw.prompt_template_required")
+    if len(prompt_template) > PAW_MAX_PROMPT_CHARS:
+        raise ValueError("paw.prompt_template_too_large")
+    if not isinstance(processor_config, dict) or set(processor_config) - PAW_CONFIG_KEYS:
+        raise ValueError("paw.processor_config_invalid")
+    max_tokens = processor_config.get("max_tokens")
+    if max_tokens is not None and (
+        isinstance(max_tokens, bool)
+        or not isinstance(max_tokens, int)
+        or not 1 <= max_tokens <= PAW_MAX_TOKENS
+    ):
+        raise ValueError("paw.max_tokens_invalid")
+    output_schema = processor_config.get("output_schema")
+    if output_schema is not None:
+        if not isinstance(output_schema, dict):
+            raise ValueError("paw.output_schema_invalid")
+        try:
+            Draft202012Validator.check_schema(output_schema)
+        except SchemaError as exc:
+            raise ValueError("paw.output_schema_invalid") from exc
+
 
 class AIAgentCreate(BaseModel):
     name: str = Field(..., min_length=1, max_length=255)
diff --git a/backend/schemas/automation.py b/backend/schemas/automation.py
index cf97b351..c51039d7 100644
--- a/backend/schemas/automation.py
+++ b/backend/schemas/automation.py
@@ -1,15 +1,24 @@
 from datetime import datetime
 from typing import Literal
 
-from pydantic import BaseModel, Field
+from pydantic import BaseModel, Field, field_validator, model_validator
 
+from backend.automation_schedule import parse_automation_schedule, parse_automation_timezone
 from backend.schemas.common import UTCModel
 
 SessionMode = Literal["fresh", "reuse"]
 ApprovalMode = Literal["observe_only", "suggest_changes", "low_risk_automatic"]
+StarterKey = Literal["daily-run-brief", "weekly-system-review", "anomaly-follow-up"]
+STARTER_KEYS: tuple[str, ...] = (
+    "daily-run-brief",
+    "weekly-system-review",
+    "anomaly-follow-up",
+)
 
 
 class AutomationCreate(BaseModel):
+    operations_agent_id: str | None = None
+    operations_agent_version: int | None = Field(default=None, ge=1)
     name: str = Field(min_length=1, max_length=255)
     prompt: str = Field(min_length=1, max_length=20000)
     precheck: str | None = Field(default=None, max_length=4000)
@@ -20,9 +29,32 @@ class AutomationCreate(BaseModel):
     approval_mode: ApprovalMode = "suggest_changes"
     project: dict = Field(default_factory=dict)
     enabled: bool = True
+    starter_key: StarterKey | None = None
 
+    @field_validator("schedule")
+    @classmethod
+    def schedule_is_supported(cls, value: str) -> str:
+        parse_automation_schedule(value)
+        return value
+
+    @field_validator("timezone")
+    @classmethod
+    def timezone_is_supported(cls, value: str) -> str:
+        parse_automation_timezone(value)
+        return value
+
+    @model_validator(mode="after")
+    def enabled_automation_has_pinned_agent(self):
+        paired = self.operations_agent_id is not None and self.operations_agent_version is not None
+        if (self.operations_agent_id is None) != (self.operations_agent_version is None):
+            raise ValueError("operations_agent_id and operations_agent_version must be set together")
+        if self.enabled and not paired:
+            raise ValueError("enabled Automation requires a pinned published Operations Agent")
+        return self
 
 class AutomationUpdate(BaseModel):
+    operations_agent_id: str | None = None
+    operations_agent_version: int | None = Field(default=None, ge=1)
     name: str | None = Field(default=None, min_length=1, max_length=255)
     prompt: str | None = Field(default=None, min_length=1, max_length=20000)
     precheck: str | None = Field(default=None, max_length=4000)
@@ -34,10 +66,40 @@ class AutomationUpdate(BaseModel):
     project: dict | None = None
     enabled: bool | None = None
 
+    @field_validator("schedule")
+    @classmethod
+    def updated_schedule_is_supported(cls, value: str | None) -> str | None:
+        if value is not None:
+            parse_automation_schedule(value)
+        return value
+
+    @field_validator("timezone")
+    @classmethod
+    def updated_timezone_is_supported(cls, value: str | None) -> str | None:
+        if value is not None:
+            parse_automation_timezone(value)
+        return value
+
+    @model_validator(mode="after")
+    def updated_agent_binding_is_paired(self):
+        fields = self.model_fields_set
+        binding_fields = {"operations_agent_id", "operations_agent_version"}
+        if fields & binding_fields and not binding_fields <= fields:
+            raise ValueError("operations_agent_id and operations_agent_version must be updated together")
+        if binding_fields <= fields and (
+            (self.operations_agent_id is None) != (self.operations_agent_version is None)
+        ):
+            raise ValueError("operations_agent_id and operations_agent_version must both be set or null")
+        return self
+
 
 class AutomationRead(UTCModel):
     id: str
     workspace_id: str
+    starter_key: StarterKey | None
+    revision: int
+    operations_agent_id: str | None
+    operations_agent_version: int | None
     name: str
     prompt: str
     precheck: str | None
@@ -53,3 +115,41 @@ class AutomationRead(UTCModel):
     updated_at: datetime
 
     model_config = {"from_attributes": True}
+
+
+class AutomationSchedulerTickRequest(BaseModel):
+    fired_at: datetime
+
+    @field_validator("fired_at")
+    @classmethod
+    def fired_at_is_timezone_aware(cls, value: datetime) -> datetime:
+        if value.tzinfo is None:
+            raise ValueError("fired_at must include a timezone")
+        return value
+
+
+class AutomationSchedulerTickResult(BaseModel):
+    run_ids: list[str]
+    occurrence_references: list[str]
+    queued_run_ids: list[str]
+
+
+class StarterPreviewItem(BaseModel):
+    key: StarterKey
+    name: str
+    installed: bool
+    automation_id: str | None = None
+
+
+class StarterInstallationPreview(BaseModel):
+    workspace_id: str
+    starters: list[StarterPreviewItem]
+    missing_count: int
+    installed_count: int
+
+
+class StarterInstallationResult(StarterInstallationPreview):
+    created_count: int
+    skipped_count: int
+
+    model_config = {"from_attributes": True}
diff --git a/backend/schemas/edge_node.py b/backend/schemas/edge_node.py
index 677df308..9a05ad3f 100644
--- a/backend/schemas/edge_node.py
+++ b/backend/schemas/edge_node.py
@@ -15,6 +15,7 @@ class EdgeNodeRead(BaseModel):
     last_seen_at: datetime | None
     ip: str | None
     runtimes: list[str] | None = None
+    runtime_capabilities: dict[str, list[str]] | None = None
     created_at: datetime
     updated_at: datetime
 
diff --git a/backend/schemas/notification.py b/backend/schemas/notification.py
index 1e78ed70..e537d7b1 100644
--- a/backend/schemas/notification.py
+++ b/backend/schemas/notification.py
@@ -51,6 +51,7 @@ class NotificationLogRead(UTCModel):
     id: str
     rule_id: str
     record_id: str | None
+    lineage: dict[str, Any] | None
     status: str
     response_data: dict[str, Any] | None
     error_message: str | None
@@ -59,8 +60,6 @@ class NotificationLogRead(UTCModel):
     acked_at: datetime | None
     created_at: datetime
 
-    model_config = {"from_attributes": True}
-
 
 class NotificationAckRequest(BaseModel):
     status: Literal["acked", "failed"]
diff --git a/backend/schemas/operations_agent.py b/backend/schemas/operations_agent.py
index f0dc1197..f823eb14 100644
--- a/backend/schemas/operations_agent.py
+++ b/backend/schemas/operations_agent.py
@@ -13,20 +13,36 @@
 
 AGENT_CONTRACT_CONFIGURATION_KEY = "agent_contract"
 AGENT_RUNTIME_BINDING_CONFIGURATION_KEY = "runtime_binding"
+DEFAULT_DEEP_RUN_TIMEOUT_SECONDS = 1800
+MAX_DEEP_RUN_TIMEOUT_SECONDS = 3600
 MAX_AGENT_SCHEMA_BYTES = 65_536
 MAX_AGENT_SCHEMA_DEPTH = 32
 MAX_AGENT_MODEL_CONFIGURATION_BYTES = 262_144
 
 
-class AgentContractV1(BaseModel):
-    """Versioned input, output, and state boundary for an Operations Agent."""
+class AgentQualityGateV1(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    id: str = Field(min_length=1, max_length=100)
+    required: bool = True
+    config: dict[str, JsonValue] = Field(default_factory=dict)
+
+
+class AgentContractV2(BaseModel):
+    """Runtime-neutral business role, I/O, policy, and evidence boundary."""
 
     model_config = ConfigDict(extra="forbid")
 
-    schema_version: Literal["agent.contract.v1"]
+    schema_version: Literal["agent.contract.v2"]
+    role: str = Field(min_length=1, max_length=100)
     input_schema: dict[str, JsonValue]
     output_schema: dict[str, JsonValue]
     state_schema: dict[str, JsonValue]
+    required_capabilities: list[str] = Field(default_factory=list, max_length=64)
+    tool_policy: dict[str, JsonValue] = Field(default_factory=dict)
+    budget: dict[str, JsonValue] = Field(default_factory=dict)
+    quality_gates: list[AgentQualityGateV1] = Field(default_factory=list, max_length=64)
+    evidence_requirements: list[str] = Field(default_factory=list, max_length=64)
 
     @field_validator("input_schema", "output_schema", "state_schema")
     @classmethod
@@ -52,18 +68,76 @@ def schemas_are_valid_json_schema(cls, schema: dict[str, JsonValue]) -> dict[str
             raise ValueError(f"invalid JSON Schema: {exc.message}") from exc
         return schema
 
+    @field_validator("required_capabilities", "evidence_requirements")
+    @classmethod
+    def identifiers_are_normalized(cls, values: list[str]) -> list[str]:
+        normalized = [value.strip() for value in values]
+        if any(
+            not value
+            or len(value) > 64
+            or not value.replace(".", "_").replace("-", "_").replace("_", "").isalnum()
+            for value in normalized
+        ):
+            raise ValueError("capability and evidence identifiers must be simple names")
+        if len(set(normalized)) != len(normalized):
+            raise ValueError("capability and evidence identifiers must be unique")
+        return normalized
+
+    @model_validator(mode="after")
+    def quality_gate_ids_are_unique(self):
+        ids = [gate.id for gate in self.quality_gates]
+        if len(set(ids)) != len(ids):
+            raise ValueError("quality gate ids must be unique")
+        return self
 
-class AgentRuntimeBindingV1(BaseModel):
-    """Published binding from one Operations Agent version to an existing edge runtime."""
+
+class AgentModelBindingV1(BaseModel):
+    """Non-secret model selection; credentials are resolved by the edge runtime."""
+
+    model_config = ConfigDict(extra="forbid")
+
+    schema_version: Literal["agent.model-binding.v1"]
+    provider: str = Field(min_length=1, max_length=100)
+    model: str = Field(min_length=1, max_length=255)
+    auth_profile: str | None = Field(default=None, min_length=1, max_length=255)
+
+
+class AgentRuntimeBindingV2(BaseModel):
+    """Capability policy for selecting an edge runtime at dispatch time."""
 
     model_config = ConfigDict(extra="forbid")
 
-    schema_version: Literal["agent.runtime-binding.v1"]
-    agent_url: str = Field(min_length=1, max_length=512)
-    runtime: Literal["pi"]
+    schema_version: Literal["agent.runtime-binding.v2"]
     workflow: str = Field(min_length=1, max_length=255)
+    preferred_agent_urls: list[str] = Field(default_factory=list, max_length=32)
+    preferred_runtimes: list[str] = Field(default_factory=list, max_length=32)
+    model_binding: AgentModelBindingV1 | None = None
     config: dict[str, JsonValue] = Field(default_factory=dict)
-    dispatch_timeout_seconds: int = Field(default=600, ge=1, le=3600)
+    dispatch_timeout_seconds: int = Field(
+        default=DEFAULT_DEEP_RUN_TIMEOUT_SECONDS,
+        ge=1,
+        le=MAX_DEEP_RUN_TIMEOUT_SECONDS,
+    )
+
+    @field_validator("preferred_agent_urls")
+    @classmethod
+    def agent_urls_are_http(cls, values: list[str]) -> list[str]:
+        normalized = [value.rstrip("/") for value in values]
+        if any(not value.startswith(("http://", "https://")) for value in normalized):
+            raise ValueError("preferred_agent_urls must contain only http/https URLs")
+        if len(set(normalized)) != len(normalized):
+            raise ValueError("preferred_agent_urls must be unique")
+        return normalized
+
+    @field_validator("preferred_runtimes")
+    @classmethod
+    def runtimes_are_normalized(cls, values: list[str]) -> list[str]:
+        normalized = [value.strip() for value in values]
+        if any(not value or len(value) > 100 for value in normalized):
+            raise ValueError("preferred_runtimes must contain non-empty names")
+        if len(set(normalized)) != len(normalized):
+            raise ValueError("preferred_runtimes must be unique")
+        return normalized
 
     @field_validator("config")
     @classmethod
@@ -78,31 +152,26 @@ def config_is_task_scoped(cls, value: dict[str, JsonValue]) -> dict[str, JsonVal
         if timeout is not None and (
             not isinstance(timeout, (int, float))
             or isinstance(timeout, bool)
-            or not 1 <= timeout <= 3600
+            or not 1 <= timeout <= MAX_DEEP_RUN_TIMEOUT_SECONDS
         ):
-            raise ValueError("config.timeout_seconds must be between 1 and 3600")
+            raise ValueError(
+                "config.timeout_seconds must be between 1 and "
+                f"{MAX_DEEP_RUN_TIMEOUT_SECONDS}"
+            )
         return value
 
-    @field_validator("agent_url")
-    @classmethod
-    def agent_url_is_http(cls, value: str) -> str:
-        normalized = value.rstrip("/")
-        if not normalized.startswith(("http://", "https://")):
-            raise ValueError("agent_url must be an http/https URL")
-        return normalized
-
 
 def validated_agent_model_configuration(model_configuration: dict[str, Any]) -> dict[str, Any]:
     """Return a detached configuration with canonical versioned agent contracts."""
 
     configuration = deepcopy(model_configuration)
     if AGENT_CONTRACT_CONFIGURATION_KEY in configuration:
-        configuration[AGENT_CONTRACT_CONFIGURATION_KEY] = AgentContractV1.model_validate(
+        configuration[AGENT_CONTRACT_CONFIGURATION_KEY] = AgentContractV2.model_validate(
             configuration[AGENT_CONTRACT_CONFIGURATION_KEY]
         ).model_dump(mode="json")
     if AGENT_RUNTIME_BINDING_CONFIGURATION_KEY in configuration:
         configuration[AGENT_RUNTIME_BINDING_CONFIGURATION_KEY] = (
-            AgentRuntimeBindingV1.model_validate(
+            AgentRuntimeBindingV2.model_validate(
                 configuration[AGENT_RUNTIME_BINDING_CONFIGURATION_KEY]
             ).model_dump(mode="json")
         )
@@ -116,20 +185,20 @@ def validated_agent_model_configuration(model_configuration: dict[str, Any]) ->
 
 def agent_contract_from_model_configuration(
     model_configuration: dict[str, Any],
-) -> AgentContractV1 | None:
+) -> AgentContractV2 | None:
     contract = model_configuration.get(AGENT_CONTRACT_CONFIGURATION_KEY)
-    return None if contract is None else AgentContractV1.model_validate(contract)
+    return None if contract is None else AgentContractV2.model_validate(contract)
 
 
 def agent_runtime_binding_from_model_configuration(
     model_configuration: dict[str, Any],
-) -> AgentRuntimeBindingV1 | None:
+) -> AgentRuntimeBindingV2 | None:
     binding = model_configuration.get(AGENT_RUNTIME_BINDING_CONFIGURATION_KEY)
-    return None if binding is None else AgentRuntimeBindingV1.model_validate(binding)
+    return None if binding is None else AgentRuntimeBindingV2.model_validate(binding)
 
 
 def validate_agent_contract_payload(
-    contract: AgentContractV1,
+    contract: AgentContractV2,
     schema_field: Literal["input_schema", "output_schema", "state_schema"],
     payload: dict[str, JsonValue],
 ) -> None:
@@ -150,7 +219,7 @@ def validate_agent_contract_payload(
 class OperationsAgentCreate(BaseModel):
     name: str = Field(min_length=1, max_length=255)
     description: str | None = Field(default=None, max_length=4000)
-    owning_team_id: str
+    owning_team_id: str | None = None
 
 
 class OperationsAgentPatch(BaseModel):
@@ -204,6 +273,20 @@ class OperationsAgentRunCreate(BaseModel):
     state_payload: dict[str, JsonValue] = Field(default_factory=dict)
 
 
+class AgentRunEvidenceEnvelopeV1(BaseModel):
+    """Runtime-independent event, artifact, evidence, lineage, and audit record."""
+
+    model_config = ConfigDict(extra="forbid")
+
+    schema_version: Literal["agent.run-evidence.v1"] = "agent.run-evidence.v1"
+    runtime: dict[str, JsonValue]
+    events: list[dict[str, JsonValue]] = Field(default_factory=list)
+    artifacts: list[dict[str, JsonValue]] = Field(default_factory=list)
+    evidence: list[dict[str, JsonValue]] = Field(default_factory=list)
+    lineage: list[dict[str, JsonValue]] = Field(default_factory=list)
+    audit: list[dict[str, JsonValue]] = Field(default_factory=list)
+
+
 class OperationsAgentRunRead(UTCModel):
     id: str
     workspace_id: str
@@ -212,11 +295,18 @@ class OperationsAgentRunRead(UTCModel):
     profile_version: int
     trigger_type: str
     trigger_reference: str | None
+    automation_id: str | None
+    automation_revision: int | None
+    automation_snapshot: dict[str, JsonValue] | None
+    scheduled_for: datetime | None
+    schedule_timezone: str | None
     target_resource_type: str
     target_resource_id: str
     input_payload: dict[str, JsonValue]
     state_payload: dict[str, JsonValue]
     output_payload: dict[str, JsonValue] | None
+    execution_binding: dict[str, JsonValue] | None
+    evidence_payload: AgentRunEvidenceEnvelopeV1 | None
     error_message: str | None
     status: str
     started_by_user_id: str
@@ -258,6 +348,16 @@ class AgentProfileRead(UTCModel):
     model_config = {"from_attributes": True}
 
 
+class OperationsAgentTeamRead(UTCModel):
+    id: str
+    workspace_id: str
+    name: str
+    slug: str
+    created_at: datetime
+
+    model_config = {"from_attributes": True}
+
+
 class OperationsAgentRead(UTCModel):
     id: str
     workspace_id: str
diff --git a/backend/schemas/provider.py b/backend/schemas/provider.py
index f675817a..00aa0d62 100644
--- a/backend/schemas/provider.py
+++ b/backend/schemas/provider.py
@@ -4,6 +4,7 @@
 from pydantic import BaseModel, Field
 
 from backend.schemas.common import UTCModel
+from backend.schemas.provider_capacity import ProviderCapacityRead, project_provider_capacity
 
 
 class ModelProviderCreate(BaseModel):
@@ -60,6 +61,7 @@ class ModelProviderRead(UTCModel):
     default_model: Optional[str]
     notes: Optional[str]
     enabled: bool
+    capacity: ProviderCapacityRead
     created_at: datetime
     updated_at: datetime
 
@@ -86,6 +88,7 @@ def from_model(cls, provider: Any) -> "ModelProviderRead":
                 "default_model": provider.default_model,
                 "notes": provider.notes,
                 "enabled": provider.enabled,
+                "capacity": project_provider_capacity(provider),
                 "created_at": provider.created_at,
                 "updated_at": provider.updated_at,
             }
diff --git a/backend/schemas/provider_capacity.py b/backend/schemas/provider_capacity.py
new file mode 100644
index 00000000..4301c516
--- /dev/null
+++ b/backend/schemas/provider_capacity.py
@@ -0,0 +1,96 @@
+"""Provider capacity projections with an explicit, honest availability state.
+
+Capacity is intentionally not a quota calculator.  A provider-specific usage
+adapter may supply opaque usage data when it has a documented endpoint; when
+there is no such adapter the projection remains ``unavailable``.  No elapsed
+runtime, request count, or other local observation is used to manufacture a
+remaining percentage.
+"""
+
+from __future__ import annotations
+
+from datetime import datetime
+from enum import StrEnum
+from typing import Any, Mapping
+
+from pydantic import BaseModel, ConfigDict, Field, JsonValue, model_validator
+
+
+class ProviderCapacityState(StrEnum):
+    """Availability of provider capacity evidence."""
+
+    MEASURED = "measured"
+    UNAVAILABLE = "unavailable"
+    NOT_APPLICABLE = "not_applicable"
+
+
+class ProviderCapacityRead(BaseModel):
+    """Serialized provider capacity evidence.
+
+    ``usage`` is deliberately opaque: providers do not share a quota schema,
+    so an adapter owns the shape of its documented response.  In particular,
+    this model has no derived ``remaining_percent`` field.
+    """
+
+    model_config = ConfigDict(extra="forbid")
+
+    state: ProviderCapacityState
+    usage: dict[str, JsonValue] | None = None
+    measured_at: datetime | None = None
+    source: str | None = Field(default=None, min_length=1, max_length=255)
+    reason: str | None = Field(default=None, min_length=1, max_length=1000)
+
+    @model_validator(mode="after")
+    def validate_evidence(self) -> ProviderCapacityRead:
+        if self.state is ProviderCapacityState.MEASURED:
+            if self.usage is None:
+                raise ValueError("measured capacity requires adapter usage data")
+            if self.reason is not None:
+                raise ValueError("measured capacity cannot include an unavailable reason")
+            return self
+        if self.usage is not None or self.measured_at is not None:
+            raise ValueError("unavailable capacity must not include measured usage data")
+        return self
+
+    @classmethod
+    def unavailable(cls, *, reason: str = "No supported provider usage endpoint") -> ProviderCapacityRead:
+        return cls(state=ProviderCapacityState.UNAVAILABLE, reason=reason)
+
+    @classmethod
+    def not_applicable(cls, *, reason: str = "Runtime has no provider quota semantics") -> ProviderCapacityRead:
+        return cls(state=ProviderCapacityState.NOT_APPLICABLE, reason=reason)
+
+    @classmethod
+    def measured(
+        cls,
+        usage: Mapping[str, JsonValue],
+        *,
+        source: str,
+        measured_at: datetime | None = None,
+    ) -> ProviderCapacityRead:
+        """Build a measured projection from an explicit provider adapter result."""
+        return cls(
+            state=ProviderCapacityState.MEASURED,
+            usage=dict(usage),
+            measured_at=measured_at,
+            source=source,
+        )
+
+
+def project_provider_capacity(provider: Any) -> ProviderCapacityRead:
+    """Project only explicit adapter evidence from a provider-like object.
+
+    Existing ``ModelProvider`` rows do not carry usage evidence, so they
+    serialize as ``unavailable``.  A future documented adapter can attach a
+    ``capacity`` projection (or its serialized mapping) without changing this
+    API; no other provider fields are consulted.
+    """
+
+    value = getattr(provider, "capacity", None)
+    if value is None:
+        return ProviderCapacityRead.unavailable()
+    if isinstance(value, ProviderCapacityRead):
+        return value
+    if isinstance(value, Mapping):
+        return ProviderCapacityRead.model_validate(value)
+    raise TypeError("provider capacity adapter result must be a mapping or ProviderCapacityRead")
diff --git a/backend/schemas/record.py b/backend/schemas/record.py
index 84d1ba7f..907de49c 100644
--- a/backend/schemas/record.py
+++ b/backend/schemas/record.py
@@ -12,6 +12,7 @@ class CollectedRecordRead(UTCModel):
     source_id: str
     workflow_id: str | None
     workflow_run_id: str | None
+    lineage: dict[str, Any] | None
     raw_data: dict[str, Any]
     normalized_data: dict[str, Any]
     ai_enrichment: dict[str, Any] | None
@@ -20,7 +21,6 @@ class CollectedRecordRead(UTCModel):
     error_message: str | None
     created_at: datetime
     updated_at: datetime
-
     model_config = {"from_attributes": True}
 
 
diff --git a/backend/security/fleet_auth.py b/backend/security/fleet_auth.py
index 94642385..df3e463e 100644
--- a/backend/security/fleet_auth.py
+++ b/backend/security/fleet_auth.py
@@ -67,6 +67,7 @@
 from collections.abc import Sequence
 from urllib.parse import parse_qs
 
+from jose import JWTError, jwt
 from starlette.datastructures import Headers
 from starlette.responses import JSONResponse
 from starlette.types import ASGIApp, Receive, Scope, Send
@@ -77,6 +78,10 @@
 #: Path prefixes guarded by :class:`FleetAuthMiddleware`.
 PROTECTED_PREFIXES = ("/api", "/mcp")
 
+# Local login is intentionally the only unauthenticated API route. Once the
+# user has a local bearer session, the identity dependency authenticates it.
+PUBLIC_PATHS = frozenset({"/api/v1/auth/login"})
+
 _LOCALHOST_HOSTS = frozenset({"localhost", "::1"})
 
 
@@ -120,6 +125,17 @@ def enforce_bind_guard(host: str, token: str) -> None:
         "local development."
     )
 
+def _is_local_session(credential: str) -> bool:
+    try:
+        claims = jwt.decode(
+            credential,
+            get_settings().secret_key,
+            algorithms=["HS256"],
+        )
+    except JWTError:
+        return False
+    return claims.get("auth_method") == "local" and claims.get("sub") == "local-admin"
+
 
 def _token_matches(candidate: str, token: str) -> bool:
     """Constant-time comparison of a caller-supplied credential against *token*."""
@@ -147,30 +163,34 @@ class FleetAuthMiddleware:
     /health exemption rationale.
     """
 
+
     def __init__(self, app: ASGIApp) -> None:
         self.app = app
 
     async def __call__(self, scope: Scope, receive: Receive, send: Send) -> None:
-        if scope["type"] not in ("http", "websocket") or not scope["path"].startswith(
-            PROTECTED_PREFIXES
+        if (
+            scope["type"] not in ("http", "websocket")
+            or not scope["path"].startswith(PROTECTED_PREFIXES)
+            or scope["path"] in PUBLIC_PATHS
         ):
             await self.app(scope, receive, send)
             return
 
-        # Read per request: get_settings() is lru_cached (cheap), but
-        # api/v1/system.py may cache_clear() it at runtime after a config
-        # patch, so don't freeze the token at middleware construction time.
+        # Read per request so a runtime configuration update is respected.
         token = get_settings().api_auth_token
         if not token:
-            # Dev posture: no token configured -> API open. Only reachable on
-            # a localhost bind thanks to enforce_bind_guard at startup.
+            # No fleet token configured: local deployments rely on the identity
+            # dependency and the bind guard limits this posture to localhost.
             await self.app(scope, receive, send)
             return
 
         if scope["type"] == "websocket":
             headers = Headers(scope=scope)
-            credential = _bearer_credential(headers) or _query_token(scope.get("query_string", b""))
-            if credential and _token_matches(credential, token):
+            bearer = _bearer_credential(headers)
+            credential = bearer or _query_token(scope.get("query_string", b""))
+            if _is_local_session(bearer) or (
+                credential and _token_matches(credential, token)
+            ):
                 await self.app(scope, receive, send)
                 return
             await WebSocketClose(code=4401, reason="Invalid or missing API token")(
@@ -179,8 +199,9 @@ async def __call__(self, scope: Scope, receive: Receive, send: Send) -> None:
             return
 
         headers = Headers(scope=scope)
-        credential = headers.get("x-api-token", "") or _bearer_credential(headers)
-        if credential and _token_matches(credential, token):
+        bearer = _bearer_credential(headers)
+        credential = headers.get("x-api-token", "") or bearer
+        if _is_local_session(bearer) or (credential and _token_matches(credential, token)):
             await self.app(scope, receive, send)
             return
 
diff --git a/backend/security/identity.py b/backend/security/identity.py
index 09e79fbd..2564c7c1 100644
--- a/backend/security/identity.py
+++ b/backend/security/identity.py
@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import os
 import hmac
 from collections.abc import Mapping
 from dataclasses import dataclass
@@ -20,20 +21,18 @@ class IdentitySettings:
     audience: str
     jwks_url: str = ""
     bootstrap_admin_token: str = ""
+    secret_key: str = "change-me-in-production"
 
     @classmethod
     def from_env(cls) -> IdentitySettings:
-        # Sourced from Settings (backend/config.py), not raw os.getenv(): the
-        # process environment alone is not a reliable source for these — under
-        # plain `uv run uvicorn ...` uv does not inject .env into os.environ,
-        # while Settings parses .env directly via pydantic-settings regardless
-        # of what the launching process actually exported.
-        settings = get_settings()
+        from backend.config import get_settings
+
         return cls(
-            issuer=settings.oidc_issuer.rstrip("/"),
-            audience=settings.oidc_audience,
-            jwks_url=settings.oidc_jwks_url,
-            bootstrap_admin_token=settings.bootstrap_admin_token,
+            issuer=os.getenv("OIDC_ISSUER", "").rstrip("/"),
+            audience=os.getenv("OIDC_AUDIENCE", ""),
+            jwks_url=os.getenv("OIDC_JWKS_URL", ""),
+            bootstrap_admin_token=os.getenv("BOOTSTRAP_ADMIN_TOKEN", ""),
+            secret_key=get_settings().secret_key,
         )
 
 
@@ -136,6 +135,18 @@ async def get_request_identity(request: Request) -> RequestIdentity:
                 "Bearer token required",
                 headers={"WWW-Authenticate": "Bearer"},
             )
+        try:
+            local_claims = jwt.decode(token, resolved.secret_key, algorithms=["HS256"])
+        except JWTError:
+            local_claims = None
+        if local_claims and local_claims.get("auth_method") == "local":
+            return RequestIdentity(
+                subject="local-admin",
+                name=local_claims.get("name") or "本地管理员",
+                username=local_claims.get("username"),
+                is_platform_admin=True,
+                auth_method="local",
+            )
         if resolved.bootstrap_admin_token and hmac.compare_digest(
             token, resolved.bootstrap_admin_token
         ):
diff --git a/backend/security/internal_service.py b/backend/security/internal_service.py
new file mode 100644
index 00000000..c7f898c6
--- /dev/null
+++ b/backend/security/internal_service.py
@@ -0,0 +1,20 @@
+"""Authentication dependency for scheduler/worker calls into API-owned runtime paths."""
+
+import hmac
+
+from fastapi import Header, HTTPException, status
+
+from backend.config import get_settings
+
+
+def require_internal_service_token(
+    x_api_token: str | None = Header(default=None, alias="X-API-Token"),
+) -> None:
+    configured = get_settings().api_auth_token
+    if not configured:
+        raise HTTPException(
+            status.HTTP_503_SERVICE_UNAVAILABLE,
+            "Internal service token is not configured",
+        )
+    if not x_api_token or not hmac.compare_digest(x_api_token, configured):
+        raise HTTPException(status.HTTP_401_UNAUTHORIZED, "Invalid internal service token")
diff --git a/backend/security/local_auth.py b/backend/security/local_auth.py
new file mode 100644
index 00000000..428449b9
--- /dev/null
+++ b/backend/security/local_auth.py
@@ -0,0 +1,88 @@
+"""Local-first administrator password and session helpers."""
+
+from __future__ import annotations
+
+import base64
+import hashlib
+import hmac
+import os
+import re
+import secrets
+from datetime import UTC, datetime, timedelta
+from pathlib import Path
+
+from jose import jwt
+
+
+def hash_password(password: str) -> str:
+    salt = secrets.token_bytes(16)
+    n, r, p = 16384, 8, 1
+    digest = hashlib.scrypt(password.encode("utf-8"), salt=salt, n=n, r=r, p=p)
+    encode = base64.urlsafe_b64encode
+    return f"scrypt${n}${r}${p}${encode(salt).decode()}${encode(digest).decode()}"
+
+DEFAULT_LOCAL_ADMIN_PASSWORD_HASH = hash_password("admin")
+
+
+def verify_password(password: str, encoded: str) -> bool:
+    try:
+        scheme, n, r, p, salt_text, digest_text = encoded.split("$", 5)
+        if scheme != "scrypt":
+            return False
+        salt = base64.urlsafe_b64decode(salt_text.encode("ascii"))
+        expected = base64.urlsafe_b64decode(digest_text.encode("ascii"))
+        actual = hashlib.scrypt(
+            password.encode("utf-8"),
+            salt=salt,
+            n=int(n),
+            r=int(r),
+            p=int(p),
+            maxmem=64 * 1024 * 1024,
+        )
+        return hmac.compare_digest(actual, expected)
+    except (ValueError, TypeError, UnicodeError):
+        return False
+
+
+def issue_local_token(username: str, secret_key: str) -> str:
+    now = datetime.now(UTC)
+    return jwt.encode(
+        {
+            "sub": "local-admin",
+            "name": "本地管理员",
+            "username": username,
+            "is_platform_admin": True,
+            "auth_method": "local",
+            "iat": now,
+            "exp": now + timedelta(days=30),
+        },
+        secret_key,
+        algorithm="HS256",
+    )
+
+
+def persist_password_hash(password_hash: str) -> None:
+    """Persist the local password in the deployment .env and current process."""
+
+    os.environ["LOCAL_ADMIN_PASSWORD_HASH"] = password_hash
+    path = os.environ.get("ENV_FILE_PATH")
+    if not path:
+        for candidate in (Path("/app/.env"), Path(__file__).resolve().parents[2] / ".env"):
+            if candidate.exists():
+                path = str(candidate)
+                break
+        else:
+            path = str(Path(__file__).resolve().parents[2] / ".env")
+
+    env_path = Path(path)
+    try:
+        content = env_path.read_text(encoding="utf-8")
+    except FileNotFoundError:
+        content = ""
+    new_line = f"LOCAL_ADMIN_PASSWORD_HASH={password_hash}"
+    pattern = r"^LOCAL_ADMIN_PASSWORD_HASH=.*$"
+    if re.search(pattern, content, re.MULTILINE):
+        content = re.sub(pattern, new_line, content, flags=re.MULTILINE)
+    else:
+        content = content.rstrip("\n") + f"\n{new_line}\n"
+    env_path.write_text(content, encoding="utf-8")
diff --git a/backend/services/agent_runtime_selection.py b/backend/services/agent_runtime_selection.py
new file mode 100644
index 00000000..660c70d9
--- /dev/null
+++ b/backend/services/agent_runtime_selection.py
@@ -0,0 +1,111 @@
+"""Select connected edge runtimes from declared capabilities, never product branches."""
+
+from typing import Any
+
+from sqlalchemy import select
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from backend.models.edge_node import EdgeNode
+from backend.schemas.operations_agent import AgentContractV2, AgentRuntimeBindingV2
+from backend import ws_agent_manager
+
+
+class RuntimeSelectionError(RuntimeError):
+    """No connected Fleet runtime satisfies the published Agent contract."""
+
+
+def _preference_rank(value: str, preferred: list[str]) -> tuple[int, int | str]:
+    if value in preferred:
+        return (0, preferred.index(value))
+    return (1, value)
+
+
+async def select_agent_runtime(
+    db: AsyncSession,
+    *,
+    contract: AgentContractV2,
+    binding: AgentRuntimeBindingV2,
+    require_connected: bool = True,
+) -> dict[str, Any]:
+    """Return a non-secret immutable selection snapshot for one run."""
+
+    required = set(contract.required_capabilities)
+    if binding.model_binding is not None:
+        required.add("model_selection")
+    if contract.tool_policy:
+        required.add("tool_policy")
+    if contract.budget:
+        required.add("budget_control")
+    if contract.quality_gates:
+        required.add("quality_gates")
+
+    nodes = list(
+        (
+            await db.execute(
+                select(EdgeNode).where(
+                    EdgeNode.protocol == "ws",
+                    *(
+                        (EdgeNode.status == "online",)
+                        if require_connected
+                        else ()
+                    ),
+                )
+            )
+        )
+        .scalars()
+        .all()
+    )
+    candidates: list[
+        tuple[
+            tuple[int, int | str],
+            tuple[int, int | str],
+            str,
+            str,
+            EdgeNode,
+            list[str],
+        ]
+    ] = []
+    for node in nodes:
+        if require_connected and not ws_agent_manager.is_connected(node.url):
+            continue
+        manifests = node.runtime_capabilities or {}
+        if not isinstance(manifests, dict):
+            continue
+        for runtime, advertised in manifests.items():
+            if not isinstance(runtime, str) or not isinstance(advertised, list):
+                continue
+            capabilities = sorted({item for item in advertised if isinstance(item, str)})
+            if not required.issubset(capabilities):
+                continue
+            candidates.append(
+                (
+                    _preference_rank(node.url, binding.preferred_agent_urls),
+                    _preference_rank(runtime, binding.preferred_runtimes),
+                    node.id,
+                    runtime,
+                    node,
+                    capabilities,
+                )
+            )
+
+    if not candidates:
+        names = ", ".join(sorted(required)) or "none"
+        raise RuntimeSelectionError(
+            f"no connected Agent Runtime satisfies required capabilities: {names}"
+        )
+
+    _, _, _, runtime, node, capabilities = min(
+        candidates,
+        key=lambda item: (item[0], item[1], item[2]),
+    )
+    model_binding = binding.model_binding
+    return {
+        "schema_version": "agent.runtime-selection.v1",
+        "agent_url": node.url,
+        "runtime": runtime,
+        "workflow": binding.workflow,
+        "capabilities": capabilities,
+        "provider": model_binding.provider if model_binding else None,
+        "model": model_binding.model if model_binding else None,
+        "auth_profile": model_binding.auth_profile if model_binding else None,
+    }
diff --git a/backend/services/automation_schedule_service.py b/backend/services/automation_schedule_service.py
new file mode 100644
index 00000000..adbd4647
--- /dev/null
+++ b/backend/services/automation_schedule_service.py
@@ -0,0 +1,400 @@
+"""Durable Automation-to-Operations-Agent scheduling and run lineage."""
+
+import logging
+from datetime import UTC, datetime, timedelta
+from typing import Any
+
+from pydantic import ValidationError
+from sqlalchemy import select
+from sqlalchemy.exc import IntegrityError
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from backend.automation_schedule import automation_fire_times
+from backend.database import AsyncSessionLocal, commit_session, queue_after_commit
+from backend.models.automation import Automation
+from backend.models.operations_agent import (
+    AgentPermissionProfile,
+    AgentProfileMode,
+    OperationsAgentIdentity,
+    OperationsAgentRun,
+    PublishedOperationsAgentVersion,
+)
+from backend.schemas.operations_agent import (
+    agent_contract_from_model_configuration,
+    agent_runtime_binding_from_model_configuration,
+    validate_agent_contract_payload,
+)
+from backend.services.agent_runtime_selection import (
+    RuntimeSelectionError,
+    select_agent_runtime,
+)
+from backend.services.operations_agent_runtime_service import schedule_operations_agent_run
+
+logger = logging.getLogger(__name__)
+
+
+class AutomationBindingError(ValueError):
+    """The persisted Automation binding cannot safely produce a run."""
+
+
+def automation_snapshot(automation: Automation) -> dict[str, Any]:
+    """Freeze every mutable Automation execution field into run lineage."""
+    return {
+        "id": automation.id,
+        "revision": automation.revision,
+        "name": automation.name,
+        "prompt": automation.prompt,
+        "precheck": automation.precheck,
+        "executor": automation.executor,
+        "schedule": automation.schedule,
+        "timezone": automation.timezone,
+        "session_mode": automation.session_mode,
+        "approval_mode": automation.approval_mode,
+        "project": automation.project,
+        "enabled": automation.enabled,
+        "operations_agent_id": automation.operations_agent_id,
+        "operations_agent_version": automation.operations_agent_version,
+    }
+
+
+def automation_run_input(
+    automation: Automation,
+    *,
+    scheduled_for: datetime | None,
+) -> dict[str, Any]:
+    snapshot = automation_snapshot(automation)
+    return {
+        "automation": snapshot,
+        "prompt": automation.prompt,
+        "scheduled_for": scheduled_for.astimezone(UTC).isoformat() if scheduled_for else None,
+    }
+
+
+async def validate_automation_binding(
+    session: AsyncSession,
+    automation: Automation,
+    *,
+    scheduled_for: datetime | None = None,
+    require_online: bool = False,
+) -> tuple[
+    OperationsAgentIdentity,
+    PublishedOperationsAgentVersion,
+    AgentPermissionProfile,
+    dict[str, Any],
+]:
+    """Resolve and validate the exact Agent/version/profile/runtime contract."""
+    if automation.operations_agent_id is None or automation.operations_agent_version is None:
+        raise AutomationBindingError("Automation requires a pinned Operations Agent version")
+
+    agent = await session.scalar(
+        select(OperationsAgentIdentity).where(
+            OperationsAgentIdentity.id == automation.operations_agent_id,
+            OperationsAgentIdentity.workspace_id == automation.workspace_id,
+        )
+    )
+    if agent is None:
+        raise AutomationBindingError("Bound Operations Agent must belong to Automation Workspace")
+    if agent.disabled:
+        raise AutomationBindingError("Bound Operations Agent is disabled")
+
+    version = await session.scalar(
+        select(PublishedOperationsAgentVersion).where(
+            PublishedOperationsAgentVersion.operations_agent_id == agent.id,
+            PublishedOperationsAgentVersion.version == automation.operations_agent_version,
+        )
+    )
+    if version is None:
+        raise AutomationBindingError("Bound Operations Agent version is not published")
+
+    profile = await session.scalar(
+        select(AgentPermissionProfile).where(
+            AgentPermissionProfile.operations_agent_id == agent.id,
+            AgentPermissionProfile.version == agent.current_profile_version,
+        )
+    )
+    if profile is None:
+        raise AutomationBindingError("Bound Operations Agent profile is missing")
+    if profile.mode == AgentProfileMode.LOW_RISK_AUTOMATIC:
+        raise AutomationBindingError("Automation cannot bind a Low-Risk Automatic profile")
+    if profile.mode != automation.approval_mode:
+        raise AutomationBindingError(
+            "Automation approval_mode must match the bound Operations Agent profile"
+        )
+
+    try:
+        binding = agent_runtime_binding_from_model_configuration(version.model_configuration)
+        contract = agent_contract_from_model_configuration(version.model_configuration)
+    except ValidationError as exc:
+        raise AutomationBindingError("Bound Operations Agent contract is invalid") from exc
+    if binding is None or contract is None:
+        raise AutomationBindingError(
+            "Bound Operations Agent version requires contract and Runtime Binding"
+        )
+    try:
+        execution_binding = await select_agent_runtime(
+            session,
+            contract=contract,
+            binding=binding,
+            require_connected=require_online,
+        )
+    except RuntimeSelectionError as exc:
+        raise AutomationBindingError(str(exc)) from exc
+
+    try:
+        validate_agent_contract_payload(
+            contract,
+            "input_schema",
+            automation_run_input(automation, scheduled_for=scheduled_for),
+        )
+        validate_agent_contract_payload(contract, "state_schema", {})
+    except ValueError as exc:
+        raise AutomationBindingError(
+            f"Automation payload is incompatible with bound AgentContractV2: {exc}"
+        ) from exc
+    return agent, version, profile, execution_binding
+
+
+async def create_bound_automation_run(
+    session: AsyncSession,
+    automation: Automation,
+    *,
+    trigger_type: str,
+    started_by_user_id: str,
+    scheduled_for: datetime | None = None,
+) -> tuple[OperationsAgentRun, bool]:
+    """Create one pinned run, deduplicating scheduled occurrences in the DB."""
+    agent, version, profile, execution_binding = await validate_automation_binding(
+        session,
+        automation,
+        scheduled_for=scheduled_for,
+        require_online=trigger_type == "manual",
+    )
+    normalized_fire = (
+        scheduled_for.astimezone(UTC).replace(second=0, microsecond=0)
+        if scheduled_for is not None
+        else None
+    )
+    trigger_reference = (
+        f"automation:{automation.id}:{normalized_fire.strftime('%Y%m%dT%H%MZ')}"
+        if normalized_fire is not None
+        else None
+    )
+    run = OperationsAgentRun(
+        workspace_id=automation.workspace_id,
+        operations_agent_id=agent.id,
+        published_version=version.version,
+        profile_version=profile.version,
+        trigger_type=trigger_type,
+        trigger_reference=trigger_reference,
+        automation_id=automation.id,
+        automation_revision=automation.revision,
+        automation_snapshot=automation_snapshot(automation),
+        scheduled_for=normalized_fire,
+        schedule_timezone=automation.timezone,
+        target_resource_type="automation",
+        target_resource_id=automation.id,
+        input_payload=automation_run_input(automation, scheduled_for=normalized_fire),
+        state_payload={},
+        execution_binding=execution_binding,
+        evidence_payload=None,
+        status="queued",
+        started_by_user_id=started_by_user_id,
+    )
+
+    try:
+        async with session.begin_nested():
+            session.add(run)
+            await session.flush()
+    except IntegrityError:
+        if normalized_fire is None:
+            raise
+        existing = await session.scalar(
+            select(OperationsAgentRun).where(
+                OperationsAgentRun.automation_id == automation.id,
+                OperationsAgentRun.scheduled_for == normalized_fire,
+            )
+        )
+        if existing is None:
+            raise
+        return existing, False
+
+    if trigger_type == "manual":
+        queue_after_commit(
+            session,
+            lambda run_id=run.id: schedule_operations_agent_run(run_id),
+        )
+    return run, True
+
+
+async def create_failed_scheduled_automation_run(
+    session: AsyncSession,
+    automation: Automation,
+    *,
+    scheduled_for: datetime,
+    error_message: str,
+) -> tuple[OperationsAgentRun | None, bool]:
+    """Persist a terminal occurrence when a previously enabled binding drifts."""
+    if automation.operations_agent_id is None or automation.operations_agent_version is None:
+        return None, False
+    agent = await session.scalar(
+        select(OperationsAgentIdentity).where(
+            OperationsAgentIdentity.id == automation.operations_agent_id,
+            OperationsAgentIdentity.workspace_id == automation.workspace_id,
+        )
+    )
+    if agent is None:
+        return None, False
+
+    normalized_fire = scheduled_for.astimezone(UTC).replace(second=0, microsecond=0)
+    run = OperationsAgentRun(
+        workspace_id=automation.workspace_id,
+        operations_agent_id=agent.id,
+        published_version=automation.operations_agent_version,
+        profile_version=agent.current_profile_version,
+        trigger_type="scheduled",
+        trigger_reference=(
+            f"automation:{automation.id}:{normalized_fire.strftime('%Y%m%dT%H%MZ')}"
+        ),
+        automation_id=automation.id,
+        automation_revision=automation.revision,
+        automation_snapshot=automation_snapshot(automation),
+        scheduled_for=normalized_fire,
+        schedule_timezone=automation.timezone,
+        target_resource_type="automation",
+        target_resource_id=automation.id,
+        input_payload=automation_run_input(
+            automation,
+            scheduled_for=normalized_fire,
+        ),
+        state_payload={},
+        output_payload=None,
+        error_message=error_message[:4000],
+        status="failed",
+        started_by_user_id=automation.created_by_user_id,
+    )
+    try:
+        async with session.begin_nested():
+            session.add(run)
+            await session.flush()
+    except IntegrityError:
+        existing = await session.scalar(
+            select(OperationsAgentRun).where(
+                OperationsAgentRun.automation_id == automation.id,
+                OperationsAgentRun.scheduled_for == normalized_fire,
+            )
+        )
+        if existing is None:
+            raise
+        return existing, False
+    return run, True
+
+
+async def claim_scheduled_automation_run(
+    automation_id: str,
+    scheduled_for: datetime,
+) -> tuple[OperationsAgentRun | None, bool]:
+    """Lock/recheck one enabled Automation and durably claim its occurrence."""
+    normalized_fire = scheduled_for.astimezone(UTC).replace(second=0, microsecond=0)
+    async with AsyncSessionLocal() as session:
+        automation = await session.scalar(
+            select(Automation).where(Automation.id == automation_id).with_for_update()
+        )
+        if automation is None or not automation.enabled:
+            return None, False
+        current_occurrences = automation_fire_times(
+            automation.schedule,
+            automation.timezone,
+            normalized_fire - timedelta(seconds=1),
+            normalized_fire,
+        )
+        if normalized_fire not in current_occurrences:
+            logger.info(
+                "Skipping stale Automation occurrence automation_id=%s scheduled_for=%s",
+                automation_id,
+                normalized_fire.isoformat(),
+            )
+            return None, False
+        try:
+            run, created = await create_bound_automation_run(
+                session,
+                automation,
+                trigger_type="scheduled",
+                started_by_user_id=automation.created_by_user_id,
+                scheduled_for=normalized_fire,
+            )
+        except AutomationBindingError as exc:
+            run, created = await create_failed_scheduled_automation_run(
+                session,
+                automation,
+                scheduled_for=normalized_fire,
+                error_message=str(exc),
+            )
+            if run is None:
+                logger.error(
+                    "Automation occurrence could not persist binding failure "
+                    "automation_id=%s scheduled_for=%s: %s",
+                    automation_id,
+                    normalized_fire.isoformat(),
+                    exc,
+                )
+                return None, False
+            logger.warning(
+                "Automation occurrence persisted binding failure automation_id=%s "
+                "run_id=%s scheduled_for=%s: %s",
+                automation_id,
+                run.id,
+                normalized_fire.isoformat(),
+                exc,
+            )
+        await commit_session(session)
+
+    if created:
+        logger.info(
+            "Scheduled Automation claimed automation_id=%s run_id=%s scheduled_for=%s "
+            "trigger_reference=%s revision=%s agent_id=%s agent_version=%s profile_version=%s",
+            automation_id,
+            run.id,
+            normalized_fire.isoformat(),
+            run.trigger_reference,
+            run.automation_revision,
+            run.operations_agent_id,
+            run.published_version,
+            run.profile_version,
+        )
+    else:
+        logger.info(
+            "Scheduled Automation occurrence already claimed automation_id=%s "
+            "run_id=%s scheduled_for=%s",
+            automation_id,
+            run.id,
+            normalized_fire.isoformat(),
+        )
+    return run, created
+
+
+async def dispatch_due_automations(
+    window_start: datetime,
+    window_end: datetime,
+) -> list[OperationsAgentRun]:
+    """Find due enabled Automations and claim every distinct occurrence once."""
+    async with AsyncSessionLocal() as session:
+        rows = (
+            await session.execute(
+                select(Automation.id, Automation.schedule, Automation.timezone).where(
+                    Automation.enabled.is_(True)
+                )
+            )
+        ).all()
+
+    claimed: list[OperationsAgentRun] = []
+    for automation_id, schedule, timezone_name in rows:
+        for scheduled_for in automation_fire_times(
+            schedule,
+            timezone_name,
+            window_start,
+            window_end,
+        ):
+            run, created = await claim_scheduled_automation_run(automation_id, scheduled_for)
+            if run is not None and created:
+                claimed.append(run)
+    return claimed
diff --git a/backend/services/automation_starter_service.py b/backend/services/automation_starter_service.py
new file mode 100644
index 00000000..523d241c
--- /dev/null
+++ b/backend/services/automation_starter_service.py
@@ -0,0 +1,160 @@
+"""First-party Agent Starter installation for Workspace automations."""
+
+from dataclasses import dataclass
+
+from sqlalchemy import select
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from backend.models.automation import Automation
+from backend.schemas.automation import (
+    StarterInstallationPreview,
+    StarterInstallationResult,
+    StarterPreviewItem,
+)
+
+
+@dataclass(frozen=True)
+class StarterDefinition:
+    key: str
+    name: str
+    prompt: str
+    schedule: str
+    precheck: str | None = None
+    executor: str = "codex"
+    timezone: str = "UTC"
+    session_mode: str = "fresh"
+    approval_mode: str = "suggest_changes"
+
+    def project(self) -> dict[str, str]:
+        return {
+            "starter_key": self.key,
+            "lineage": "first-party-agent-starter",
+        }
+
+
+STARTER_DEFINITIONS: tuple[StarterDefinition, ...] = (
+    StarterDefinition(
+        key="daily-run-brief",
+        name="运行简报 Agent",
+        prompt="Prepare a concise daily run brief from the latest workspace activity and open work.",
+        schedule="daily@09:00",
+    ),
+    StarterDefinition(
+        key="weekly-system-review",
+        name="系统回顾 Agent",
+        prompt="Review the workspace system state, summarize trends, and identify actionable improvements.",
+        schedule="weekly@09:00",
+    ),
+    StarterDefinition(
+        key="anomaly-follow-up",
+        name="异常跟进 Agent",
+        prompt="Review unresolved anomalies, gather evidence, and propose the next safe follow-up actions.",
+        schedule="weekdays@09:00",
+    ),
+)
+
+STARTER_KEYS: tuple[str, ...] = tuple(
+    definition.key for definition in STARTER_DEFINITIONS
+)
+AGENT_STARTERS = STARTER_DEFINITIONS
+
+
+
+def _preview(
+    workspace_id: str,
+    installed_by_key: dict[str, Automation],
+) -> StarterInstallationPreview:
+    starters = [
+        StarterPreviewItem(
+            key=definition.key,
+            name=definition.name,
+            installed=definition.key in installed_by_key,
+            automation_id=(
+                installed_by_key[definition.key].id
+                if definition.key in installed_by_key
+                else None
+            ),
+        )
+        for definition in STARTER_DEFINITIONS
+    ]
+    installed_count = sum(item.installed for item in starters)
+    return StarterInstallationPreview(
+        workspace_id=workspace_id,
+        starters=starters,
+        missing_count=len(starters) - installed_count,
+        installed_count=installed_count,
+    )
+
+
+async def preview_starter_installation(
+    session: AsyncSession,
+    *,
+    workspace_id: str,
+) -> StarterInstallationPreview:
+    rows = (
+        await session.scalars(
+            select(Automation).where(
+                Automation.workspace_id == workspace_id,
+                Automation.starter_key.in_(STARTER_KEYS),
+            )
+        )
+    ).all()
+    return _preview(workspace_id, {row.starter_key: row for row in rows if row.starter_key})
+
+
+async def install_starters(
+    session: AsyncSession,
+    *,
+    workspace_id: str,
+    created_by_user_id: str,
+) -> StarterInstallationResult:
+    """Install missing starters atomically and return the resulting inventory.
+
+    A nested transaction keeps a failed pack installation from leaving a partial
+    set of rows behind. The unique workspace/starter key constraint is the final
+    guard against duplicate rows when requests race.
+    """
+
+    async with session.begin_nested():
+        rows = (
+            await session.scalars(
+                select(Automation)
+                .where(
+                    Automation.workspace_id == workspace_id,
+                    Automation.starter_key.in_(STARTER_KEYS),
+                )
+                .with_for_update()
+            )
+        ).all()
+        installed_by_key = {row.starter_key: row for row in rows if row.starter_key}
+        skipped_count = len(installed_by_key)
+        created_count = 0
+        for definition in STARTER_DEFINITIONS:
+            if definition.key in installed_by_key:
+                continue
+            row = Automation(
+                workspace_id=workspace_id,
+                starter_key=definition.key,
+                name=definition.name,
+                prompt=definition.prompt,
+                precheck=definition.precheck,
+                executor=definition.executor,
+                schedule=definition.schedule,
+                timezone=definition.timezone,
+                session_mode=definition.session_mode,
+                approval_mode=definition.approval_mode,
+                project=definition.project(),
+                enabled=False,
+                created_by_user_id=created_by_user_id,
+            )
+            session.add(row)
+            await session.flush()
+            installed_by_key[definition.key] = row
+            created_count += 1
+
+    preview = _preview(workspace_id, installed_by_key)
+    return StarterInstallationResult(
+        **preview.model_dump(),
+        created_count=created_count,
+        skipped_count=skipped_count,
+    )
diff --git a/backend/services/operations_agent_runtime_service.py b/backend/services/operations_agent_runtime_service.py
index fe0b0e80..45bbd1fc 100644
--- a/backend/services/operations_agent_runtime_service.py
+++ b/backend/services/operations_agent_runtime_service.py
@@ -15,10 +15,13 @@
     PublishedOperationsAgentVersion,
 )
 from backend.schemas.operations_agent import (
+    AgentRunEvidenceEnvelopeV1,
+    AgentContractV2,
     agent_contract_from_model_configuration,
     agent_runtime_binding_from_model_configuration,
     validate_agent_contract_payload,
 )
+from backend.services.agent_runtime_selection import select_agent_runtime
 from backend.ws_agent_manager import send_agent_task
 
 logger = logging.getLogger(__name__)
@@ -27,6 +30,44 @@
 # broker if Operations Agent runs must survive API process loss mid-flight.
 _ACTIVE_DISPATCHES: dict[str, asyncio.Task[None]] = {}
 
+_SENSITIVE_EVENT_KEYS = frozenset(
+    {
+        "access_token",
+        "api_key",
+        "apikey",
+        "authorization",
+        "credential",
+        "password",
+        "refresh_token",
+        "secret",
+    }
+)
+
+
+def _redact_runtime_value(value: Any) -> Any:
+    if isinstance(value, dict):
+        return {
+            key: (
+                "[REDACTED]"
+                if key.lower().replace("-", "_") in _SENSITIVE_EVENT_KEYS
+                else _redact_runtime_value(child)
+            )
+            for key, child in value.items()
+        }
+    if isinstance(value, list):
+        return [_redact_runtime_value(child) for child in value]
+    return value
+
+
+def _event_record(sequence: int, event: dict[str, Any]) -> dict[str, Any]:
+    return {
+        "sequence": sequence,
+        "type": str(event.get("type") or "unknown"),
+        "payload": _redact_runtime_value(
+            {key: value for key, value in event.items() if key not in {"type", "task_id"}}
+        ),
+    }
+
 
 def schedule_operations_agent_run(run_id: str) -> None:
     task = asyncio.create_task(dispatch_operations_agent_run(run_id))
@@ -40,12 +81,16 @@ def cancel_operations_agent_run(run_id: str) -> None:
         task.cancel()
 
 
+
+
 def _forget_dispatch(run_id: str, task: asyncio.Task[None]) -> None:
     if _ACTIVE_DISPATCHES.get(run_id) is task:
         _ACTIVE_DISPATCHES.pop(run_id, None)
 
 
 async def dispatch_operations_agent_run(run_id: str) -> None:
+    recorded_events: list[dict[str, Any]] = []
+    selection: dict[str, Any] | None = None
     try:
         async with AsyncSessionLocal() as session:
             claimed = await session.execute(
@@ -65,7 +110,8 @@ async def dispatch_operations_agent_run(run_id: str) -> None:
             version = await session.scalar(
                 select(PublishedOperationsAgentVersion)
                 .where(
-                    PublishedOperationsAgentVersion.operations_agent_id == run.operations_agent_id
+                    PublishedOperationsAgentVersion.operations_agent_id
+                    == run.operations_agent_id
                 )
                 .where(PublishedOperationsAgentVersion.version == run.published_version)
             )
@@ -94,83 +140,281 @@ async def dispatch_operations_agent_run(run_id: str) -> None:
             except ValidationError:
                 await _fail_run(run_id, "Published Operations Agent configuration is invalid")
                 return
-            if binding is None:
-                await _fail_run(run_id, "Published Agent Version has no Runtime Binding")
+            if binding is None or contract is None:
+                await _fail_run(
+                    run_id,
+                    "Published Agent Version requires contract and runtime binding",
+                )
                 return
 
+            selection = await select_agent_runtime(
+                session,
+                contract=contract,
+                binding=binding,
+            )
+            run.execution_binding = selection
+            await session.commit()
+
             runtime_input = cast(dict[str, Any], run.input_payload)
             runtime_config = dict(binding.config)
+            configured_timeout = runtime_config.get("timeout_seconds")
+            if (
+                not isinstance(configured_timeout, (int, float))
+                or isinstance(configured_timeout, bool)
+                or configured_timeout < binding.dispatch_timeout_seconds
+            ):
+                runtime_config["timeout_seconds"] = binding.dispatch_timeout_seconds
             runtime_config["permission_mode"] = profile.mode
+            permissions = {
+                "mode": profile.mode,
+                "tool_scope": list(profile.tool_scope),
+                "resource_scope": list(profile.resource_scope),
+                "action_scope": list(profile.action_scope),
+                "tool_policy": contract.tool_policy,
+            }
 
         state_contract_error: str | None = None
 
         async def on_event(event: dict[str, Any]) -> None:
             nonlocal state_contract_error
+            recorded_events.append(_event_record(len(recorded_events) + 1, event))
             if event.get("type") != "state":
                 return
             state = event.get("state")
             if not isinstance(state, dict):
                 state_contract_error = "Runtime state event must contain an object"
                 return
-            if contract is not None:
-                try:
-                    validate_agent_contract_payload(
-                        contract,
-                        "state_schema",
-                        cast(dict[str, JsonValue], state),
-                    )
-                except ValueError as exc:
-                    state_contract_error = str(exc)
-                    return
+            try:
+                validate_agent_contract_payload(
+                    contract,
+                    "state_schema",
+                    cast(dict[str, JsonValue], state),
+                )
+            except ValueError as exc:
+                state_contract_error = str(exc)
+                return
             await _persist_state(run_id, state)
 
         terminal = await send_agent_task(
-            binding.agent_url,
+            str(selection["agent_url"]),
             {
-                "runtime": binding.runtime,
-                "workflow": binding.workflow,
+                "runtime": selection["runtime"],
+                "workflow": selection["workflow"],
                 "instructions": version.instructions,
                 "input": runtime_input,
                 "config": runtime_config,
                 "session_id": None,
+                "provider": selection.get("provider"),
+                "model": selection.get("model"),
+                "required_capabilities": contract.required_capabilities,
+                "permissions": permissions,
+                "budget": contract.budget,
+                "evidence_requirements": contract.evidence_requirements,
             },
             on_event,
             timeout=float(binding.dispatch_timeout_seconds),
         )
+        if not recorded_events or recorded_events[-1].get("type") != terminal.get("type"):
+            recorded_events.append(_event_record(len(recorded_events) + 1, terminal))
 
+        evidence_payload = _build_evidence_payload(
+            selection,
+            recorded_events,
+            terminal.get("result") if terminal.get("type") == "done" else None,
+        )
         if state_contract_error is not None:
             await _fail_run(
                 run_id,
-                f"Runtime state violates AgentContractV1: {state_contract_error}",
+                f"Runtime state violates AgentContractV2: {state_contract_error}",
+                evidence_payload=evidence_payload,
             )
             return
         if terminal.get("type") == "error":
-            await _fail_run(run_id, str(terminal.get("message") or "Runtime failed"))
+            await _fail_run(
+                run_id,
+                str(terminal.get("message") or "Runtime failed"),
+                evidence_payload=evidence_payload,
+            )
             return
         if terminal.get("type") != "done":
-            await _fail_run(run_id, "Runtime returned no terminal done/error event")
+            await _fail_run(
+                run_id,
+                "Runtime returned no terminal done/error event",
+                evidence_payload=evidence_payload,
+            )
             return
         output = terminal.get("result") or {}
         if not isinstance(output, dict):
-            await _fail_run(run_id, "Runtime output must be an object")
+            await _fail_run(
+                run_id,
+                "Runtime output must be an object",
+                evidence_payload=evidence_payload,
+            )
             return
-        if contract is not None:
-            try:
-                validate_agent_contract_payload(
-                    contract,
-                    "output_schema",
-                    cast(dict[str, JsonValue], output),
-                )
-            except ValueError as exc:
-                await _fail_run(
-                    run_id,
-                    f"Runtime output violates AgentContractV1: {exc}",
-                )
-                return
-        await _complete_run(run_id, output)
+        try:
+            validate_agent_contract_payload(
+                contract,
+                "output_schema",
+                cast(dict[str, JsonValue], output),
+            )
+        except ValueError as exc:
+            await _fail_run(
+                run_id,
+                f"Runtime output violates AgentContractV2: {exc}",
+                evidence_payload=evidence_payload,
+            )
+            return
+        gate_failures = _quality_gate_failures(contract, output)
+        if gate_failures:
+            await _fail_run(
+                run_id,
+                "Runtime output failed required quality gates: "
+                + ", ".join(gate_failures),
+                evidence_payload=evidence_payload,
+            )
+            return
+        missing_evidence = _missing_evidence_requirements(
+            contract,
+            evidence_payload,
+        )
+        if missing_evidence:
+            await _fail_run(
+                run_id,
+                "Runtime output is missing required evidence: "
+                + ", ".join(missing_evidence),
+                evidence_payload=evidence_payload,
+            )
+            return
+        await _complete_run(run_id, output, evidence_payload)
     except Exception as exc:
         logger.exception("Operations Agent run dispatch failed | run_id=%s", run_id)
-        await _fail_run(run_id, str(exc))
+        await _fail_run(
+            run_id,
+            str(exc),
+            evidence_payload=(
+                _build_evidence_payload(selection, recorded_events, None)
+                if selection is not None
+                else None
+            ),
+        )
+
+
+def _output_objects(output: dict[str, Any] | None, key: str) -> list[dict[str, Any]]:
+    if not isinstance(output, dict):
+        return []
+    values = output.get(key)
+    if not isinstance(values, list):
+        return []
+    return [value for value in values if isinstance(value, dict)]
+
+
+def _quality_gate_failures(
+    contract: AgentContractV2,
+    output: dict[str, Any],
+) -> list[str]:
+    results = {
+        str(item.get("id")): item
+        for item in _output_objects(output, "quality_gates")
+        if item.get("id")
+    }
+    failures: list[str] = []
+    for gate in contract.quality_gates:
+        if not gate.required:
+            continue
+        result = results.get(gate.id)
+        if result is None or not (
+            result.get("passed") is True
+            or result.get("status") in {"passed", "completed"}
+        ):
+            failures.append(gate.id)
+    return failures
+
+
+def _missing_evidence_requirements(
+    contract: AgentContractV2,
+    envelope: dict[str, Any],
+) -> list[str]:
+    available = {"runtime_events"} if envelope.get("events") else set()
+    for category in ("artifacts", "evidence", "lineage", "audit"):
+        values = envelope.get(category)
+        if not isinstance(values, list) or not values:
+            continue
+        available.add(category)
+        available.add(category.removesuffix("s"))
+        for value in values:
+            if not isinstance(value, dict):
+                continue
+            for key in ("id", "kind", "type", "schema"):
+                identifier = value.get(key)
+                if isinstance(identifier, str):
+                    available.add(identifier)
+    return sorted(set(contract.evidence_requirements) - available)
+
+
+def _build_evidence_payload(
+    selection: dict[str, Any],
+    events: list[dict[str, Any]],
+    output: dict[str, Any] | None,
+) -> dict[str, Any]:
+    artifacts = [
+        record["payload"]["artifact"]
+        for record in events
+        if record.get("type") == "artifact"
+        and isinstance(record.get("payload"), dict)
+        and isinstance(record["payload"].get("artifact"), dict)
+    ]
+    evidence = [
+        record["payload"]["evidence"]
+        for record in events
+        if record.get("type") == "evidence"
+        and isinstance(record.get("payload"), dict)
+        and isinstance(record["payload"].get("evidence"), dict)
+    ]
+    audit = [
+        {
+            "type": "runtime_selected",
+            "runtime": selection["runtime"],
+            "agent_url": selection["agent_url"],
+            "capabilities": selection["capabilities"],
+        },
+        *[
+            record["payload"]["audit"]
+            for record in events
+            if record.get("type") == "audit"
+            and isinstance(record.get("payload"), dict)
+            and isinstance(record["payload"].get("audit"), dict)
+        ],
+    ]
+    lineage = [
+        {
+            "type": "agent_runtime",
+            "runtime": selection["runtime"],
+            "workflow": selection["workflow"],
+            "provider": selection.get("provider"),
+            "model": selection.get("model"),
+        }
+    ]
+    envelope = AgentRunEvidenceEnvelopeV1(
+        runtime=cast(dict[str, JsonValue], _redact_runtime_value(selection)),
+        events=cast(list[dict[str, JsonValue]], events),
+        artifacts=cast(
+            list[dict[str, JsonValue]],
+            [*artifacts, *_output_objects(output, "artifacts")],
+        ),
+        evidence=cast(
+            list[dict[str, JsonValue]],
+            [*evidence, *_output_objects(output, "evidence")],
+        ),
+        lineage=cast(
+            list[dict[str, JsonValue]],
+            [*lineage, *_output_objects(output, "lineage")],
+        ),
+        audit=cast(
+            list[dict[str, JsonValue]],
+            [*audit, *_output_objects(output, "audit")],
+        ),
+    )
+    return envelope.model_dump(mode="json")
 
 
 async def _persist_state(run_id: str, state: dict[str, Any]) -> None:
@@ -184,7 +428,11 @@ async def _persist_state(run_id: str, state: dict[str, Any]) -> None:
         await session.commit()
 
 
-async def _complete_run(run_id: str, output: dict[str, Any]) -> None:
+async def _complete_run(
+    run_id: str,
+    output: dict[str, Any],
+    evidence_payload: dict[str, Any],
+) -> None:
     async with AsyncSessionLocal() as session:
         await session.execute(
             update(OperationsAgentRun)
@@ -192,6 +440,7 @@ async def _complete_run(run_id: str, output: dict[str, Any]) -> None:
             .where(OperationsAgentRun.status == "running")
             .values(
                 output_payload=output,
+                evidence_payload=evidence_payload,
                 error_message=None,
                 status="completed",
             )
@@ -199,12 +448,21 @@ async def _complete_run(run_id: str, output: dict[str, Any]) -> None:
         await session.commit()
 
 
-async def _fail_run(run_id: str, message: str) -> None:
+async def _fail_run(
+    run_id: str,
+    message: str,
+    *,
+    evidence_payload: dict[str, Any] | None = None,
+) -> None:
     async with AsyncSessionLocal() as session:
         await session.execute(
             update(OperationsAgentRun)
             .where(OperationsAgentRun.id == run_id)
             .where(OperationsAgentRun.status.in_(("queued", "running")))
-            .values(error_message=message[:4000], status="failed")
+            .values(
+                error_message=message[:4000],
+                evidence_payload=evidence_payload,
+                status="failed",
+            )
         )
         await session.commit()
diff --git a/backend/services/scheduled_run_recovery.py b/backend/services/scheduled_run_recovery.py
new file mode 100644
index 00000000..37eddf97
--- /dev/null
+++ b/backend/services/scheduled_run_recovery.py
@@ -0,0 +1,53 @@
+"""Recovery queue for durable scheduled Operations Agent claims."""
+
+from sqlalchemy import select, update
+
+from backend.database import AsyncSessionLocal
+from backend.models.operations_agent import OperationsAgentRun
+from backend.services.operations_agent_runtime_service import schedule_operations_agent_run
+
+
+async def list_queued_scheduled_run_ids(limit: int = 200) -> list[str]:
+    async with AsyncSessionLocal() as session:
+        rows = await session.scalars(
+            select(OperationsAgentRun.id)
+            .where(
+                OperationsAgentRun.trigger_type == "scheduled",
+                OperationsAgentRun.status == "queued",
+            )
+            .order_by(OperationsAgentRun.created_at)
+            .limit(limit)
+        )
+        return list(rows)
+
+
+async def recover_queued_scheduled_runs_local() -> list[str]:
+    run_ids = await list_queued_scheduled_run_ids()
+    for run_id in run_ids:
+        schedule_operations_agent_run(run_id)
+    return run_ids
+
+
+async def recover_operations_agent_runs_on_startup() -> None:
+    """Preserve queued scheduled claims; fail only work that cannot be resumed."""
+    async with AsyncSessionLocal() as session:
+        await session.execute(
+            update(OperationsAgentRun)
+            .where(OperationsAgentRun.status == "running")
+            .values(
+                status="failed",
+                error_message="Operations Agent run interrupted by server restart",
+            )
+        )
+        await session.execute(
+            update(OperationsAgentRun)
+            .where(
+                OperationsAgentRun.status == "queued",
+                OperationsAgentRun.trigger_type != "scheduled",
+            )
+            .values(
+                status="failed",
+                error_message="Operations Agent run interrupted by server restart",
+            )
+        )
+        await session.commit()
\ No newline at end of file
diff --git a/backend/worker/celery_app.py b/backend/worker/celery_app.py
index ac45e3a3..8a460540 100644
--- a/backend/worker/celery_app.py
+++ b/backend/worker/celery_app.py
@@ -28,4 +28,10 @@
     # that restart and, as it turned out, was never even wired in here.
     beat_scheduler="redbeat.RedBeatScheduler",
     redbeat_redis_url=settings.redis_url,
+    beat_schedule={
+        "scheduled-automation-tick": {
+            "task": "run_automation_scheduler_tick",
+            "schedule": 60.0,
+        }
+    },
 )
diff --git a/backend/worker/control_plane_client.py b/backend/worker/control_plane_client.py
new file mode 100644
index 00000000..89e6a4ae
--- /dev/null
+++ b/backend/worker/control_plane_client.py
@@ -0,0 +1,19 @@
+"""Authenticated internal API client used by durable Celery dispatch tasks."""
+
+import httpx
+
+from backend.config import get_settings
+
+
+def post_control_plane(path: str, payload: dict | None = None) -> dict:
+    settings = get_settings()
+    if not settings.api_auth_token:
+        raise RuntimeError("API_AUTH_TOKEN is required for internal scheduler dispatch")
+    with httpx.Client(timeout=90, trust_env=False) as client:
+        response = client.post(
+            settings.control_plane_url.rstrip("/") + path,
+            headers={"X-API-Token": settings.api_auth_token},
+            json=payload,
+        )
+        response.raise_for_status()
+        return response.json()
diff --git a/backend/worker/tasks.py b/backend/worker/tasks.py
index 30b3f14a..914c2072 100644
--- a/backend/worker/tasks.py
+++ b/backend/worker/tasks.py
@@ -1,11 +1,13 @@
 """Celery tasks for async pipeline execution."""
 
 import asyncio
+from datetime import UTC, datetime
 import logging
 from typing import Any
 
 from celery import Task
 
+from backend.worker.control_plane_client import post_control_plane
 from backend.worker.celery_app import celery_app
 
 logger = logging.getLogger(__name__)
@@ -304,6 +306,35 @@ def run_scheduled_collection(
     )
 
 
+@celery_app.task(
+    name="dispatch_scheduled_operations_agent_run",
+    autoretry_for=(Exception,),
+    retry_backoff=True,
+    retry_backoff_max=300,
+    retry_jitter=True,
+    max_retries=5,
+)
+def dispatch_scheduled_operations_agent_run(run_id: str) -> dict:
+    """Durably ask the API process that owns Fleet WS state to execute one run."""
+    return post_control_plane(
+        f"/api/v1/internal/operations-agent-runs/{run_id}/dispatch"
+    )
+
+
+@celery_app.task(name="run_automation_scheduler_tick")
+def run_automation_scheduler_tick() -> dict:
+    """Claim due Automations and enqueue every queued scheduled run for dispatch."""
+    result = post_control_plane(
+        "/api/v1/internal/automations/scheduler/tick",
+        {"fired_at": datetime.now(UTC).isoformat()},
+    )
+    queued_run_ids = result.get("data", {}).get("queued_run_ids", [])
+    for run_id in queued_run_ids:
+        dispatch_scheduled_operations_agent_run.delay(run_id)
+    result["dispatch_enqueued_run_ids"] = queued_run_ids
+    return result
+
+
 @celery_app.task(name="send_notification")
 def send_notification(rule_id: str, record_id: str) -> dict:
     """Send a single notification for a rule/record pair."""
@@ -340,6 +371,7 @@ async def _send_notification_async(rule_id: str, record_id: str) -> dict:
             event=rule.trigger_event,
             source_id=record.source_id,
             record_id=record.id,
+            lineage=record.lineage,
             data=record.normalized_data,
             ai_enrichment=record.ai_enrichment,
         )
diff --git a/backend/workflow/capability_projection.py b/backend/workflow/capability_projection.py
index e1a3394e..c353e283 100644
--- a/backend/workflow/capability_projection.py
+++ b/backend/workflow/capability_projection.py
@@ -113,9 +113,7 @@ def _capability(
     )
 
 
-def _catalog_capabilities(
-    *, dify_runtime_ready: bool = False
-) -> list[WorkflowRuntimeCapability]:
+def _catalog_capabilities(*, dify_runtime_ready: bool = False) -> list[WorkflowRuntimeCapability]:
     expected_native_children = {
         action: f"tool.intelligence.native.{action}"
         for action in NATIVE_INTELLIGENCE_LIFECYCLE_ACTIONS
@@ -131,9 +129,7 @@ def _catalog_capabilities(
         )
     ]
     native_actions = [
-        action
-        if isinstance(action := tool.executor.params.get("action"), str)
-        else "<missing>"
+        action if isinstance(action := tool.executor.params.get("action"), str) else "<missing>"
         for tool in native_tools
     ]
     native_ids = [tool.id for tool in native_tools]
@@ -141,17 +137,11 @@ def _catalog_capabilities(
     id_counts = Counter(native_ids)
     missing_actions = sorted(set(expected_native_children) - set(native_actions))
     extra_actions = sorted(set(native_actions) - set(expected_native_children))
-    duplicate_actions = sorted(
-        action for action, count in action_counts.items() if count > 1
-    )
+    duplicate_actions = sorted(action for action, count in action_counts.items() if count > 1)
     missing_tool_ids = sorted(expected_native_ids - set(native_ids))
     extra_tool_ids = sorted(set(native_ids) - expected_native_ids)
-    duplicate_tool_ids = sorted(
-        tool_id for tool_id, count in id_counts.items() if count > 1
-    )
-    expected_pairs = {
-        (tool_id, action) for action, tool_id in expected_native_children.items()
-    }
+    duplicate_tool_ids = sorted(tool_id for tool_id, count in id_counts.items() if count > 1)
+    expected_pairs = {(tool_id, action) for action, tool_id in expected_native_children.items()}
     actual_pairs = set(zip(native_ids, native_actions, strict=True))
     missing_children = [
         {"id": tool_id, "action": action}
@@ -173,9 +163,7 @@ def _catalog_capabilities(
         or extra_children
     )
     native_package_status: WorkflowCapabilityStatus = (
-        "runnable"
-        if not native_action_set_invalid and not native_blocked
-        else "blocked"
+        "runnable" if not native_action_set_invalid and not native_blocked else "blocked"
     )
     native_package_missing = [
         reason
@@ -336,6 +324,73 @@ def _catalog_capabilities(
                 runtime_binding=SOURCE_FETCH_BINDING_ID,
             ),
         ),
+        _capability(
+            id="intelligence.source.doubao-research",
+            label="Doubao Research Capture",
+            surface="catalog",
+            status="runnable",
+            backend_available=True,
+            kind="source",
+            capability="fetch",
+            provider="doubao_research",
+            channel_type="doubao_research",
+            runtime_binding=SOURCE_FETCH_BINDING_ID,
+            reason=(
+                "Canvas Run captures a live Doubao answer through the existing "
+                "session-bound doubao_research channel and preserves its evidence."
+            ),
+            tags=["source", "doubao", "research", "chat-ai.capture", "live"],
+            source="backend.workflow.gaojixing_runtime",
+            manifest={
+                **_manifest(
+                    schema="capability.source.doubao-research.v1",
+                    input_ports=[_port("in", "trigger")],
+                    output_ports=[_port("out", "items[]")],
+                    resources=["opencli_session", "browser_profile"],
+                    permissions=["network.fetch", "canFetchNetwork"],
+                    runtime_binding=SOURCE_FETCH_BINDING_ID,
+                    trace_events=["partial:gaojixing.capture", "completed"],
+                    probes=["doubao_research_channel_available", "session_authenticated"],
+                ),
+                "canvas": {"node": True},
+                "nodeCatalog": {
+                    "authority": "backend",
+                    "origin": "source-preset",
+                    "category": "source",
+                    "kind": "source",
+                    "capability": "fetch",
+                    "adapter": {
+                        "id": "source-doubao-research-capture",
+                        "type": "source",
+                        "provider": "doubao_research",
+                        "mode": "live",
+                        "config": {
+                            "channelType": "doubao_research",
+                            "liveMode": "live",
+                            "settle_seconds": 150,
+                        },
+                    },
+                },
+                "presentation": {
+                    "icon": "Search",
+                    "description": "通过已认证的豆包研究会话捕获问题回答与证据。",
+                    "parameters": [
+                        {
+                            "name": "question",
+                            "label": "研究问题 / Research question",
+                            "type": "string",
+                            "required": True,
+                        },
+                        {
+                            "name": "sourceGroup",
+                            "label": "来源分组 / Source group",
+                            "type": "string",
+                            "default": "doubao-research",
+                        },
+                    ],
+                },
+            },
+        ),
         _capability(
             id="intelligence.source.searxng",
             label="SearXNG 元搜索 / Metasearch",
@@ -472,8 +527,7 @@ def _catalog_capabilities(
                 "presentation": {
                     "icon": "Rss",
                     "description": (
-                        "读取自托管 RSSHub 路由或已登记的 Feed Provider，"
-                        "保留来源分组与运行血缘。"
+                        "读取自托管 RSSHub 路由或已登记的 Feed Provider，保留来源分组与运行血缘。"
                     ),
                     "parameters": [
                         {
@@ -1077,7 +1131,7 @@ def _catalog_capabilities(
                     "duplicateToolIds": duplicate_tool_ids,
                     "missingChildren": missing_children,
                     "extraChildren": extra_children,
-                }
+                },
             },
         ),
         _blocked_catalog(
@@ -1189,15 +1243,11 @@ def _data_operator_capabilities() -> list[WorkflowRuntimeCapability]:
                         ],
                         probes=["data_operator_registry"],
                     ),
-                    "operatorIds": list(
-                        dict.fromkeys(operator["id"] for operator in operators)
-                    ),
+                    "operatorIds": list(dict.fromkeys(operator["id"] for operator in operators)),
                     "operators": operators,
                     "packs": sorted({operator["packId"] for operator in operators}),
                     "params": list(
-                        dict.fromkeys(
-                            key for spec in specs for key in spec.config_keys
-                        )
+                        dict.fromkeys(key for spec in specs for key in spec.config_keys)
                     ),
                     "artifacts": [
                         "recordCandidate[]",
@@ -1208,6 +1258,8 @@ def _data_operator_capabilities() -> list[WorkflowRuntimeCapability]:
             )
         )
     return rows
+
+
 def _dify_runtime_ready(installations: list[PluginInstallationRead]) -> bool:
     return any(
         installation.id == "bundled:dify-graphon-runtime" and installation.runtime_status == "READY"
@@ -1350,8 +1402,7 @@ def _primitive_capabilities() -> list[WorkflowRuntimeCapability]:
             kind="schedule",
             capability="trigger",
             runtime_binding=WEBHOOK_TRIGGER_BINDING_ID,
-            reason="The workflow webhook input contract has HTTP ingress "
-            "and run dispatch.",
+            reason="The workflow webhook input contract has HTTP ingress and run dispatch.",
             missing=[],
             tags=["primitive", "webhook", "trigger"],
         ),
@@ -1364,8 +1415,7 @@ def _primitive_capabilities() -> list[WorkflowRuntimeCapability]:
             kind="schedule",
             capability="trigger",
             runtime_binding=WEBHOOK_TRIGGER_BINDING_ID,
-            reason="The workflow webhook input contract has HTTP ingress "
-            "and run dispatch.",
+            reason="The workflow webhook input contract has HTTP ingress and run dispatch.",
             missing=[],
             tags=["primitive", "webhook", "trigger"],
         ),
@@ -1536,8 +1586,7 @@ def _trigger_capabilities() -> list[WorkflowRuntimeCapability]:
             kind="schedule",
             capability="trigger",
             runtime_binding=WEBHOOK_TRIGGER_BINDING_ID,
-            reason="The workflow webhook input contract has HTTP ingress "
-            "and run dispatch.",
+            reason="The workflow webhook input contract has HTTP ingress and run dispatch.",
             missing=[],
             tags=["trigger", "webhook"],
         ),
@@ -1774,11 +1823,7 @@ def _tool_catalog_capabilities() -> list[WorkflowRuntimeCapability]:
                 provider=tool.provider,
                 runtime_binding=_read_manifest_runtime_binding(tool.manifest),
                 reason=tool.description,
-                missing=(
-                    []
-                    if tool.status == "runnable"
-                    else ["tool_capability_unavailable"]
-                ),
+                missing=([] if tool.status == "runnable" else ["tool_capability_unavailable"]),
                 tags=["catalog", "tool-capability", *tool.tags],
                 source="backend.workflow.tool_capabilities",
                 manifest=manifest,
diff --git a/backend/workflow/compiler.py b/backend/workflow/compiler.py
index 29ea1a7a..33e701f5 100644
--- a/backend/workflow/compiler.py
+++ b/backend/workflow/compiler.py
@@ -84,6 +84,10 @@ class _PortContract:
         [_PortContract("in", "input", "trigger", required=False)],
         [_PortContract("out", "output", "items[]")],
     ),
+    "intelligence.source.doubao-research": (
+        [_PortContract("in", "input", "trigger", required=False)],
+        [_PortContract("out", "output", "items[]")],
+    ),
     "intelligence.source.pool": (
         [_PortContract("in", "input", "trigger", required=False)],
         [_PortContract("out", "output", "trigger")],
@@ -506,9 +510,7 @@ def _validate_data_operator_node(
     catalog_id = _read_string((node.ui or {}).get("catalogId"))
     prefix = "intelligence.data."
     expected_kind = (
-        catalog_id.removeprefix(prefix)
-        if catalog_id and catalog_id.startswith(prefix)
-        else None
+        catalog_id.removeprefix(prefix) if catalog_id and catalog_id.startswith(prefix) else None
     )
     operator_id = _read_string(node.params.get("operatorId"))
     if expected_kind not in {"generate", "filter", "evaluate", "refine"}:
@@ -549,15 +551,10 @@ def _validate_data_operator_node(
                 path=[*path_prefix, "params", "packVersion"],
             )
         ]
-    resolved_pack_version = (
-        requested_pack_version or _LEGACY_DATA_OPERATOR_PACK_VERSION
-    )
+    resolved_pack_version = requested_pack_version or _LEGACY_DATA_OPERATOR_PACK_VERSION
     spec = resolve_data_operator(operator_id, resolved_pack_version)
     if spec is None:
-        if any(
-            registered.id == operator_id
-            for registered in list_data_operator_specs()
-        ):
+        if any(registered.id == operator_id for registered in list_data_operator_specs()):
             return [
                 WorkflowCompileError(
                     code="unsupported_data_operator_version",
@@ -574,8 +571,7 @@ def _validate_data_operator_node(
             WorkflowCompileError(
                 code="unknown_data_operator",
                 message=(
-                    f'Workflow data node "{node.id}" references unknown operator '
-                    f'"{operator_id}"'
+                    f'Workflow data node "{node.id}" references unknown operator "{operator_id}"'
                 ),
                 node_id=node.id,
                 path=[*path_prefix, "params", "operatorId"],
@@ -594,6 +590,8 @@ def _validate_data_operator_node(
             )
         ]
     return []
+
+
 def _validate_node_capability_gaps(
     node: WorkflowProjectNode,
     path_prefix: list[str],
@@ -658,8 +656,7 @@ def _validate_typed_edges(
         target_port = _resolve_input_port(target_contract[0], edge.targetPort)
         if (
             target_port is None
-            and _read_string((target_node.ui or {}).get("catalogId"))
-            == "intelligence.flow.merge"
+            and _read_string((target_node.ui or {}).get("catalogId")) == "intelligence.flow.merge"
             and edge.targetPort
             and re.fullmatch(r"in\d+", edge.targetPort)
         ):
@@ -825,14 +822,8 @@ def _native_intelligence_port_contracts(
     if contract is None:
         return None
     return (
-        [
-            _PortContract(name, "input", type_)
-            for name, type_ in contract.input_ports
-        ],
-        [
-            _PortContract(name, "output", type_)
-            for name, type_ in contract.output_ports
-        ],
+        [_PortContract(name, "input", type_) for name, type_ in contract.input_ports],
+        [_PortContract(name, "output", type_) for name, type_ in contract.output_ports],
     )
 
 
@@ -1551,11 +1542,7 @@ def _compile_node(
         "capability": node.capability,
         "dispatch": "preview",
         "origin": resolve_node_origin(node).model_dump(exclude_none=True),
-        **(
-            {"proposal_state": node.proposalState}
-            if node.proposalState is not None
-            else {}
-        ),
+        **({"proposal_state": node.proposalState} if node.proposalState is not None else {}),
     }
     if runtime:
         runtime_metadata.update(runtime)
@@ -1591,7 +1578,6 @@ def _compile_node(
     )
 
 
-
 def _widen_merge_fan_in(plan: PlanGraph) -> None:
     """Grow merge nodes' declared inputs to cover every in<N> edge that
     actually targets them. The static contract names only in1/in2, but the
@@ -1618,8 +1604,7 @@ def _to_plan_ir(project: WorkflowProject) -> PlanGraph:
         name=project.name,
         draft=True,
         nodes=[
-            _to_plan_node(node, merge_fan_in=fan_in_by_node.get(node.id))
-            for node in project.nodes
+            _to_plan_node(node, merge_fan_in=fan_in_by_node.get(node.id)) for node in project.nodes
         ],
         edges=[
             PlanEdge(
diff --git a/backend/workflow/demand_assembler.py b/backend/workflow/demand_assembler.py
index 1882cffa..9d979266 100644
--- a/backend/workflow/demand_assembler.py
+++ b/backend/workflow/demand_assembler.py
@@ -29,7 +29,7 @@ def draft_workflow_demand(body: WorkflowDemandDraftRequest) -> WorkflowPatchResp
     capabilities that the Canvas can review before materialization.
     """
 
-    sources = _source_slots_for_need(body.text)
+    sources = _doubao_research_slots_for_need(body.text) or _source_slots_for_need(body.text)
     if not sources:
         return preview_workflow_patch(
             body.project,
@@ -71,7 +71,14 @@ def _native_first_loop_operations(
 
     for index, source in enumerate(sources):
         source_slug = _read_string(source.get("id")) or f"source-{index + 1}"
-        adapter_id = _unique_id(used_adapter_ids, f"opencli-{source_slug}")
+        adapter_id = _unique_id(
+            used_adapter_ids,
+            _read_string(source.get("adapterId")) or f"opencli-{source_slug}",
+        )
+        provider = _read_string(source.get("provider")) or "opencli"
+        adapter_config = source.get("adapterConfig")
+        if not isinstance(adapter_config, dict):
+            adapter_config = {"channel": "opencli"}
         if adapter_id not in {adapter.id for adapter in project.adapters}:
             operations.append(
                 WorkflowPatchOperation(
@@ -79,9 +86,9 @@ def _native_first_loop_operations(
                     adapter=WorkflowAdapterBinding(
                         id=adapter_id,
                         type="source",
-                        provider="opencli",
-                        mode="live",
-                        config={"channel": "opencli"},
+                        provider=provider,
+                        mode=_read_string(source.get("mode")) or "live",
+                        config=adapter_config,
                     ),
                 )
             )
@@ -89,6 +96,24 @@ def _native_first_loop_operations(
         source_id = _unique_id(used_node_ids, f"source-{source_slug}")
         normalize_id = _unique_id(used_node_ids, f"normalize-{source_slug}")
         normalize_ids.append(normalize_id)
+        source_params = source.get("params")
+        if not isinstance(source_params, dict):
+            source_params = {
+                "site": source["site"],
+                "command": source["command"],
+                "args": source.get("args", {}),
+                "sourceGroup": source.get("sourceGroup"),
+            }
+        source_params = {
+            **source_params,
+            "demand": {
+                "text": demand_text,
+                "locale": locale,
+                "source": "ai_plan_draft",
+                "labels": _demand_labels(demand_text),
+                "labelSource": "demand",
+            },
+        }
         operations.extend(
             [
                 WorkflowPatchOperation(
@@ -98,19 +123,11 @@ def _native_first_loop_operations(
                         kind="source",
                         capability="fetch",
                         adapter=adapter_id,
-                        params={
-                            "site": source["site"],
-                            "command": source["command"],
-                            "args": source.get("args", {}),
-                            "sourceGroup": source.get("sourceGroup"),
-                            "demand": {
-                                "text": demand_text,
-                                "locale": locale,
-                                "source": "ai_plan_draft",
-                            },
-                        },
+                        params=source_params,
                         ui={
-                            "catalogId": "intelligence.source.opencli-slot",
+                            "catalogId": source.get(
+                                "catalogId", "intelligence.source.opencli-slot"
+                            ),
                             "label": source.get("label", source_slug),
                             "position": {"x": 180, "y": 180 + index * 120},
                         },
@@ -544,6 +561,74 @@ def _research_dimensions_for_need(text: str) -> list[str]:
     return list(dict.fromkeys(value for value in values if value))[:12]
 
 
+_DOUBAO_RESEARCH_DEMAND_MARKERS = ("竞品", "同类产品", "产品溯源", "grounded")
+_DEMAND_LABEL_MARKERS = (
+    "竞品",
+    "品牌",
+    "产品",
+    "配方",
+    "价格",
+    "渠道",
+    "监管",
+    "投诉",
+    "召回",
+    "检测",
+    "风险",
+    "grounded",
+)
+
+
+def _doubao_research_slots_for_need(text: str) -> list[dict[str, Any]]:
+    normalized = text.casefold()
+    if not any(marker.casefold() in normalized for marker in _DOUBAO_RESEARCH_DEMAND_MARKERS):
+        return []
+    target = _demand_target(text)
+    batch_match = re.search(r"每批\s*(\d{1,2})\s*条", text)
+    batch_size = int(batch_match.group(1)) if batch_match else 5
+    return [
+        {
+            "id": "doubao-research",
+            "adapterId": "source-doubao-research-capture",
+            "label": f"{target} · 需求研究",
+            "provider": "doubao_research",
+            "mode": "live",
+            "catalogId": "intelligence.source.doubao-research",
+            "adapterConfig": {
+                "channelType": "doubao_research",
+                "liveMode": "live",
+                "site_session": "persistent",
+                "settle_seconds": 240,
+                "capabilityId": "chat-ai.capture",
+                "extract_citations": True,
+                "capture_conversation_url": True,
+            },
+            "params": {
+                "question": text,
+                "sourceGroup": "demand-competitor-research",
+                "rotationBatchSize": batch_size,
+            },
+        }
+    ]
+
+
+def _demand_target(text: str) -> str:
+    match = re.search(r"围绕\s*([^，,；;。\n]+)", text, flags=re.IGNORECASE)
+    if match:
+        return match.group(1).strip()
+    return _keyword_from_need(text)[:48]
+
+
+def _demand_labels(text: str) -> list[str]:
+    labels = [_demand_target(text)]
+    normalized = text.casefold()
+    labels.extend(
+        marker
+        for marker in _DEMAND_LABEL_MARKERS
+        if marker.casefold() in normalized and marker not in labels
+    )
+    return labels[:12]
+
+
 def _source_slots_for_need(text: str) -> list[dict[str, Any]]:
     """Resolve a collection need to native OpenCLI source slots.
 
diff --git a/backend/workflow/gaojixing_runtime.py b/backend/workflow/gaojixing_runtime.py
new file mode 100644
index 00000000..67724d1b
--- /dev/null
+++ b/backend/workflow/gaojixing_runtime.py
@@ -0,0 +1,314 @@
+"""Gaojixing live Doubao source contract for WorkflowProject runs."""
+
+from __future__ import annotations
+
+import hashlib
+import json
+from collections.abc import Mapping
+from copy import deepcopy
+from dataclasses import dataclass
+from types import MappingProxyType
+from typing import Any
+
+from backend.channels.base import ChannelResult
+from backend.channels.doubao_research_channel import DoubaoResearchChannel
+
+GAOJIXING_CAPABILITY_ID = "chat-ai.capture"
+GAOJIXING_CHANNEL_TYPE = "doubao_research"
+GAOJIXING_LIVE_MODE = "live"
+GAOJIXING_FIXTURE_MODE = "fixture"
+GAOJIXING_MOCK_MODE = "mock"
+GAOJIXING_EXECUTION_MODES = frozenset(
+    {GAOJIXING_LIVE_MODE, GAOJIXING_FIXTURE_MODE, GAOJIXING_MOCK_MODE}
+)
+GAOJIXING_PACKAGE_SCHEMA = "gaojixing.question-package.v1"
+GAOJIXING_EVIDENCE_SCHEMA = "gaojixing.capture-evidence.v1"
+
+
+class GaojixingReadinessError(RuntimeError):
+    """Typed fail-closed blocker for a live Gaojixing source."""
+
+    def __init__(self, code: str, message: str, *, details: dict[str, Any] | None = None) -> None:
+        super().__init__(message)
+        self.code = code
+        self.message = message
+        self.details = details or {}
+
+
+@dataclass(frozen=True)
+class GaojixingQuestionPackage:
+    schema: str
+    question: str
+    options: Mapping[str, Any]
+    digest: str
+
+    def to_dict(self) -> dict[str, Any]:
+        return {
+            "schema": self.schema,
+            "question": self.question,
+            "options": _thaw_json(self.options),
+            "digest": self.digest,
+        }
+
+
+def build_question_package(
+    *,
+    node_params: dict[str, Any],
+    adapter_config: dict[str, Any],
+    runtime_payload: dict[str, Any],
+) -> GaojixingQuestionPackage:
+    """Resolve the effective question once and hash its canonical snapshot."""
+
+    question = (
+        _string(runtime_payload.get("question"))
+        or _string(runtime_payload.get("query"))
+        or _string(node_params.get("question"))
+    )
+    if question is None:
+        question = _string(adapter_config.get("question"))
+    if question is None:
+        raise GaojixingReadinessError(
+            "gaojixing_question_required",
+            (
+                "A live Gaojixing run requires an effective question in run input, "
+                "node params, or adapter config."
+            ),
+            details={"required": "question"},
+        )
+
+    option_keys = (
+        "extract_citations",
+        "capture_conversation_url",
+        "site_session",
+        "settle_seconds",
+        "capabilityId",
+        "sourceGroup",
+    )
+    options = {
+        key: value
+        for key in option_keys
+        for value in [
+            _json_safe(runtime_payload.get(key, node_params.get(key, adapter_config.get(key))))
+        ]
+        if value is not None
+    }
+    canonical = {"schema": GAOJIXING_PACKAGE_SCHEMA, "question": question, "options": options}
+    encoded = json.dumps(
+        canonical, ensure_ascii=False, sort_keys=True, separators=(",", ":")
+    ).encode()
+    return GaojixingQuestionPackage(
+        schema=GAOJIXING_PACKAGE_SCHEMA,
+        question=question,
+        options=_freeze_json(options),
+        digest=hashlib.sha256(encoded).hexdigest(),
+    )
+
+
+async def capture_live_doubao(
+    *,
+    package: GaojixingQuestionPackage,
+    node_params: dict[str, Any],
+    adapter_config: dict[str, Any],
+    network_allowed: bool,
+) -> ChannelResult:
+    """Preflight and execute the existing Doubao channel; never use fixtures."""
+
+    capability_id = (
+        _string(node_params.get("capabilityId"))
+        or _string(adapter_config.get("capabilityId"))
+        or GAOJIXING_CAPABILITY_ID
+    )
+    if capability_id not in {GAOJIXING_CAPABILITY_ID, "doubao.ask"}:
+        raise GaojixingReadinessError(
+            "gaojixing_capability_missing",
+            f'Live Gaojixing capability "{capability_id}" is not registered.',
+            details={"capabilityId": capability_id},
+        )
+    if (
+        node_params.get("capabilityAvailable") is False
+        or adapter_config.get("capabilityAvailable") is False
+    ):
+        raise GaojixingReadinessError(
+            "gaojixing_capability_missing",
+            "The live Gaojixing chat-ai.capture/Doubao capability is unavailable.",
+            details={"capabilityId": capability_id},
+        )
+    _raise_configured_readiness_blocker(adapter_config)
+    if not network_allowed:
+        raise GaojixingReadinessError(
+            "gaojixing_network_denied",
+            "Live Gaojixing capture requires workflow network permission.",
+            details={"requiredPermission": "canFetchNetwork"},
+        )
+
+    channel = DoubaoResearchChannel()
+    healthy = await channel.health_check(adapter_config)
+    if not healthy:
+        readiness_code = await channel.readiness_code(adapter_config)
+        raise GaojixingReadinessError(
+            f"gaojixing_{readiness_code or 'session_unavailable'}",
+            _readiness_message(readiness_code),
+            details={"site": "doubao", "session": adapter_config.get("site_session", "persistent")},
+        )
+
+    config = {
+        **adapter_config,
+        "question": package.question,
+        "site_session": adapter_config.get("site_session", "persistent"),
+        "extract_citations": adapter_config.get("extract_citations", True),
+        "capture_conversation_url": adapter_config.get("capture_conversation_url", True),
+    }
+    return await channel.collect(config, {"question": package.question})
+
+
+def map_capture_item(
+    item: dict[str, Any],
+    *,
+    package: GaojixingQuestionPackage,
+    workflow_id: str,
+    run_id: str,
+    node_id: str,
+    artifact_id: str,
+    mode: str = GAOJIXING_LIVE_MODE,
+    provenance: str | None = None,
+) -> dict[str, Any]:
+    """Attach separate answer/citation/conversation evidence to one raw item."""
+    if mode not in GAOJIXING_EXECUTION_MODES:
+        raise ValueError(f"Unsupported Gaojixing execution mode: {mode}")
+    provenance = provenance or (
+        "opencli:doubao" if mode == GAOJIXING_LIVE_MODE else f"{mode}:unspecified"
+    )
+
+    answer = _string(item.get("content"))
+    citations = item.get("citations") if isinstance(item.get("citations"), list) else []
+    conversation_url = _string(item.get("conversation_url"))
+    evidence = {
+        "schema": GAOJIXING_EVIDENCE_SCHEMA,
+        "mode": mode,
+        "provenance": provenance,
+        "packageDigest": package.digest,
+        "runId": run_id,
+        "workflowId": workflow_id,
+        "nodeId": node_id,
+        "answer": {
+            "status": "captured" if answer else "unavailable",
+            "artifactId": artifact_id,
+            "text": answer,
+        },
+        "citations": {
+            "status": "captured" if citations else "empty",
+            "capture": item.get("citation_capture", "answer_url_extraction"),
+            "verified": False,
+            "items": citations,
+        },
+        "conversation": {
+            "status": "captured" if conversation_url else "unknown",
+            "url": conversation_url,
+        },
+    }
+    mapped = {
+        **item,
+        "gaojixing": {
+            "mode": mode,
+            "provenance": provenance,
+            "capabilityId": GAOJIXING_CAPABILITY_ID,
+            "package": package.to_dict(),
+            "artifactId": artifact_id,
+            "evidence": evidence,
+        },
+        "packageDigest": package.digest,
+        "questionPackage": package.to_dict(),
+        "mode": mode,
+        "provenance": provenance,
+        "answerArtifactId": artifact_id,
+    }
+    if conversation_url:
+        mapped["dedupe"] = {
+            "type": "source-identity",
+            "field": "conversation_url",
+            "value": conversation_url,
+            "status": "unique",
+        }
+    return mapped
+
+
+def _string(value: Any) -> str | None:
+    return value.strip() if isinstance(value, str) and value.strip() else None
+
+
+def _json_safe(value: Any) -> Any:
+    if value is None or isinstance(value, (str, int, float, bool)):
+        return value
+    if isinstance(value, dict):
+        return {str(key): _json_safe(item) for key, item in value.items()}
+    if isinstance(value, list):
+        return [_json_safe(item) for item in value]
+    return str(value)
+
+
+def _raise_configured_readiness_blocker(adapter_config: dict[str, Any]) -> None:
+    if adapter_config.get("adapterAvailable") is False:
+        raise GaojixingReadinessError(
+            "gaojixing_adapter_missing",
+            "The Doubao OpenCLI adapter is unavailable.",
+            details={"requiredAdapter": "opencli:doubao"},
+        )
+    if (
+        adapter_config.get("authenticated") is False
+        or adapter_config.get("authenticationAvailable") is False
+    ):
+        raise GaojixingReadinessError(
+            "gaojixing_authentication_required",
+            "The configured Doubao session is not authenticated.",
+            details={"site": "doubao"},
+        )
+    if adapter_config.get("sessionAvailable") is False:
+        raise GaojixingReadinessError(
+            "gaojixing_session_unavailable",
+            "The configured Doubao OpenCLI session is unavailable.",
+            details={"site": "doubao", "session": adapter_config.get("site_session", "persistent")},
+        )
+
+
+def _readiness_message(code: str | None) -> str:
+    messages = {
+        "adapter_missing": "The Doubao OpenCLI adapter is unavailable.",
+        "authentication_required": "The configured Doubao session is not authenticated.",
+        "captcha_challenge": "The Doubao session is blocked by a CAPTCHA challenge.",
+    }
+    return messages.get(
+        code or "",
+        "The Doubao OpenCLI session is unavailable or not logged in.",
+    )
+
+
+def _freeze_json(value: Any) -> Any:
+    if isinstance(value, dict):
+        return MappingProxyType({str(key): _freeze_json(item) for key, item in value.items()})
+    if isinstance(value, list):
+        return tuple(_freeze_json(item) for item in value)
+    return deepcopy(value)
+
+
+def _thaw_json(value: Any) -> Any:
+    if isinstance(value, Mapping):
+        return {str(key): _thaw_json(item) for key, item in value.items()}
+    if isinstance(value, tuple):
+        return [_thaw_json(item) for item in value]
+    return deepcopy(value)
+
+
+__all__ = [
+    "GAOJIXING_CAPABILITY_ID",
+    "GAOJIXING_CHANNEL_TYPE",
+    "GAOJIXING_EVIDENCE_SCHEMA",
+    "GAOJIXING_LIVE_MODE",
+    "GAOJIXING_EXECUTION_MODES",
+    "GAOJIXING_FIXTURE_MODE",
+    "GaojixingQuestionPackage",
+    "GaojixingReadinessError",
+    "build_question_package",
+    "capture_live_doubao",
+    "GAOJIXING_MOCK_MODE",
+    "map_capture_item",
+]
diff --git a/backend/workflow/node_registry.py b/backend/workflow/node_registry.py
index afe7512e..766b7e11 100644
--- a/backend/workflow/node_registry.py
+++ b/backend/workflow/node_registry.py
@@ -26,6 +26,7 @@
     "intelligence.source.rss-bridge",
     "intelligence.source.rsshub",
     "intelligence.source.searxng",
+    "intelligence.source.doubao-research",
     "intelligence.source.pool",
     "intelligence.source.opencli-slot",
     "intelligence.processing.normalize",
@@ -51,7 +52,6 @@
     "package.collection.pipeline",
     "package.processing.record-hygiene",
     "package.opencli.multi-source-hda",
-    "package.processing.record-hygiene",
     "package.intelligence.situation-awareness",
     "package.simulation.swarm-forecast",
     "package.intelligence.native-lifecycle",
diff --git a/backend/workflow/opencli_hda_tracer.py b/backend/workflow/opencli_hda_tracer.py
index e0f7e83f..849f13ce 100644
--- a/backend/workflow/opencli_hda_tracer.py
+++ b/backend/workflow/opencli_hda_tracer.py
@@ -4,6 +4,7 @@
 
 import asyncio
 import uuid
+from copy import deepcopy
 from dataclasses import dataclass
 from datetime import UTC, datetime
 from inspect import signature
@@ -23,6 +24,7 @@
 from backend.models.workflow_run import WorkflowRun as WorkflowRunRow
 from backend.models.workflow_run import WorkflowRunEvent as WorkflowRunEventRow
 from backend.pipeline.normalizer import normalize_item
+from backend.pipeline.sinks.base import CollectionLineage
 from backend.pipeline.storer import store_records
 from backend.schemas.workflow import (
     CompiledWorkflowNode,
@@ -68,6 +70,15 @@
 from backend.workflow.dify_graphon_client import DifyGraphonClient
 from backend.workflow.event_mirror import publish_workflow_run_event_mirror
 from backend.workflow.fleet_inventory import match_workflow_fleet_capability
+from backend.workflow.gaojixing_runtime import (
+    GAOJIXING_CHANNEL_TYPE,
+    GAOJIXING_EXECUTION_MODES,
+    GAOJIXING_LIVE_MODE,
+    GaojixingReadinessError,
+    build_question_package,
+    capture_live_doubao,
+    map_capture_item,
+)
 from backend.workflow.http_source_executor import (
     WorkflowHTTPSourceExecutionError,
     execute_workflow_http_source,
@@ -298,12 +309,14 @@ async def start_workflow_run(
     workflow_version_id: str | None = None,
     studio_workflow_version_id: str | None = None,
     graphon_client: DifyGraphonClient | None = None,
+    replay_source_node_ids: set[str] | None = None,
 ) -> WorkflowRunProjection:
     """Create a replayable workflow run projection from a compiled WorkflowProject."""
 
     run_id = body.runId or str(uuid.uuid4())
     trace_id = body.traceId or str(uuid.uuid4())
     started_at = _utcnow()
+    replay_source_node_ids = replay_source_node_ids or set()
     prior_events = list(existing_events or [])
     # Source-level trigger scope selection runs before authoritative compilation
     # so a disconnected, incomplete canvas node cannot block a valid
@@ -692,12 +705,68 @@ async def start_workflow_run(
                     "runId": run_id,
                     "nodeId": node.id,
                     "sourceId": body.input.sourceId or body.input.source,
-                    "requestId": body.trigger.requestId,
                     "runtimeInputEnvelope": envelope,
                 },
             )
             emitter.emit(node, "completed", message="Workflow webhook input completed")
             continue
+        if node.id in replay_source_node_ids and _is_gaojixing_source_node(node):
+            resumed_items = _read_dict_list(body.sourceOutputs.get(node.id))
+            if not resumed_items:
+                raise ValueError(
+                    f"Persisted Gaojixing replay source {node.id} has no source output"
+                )
+            raw = _read_dict(resumed_items[0].get("raw"))
+            gaojixing = _read_dict(raw.get("gaojixing"))
+            package = _read_dict(gaojixing.get("package"))
+            artifact_id = _read_string(gaojixing.get("artifactId"))
+            package_digest = _read_string(package.get("digest"))
+            if not artifact_id or not package_digest:
+                raise ValueError(f"Persisted Gaojixing replay source {node.id} lacks evidence")
+            outputs_by_node[node.id] = resumed_items
+            emitter.emit(node, "started", message="Persisted Gaojixing source replay started")
+            emitter.emit(
+                node,
+                "partial",
+                message="Persisted Gaojixing source evidence loaded",
+                batch=_node_batch_reference(
+                    body.project.id, run_id, node, item_count=len(resumed_items)
+                ),
+                details={
+                    "bindingId": SOURCE_FETCH_BINDING_ID,
+                    "channelType": GAOJIXING_CHANNEL_TYPE,
+                    "mode": "persisted-replay",
+                    "sourceRunId": body.input.sourceId,
+                    "packageDigest": package_digest,
+                    "artifactId": artifact_id,
+                    "evidence": _read_dict(gaojixing.get("evidence")),
+                    "lineage": _lineage_pointer(node),
+                },
+            )
+            emitter.emit(node, "completed", message="Persisted Gaojixing source replay completed")
+            continue
+        if _is_gaojixing_source_node(node) and _source_live_mode(node):
+            await _execute_gaojixing_source(
+                node,
+                body=body,
+                run_id=run_id,
+                workflow_id=body.project.id,
+                trace_id=trace_id,
+                outputs_by_node=outputs_by_node,
+                emitter=emitter,
+                session=session,
+            )
+            continue
+        if _is_gaojixing_source_node(node):
+            await _execute_gaojixing_fixture_source(
+                node,
+                body=body,
+                run_id=run_id,
+                workflow_id=body.project.id,
+                outputs_by_node=outputs_by_node,
+                emitter=emitter,
+            )
+            continue
 
         resumed_assets = _read_dict_list(body.sourceOutputs.get(node.id))
         if _binding_id(node) == IMAGE_GENERATION_BINDING_ID and resumed_assets:
@@ -1096,9 +1165,7 @@ async def start_workflow_run(
                     details=_tool_call_trace_details(
                         _external_tool_call_details(
                             node,
-                            input_item_count=len(
-                                _upstream_outputs(node, outputs_by_node)
-                            ),
+                            input_item_count=len(_upstream_outputs(node, outputs_by_node)),
                             output_item_count=0,
                         )
                     ),
@@ -1190,9 +1257,7 @@ async def start_workflow_run(
                     # Data-operator config/runtime errors must surface with the
                     # data_operator_execution_failed contract, not the native
                     # intelligence block shape this handler emits.
-                    binding_input = _read_dict(
-                        _read_dict(node.runtime.get("binding")).get("input")
-                    )
+                    binding_input = _read_dict(_read_dict(node.runtime.get("binding")).get("input"))
                     reason = WorkflowRunBlockReason(
                         code="data_operator_execution_failed",
                         message="Data operator execution failed",
@@ -1250,9 +1315,7 @@ async def start_workflow_run(
             except Exception as exc:
                 if _binding_id(node) not in _DATA_OPERATOR_BINDING_IDS:
                     raise
-                binding_input = _read_dict(
-                    _read_dict(node.runtime.get("binding")).get("input")
-                )
+                binding_input = _read_dict(_read_dict(node.runtime.get("binding")).get("input"))
                 reason = WorkflowRunBlockReason(
                     code="data_operator_execution_failed",
                     message="Data operator execution failed",
@@ -1329,9 +1392,7 @@ async def start_workflow_run(
             node,
             "started",
             message=(
-                "OpenCLI action dispatch started"
-                if is_write
-                else "OpenCLI source dispatch started"
+                "OpenCLI action dispatch started" if is_write else "OpenCLI source dispatch started"
             ),
         )
         fleet_match = await _match_dispatch_fleet_target(
@@ -1384,11 +1445,7 @@ async def start_workflow_run(
                 details={
                     "functionId": OPENCLI_FUNCTION_ID,
                     "worker": OPENCLI_WORKER,
-                    **(
-                        {"fleetMatch": fleet_match_details}
-                        if fleet_match_details
-                        else {}
-                    ),
+                    **({"fleetMatch": fleet_match_details} if fleet_match_details else {}),
                     **resource_details,
                 },
             )
@@ -1528,9 +1585,7 @@ async def start_workflow_run(
         internal_reasons = blocked_by_package.get(package_node.id, [])
         if internal_reasons:
             descendant_ids = {
-                node.id
-                for node in runtime_nodes
-                if package_node.id in _package_ancestor_ids(node)
+                node.id for node in runtime_nodes if package_node.id in _package_ancestor_ids(node)
             }
             source_node_ids = {
                 node.id
@@ -1543,10 +1598,11 @@ async def start_workflow_run(
             }
             terminal_events_by_node: dict[str, WorkflowNodeRunEvent] = {}
             for event in emitter.events:
-                if (
-                    event.nodeId in descendant_ids
-                    and event.eventType in {"completed", "failed", "blocked"}
-                ):
+                if event.nodeId in descendant_ids and event.eventType in {
+                    "completed",
+                    "failed",
+                    "blocked",
+                }:
                     terminal_events_by_node[event.nodeId] = event
             source_terminal_events = [
                 event
@@ -1557,12 +1613,10 @@ async def start_workflow_run(
                 event.eventType == "completed" for event in source_terminal_events
             )
             has_source_failure = any(
-                event.eventType in {"failed", "blocked"}
-                for event in source_terminal_events
+                event.eventType in {"failed", "blocked"} for event in source_terminal_events
             )
             has_non_source_failure = any(
-                event.eventType in {"failed", "blocked"}
-                and event.nodeId not in source_node_ids
+                event.eventType in {"failed", "blocked"} and event.nodeId not in source_node_ids
                 for event in terminal_events_by_node.values()
             )
             tolerates_internal_reasons = (
@@ -1635,8 +1689,7 @@ async def start_workflow_run(
                 workflow_id=body.project.id,
                 run_id=run_id,
                 trace_id=trace_id,
-                package_node_id=(trace.packageNodeId if trace else None)
-                or body.packageNodeId,
+                package_node_id=(trace.packageNodeId if trace else None) or body.packageNodeId,
                 started_at=started_at,
                 valid=compile_result.valid,
                 errors=list(compile_result.errors) if trace is None else compile_result.errors,
@@ -1752,10 +1805,7 @@ async def list_workflow_run_events(
         if limit is not None:
             statement = statement.limit(limit)
         rows = (await session.execute(statement)).scalars().all()
-        return [
-            WorkflowNodeRunEvent.model_validate(event_row.payload)
-            for event_row in rows
-        ]
+        return [WorkflowNodeRunEvent.model_validate(event_row.payload) for event_row in rows]
 
     stored = _RUNS.get(run_id)
     if not stored:
@@ -1785,6 +1835,155 @@ async def get_workflow_run_checkpoint(
     return _build_checkpoint(stored.request, stored.projection, stored.events)
 
 
+async def replay_downstream_from_persisted_gaojixing_source(
+    source_run_id: str,
+    *,
+    expected_workflow_id: str,
+    expected_studio_workflow_version_id: str,
+    session: AsyncSession,
+) -> WorkflowRunProjection:
+    """Replay only a completed Gaojixing source's persisted downstream path."""
+    source_run = await _load_workflow_run(source_run_id, session=session)
+    if source_run is None:
+        raise ValueError("Workflow run not found")
+    if (
+        source_run.projection.status != "completed"
+        or source_run.projection.workflowId != expected_workflow_id
+        or source_run.studio_workflow_version_id != expected_studio_workflow_version_id
+    ):
+        raise ValueError("Persisted source run is not replayable for this workflow version")
+
+    replay_run_id = _stable_id(
+        "downstream-replay", source_run_id, expected_studio_workflow_version_id
+    )
+    existing_replay = await _load_workflow_run(replay_run_id, session=session)
+    if existing_replay is not None:
+        if (
+            existing_replay.projection.workflowId != expected_workflow_id
+            or existing_replay.studio_workflow_version_id != expected_studio_workflow_version_id
+            or existing_replay.request.input.sourceId != source_run_id
+        ):
+            raise ValueError("Persisted replay identity conflicts with another workflow run")
+        return existing_replay.projection
+
+    compiled = compile_workflow_project(source_run.request.project)
+    if not compiled.valid or compiled.plan is None:
+        raise ValueError("Persisted source workflow no longer compiles")
+    source_nodes = {
+        node.id: node
+        for node in compiled.plan.runtime.nodes
+        if _is_gaojixing_source_node(node)
+    }
+    if not source_nodes:
+        raise ValueError("Persisted workflow has no Gaojixing source")
+
+    source_outputs: dict[str, list[dict[str, Any]]] = {}
+    for node_id, node in source_nodes.items():
+        partial = next(
+            (
+                event
+                for event in source_run.events
+                if event.nodeId == node_id
+                and event.eventType == "partial"
+                and _read_string(_read_dict(event.details).get("channelType"))
+                == GAOJIXING_CHANNEL_TYPE
+            ),
+            None,
+        )
+        completed = any(
+            event.nodeId == node_id and event.eventType == "completed"
+            for event in source_run.events
+        )
+        details = _read_dict(partial.details) if partial is not None else {}
+        package = _read_dict(details.get("package"))
+        evidence = _read_dict(details.get("evidence"))
+        package_digest = _read_string(package.get("digest"))
+        artifact_id = _read_string(details.get("artifactId"))
+        answer = _read_dict(evidence.get("answer"))
+        citations = _read_dict(evidence.get("citations"))
+        conversation = _read_dict(evidence.get("conversation"))
+        answer_text = _read_string(answer.get("text"))
+        if (
+            not completed
+            or not package_digest
+            or not artifact_id
+            or not answer_text
+            or evidence.get("packageDigest") != package_digest
+            or evidence.get("runId") != source_run_id
+            or evidence.get("workflowId") != expected_workflow_id
+            or evidence.get("nodeId") != node_id
+            or answer.get("artifactId") != artifact_id
+        ):
+            raise ValueError(f"Persisted Gaojixing source {node_id} lacks completed evidence")
+
+        conversation_url = _read_string(conversation.get("url"))
+        raw: dict[str, Any] = {
+            "content": answer_text,
+            "citations": deepcopy(citations.get("items", [])),
+            "conversation_url": conversation_url or "",
+            "gaojixing": {
+                "mode": evidence.get("mode"),
+                "provenance": evidence.get("provenance"),
+                "capabilityId": details.get("capabilityId"),
+                "package": deepcopy(package),
+                "artifactId": artifact_id,
+                "evidence": deepcopy(evidence),
+            },
+            "packageDigest": package_digest,
+            "questionPackage": deepcopy(package),
+            "answerArtifactId": artifact_id,
+            "mode": evidence.get("mode"),
+            "provenance": evidence.get("provenance"),
+        }
+        if conversation_url:
+            raw["dedupe"] = {
+                "type": "source-identity",
+                "field": "conversation_url",
+                "value": conversation_url,
+                "status": "unique",
+            }
+        source_outputs[node_id] = [
+            {
+                "raw": raw,
+                "lineage": [
+                    {
+                        "nodeId": node_id,
+                        "sourceGroup": _source_group(node, node_id),
+                        "artifact": "gaojixing.capture",
+                        "artifactId": artifact_id,
+                        "packageDigest": package_digest,
+                        "runId": source_run_id,
+                        "workflowId": expected_workflow_id,
+                        "mode": "persisted-replay",
+                        "provenance": evidence.get("provenance"),
+                        "index": 0,
+                    }
+                ],
+            }
+        ]
+
+    request = source_run.request.model_copy(
+        update={
+            "runId": replay_run_id,
+            "traceId": _stable_id(
+                "downstream-replay-trace", source_run_id, expected_studio_workflow_version_id
+            ),
+            "sourceOutputs": source_outputs,
+            "input": source_run.request.input.model_copy(
+                update={"source": "agent", "sourceId": source_run_id}
+            ),
+        },
+        deep=True,
+    )
+    return await start_workflow_run(
+        request,
+        session=session,
+        workflow_version_id=source_run.workflow_version_id,
+        studio_workflow_version_id=expected_studio_workflow_version_id,
+        replay_source_node_ids=set(source_outputs),
+    )
+
+
 async def continue_workflow_run_with_source_outputs(
     run_id: str,
     body: WorkflowRunSourceOutputsRequest,
@@ -2324,9 +2523,7 @@ async def _dispatch_opencli_source_to_fleet(
         payload = _read_dict(dispatch.iii.get("payload"))
         dispatch_policy = _read_string(payload.get("dispatch_policy"))
         adapter_node_id = (
-            _read_string(node.params.get("opencliAdapterNodeId"))
-            if node is not None
-            else None
+            _read_string(node.params.get("opencliAdapterNodeId")) if node is not None else None
         )
         local_adapter = False
         if adapter_node_id:
@@ -2334,11 +2531,7 @@ async def _dispatch_opencli_source_to_fleet(
 
             adapter_node = resolve_opencli_adapter_node(adapter_node_id)
             local_adapter = adapter_node is not None and not adapter_node.browser
-        if (
-            dispatch.packageNodeId is not None
-            and dispatch_policy != "inline"
-            and not local_adapter
-        ):
+        if dispatch.packageNodeId is not None and dispatch_policy != "inline" and not local_adapter:
             # Packaged HDA fanout retains its asynchronous worker-envelope
             # contract unless the package explicitly exposes raw items to a
             # downstream node in the same run.
@@ -2543,9 +2736,7 @@ def _run_status(
         else set()
     )
     effective_statuses = {
-        state.status
-        for state in node_states
-        if state.nodeId not in tolerated_source_failure_ids
+        state.status for state in node_states if state.nodeId not in tolerated_source_failure_ids
     }
     if runtime_nodes:
         terminal_ids = {node.id for node in runtime_nodes if _is_builder_output(node)}
@@ -2855,10 +3046,7 @@ def _opentabs_tool_block_reason(
             return None
         return WorkflowRunBlockReason(
             code=FETCH_PERMISSION_REQUIRED,
-            message=(
-                "OpenTabs read tool is bound, but "
-                "agentPermissions.canFetchNetwork is false."
-            ),
+            message=("OpenTabs read tool is bound, but agentPermissions.canFetchNetwork is false."),
             source="workflow_permissions",
             details={
                 "nodeId": node.id,
@@ -2910,10 +3098,7 @@ def _bbx_tool_block_reason(
             return None
         return WorkflowRunBlockReason(
             code=FETCH_PERMISSION_REQUIRED,
-            message=(
-                "BBX read tool is bound, but "
-                "agentPermissions.canFetchNetwork is false."
-            ),
+            message=("BBX read tool is bound, but agentPermissions.canFetchNetwork is false."),
             source="workflow_permissions",
             details={
                 "nodeId": node.id,
@@ -2938,8 +3123,7 @@ def _bbx_tool_block_reason(
         return WorkflowRunBlockReason(
             code=OPENCLI_WRITE_PERMISSION_REQUIRED,
             message=(
-                "BBX write tool is accepted, but "
-                "agentPermissions.canMutateExternalSites is false."
+                "BBX write tool is accepted, but agentPermissions.canMutateExternalSites is false."
             ),
             source="workflow_permissions",
             details={
@@ -2951,6 +3135,294 @@ def _bbx_tool_block_reason(
     return None
 
 
+async def _execute_gaojixing_fixture_source(
+    node: CompiledWorkflowNode,
+    *,
+    body: WorkflowRunStartRequest,
+    run_id: str,
+    workflow_id: str,
+    outputs_by_node: dict[str, list[dict[str, Any]]],
+    emitter: Any,
+) -> None:
+    """Materialize explicit fixture/mock Gaojixing input without live dispatch."""
+    binding_input = _binding_input(node)
+    adapter_config = _read_dict(binding_input.get("adapterConfig")) or binding_input
+    mode = _gaojixing_execution_mode(node)
+    if mode not in GAOJIXING_EXECUTION_MODES - {GAOJIXING_LIVE_MODE}:
+        reason = WorkflowRunBlockReason(
+            code="gaojixing_execution_mode_invalid",
+            message=f'Gaojixing source mode "{mode}" is not supported.',
+            source="gaojixing_fixture",
+            details={"nodeId": node.id, "mode": mode},
+        )
+        emitter.emit(
+            node, "blocked", message=reason.message, block_reason=reason, details=reason.details
+        )
+        outputs_by_node[node.id] = []
+        return
+    raw_items = _read_dict_list(body.sourceOutputs.get(node.id))
+    source = "sourceOutputs"
+    if not raw_items:
+        raw_items = _read_dict_list(node.params.get("fixtureItems"))
+        source = "fixtureItems"
+    if not raw_items:
+        raw_items = _read_dict_list(node.params.get("sampleItems", node.params.get("items")))
+        source = "sampleItems"
+    provenance = (
+        _read_string(adapter_config.get("fixtureProvenance"))
+        or _read_string(adapter_config.get("provenance"))
+        or f"{mode}:{source}"
+    )
+    try:
+        package = build_question_package(
+            node_params=dict(node.params),
+            adapter_config=adapter_config,
+            runtime_payload=body.input.payload,
+        )
+    except GaojixingReadinessError as exc:
+        reason = WorkflowRunBlockReason(
+            code=exc.code,
+            message=exc.message,
+            source="gaojixing_fixture",
+            details={"nodeId": node.id, "mode": mode, "provenance": provenance, **exc.details},
+        )
+        emitter.emit(
+            node, "blocked", message=exc.message, block_reason=reason, details=reason.details
+        )
+        outputs_by_node[node.id] = []
+        return
+    if not raw_items:
+        reason = WorkflowRunBlockReason(
+            code="gaojixing_fixture_output_required",
+            message=(
+                f"{mode.capitalize()} Gaojixing execution requires explicit fixture/mock input."
+            ),
+            source="gaojixing_fixture",
+            details={
+                "nodeId": node.id,
+                "mode": mode,
+                "provenance": provenance,
+                "packageDigest": package.digest,
+            },
+        )
+        emitter.emit(
+            node, "blocked", message=reason.message, block_reason=reason, details=reason.details
+        )
+        outputs_by_node[node.id] = []
+        return
+    mapped_items = []
+    source_group = _source_group(node, node.id)
+    for index, raw_item in enumerate(raw_items):
+        artifact_id = _stable_id(
+            "gaojixing-answer-artifact", mode, run_id, node.id, package.digest, str(index)
+        )
+        mapped = map_capture_item(
+            raw_item,
+            package=package,
+            workflow_id=workflow_id,
+            run_id=run_id,
+            node_id=node.id,
+            artifact_id=artifact_id,
+            mode=mode,
+            provenance=provenance,
+        )
+        mapped_items.append(
+            {
+                "raw": mapped,
+                "lineage": [
+                    {
+                        "nodeId": node.id,
+                        "sourceGroup": source_group,
+                        "artifact": "gaojixing.capture",
+                        "artifactId": artifact_id,
+                        "packageDigest": package.digest,
+                        "runId": run_id,
+                        "workflowId": workflow_id,
+                        "mode": mode,
+                        "provenance": provenance,
+                        "index": index,
+                    }
+                ],
+            }
+        )
+    outputs_by_node[node.id] = mapped_items
+    batch = _node_batch_reference(workflow_id, run_id, node, item_count=len(mapped_items))
+    emitter.emit(node, "started", message=f"{mode.capitalize()} Gaojixing source started")
+    emitter.emit(
+        node,
+        "partial",
+        message=f"{mode.capitalize()} Gaojixing evidence captured",
+        batch=batch,
+        details={
+            "bindingId": SOURCE_FETCH_BINDING_ID,
+            "channelType": GAOJIXING_CHANNEL_TYPE,
+            "mode": mode,
+            "provenance": provenance,
+            "package": package.to_dict(),
+            "artifacts": [
+                item["raw"]["gaojixing"]["artifactId"] for item in mapped_items
+            ],
+            "evidence": [
+                item["raw"]["gaojixing"]["evidence"] for item in mapped_items
+            ],
+            "lineage": _lineage_pointer(node),
+            "liveAccepted": False,
+        },
+    )
+    emitter.emit(node, "completed", message=f"{mode.capitalize()} Gaojixing source completed")
+
+
+async def _execute_gaojixing_source(
+    node: CompiledWorkflowNode,
+    *,
+    body: WorkflowRunStartRequest,
+    run_id: str,
+    workflow_id: str,
+    trace_id: str,
+    outputs_by_node: dict[str, list[dict[str, Any]]],
+    emitter: Any,
+    session: AsyncSession | None,
+) -> None:
+    """Run live Gaojixing through the existing channel and source path."""
+    del trace_id, session
+    binding_input = _binding_input(node)
+    adapter_config = _read_dict(binding_input.get("adapterConfig"))
+    if not adapter_config:
+        adapter_config = binding_input
+    try:
+        package = build_question_package(
+            node_params=dict(node.params),
+            adapter_config=adapter_config,
+            runtime_payload=body.input.payload,
+        )
+        result = await capture_live_doubao(
+            package=package,
+            node_params=dict(node.params),
+            adapter_config=adapter_config,
+            network_allowed=body.project.agentPermissions.canFetchNetwork,
+        )
+    except GaojixingReadinessError as exc:
+        reason = WorkflowRunBlockReason(
+            code=exc.code,
+            message=exc.message,
+            source="gaojixing_readiness",
+            details={"nodeId": node.id, **exc.details},
+        )
+        emitter.emit(
+            node, "blocked", message=exc.message, block_reason=reason, details=reason.details
+        )
+        outputs_by_node[node.id] = []
+        return
+
+    if not result.success:
+        code = result.error_type or "gaojixing_capture_failed"
+        reason = WorkflowRunBlockReason(
+            code=code,
+            message=result.error or "Live Gaojixing capture failed.",
+            source="doubao_research_channel",
+            details={"nodeId": node.id, "mode": "live", "packageDigest": package.digest},
+        )
+        emitter.emit(
+            node, "failed", message=reason.message, block_reason=reason, details=reason.details
+        )
+        outputs_by_node[node.id] = []
+        return
+
+    raw_item = next(
+        (
+            item
+            for item in result.items
+            if isinstance(item, dict) and _read_string(item.get("content"))
+        ),
+        None,
+    )
+    if raw_item is None:
+        reason = WorkflowRunBlockReason(
+            code="gaojixing_answer_missing",
+            message="Live Gaojixing capture returned no assistant answer.",
+            source="gaojixing_evidence",
+            details={"nodeId": node.id, "packageDigest": package.digest},
+        )
+        emitter.emit(
+            node, "failed", message=reason.message, block_reason=reason, details=reason.details
+        )
+        outputs_by_node[node.id] = []
+        return
+
+    artifact_id = _stable_id("gaojixing-answer-artifact", run_id, node.id, package.digest)
+    mapped = map_capture_item(
+        raw_item,
+        package=package,
+        workflow_id=workflow_id,
+        run_id=run_id,
+        node_id=node.id,
+        artifact_id=artifact_id,
+    )
+    source_group = _source_group(node, node.id)
+    lineage = [
+        {
+            "nodeId": node.id,
+            "sourceGroup": source_group,
+            "artifact": "gaojixing.capture",
+            "artifactId": artifact_id,
+            "packageDigest": package.digest,
+            "runId": run_id,
+            "workflowId": workflow_id,
+            "mode": "live",
+            "provenance": mapped["gaojixing"]["provenance"],
+            "index": 0,
+        }
+    ]
+    outputs_by_node[node.id] = [{"raw": mapped, "lineage": lineage}]
+    evidence = mapped["gaojixing"]["evidence"]
+    batch = _node_batch_reference(workflow_id, run_id, node, item_count=1)
+    emitter.emit(
+        node,
+        "partial",
+        message="Live Gaojixing answer and evidence captured",
+        batch=batch,
+        details={
+            "bindingId": SOURCE_FETCH_BINDING_ID,
+            "channelType": GAOJIXING_CHANNEL_TYPE,
+            "capabilityId": mapped["gaojixing"]["capabilityId"],
+            "mode": "live",
+            "provenance": mapped["gaojixing"]["provenance"],
+            "package": package.to_dict(),
+            "artifactId": artifact_id,
+            "evidence": evidence,
+            "lineage": _lineage_pointer(node),
+        },
+    )
+    emitter.emit(node, "completed", message="Live Gaojixing source completed")
+
+
+def _is_gaojixing_source_node(node: CompiledWorkflowNode) -> bool:
+    binding_input = _binding_input(node)
+    return (
+        node.kind == "source"
+        and (
+            _read_string(binding_input.get("channelType"))
+            or _read_string(node.params.get("channelType"))
+        )
+        == GAOJIXING_CHANNEL_TYPE
+    )
+
+
+def _gaojixing_execution_mode(node: CompiledWorkflowNode) -> str:
+    binding_input = _binding_input(node)
+    adapter_config = _read_dict(binding_input.get("adapterConfig"))
+    return (
+        _read_string(binding_input.get("liveMode"))
+        or _read_string(adapter_config.get("liveMode"))
+        or _read_string(node.params.get("liveMode"))
+        or GAOJIXING_LIVE_MODE
+    )
+
+
+def _source_live_mode(node: CompiledWorkflowNode) -> bool:
+    return _gaojixing_execution_mode(node) == GAOJIXING_LIVE_MODE
+
+
 def _is_first_loop_native_node(node: CompiledWorkflowNode) -> bool:
     binding = node.runtime.get("binding")
     if not isinstance(binding, dict):
@@ -3259,8 +3731,7 @@ async def _execute_native_node(
         if operator_id is None:
             raise ValueError(f"Data operator binding {binding_id} is missing operatorId")
         pack_version = (
-            _read_string(binding_input.get("packVersion"))
-            or _LEGACY_DATA_OPERATOR_PACK_VERSION
+            _read_string(binding_input.get("packVersion")) or _LEGACY_DATA_OPERATOR_PACK_VERSION
         )
         config = binding_input.get("config")
         if not isinstance(config, dict):
@@ -3298,9 +3769,7 @@ async def _execute_native_node(
                 "outputPort": binding_input.get("outputPort", "recordCandidate[]"),
                 "inputItemCount": len(input_items),
                 "outputItemCount": len(output_items),
-                "rejectedCount": result.metrics.get(
-                    "rejectedCount", len(rejected_candidate_ids)
-                ),
+                "rejectedCount": result.metrics.get("rejectedCount", len(rejected_candidate_ids)),
                 "rejectedCandidateIds": rejected_candidate_ids[:100],
                 "rejectedCandidateIdsTruncated": len(rejected_candidate_ids) > 100,
                 "metrics": result.metrics,
@@ -3412,8 +3881,7 @@ async def _execute_native_node(
             materialized_source_tasks=materialized_source_tasks or {},
         )
         stored_refs = [
-            _append_lineage(item, node, step="store", run_id=run_id)
-            for item in stored_refs
+            _append_lineage(item, node, step="store", run_id=run_id) for item in stored_refs
         ]
         return (
             {
@@ -3421,7 +3889,9 @@ async def _execute_native_node(
                 "target": target,
                 "writeMode": binding_input.get("writeMode", "append"),
                 "inputRecordCount": len(input_items),
-                "storedRecordCount": len(stored_refs),
+                "storedRecordCount": sum(
+                    reference.get("outcome") != "skipped" for reference in stored_refs
+                ),
                 "skippedRecordCount": skipped_count,
                 "storedRefs": stored_refs,
                 "lineage": _lineage_pointer(node),
@@ -3561,9 +4031,7 @@ async def _execute_external_tool_capability(
     if binding_input.get("executorMode") == NATIVE_INTELLIGENCE_EXECUTOR:
         tool_id = binding_input.get("toolCapabilityId")
         action = (
-            NATIVE_INTELLIGENCE_ACTION_BY_TOOL_ID.get(tool_id)
-            if isinstance(tool_id, str)
-            else None
+            NATIVE_INTELLIGENCE_ACTION_BY_TOOL_ID.get(tool_id) if isinstance(tool_id, str) else None
         )
         if action is None:
             raise ValueError("native_intelligence_action_not_registered")
@@ -3774,9 +4242,7 @@ def _resolved_kats_params(
     binding_input: dict[str, Any],
 ) -> tuple[str, dict[str, Any]]:
     params = _merged_tool_params(binding_input)
-    operation = _read_string(
-        _read_dict(binding_input.get("executorParams")).get("operation")
-    )
+    operation = _read_string(_read_dict(binding_input.get("executorParams")).get("operation"))
     params["operation"] = operation
     return operation, params
 
@@ -3926,7 +4392,9 @@ async def _store_record_sink_outputs(
             0,
         )
 
-    triples_by_source_node: dict[str, list[tuple[dict, dict, str, list[dict[str, Any]]]]] = {}
+    triples_by_source_node: dict[
+        str, list[tuple[dict, dict, str, list[dict[str, Any]], str | None]]
+    ] = {}
     for item in input_items:
         source_node_id = _origin_source_node_id(item, runtime_nodes_by_id)
         if not source_node_id:
@@ -3950,9 +4418,18 @@ async def _store_record_sink_outputs(
             {key: value for key, value in accepted_normalized.items() if key not in {"source_id"}}
         )
         normalized["source_id"] = source_id
+        gaojixing = _read_dict(raw.get("gaojixing"))
+        if gaojixing:
+            package = _read_dict(gaojixing.get("package"))
+            normalized["packageDigest"] = _read_string(package.get("digest"))
+            normalized["answerArtifactId"] = _read_string(gaojixing.get("artifactId"))
+            normalized["evidenceRefs"] = _read_dict(gaojixing.get("evidence"))
         content_hash = _read_string(item.get("contentHash")) or content_hash
+        dedupe_identity = _dedupe_identity(item)
+        if dedupe_identity:
+            normalized["dedupeIdentity"] = dedupe_identity
         triples_by_source_node.setdefault(source_node_id, []).append(
-            (raw, normalized, content_hash, lineage)
+            (raw, normalized, content_hash, lineage, dedupe_identity)
         )
 
     stored_refs: list[dict[str, Any]] = []
@@ -3972,21 +4449,55 @@ async def _store_record_sink_outputs(
             source_id,
             [
                 (raw, normalized, content_hash)
-                for raw, normalized, content_hash, _lineage in triples_with_lineage
+                for raw, normalized, content_hash, _lineage, _identity in triples_with_lineage
             ],
             channel_type=_workflow_source_channel_type(runtime_nodes_by_id[source_node_id]),
             forward_to_odp=False,
             workflow_id=workflow_id,
             workflow_run_id=run_id,
+            identities=[
+                identity for _raw, _normalized, _hash, _lineage, identity in triples_with_lineage
+            ],
+            lineage=_record_lineage_envelope(
+                triples_with_lineage[0][0],
+                triples_with_lineage[0][3],
+                workflow_id=workflow_id,
+                run_id=run_id,
+                source_id=source_id,
+                task_id=task_id,
+                source_node_id=source_node_id,
+            ),
         )
         skipped_total += skipped
-        for record, (raw, normalized, content_hash, lineage) in zip(
-            records, triples_with_lineage, strict=False
-        ):
+        persisted_ids = {record.id for record in records}
+        result = await session.execute(
+            select(CollectedRecord).where(
+                CollectedRecord.source_id == source_id,
+                CollectedRecord.content_hash.in_(
+                    [
+                        content_hash
+                        for (
+                            _raw,
+                            _normalized,
+                            content_hash,
+                            _lineage,
+                            _identity,
+                        ) in triples_with_lineage
+                    ]
+                ),
+            )
+        )
+        records_by_hash = {record.content_hash: record for record in result.scalars()}
+        for raw, normalized, content_hash, lineage, dedupe_identity in triples_with_lineage:
+            record = records_by_hash.get(content_hash)
+            if record is None:
+                raise RuntimeError("record sink did not resolve durable record reference")
             stored_refs.append(
                 {
                     "recordId": record.id,
                     "target": target,
+                    "outcome": "stored" if record.id in persisted_ids else "skipped",
+                    "dedupeIdentity": dedupe_identity,
                     "sourceId": source_id,
                     "taskId": task_id,
                     "raw": raw,
@@ -4000,6 +4511,55 @@ async def _store_record_sink_outputs(
     return stored_refs, skipped_total
 
 
+def _record_lineage_envelope(
+    raw: dict[str, Any],
+    lineage: list[dict[str, Any]],
+    *,
+    workflow_id: str,
+    run_id: str,
+    source_id: str,
+    task_id: str,
+    source_node_id: str,
+) -> dict[str, Any]:
+    gaojixing = _read_dict(raw.get("gaojixing"))
+    package = _read_dict(gaojixing.get("package"))
+    evidence = _read_dict(gaojixing.get("evidence"))
+    artifact_id = _read_string(gaojixing.get("artifactId"))
+    artifact_refs: list[dict[str, Any]] = [
+        {
+            "kind": "workflow-lineage",
+            "nodeId": source_node_id,
+            "runId": run_id,
+            "lineage": lineage,
+        }
+    ]
+    if artifact_id:
+        artifact_refs.append(
+            {
+                "kind": "gaojixing.capture",
+                "artifactId": artifact_id,
+                "packageDigest": _read_string(package.get("digest")),
+                "evidence": evidence,
+            }
+        )
+    return CollectionLineage(
+        task_id=task_id,
+        source_id=source_id,
+        provider=GAOJIXING_CHANNEL_TYPE if gaojixing else "workflow",
+        ingest_mode="snapshot",
+        collection_run_id=run_id,
+        project_id=workflow_id,
+        runtime_id="workflow.opencli_hda",
+        artifact_refs=artifact_refs,
+    ).to_dict()
+
+
+def _dedupe_identity(item: dict[str, Any]) -> str | None:
+    dedupe = _read_dict(item.get("dedupe"))
+    identity = _read_string(dedupe.get("identity"))
+    return identity if dedupe.get("status") == "unique" and identity else None
+
+
 async def _materialize_source_task(
     session: AsyncSession,
     source_node: CompiledWorkflowNode,
@@ -4082,9 +4642,7 @@ async def _find_materialized_workflow_source(
     channel_type: str,
 ) -> DataSource | None:
     candidates = (
-        await session.scalars(
-            select(DataSource).where(DataSource.channel_type == channel_type)
-        )
+        await session.scalars(select(DataSource).where(DataSource.channel_type == channel_type))
     ).all()
     for candidate in candidates:
         config = candidate.channel_config if isinstance(candidate.channel_config, dict) else {}
@@ -4479,12 +5037,9 @@ def _notify_send_block_reason(
 
 
 def _is_opencli_write_node(node: CompiledWorkflowNode) -> bool:
-    return (
-        _binding_id(node) == OPENCLI_BINDING_ID
-        and (
-            _read_string(node.params.get("opencliAccess")) == "write"
-            or (node.kind == "action" and node.capability == "store")
-        )
+    return _binding_id(node) == OPENCLI_BINDING_ID and (
+        _read_string(node.params.get("opencliAccess")) == "write"
+        or (node.kind == "action" and node.capability == "store")
     )
 
 
diff --git a/backend/workflow/record_hygiene/engine.py b/backend/workflow/record_hygiene/engine.py
index a5ff1fd3..37fe97f2 100644
--- a/backend/workflow/record_hygiene/engine.py
+++ b/backend/workflow/record_hygiene/engine.py
@@ -132,6 +132,9 @@ def _normalize(
             "quality": _quality(item, normalized),
             "lineage": lineage,
         }
+        if canonical_dedupe := _canonical_source_dedupe(raw, normalized):
+            candidate["dedupe"] = canonical_dedupe
+            candidate["evidence"] = [canonical_dedupe]
         if language:
             candidate["normalization"] = {"language": language}
         if preserve_source_refs:
@@ -178,6 +181,7 @@ def _dedupe(
             "window": window_label,
             "windowHours": window_hours,
         }
+        dedupe_evidence = _canonical_dedupe_evidence(item, evidence)
         updated = deepcopy(item)
         updated["lineage"] = [
             *_lineage(item, index),
@@ -188,12 +192,16 @@ def _dedupe(
         if winner is not None:
             duplicate_of = _record_identity(winner)
             updated["duplicateOf"] = duplicate_of
-            updated["dedupe"] = {**evidence, "status": "duplicate", "duplicateOf": duplicate_of}
+            updated["dedupe"] = {
+                **dedupe_evidence,
+                "status": "duplicate",
+                "duplicateOf": duplicate_of,
+            }
             updated["rejection"] = {"code": "duplicate", "duplicateOf": duplicate_of}
             rejected.append(updated)
             continue
 
-        updated["dedupe"] = {**evidence, "status": "unique"}
+        updated["dedupe"] = {**dedupe_evidence, "status": "unique"}
         records.append(updated)
         canonical.append((updated, identity, published_at, published_raw))
 
@@ -483,7 +491,87 @@ def _quality(item: Mapping[str, Any], normalized: Mapping[str, Any]) -> float:
 
 def _has_unique_dedupe_evidence(item: Mapping[str, Any]) -> bool:
     dedupe = item.get("dedupe")
-    return isinstance(dedupe, Mapping) and dedupe.get("status") == "unique"
+    return (
+        isinstance(dedupe, Mapping)
+        and dedupe.get("status") == "unique"
+        and isinstance(dedupe.get("identity"), str)
+        and bool(dedupe["identity"])
+        and isinstance(dedupe.get("fingerprint"), str)
+        and bool(dedupe["fingerprint"])
+    )
+
+
+def _canonical_source_dedupe(
+    raw: Mapping[str, Any], normalized: Mapping[str, Any]
+) -> dict[str, Any] | None:
+    source_dedupe = raw.get("dedupe")
+    package_digest = _package_digest(raw)
+    artifact_id = _artifact_id(raw)
+    if (
+        not isinstance(source_dedupe, Mapping)
+        or source_dedupe.get("type") != "source-identity"
+        or source_dedupe.get("status") != "unique"
+        or not isinstance(source_dedupe.get("field"), str)
+        or not source_dedupe["field"].strip()
+        or not isinstance(source_dedupe.get("value"), str)
+        or not source_dedupe["value"].strip()
+        or package_digest is None
+        or artifact_id is None
+    ):
+        return None
+    business = {
+        "sourceId": normalized.get("source_id", ""),
+        "sourceIdentity": {
+            "field": source_dedupe["field"].strip(),
+            "value": source_dedupe["value"].strip(),
+        },
+        "url": normalized.get("url", ""),
+        "title": normalized.get("title", ""),
+        "publishedAt": normalized.get("published_at", ""),
+    }
+    basis = {"business": business, "packageDigest": package_digest}
+    fingerprint = _stable_json_hash(basis)
+    return {
+        "type": "canonical-dedupe",
+        "status": "unique",
+        "identity": f"dedupe-{fingerprint[:24]}",
+        "fingerprint": fingerprint,
+        "basis": basis,
+        "artifact": {"artifactId": artifact_id, "packageDigest": package_digest},
+    }
+
+
+def _canonical_dedupe_evidence(
+    item: Mapping[str, Any], operation: Mapping[str, Any]
+) -> dict[str, Any]:
+    existing = item.get("dedupe")
+    if _has_unique_dedupe_evidence({"dedupe": existing}):
+        return {**dict(existing), "operation": dict(operation)}
+    basis = {"key": operation["key"], "values": operation["values"]}
+    fingerprint = _stable_json_hash(basis)
+    return {
+        **dict(operation),
+        "identity": f"dedupe-{fingerprint[:24]}",
+        "fingerprint": fingerprint,
+        "basis": basis,
+    }
+
+
+def _package_digest(raw: Mapping[str, Any]) -> str | None:
+    package = raw.get("questionPackage")
+    if not isinstance(package, Mapping):
+        gaojixing = raw.get("gaojixing")
+        package = gaojixing.get("package") if isinstance(gaojixing, Mapping) else None
+    digest = package.get("digest") if isinstance(package, Mapping) else raw.get("packageDigest")
+    return digest.strip() if isinstance(digest, str) and digest.strip() else None
+
+
+def _artifact_id(raw: Mapping[str, Any]) -> str | None:
+    artifact_id = raw.get("answerArtifactId")
+    if not isinstance(artifact_id, str):
+        gaojixing = raw.get("gaojixing")
+        artifact_id = gaojixing.get("artifactId") if isinstance(gaojixing, Mapping) else None
+    return artifact_id.strip() if isinstance(artifact_id, str) and artifact_id.strip() else None
 
 
 def _record_identity(item: Mapping[str, Any]) -> str:
diff --git a/backend/workflow/webhook_delivery.py b/backend/workflow/webhook_delivery.py
index f5b2ecaa..e60a9218 100644
--- a/backend/workflow/webhook_delivery.py
+++ b/backend/workflow/webhook_delivery.py
@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import uuid
 from typing import Any
 
 import httpx
@@ -32,57 +33,89 @@ async def execute_workflow_webhook_delivery(
 ) -> dict[str, Any]:
     config = _webhook_config(binding_input)
     target = _read_string(binding_input.get("target")) or "webhook"
+    gaojixing = _gaojixing_delivery_context(
+        input_items,
+        workflow_id=workflow_id,
+        run_id=run_id,
+        node_id=node_id,
+    )
+    payload_data = {
+        "schema": WEBHOOK_DELIVERY_PAYLOAD_SCHEMA,
+        "workflowId": workflow_id,
+        "workflowRunId": run_id,
+        "nodeId": node_id,
+        "target": target,
+        "itemCount": len(input_items),
+        "items": [_safe_delivery_item(item) for item in input_items],
+    }
+    if gaojixing:
+        payload_data["packageDigest"] = gaojixing["packageDigest"]
+        payload_data["lineage"] = gaojixing["lineage"]
+        payload_data["mode"] = gaojixing["mode"]
+        payload_data["provenance"] = gaojixing["provenance"]
     payload = NotificationPayload(
         event=WEBHOOK_DELIVERY_EVENT,
         source_id=workflow_id,
+        delivery_id=gaojixing["deliveryAttemptId"] if gaojixing else None,
         record_id=run_id,
-        data={
-            "schema": WEBHOOK_DELIVERY_PAYLOAD_SCHEMA,
-            "workflowId": workflow_id,
-            "workflowRunId": run_id,
-            "nodeId": node_id,
-            "target": target,
-            "itemCount": len(input_items),
-            "items": [_safe_delivery_item(item) for item in input_items],
-        },
+        data=payload_data,
+        lineage=gaojixing["lineage"] if gaojixing else None,
     )
 
     try:
-        delivered, _ = _normalize_send_result(
+        delivered, response_data = _normalize_send_result(
             await get_notifier("webhook").send(config, payload)
         )
     except WorkflowWebhookDeliveryError:
         raise
     except (httpx.HTTPError, OSError) as exc:
-        # Network-level failures (connect errors, timeouts, DNS failures,
-        # etc.) from the notifier's underlying HTTP client are not part of
-        # its bool contract — without this, an unreachable webhook raises an
-        # untyped exception here that isn't caught by the per-node
-        # WorkflowWebhookDeliveryError handler in opencli_hda_tracer.py and
-        # 500s the entire workflow run instead of failing just this node.
+        details = {
+            "nodeId": node_id,
+            "target": target,
+            "itemCount": len(input_items),
+            "payloadSchema": WEBHOOK_DELIVERY_PAYLOAD_SCHEMA,
+        }
+        if gaojixing:
+            details.update(
+                {
+                    "deliveryAttemptId": gaojixing["deliveryAttemptId"],
+                    "transportStatus": "failed",
+                    "businessOutcome": "unknown",
+                    "lineage": gaojixing["lineage"],
+                    "mode": gaojixing["mode"],
+                    "provenance": gaojixing["provenance"],
+                }
+            )
         raise WorkflowWebhookDeliveryError(
             code="webhook_delivery_network_error",
             message=f"Webhook delivery failed due to a network error: {exc}",
-            details={
-                "nodeId": node_id,
-                "target": target,
-                "itemCount": len(input_items),
-                "payloadSchema": WEBHOOK_DELIVERY_PAYLOAD_SCHEMA,
-            },
+            details=details,
         ) from exc
     if not delivered:
+        details = {
+            "nodeId": node_id,
+            "target": target,
+            "itemCount": len(input_items),
+            "payloadSchema": WEBHOOK_DELIVERY_PAYLOAD_SCHEMA,
+        }
+        if gaojixing:
+            details.update(
+                {
+                    "deliveryAttemptId": gaojixing["deliveryAttemptId"],
+                    "transportStatus": "failed",
+                    "businessOutcome": "unknown",
+                    "lineage": gaojixing["lineage"],
+                    "mode": gaojixing["mode"],
+                    "provenance": gaojixing["provenance"],
+                }
+            )
         raise WorkflowWebhookDeliveryError(
             code="webhook_delivery_failed",
             message="Webhook delivery attempted but the notifier returned a failure.",
-            details={
-                "nodeId": node_id,
-                "target": target,
-                "itemCount": len(input_items),
-                "payloadSchema": WEBHOOK_DELIVERY_PAYLOAD_SCHEMA,
-            },
+            details=details,
         )
 
-    return {
+    result = {
         "notifierType": "webhook",
         "target": target,
         "deliveryAttempted": True,
@@ -91,6 +124,237 @@ async def execute_workflow_webhook_delivery(
         "payloadSchema": WEBHOOK_DELIVERY_PAYLOAD_SCHEMA,
         "itemCount": len(input_items),
     }
+    if gaojixing:
+        ack = _destination_ack(response_data, gaojixing["deliveryAttemptId"])
+        matching_ack = bool(ack and ack["matchesDeliveryAttempt"])
+        is_live = gaojixing["mode"] == "live"
+        result.update(
+            {
+                "deliveryAttemptId": gaojixing["deliveryAttemptId"],
+                "transportStatus": "accepted",
+                "businessOutcome": (
+                    "confirmed"
+                    if is_live and matching_ack
+                    else "unconfirmed"
+                    if is_live
+                    else gaojixing["mode"]
+                ),
+                "ackEvidence": ack,
+                "packageDigest": gaojixing["packageDigest"],
+                "mode": gaojixing["mode"],
+                "provenance": gaojixing["provenance"],
+                "liveAccepted": is_live and matching_ack,
+                "lineage": gaojixing["lineage"],
+            }
+        )
+    return result
+
+def _gaojixing_delivery_context(
+    input_items: list[dict[str, Any]],
+    *,
+    workflow_id: str,
+    run_id: str,
+    node_id: str,
+) -> dict[str, Any] | None:
+    contexts: list[dict[str, Any]] = []
+    for index, item in enumerate(input_items):
+        raw = _read_dict(item.get("raw"))
+        gaojixing = _read_dict(raw.get("gaojixing"))
+        if not gaojixing:
+            continue
+        package = _read_dict(gaojixing.get("package"))
+        evidence = _read_dict(gaojixing.get("evidence"))
+        package_digest = _read_string(package.get("digest"))
+        mode = _read_string(gaojixing.get("mode"))
+        provenance = _read_string(gaojixing.get("provenance"))
+        evidence_digest = _read_string(evidence.get("packageDigest"))
+        evidence_mode = _read_string(evidence.get("mode"))
+        evidence_provenance = _read_string(evidence.get("provenance"))
+        context_details = {
+            "workflowId": workflow_id,
+            "runId": run_id,
+            "nodeId": node_id,
+            "itemIndex": index,
+        }
+        if not package_digest or not mode or not provenance:
+            raise WorkflowWebhookDeliveryError(
+                code="gaojixing_delivery_context_incomplete",
+                message="Gaojixing delivery input lacks package, mode, or provenance.",
+                details=context_details,
+            )
+        if (
+            evidence_digest != package_digest
+            or evidence_mode != mode
+            or evidence_provenance != provenance
+        ):
+            raise WorkflowWebhookDeliveryError(
+                code="gaojixing_delivery_evidence_mismatch",
+                message="Gaojixing delivery evidence contradicts its source envelope.",
+                details={
+                    **context_details,
+                    "packageDigest": package_digest,
+                    "mode": mode,
+                    "provenance": provenance,
+                },
+            )
+        _validate_gaojixing_lineage(
+            raw=raw,
+            gaojixing=gaojixing,
+            evidence=evidence,
+            source_lineage=_read_dict_list(item.get("lineage")),
+            context_details=context_details,
+        )
+        contexts.append(
+            {
+                "artifactId": _read_string(gaojixing.get("artifactId")),
+                "itemIndex": index,
+                "mode": mode,
+                "packageDigest": package_digest,
+                "provenance": provenance,
+                "sourceLineage": _read_dict_list(item.get("lineage")),
+            }
+        )
+    if not contexts:
+        return None
+    first = contexts[0]
+    for context in contexts[1:]:
+        inconsistent = [
+            field
+            for field in ("packageDigest", "mode", "provenance")
+            if context[field] != first[field]
+        ]
+        if inconsistent:
+            raise WorkflowWebhookDeliveryError(
+                code="gaojixing_delivery_context_mismatch",
+                message="Gaojixing delivery batch mixes incompatible source contexts.",
+                details={
+                    "workflowId": workflow_id,
+                    "runId": run_id,
+                    "nodeId": node_id,
+                    "itemIndex": context["itemIndex"],
+                    "inconsistentFields": inconsistent,
+                },
+            )
+    lineage = {
+        "workflowId": workflow_id,
+        "workflowRunId": run_id,
+        "nodeId": node_id,
+        "packageDigest": first["packageDigest"],
+        "artifactId": first["artifactId"],
+        "sourceLineage": first["sourceLineage"],
+        "mode": first["mode"],
+        "provenance": first["provenance"],
+    }
+    delivery_id = str(
+        uuid.uuid5(
+            uuid.NAMESPACE_URL,
+            (
+                "opencli-admin/gaojixing/delivery/"
+                f"{workflow_id}/{run_id}/{node_id}/{first['packageDigest']}"
+            ),
+        )
+    )
+    return {
+        "deliveryAttemptId": delivery_id,
+        "packageDigest": first["packageDigest"],
+        "lineage": lineage,
+        "mode": first["mode"],
+        "provenance": first["provenance"],
+    }
+
+
+def _destination_ack(
+    response_data: dict[str, Any] | None, delivery_attempt_id: str
+) -> dict[str, Any] | None:
+    if not isinstance(response_data, dict):
+        return None
+    confirmation_field: str | None = None
+    for key in ("businessAck", "business_ack", "acknowledged"):
+        if response_data.get(key) is True:
+            confirmation_field = key
+            break
+    status = _read_string(response_data.get("status"))
+    if confirmation_field is None and status in {"confirmed", "acknowledged"}:
+        confirmation_field = "status"
+    if confirmation_field is None:
+        return None
+    ack_delivery_id = _delivery_attempt_id(response_data)
+    return {
+        "status": "confirmed",
+        "source": "destination_response",
+        "field": confirmation_field,
+        "deliveryAttemptId": ack_delivery_id,
+        "matchesDeliveryAttempt": ack_delivery_id == delivery_attempt_id,
+    }
+
+
+def _validate_gaojixing_lineage(
+    *,
+    raw: dict[str, Any],
+    gaojixing: dict[str, Any],
+    evidence: dict[str, Any],
+    source_lineage: list[dict[str, Any]],
+    context_details: dict[str, Any],
+) -> None:
+    package = _read_dict(gaojixing.get("package"))
+    package_digest = _read_string(package.get("digest"))
+    artifact_id = _read_string(gaojixing.get("artifactId"))
+    expected = {
+        "packageDigest": package_digest,
+        "artifactId": artifact_id,
+        "mode": _read_string(gaojixing.get("mode")),
+        "provenance": _read_string(gaojixing.get("provenance")),
+        "runId": _read_string(evidence.get("runId")),
+        "workflowId": _read_string(evidence.get("workflowId")),
+        "nodeId": _read_string(evidence.get("nodeId")),
+    }
+    raw_digest = _read_string(raw.get("packageDigest"))
+    question_package = _read_dict(raw.get("questionPackage"))
+    mismatches = [
+        field
+        for field, actual, expected_value in (
+            ("packageDigest", raw_digest, package_digest),
+            (
+                "questionPackage.digest",
+                _read_string(question_package.get("digest")),
+                package_digest,
+            ),
+            (
+                "evidence.answer.artifactId",
+                _read_string(_read_dict(evidence.get("answer")).get("artifactId")),
+                artifact_id,
+            ),
+        )
+        if actual is not None and actual != expected_value
+    ]
+    for entry in source_lineage:
+        for field, expected_value in expected.items():
+            actual = _read_string(entry.get(field))
+            if field == "mode" and actual == "persisted-replay":
+                continue
+            if actual is not None and expected_value is not None and actual != expected_value:
+                mismatches.append(f"sourceLineage.{field}")
+    if mismatches:
+        raise WorkflowWebhookDeliveryError(
+            code="gaojixing_delivery_lineage_mismatch",
+            message="Gaojixing delivery lineage contradicts its immutable source evidence.",
+            details={**context_details, "mismatchedFields": sorted(set(mismatches))},
+        )
+
+
+def _delivery_attempt_id(response_data: dict[str, Any]) -> str | None:
+    keys = (
+        "deliveryAttemptId",
+        "delivery_attempt_id",
+        "deliveryId",
+        "delivery_id",
+        "idempotencyKey",
+    )
+    for key in keys:
+        value = _read_string(response_data.get(key))
+        if value:
+            return value
+    return None
 
 
 def _webhook_config(binding_input: dict[str, Any]) -> dict[str, Any]:
diff --git a/chrome/entrypoint.sh b/chrome/entrypoint.sh
index 864e5f33..ac232904 100644
--- a/chrome/entrypoint.sh
+++ b/chrome/entrypoint.sh
@@ -30,7 +30,7 @@ websockify --web /usr/share/novnc 6080 localhost:5900 &
 # Start Browser Bridge daemon (always enabled).
 # Listens on 0.0.0.0 so the API/worker containers can reach it via chrome-{N}:19825.
 # The extension connects to ws://localhost:19825/ext.
-DAEMON_JS="$(npm root -g)/@jackwener/opencli/dist/daemon.js"
+DAEMON_JS="$(npm root -g)/@jackwener/opencli/dist/src/daemon.js"
 if [ -f "$DAEMON_JS" ]; then
   (while true; do
     OPENCLI_DAEMON_LISTEN=0.0.0.0 node "$DAEMON_JS"
diff --git a/collaboration/Dockerfile b/collaboration/Dockerfile
new file mode 100644
index 00000000..d6b08a49
--- /dev/null
+++ b/collaboration/Dockerfile
@@ -0,0 +1,12 @@
+FROM node:22.16.0-alpine
+
+WORKDIR /app
+COPY package.json package-lock.json ./
+RUN npm ci --omit=dev --ignore-scripts
+COPY --chown=node:node server.mjs ./
+
+USER node
+EXPOSE 1234
+HEALTHCHECK --interval=10s --timeout=3s --start-period=10s --retries=5 \
+  CMD node -e "fetch('http://127.0.0.1:1234/health').then((response) => process.exit(response.ok ? 0 : 1)).catch(() => process.exit(1))"
+CMD ["npm", "start"]
diff --git a/collaboration/package-lock.json b/collaboration/package-lock.json
new file mode 100644
index 00000000..581a7f2a
--- /dev/null
+++ b/collaboration/package-lock.json
@@ -0,0 +1,541 @@
+{
+  "name": "opencli-studio-collaboration",
+  "version": "1.0.0",
+  "lockfileVersion": 3,
+  "requires": true,
+  "packages": {
+    "": {
+      "name": "opencli-studio-collaboration",
+      "version": "1.0.0",
+      "dependencies": {
+        "@y/websocket-server": "0.1.1",
+        "ws": "6.2.6",
+        "y-protocols": "1.0.6",
+        "yjs": "13.6.31"
+      },
+      "engines": {
+        "node": ">=22.0.0 <23"
+      }
+    },
+    "node_modules/@y/websocket-server": {
+      "version": "0.1.1",
+      "resolved": "https://registry.npmjs.org/@y/websocket-server/-/websocket-server-0.1.1.tgz",
+      "integrity": "sha512-pPtXm5Ceqs4orhXXHwm2I+u1mKNBDNzlrwNiI7OMwM7PlVS4WCMpiIuSB8WsYeSuISbvpXPNvaj6H1MoQBbE+g==",
+      "license": "MIT",
+      "dependencies": {
+        "lib0": "^0.2.102",
+        "y-protocols": "^1.0.5"
+      },
+      "bin": {
+        "y-websocket": "src/server.js",
+        "y-websocket-server": "src/server.js"
+      },
+      "engines": {
+        "node": ">=16.0.0",
+        "npm": ">=8.0.0"
+      },
+      "funding": {
+        "type": "GitHub Sponsors ❤",
+        "url": "https://github.com/sponsors/dmonad"
+      },
+      "optionalDependencies": {
+        "ws": "^6.2.1",
+        "y-leveldb": "^0.1.0"
+      },
+      "peerDependencies": {
+        "yjs": "^13.5.6"
+      }
+    },
+    "node_modules/abstract-leveldown": {
+      "version": "6.2.3",
+      "resolved": "https://registry.npmjs.org/abstract-leveldown/-/abstract-leveldown-6.2.3.tgz",
+      "integrity": "sha512-BsLm5vFMRUrrLeCcRc+G0t2qOaTzpoJQLOubq2XM72eNpjF5UdU5o/5NvlNhx95XHcAvcl8OMXr4mlg/fRgUXQ==",
+      "deprecated": "Superseded by abstract-level (https://github.com/Level/community#faq)",
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "buffer": "^5.5.0",
+        "immediate": "^3.2.3",
+        "level-concat-iterator": "~2.0.0",
+        "level-supports": "~1.0.0",
+        "xtend": "~4.0.0"
+      },
+      "engines": {
+        "node": ">=6"
+      }
+    },
+    "node_modules/async-limiter": {
+      "version": "1.0.1",
+      "resolved": "https://registry.npmjs.org/async-limiter/-/async-limiter-1.0.1.tgz",
+      "integrity": "sha512-csOlWGAcRFJaI6m+F2WKdnMKr4HhdhFVBk0H/QbJFMCr+uO2kwohwXQPxw/9OCxp05r5ghVBFSyioixx3gfkNQ==",
+      "license": "MIT"
+    },
+    "node_modules/base64-js": {
+      "version": "1.5.1",
+      "resolved": "https://registry.npmjs.org/base64-js/-/base64-js-1.5.1.tgz",
+      "integrity": "sha512-AKpaYlHn8t4SVbOHCy+b5+KKgvR4vrsD8vbvrbiQJps7fKDTkjkDry6ji0rUJjC0kzbNePLwzxq8iypo41qeWA==",
+      "funding": [
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/feross"
+        },
+        {
+          "type": "patreon",
+          "url": "https://www.patreon.com/feross"
+        },
+        {
+          "type": "consulting",
+          "url": "https://feross.org/support"
+        }
+      ],
+      "license": "MIT",
+      "optional": true
+    },
+    "node_modules/buffer": {
+      "version": "5.7.1",
+      "resolved": "https://registry.npmjs.org/buffer/-/buffer-5.7.1.tgz",
+      "integrity": "sha512-EHcyIPBQ4BSGlvjB16k5KgAJ27CIsHY/2JBmCRReo48y9rQ3MaUzWX3KVlBa4U7MyX02HdVj0K7C3WaB3ju7FQ==",
+      "funding": [
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/feross"
+        },
+        {
+          "type": "patreon",
+          "url": "https://www.patreon.com/feross"
+        },
+        {
+          "type": "consulting",
+          "url": "https://feross.org/support"
+        }
+      ],
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "base64-js": "^1.3.1",
+        "ieee754": "^1.1.13"
+      }
+    },
+    "node_modules/deferred-leveldown": {
+      "version": "5.3.0",
+      "resolved": "https://registry.npmjs.org/deferred-leveldown/-/deferred-leveldown-5.3.0.tgz",
+      "integrity": "sha512-a59VOT+oDy7vtAbLRCZwWgxu2BaCfd5Hk7wxJd48ei7I+nsg8Orlb9CLG0PMZienk9BSUKgeAqkO2+Lw+1+Ukw==",
+      "deprecated": "Superseded by abstract-level (https://github.com/Level/community#faq)",
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "abstract-leveldown": "~6.2.1",
+        "inherits": "^2.0.3"
+      },
+      "engines": {
+        "node": ">=6"
+      }
+    },
+    "node_modules/encoding-down": {
+      "version": "6.3.0",
+      "resolved": "https://registry.npmjs.org/encoding-down/-/encoding-down-6.3.0.tgz",
+      "integrity": "sha512-QKrV0iKR6MZVJV08QY0wp1e7vF6QbhnbQhb07bwpEyuz4uZiZgPlEGdkCROuFkUwdxlFaiPIhjyarH1ee/3vhw==",
+      "deprecated": "Superseded by abstract-level (https://github.com/Level/community#faq)",
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "abstract-leveldown": "^6.2.1",
+        "inherits": "^2.0.3",
+        "level-codec": "^9.0.0",
+        "level-errors": "^2.0.0"
+      },
+      "engines": {
+        "node": ">=6"
+      }
+    },
+    "node_modules/errno": {
+      "version": "0.1.8",
+      "resolved": "https://registry.npmjs.org/errno/-/errno-0.1.8.tgz",
+      "integrity": "sha512-dJ6oBr5SQ1VSd9qkk7ByRgb/1SH4JZjCHSW/mr63/QcXO9zLVxvJ6Oy13nio03rxpSnVDDjFor75SjVeZWPW/A==",
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "prr": "~1.0.1"
+      },
+      "bin": {
+        "errno": "cli.js"
+      }
+    },
+    "node_modules/ieee754": {
+      "version": "1.2.1",
+      "resolved": "https://registry.npmjs.org/ieee754/-/ieee754-1.2.1.tgz",
+      "integrity": "sha512-dcyqhDvX1C46lXZcVqCpK+FtMRQVdIMN6/Df5js2zouUsqG7I6sFxitIC+7KYK29KdXOLHdu9zL4sFnoVQnqaA==",
+      "funding": [
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/feross"
+        },
+        {
+          "type": "patreon",
+          "url": "https://www.patreon.com/feross"
+        },
+        {
+          "type": "consulting",
+          "url": "https://feross.org/support"
+        }
+      ],
+      "license": "BSD-3-Clause",
+      "optional": true
+    },
+    "node_modules/immediate": {
+      "version": "3.3.0",
+      "resolved": "https://registry.npmjs.org/immediate/-/immediate-3.3.0.tgz",
+      "integrity": "sha512-HR7EVodfFUdQCTIeySw+WDRFJlPcLOJbXfwwZ7Oom6tjsvZ3bOkCDJHehQC3nxJrv7+f9XecwazynjU8e4Vw3Q==",
+      "license": "MIT",
+      "optional": true
+    },
+    "node_modules/inherits": {
+      "version": "2.0.4",
+      "resolved": "https://registry.npmjs.org/inherits/-/inherits-2.0.4.tgz",
+      "integrity": "sha512-k/vGaX4/Yla3WzyMCvTQOXYeIHvqOKtnqBduzTHpzpQZzAskKMhZ2K+EnBiSM9zGSoIFeMpXKxa4dYeZIQqewQ==",
+      "license": "ISC",
+      "optional": true
+    },
+    "node_modules/isomorphic.js": {
+      "version": "0.2.5",
+      "resolved": "https://registry.npmjs.org/isomorphic.js/-/isomorphic.js-0.2.5.tgz",
+      "integrity": "sha512-PIeMbHqMt4DnUP3MA/Flc0HElYjMXArsw1qwJZcm9sqR8mq3l8NYizFMty0pWwE/tzIGH3EKK5+jes5mAr85yw==",
+      "license": "MIT",
+      "funding": {
+        "type": "GitHub Sponsors ❤",
+        "url": "https://github.com/sponsors/dmonad"
+      }
+    },
+    "node_modules/level": {
+      "version": "6.0.1",
+      "resolved": "https://registry.npmjs.org/level/-/level-6.0.1.tgz",
+      "integrity": "sha512-psRSqJZCsC/irNhfHzrVZbmPYXDcEYhA5TVNwr+V92jF44rbf86hqGp8fiT702FyiArScYIlPSBTDUASCVNSpw==",
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "level-js": "^5.0.0",
+        "level-packager": "^5.1.0",
+        "leveldown": "^5.4.0"
+      },
+      "engines": {
+        "node": ">=8.6.0"
+      },
+      "funding": {
+        "type": "opencollective",
+        "url": "https://opencollective.com/level"
+      }
+    },
+    "node_modules/level-codec": {
+      "version": "9.0.2",
+      "resolved": "https://registry.npmjs.org/level-codec/-/level-codec-9.0.2.tgz",
+      "integrity": "sha512-UyIwNb1lJBChJnGfjmO0OR+ezh2iVu1Kas3nvBS/BzGnx79dv6g7unpKIDNPMhfdTEGoc7mC8uAu51XEtX+FHQ==",
+      "deprecated": "Superseded by level-transcoder (https://github.com/Level/community#faq)",
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "buffer": "^5.6.0"
+      },
+      "engines": {
+        "node": ">=6"
+      }
+    },
+    "node_modules/level-concat-iterator": {
+      "version": "2.0.1",
+      "resolved": "https://registry.npmjs.org/level-concat-iterator/-/level-concat-iterator-2.0.1.tgz",
+      "integrity": "sha512-OTKKOqeav2QWcERMJR7IS9CUo1sHnke2C0gkSmcR7QuEtFNLLzHQAvnMw8ykvEcv0Qtkg0p7FOwP1v9e5Smdcw==",
+      "deprecated": "Superseded by abstract-level (https://github.com/Level/community#faq)",
+      "license": "MIT",
+      "optional": true,
+      "engines": {
+        "node": ">=6"
+      }
+    },
+    "node_modules/level-errors": {
+      "version": "2.0.1",
+      "resolved": "https://registry.npmjs.org/level-errors/-/level-errors-2.0.1.tgz",
+      "integrity": "sha512-UVprBJXite4gPS+3VznfgDSU8PTRuVX0NXwoWW50KLxd2yw4Y1t2JUR5In1itQnudZqRMT9DlAM3Q//9NCjCFw==",
+      "deprecated": "Superseded by abstract-level (https://github.com/Level/community#faq)",
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "errno": "~0.1.1"
+      },
+      "engines": {
+        "node": ">=6"
+      }
+    },
+    "node_modules/level-iterator-stream": {
+      "version": "4.0.2",
+      "resolved": "https://registry.npmjs.org/level-iterator-stream/-/level-iterator-stream-4.0.2.tgz",
+      "integrity": "sha512-ZSthfEqzGSOMWoUGhTXdX9jv26d32XJuHz/5YnuHZzH6wldfWMOVwI9TBtKcya4BKTyTt3XVA0A3cF3q5CY30Q==",
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "inherits": "^2.0.4",
+        "readable-stream": "^3.4.0",
+        "xtend": "^4.0.2"
+      },
+      "engines": {
+        "node": ">=6"
+      }
+    },
+    "node_modules/level-js": {
+      "version": "5.0.2",
+      "resolved": "https://registry.npmjs.org/level-js/-/level-js-5.0.2.tgz",
+      "integrity": "sha512-SnBIDo2pdO5VXh02ZmtAyPP6/+6YTJg2ibLtl9C34pWvmtMEmRTWpra+qO/hifkUtBTOtfx6S9vLDjBsBK4gRg==",
+      "deprecated": "Superseded by browser-level (https://github.com/Level/community#faq)",
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "abstract-leveldown": "~6.2.3",
+        "buffer": "^5.5.0",
+        "inherits": "^2.0.3",
+        "ltgt": "^2.1.2"
+      }
+    },
+    "node_modules/level-packager": {
+      "version": "5.1.1",
+      "resolved": "https://registry.npmjs.org/level-packager/-/level-packager-5.1.1.tgz",
+      "integrity": "sha512-HMwMaQPlTC1IlcwT3+swhqf/NUO+ZhXVz6TY1zZIIZlIR0YSn8GtAAWmIvKjNY16ZkEg/JcpAuQskxsXqC0yOQ==",
+      "deprecated": "Superseded by abstract-level (https://github.com/Level/community#faq)",
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "encoding-down": "^6.3.0",
+        "levelup": "^4.3.2"
+      },
+      "engines": {
+        "node": ">=6"
+      }
+    },
+    "node_modules/level-supports": {
+      "version": "1.0.1",
+      "resolved": "https://registry.npmjs.org/level-supports/-/level-supports-1.0.1.tgz",
+      "integrity": "sha512-rXM7GYnW8gsl1vedTJIbzOrRv85c/2uCMpiiCzO2fndd06U/kUXEEU9evYn4zFggBOg36IsBW8LzqIpETwwQzg==",
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "xtend": "^4.0.2"
+      },
+      "engines": {
+        "node": ">=6"
+      }
+    },
+    "node_modules/leveldown": {
+      "version": "5.6.0",
+      "resolved": "https://registry.npmjs.org/leveldown/-/leveldown-5.6.0.tgz",
+      "integrity": "sha512-iB8O/7Db9lPaITU1aA2txU/cBEXAt4vWwKQRrrWuS6XDgbP4QZGj9BL2aNbwb002atoQ/lIotJkfyzz+ygQnUQ==",
+      "deprecated": "Superseded by classic-level (https://github.com/Level/community#faq)",
+      "hasInstallScript": true,
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "abstract-leveldown": "~6.2.1",
+        "napi-macros": "~2.0.0",
+        "node-gyp-build": "~4.1.0"
+      },
+      "engines": {
+        "node": ">=8.6.0"
+      }
+    },
+    "node_modules/levelup": {
+      "version": "4.4.0",
+      "resolved": "https://registry.npmjs.org/levelup/-/levelup-4.4.0.tgz",
+      "integrity": "sha512-94++VFO3qN95cM/d6eBXvd894oJE0w3cInq9USsyQzzoJxmiYzPAocNcuGCPGGjoXqDVJcr3C1jzt1TSjyaiLQ==",
+      "deprecated": "Superseded by abstract-level (https://github.com/Level/community#faq)",
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "deferred-leveldown": "~5.3.0",
+        "level-errors": "~2.0.0",
+        "level-iterator-stream": "~4.0.0",
+        "level-supports": "~1.0.0",
+        "xtend": "~4.0.0"
+      },
+      "engines": {
+        "node": ">=6"
+      }
+    },
+    "node_modules/lib0": {
+      "version": "0.2.117",
+      "resolved": "https://registry.npmjs.org/lib0/-/lib0-0.2.117.tgz",
+      "integrity": "sha512-DeXj9X5xDCjgKLU/7RR+/HQEVzuuEUiwldwOGsHK/sfAfELGWEyTcf0x+uOvCvK3O2zPmZePXWL85vtia6GyZw==",
+      "license": "MIT",
+      "dependencies": {
+        "isomorphic.js": "^0.2.4"
+      },
+      "bin": {
+        "0ecdsa-generate-keypair": "bin/0ecdsa-generate-keypair.js",
+        "0gentesthtml": "bin/gentesthtml.js",
+        "0serve": "bin/0serve.js"
+      },
+      "engines": {
+        "node": ">=16"
+      },
+      "funding": {
+        "type": "GitHub Sponsors ❤",
+        "url": "https://github.com/sponsors/dmonad"
+      }
+    },
+    "node_modules/ltgt": {
+      "version": "2.2.1",
+      "resolved": "https://registry.npmjs.org/ltgt/-/ltgt-2.2.1.tgz",
+      "integrity": "sha512-AI2r85+4MquTw9ZYqabu4nMwy9Oftlfa/e/52t9IjtfG+mGBbTNdAoZ3RQKLHR6r0wQnwZnPIEh/Ya6XTWAKNA==",
+      "license": "MIT",
+      "optional": true
+    },
+    "node_modules/napi-macros": {
+      "version": "2.0.0",
+      "resolved": "https://registry.npmjs.org/napi-macros/-/napi-macros-2.0.0.tgz",
+      "integrity": "sha512-A0xLykHtARfueITVDernsAWdtIMbOJgKgcluwENp3AlsKN/PloyO10HtmoqnFAQAcxPkgZN7wdfPfEd0zNGxbg==",
+      "license": "MIT",
+      "optional": true
+    },
+    "node_modules/node-gyp-build": {
+      "version": "4.1.1",
+      "resolved": "https://registry.npmjs.org/node-gyp-build/-/node-gyp-build-4.1.1.tgz",
+      "integrity": "sha512-dSq1xmcPDKPZ2EED2S6zw/b9NKsqzXRE6dVr8TVQnI3FJOTteUMuqF3Qqs6LZg+mLGYJWqQzMbIjMtJqTv87nQ==",
+      "license": "MIT",
+      "optional": true,
+      "bin": {
+        "node-gyp-build": "bin.js",
+        "node-gyp-build-optional": "optional.js",
+        "node-gyp-build-test": "build-test.js"
+      }
+    },
+    "node_modules/prr": {
+      "version": "1.0.1",
+      "resolved": "https://registry.npmjs.org/prr/-/prr-1.0.1.tgz",
+      "integrity": "sha512-yPw4Sng1gWghHQWj0B3ZggWUm4qVbPwPFcRG8KyxiU7J2OHFSoEHKS+EZ3fv5l1t9CyCiop6l/ZYeWbrgoQejw==",
+      "license": "MIT",
+      "optional": true
+    },
+    "node_modules/readable-stream": {
+      "version": "3.6.2",
+      "resolved": "https://registry.npmjs.org/readable-stream/-/readable-stream-3.6.2.tgz",
+      "integrity": "sha512-9u/sniCrY3D5WdsERHzHE4G2YCXqoG5FTHUiCC4SIbr6XcLZBY05ya9EKjYek9O5xOAwjGq+1JdGBAS7Q9ScoA==",
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "inherits": "^2.0.3",
+        "string_decoder": "^1.1.1",
+        "util-deprecate": "^1.0.1"
+      },
+      "engines": {
+        "node": ">= 6"
+      }
+    },
+    "node_modules/safe-buffer": {
+      "version": "5.2.1",
+      "resolved": "https://registry.npmjs.org/safe-buffer/-/safe-buffer-5.2.1.tgz",
+      "integrity": "sha512-rp3So07KcdmmKbGvgaNxQSJr7bGVSVk5S9Eq1F+ppbRo70+YeaDxkw5Dd8NPN+GD6bjnYm2VuPuCXmpuYvmCXQ==",
+      "funding": [
+        {
+          "type": "github",
+          "url": "https://github.com/sponsors/feross"
+        },
+        {
+          "type": "patreon",
+          "url": "https://www.patreon.com/feross"
+        },
+        {
+          "type": "consulting",
+          "url": "https://feross.org/support"
+        }
+      ],
+      "license": "MIT",
+      "optional": true
+    },
+    "node_modules/string_decoder": {
+      "version": "1.3.0",
+      "resolved": "https://registry.npmjs.org/string_decoder/-/string_decoder-1.3.0.tgz",
+      "integrity": "sha512-hkRX8U1WjJFd8LsDJ2yQ/wWWxaopEsABU1XfkM8A+j0+85JAGppt16cr1Whg6KIbb4okU6Mql6BOj+uup/wKeA==",
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "safe-buffer": "~5.2.0"
+      }
+    },
+    "node_modules/util-deprecate": {
+      "version": "1.0.2",
+      "resolved": "https://registry.npmjs.org/util-deprecate/-/util-deprecate-1.0.2.tgz",
+      "integrity": "sha512-EPD5q1uXyFxJpCrLnCc1nHnq3gOa6DZBocAIiI2TaSCA7VCJ1UJDMagCzIkXNsUYfD1daK//LTEQ8xiIbrHtcw==",
+      "license": "MIT",
+      "optional": true
+    },
+    "node_modules/ws": {
+      "version": "6.2.6",
+      "resolved": "https://registry.npmjs.org/ws/-/ws-6.2.6.tgz",
+      "integrity": "sha512-XTrf1gv7kXoVf1hbC3PAyAiPgR8Wz1blcrYIjEsUmr08BLksT41R8KbjmS9408C2ERx7v1JDLD/BkpLEttjfKA==",
+      "license": "MIT",
+      "dependencies": {
+        "async-limiter": "~1.0.0"
+      }
+    },
+    "node_modules/xtend": {
+      "version": "4.0.2",
+      "resolved": "https://registry.npmjs.org/xtend/-/xtend-4.0.2.tgz",
+      "integrity": "sha512-LKYU1iAXJXUgAXn9URjiu+MWhyUXHsvfp7mcuYm9dSUKK0/CjtrUwFAxD82/mCWbtLsGjFIad0wIsod4zrTAEQ==",
+      "license": "MIT",
+      "optional": true,
+      "engines": {
+        "node": ">=0.4"
+      }
+    },
+    "node_modules/y-leveldb": {
+      "version": "0.1.2",
+      "resolved": "https://registry.npmjs.org/y-leveldb/-/y-leveldb-0.1.2.tgz",
+      "integrity": "sha512-6ulEn5AXfXJYi89rXPEg2mMHAyyw8+ZfeMMdOtBbV8FJpQ1NOrcgi6DTAcXof0dap84NjHPT2+9d0rb6cFsjEg==",
+      "license": "MIT",
+      "optional": true,
+      "dependencies": {
+        "level": "^6.0.1",
+        "lib0": "^0.2.31"
+      },
+      "funding": {
+        "type": "GitHub Sponsors ❤",
+        "url": "https://github.com/sponsors/dmonad"
+      },
+      "peerDependencies": {
+        "yjs": "^13.0.0"
+      }
+    },
+    "node_modules/y-protocols": {
+      "version": "1.0.6",
+      "resolved": "https://registry.npmjs.org/y-protocols/-/y-protocols-1.0.6.tgz",
+      "integrity": "sha512-vHRF2L6iT3rwj1jub/K5tYcTT/mEYDUppgNPXwp8fmLpui9f7Yeq3OEtTLVF012j39QnV+KEQpNqoN7CWU7Y9Q==",
+      "license": "MIT",
+      "dependencies": {
+        "lib0": "^0.2.85"
+      },
+      "engines": {
+        "node": ">=16.0.0",
+        "npm": ">=8.0.0"
+      },
+      "funding": {
+        "type": "GitHub Sponsors ❤",
+        "url": "https://github.com/sponsors/dmonad"
+      },
+      "peerDependencies": {
+        "yjs": "^13.0.0"
+      }
+    },
+    "node_modules/yjs": {
+      "version": "13.6.31",
+      "resolved": "https://registry.npmjs.org/yjs/-/yjs-13.6.31.tgz",
+      "integrity": "sha512-Eq+5BRfbeGyqGVrTJL3bEcr8gKkxPuyuoHmAwpk52fDb8kOVMrfVSTRPd6yiGgX5Fskb96qCRjzjbRjrL4YEnw==",
+      "license": "MIT",
+      "dependencies": {
+        "lib0": "^0.2.99"
+      },
+      "engines": {
+        "node": ">=16.0.0",
+        "npm": ">=8.0.0"
+      },
+      "funding": {
+        "type": "GitHub Sponsors ❤",
+        "url": "https://github.com/sponsors/dmonad"
+      }
+    }
+  }
+}
diff --git a/collaboration/package.json b/collaboration/package.json
new file mode 100644
index 00000000..347b53d2
--- /dev/null
+++ b/collaboration/package.json
@@ -0,0 +1,18 @@
+{
+  "name": "opencli-studio-collaboration",
+  "private": true,
+  "version": "1.0.0",
+  "type": "module",
+  "engines": {
+    "node": ">=22.0.0 <23"
+  },
+  "scripts": {
+    "start": "node server.mjs"
+  },
+  "dependencies": {
+    "@y/websocket-server": "0.1.1",
+    "ws": "6.2.6",
+    "y-protocols": "1.0.6",
+    "yjs": "13.6.31"
+  }
+}
diff --git a/collaboration/server.mjs b/collaboration/server.mjs
new file mode 100644
index 00000000..bd93c5f9
--- /dev/null
+++ b/collaboration/server.mjs
@@ -0,0 +1,182 @@
+import { createServer } from "node:http"
+import { docs, setupWSConnection } from "@y/websocket-server/utils"
+import WebSocket from "ws"
+const { Server: WebSocketServer } = WebSocket
+
+const port = parsePort(process.env.PORT ?? "1234")
+const apiUrl = requiredUrl(process.env.COLLABORATION_API_URL ?? "http://api:8000")
+const apiToken = required("API_AUTH_TOKEN")
+const debounceMs = parseBoundedInteger(process.env.SNAPSHOT_DEBOUNCE_MS ?? "250", 100, 10_000)
+const maxSnapshotBytes = parseBoundedInteger(
+  process.env.MAX_SNAPSHOT_BYTES ?? "1048576",
+  1_024,
+  10 * 1024 * 1024,
+)
+const requestTimeoutMs = parseBoundedInteger(process.env.API_REQUEST_TIMEOUT_MS ?? "5000", 100, 30_000)
+
+const websocketServer = new WebSocketServer({ noServer: true, maxPayload: maxSnapshotBytes })
+const roomHooks = new WeakSet()
+const snapshotTimers = new WeakMap()
+const snapshotInFlight = new WeakSet()
+const snapshotPending = new WeakSet()
+
+const server = createServer((request, response) => {
+  const url = new URL(request.url ?? "/", "http://localhost")
+  if (request.method === "GET" && url.pathname === "/health") {
+    response.writeHead(200, { "content-type": "application/json", "cache-control": "no-store" })
+    response.end('{"status":"ok"}')
+    return
+  }
+  response.writeHead(404, { "content-type": "application/json" })
+  response.end('{"error":"not found"}')
+})
+
+server.on("upgrade", async (request, socket, head) => {
+  try {
+    const url = new URL(request.url ?? "/", "http://localhost")
+    const room = roomFromPath(url.pathname)
+    const token = url.searchParams.get("token")
+    if (!room || !token) {
+      rejectUpgrade(socket, 401, "Unauthorized")
+      return
+    }
+
+    const authorizedRoom = await authorizeRoom(room, token)
+    if (authorizedRoom !== room) {
+      rejectUpgrade(socket, 403, "Forbidden")
+      return
+    }
+
+    websocketServer.handleUpgrade(request, socket, head, (connection) => {
+      connection.on("error", () => connection.terminate())
+      setupWSConnection(connection, request, { docName: room })
+      installSnapshotHook(docs.get(room), room)
+    })
+  } catch (error) {
+    logError("websocket upgrade rejected", error)
+    rejectUpgrade(socket, 401, "Unauthorized")
+  }
+})
+
+function installSnapshotHook(doc, room) {
+  if (!doc || roomHooks.has(doc)) return
+  roomHooks.add(doc)
+  doc.on("update", () => scheduleSnapshot(doc, room))
+}
+
+function scheduleSnapshot(doc, room) {
+  clearTimeout(snapshotTimers.get(doc))
+  snapshotTimers.set(
+    doc,
+    setTimeout(() => void sendSnapshot(doc, room), debounceMs),
+  )
+}
+
+async function sendSnapshot(doc, room) {
+  snapshotTimers.delete(doc)
+  if (snapshotInFlight.has(doc)) {
+    snapshotPending.add(doc)
+    return
+  }
+  snapshotInFlight.add(doc)
+  try {
+    const payload = JSON.stringify({
+      room,
+      data: {
+        nodes: { type: "Map", content: doc.getMap("nodes").toJSON() },
+        edges: { type: "Map", content: doc.getMap("edges").toJSON() },
+      },
+    })
+    if (Buffer.byteLength(payload) > maxSnapshotBytes) {
+      console.error("collaboration snapshot exceeds configured byte limit", { room })
+      return
+    }
+    const response = await fetch(`${apiUrl}/api/v1/internal/collaboration/snapshot`, {
+      method: "POST",
+      headers: {
+        authorization: `Bearer ${apiToken}`,
+        "content-type": "application/json",
+      },
+      body: payload,
+      signal: AbortSignal.timeout(requestTimeoutMs),
+    })
+    if (!response.ok) {
+      console.error("collaboration snapshot rejected", { room, status: response.status })
+    }
+  } catch (error) {
+    logError("collaboration snapshot failed", error)
+  } finally {
+    snapshotInFlight.delete(doc)
+    if (snapshotPending.delete(doc)) scheduleSnapshot(doc, room)
+  }
+}
+
+async function authorizeRoom(room, token) {
+  const response = await fetch(`${apiUrl}/api/v1/internal/collaboration/authorize`, {
+    method: "POST",
+    headers: {
+      authorization: `Bearer ${token}`,
+      "content-type": "application/json",
+    },
+    body: JSON.stringify({ room }),
+    signal: AbortSignal.timeout(requestTimeoutMs),
+  })
+  if (!response.ok) throw new Error(`authorization status ${response.status}`)
+  const body = await response.json()
+  const authorizedRoom = body?.data?.room
+  if (typeof authorizedRoom !== "string") throw new Error("authorization response is invalid")
+  return authorizedRoom
+}
+
+function roomFromPath(pathname) {
+  if (!pathname.startsWith("/") || pathname.length > 513) return null
+  try {
+    const room = decodeURIComponent(pathname.slice(1))
+    return room.includes("/") || !room ? null : room
+  } catch {
+    return null
+  }
+}
+
+function rejectUpgrade(socket, statusCode, reason) {
+  if (!socket.writable) return
+  socket.write(`HTTP/1.1 ${statusCode} ${reason}\r\nConnection: close\r\n\r\n`)
+  socket.destroy()
+}
+
+function required(name) {
+  const value = process.env[name]?.trim()
+  if (!value) throw new Error(`${name} must be configured`)
+  return value
+}
+
+function requiredUrl(value) {
+  const url = new URL(value)
+  if (url.protocol !== "http:" && url.protocol !== "https:") {
+    throw new Error("COLLABORATION_API_URL must use http or https")
+  }
+  return url.toString().replace(/\/$/, "")
+}
+
+function parsePort(value) {
+  return parseBoundedInteger(value, 1, 65_535)
+}
+
+function parseBoundedInteger(value, minimum, maximum) {
+  const parsed = Number.parseInt(value, 10)
+  if (!Number.isSafeInteger(parsed) || String(parsed) !== value || parsed < minimum || parsed > maximum) {
+    throw new Error(`Expected an integer from ${minimum} to ${maximum}`)
+  }
+  return parsed
+}
+
+function logError(message, error) {
+  console.error(message, {
+    message: error instanceof Error ? error.message : String(error),
+    stack: error instanceof Error ? error.stack : undefined,
+  })
+}
+
+server.listen(port, "0.0.0.0", () => {
+  console.log(`Studio collaboration service listening on :${port}`)
+})
diff --git a/compat/paw_runtime/Dockerfile b/compat/paw_runtime/Dockerfile
new file mode 100644
index 00000000..b62bd8cd
--- /dev/null
+++ b/compat/paw_runtime/Dockerfile
@@ -0,0 +1,46 @@
+FROM python:3.13-slim-bookworm
+
+ARG PAW_LLAMA_WHEEL_SHA256=bb121ab711305a6e2e44a4fce59740a3de489ae0f320d4240d890f8063bee70a
+ARG PAW_SDK_WHEEL_SHA256=c1d8bc54bc4225ff882722cc4986e7b29912225bcd5ebe215297e5dcf8cfa68c
+
+ENV HOME=/tmp \
+    PAW_OFFLINE=1 \
+    PAW_CACHE_DIR=/var/cache/paw \
+    PIP_DISABLE_PIP_VERSION_CHECK=1 \
+    PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1
+
+# The vendor wheel is deliberately used instead of a source distribution, so
+# the sidecar never compiles llama.cpp. Its 0.3.19 wheel omits RECORD, which
+# pip refuses to install; extract that signed wheel directly, then install the
+RUN mkdir -p /tmp/paw-wheels \
+    && pip download --no-deps --only-binary=:all: \
+        --extra-index-url https://pypi.programasweights.com/simple/ \
+        --dest /tmp/paw-wheels "llama-cpp-python==0.3.19" \
+    && echo "${PAW_LLAMA_WHEEL_SHA256}  /tmp/paw-wheels/llama_cpp_python-0.3.19-cp313-cp313-linux_x86_64.whl" | sha256sum -c - \
+    && python -m zipfile -e /tmp/paw-wheels/llama_cpp_python-0.3.19-cp313-cp313-linux_x86_64.whl /usr/local/lib/python3.13/site-packages \
+    && pip install --no-cache-dir \
+        "fastapi==0.141.1" \
+        "diskcache==5.6.3" \
+        "jinja2==3.1.6" \
+        "httpx==0.28.1" \
+        "numpy==2.5.2" \
+        "uvicorn==0.52.4" \
+    && pip download --no-deps --only-binary=:all: \
+        --extra-index-url https://pypi.programasweights.com/simple/ \
+        --dest /tmp/paw-wheels "programasweights==0.4.4" \
+    && echo "${PAW_SDK_WHEEL_SHA256}  /tmp/paw-wheels/programasweights-0.4.4-py3-none-any.whl" | sha256sum -c - \
+    && pip install --no-cache-dir --no-deps \
+        /tmp/paw-wheels/programasweights-0.4.4-py3-none-any.whl \
+    && rm -rf /tmp/paw-wheels
+
+RUN addgroup --system paw \
+    && adduser --system --ingroup paw --home /nonexistent paw
+
+WORKDIR /app
+COPY app.py engine.py ./
+
+USER paw
+EXPOSE 8097
+
+CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "8097", "--no-access-log"]
diff --git a/compat/paw_runtime/app.py b/compat/paw_runtime/app.py
new file mode 100644
index 00000000..da88242c
--- /dev/null
+++ b/compat/paw_runtime/app.py
@@ -0,0 +1,173 @@
+from __future__ import annotations
+
+from typing import Any
+
+from fastapi import FastAPI, Request
+from fastapi.exceptions import RequestValidationError
+from fastapi.responses import JSONResponse
+from pydantic import BaseModel, ConfigDict, Field
+from starlette.types import ASGIApp, Message, Receive, Scope, Send
+
+try:
+    from .engine import (
+        CONTRACT_VERSION,
+        MAX_INPUT_CHARS,
+        MAX_TOKENS,
+        PawRuntime,
+        PawRuntimeError,
+    )
+except ImportError:  # uvicorn runs app.py as a top-level module in the image.
+    from engine import (
+        CONTRACT_VERSION,
+        MAX_INPUT_CHARS,
+        MAX_TOKENS,
+        PawRuntime,
+        PawRuntimeError,
+    )
+
+MAX_REQUEST_BYTES = MAX_INPUT_CHARS * 12 + 1_024
+
+
+class EnrichRequest(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    programId: str = Field(min_length=1, max_length=128)
+    input: str = Field(min_length=1, max_length=MAX_INPUT_CHARS)
+    maxTokens: int = Field(default=128, ge=1, le=MAX_TOKENS)
+
+
+class RequestByteLimitMiddleware:
+    def __init__(self, app: ASGIApp, *, max_bytes: int) -> None:
+        self.app = app
+        self.max_bytes = max_bytes
+
+    async def __call__(self, scope: Scope, receive: Receive, send: Send) -> None:
+        if scope["type"] != "http":
+            await self.app(scope, receive, send)
+            return
+
+        raw_length = dict(scope.get("headers", [])).get(b"content-length")
+        if raw_length is not None:
+            try:
+                if int(raw_length) > self.max_bytes:
+                    await _request_too_large(self.max_bytes)(scope, receive, send)
+                    return
+            except ValueError:
+                pass
+
+        messages: list[Message] = []
+        received_bytes = 0
+        while True:
+            message = await receive()
+            messages.append(message)
+            if message["type"] == "http.disconnect":
+                break
+            if message["type"] != "http.request":
+                continue
+            received_bytes += len(message.get("body", b""))
+            if received_bytes > self.max_bytes:
+                await _request_too_large(self.max_bytes)(scope, receive, send)
+                return
+            if not message.get("more_body", False):
+                break
+
+        index = 0
+
+        async def replay() -> Message:
+            nonlocal index
+            if index < len(messages):
+                message = messages[index]
+                index += 1
+                return message
+            return await receive()
+
+        await self.app(scope, replay, send)
+
+
+def _request_too_large(max_bytes: int) -> JSONResponse:
+    return JSONResponse(
+        status_code=413,
+        content={
+            "error": {
+                "code": "request.too_large",
+                "message": "The request body exceeds the configured request limit.",
+                "details": {"maxBytes": max_bytes},
+            }
+        },
+    )
+
+
+def create_app(
+    runtime: PawRuntime | None = None,
+    *,
+    max_request_bytes: int = MAX_REQUEST_BYTES,
+) -> FastAPI:
+    paw_runtime = runtime or PawRuntime()
+    application = FastAPI(
+        title="OpenCLI PAW Runtime",
+        version="0.4.4",
+        docs_url=None,
+        redoc_url=None,
+    )
+    application.add_middleware(RequestByteLimitMiddleware, max_bytes=max_request_bytes)
+
+    @application.exception_handler(PawRuntimeError)
+    async def handle_runtime_error(_request: Request, error: PawRuntimeError) -> JSONResponse:
+        return JSONResponse(
+            status_code=error.status_code,
+            content={"error": {"code": error.code, "message": error.message, "details": {}}},
+        )
+
+    @application.exception_handler(RequestValidationError)
+    async def handle_validation_error(
+        _request: Request,
+        error: RequestValidationError,
+    ) -> JSONResponse:
+        return JSONResponse(
+            status_code=422,
+            content={
+                "error": {
+                    "code": "request.invalid",
+                    "message": "The request does not match the PAW runtime contract.",
+                    "details": {
+                        "issues": [
+                            {
+                                "path": "/".join(str(part) for part in issue["loc"]),
+                                "type": issue["type"],
+                            }
+                            for issue in error.errors()
+                        ]
+                    },
+                }
+            },
+        )
+
+    @application.get("/health")
+    async def health() -> JSONResponse:
+        ready = paw_runtime.is_ready()
+        return JSONResponse(
+            status_code=200 if ready else 503,
+            content={
+                "status": "ok" if ready else "not_ready",
+                "ready": ready,
+                "offline": True,
+                "contractVersion": CONTRACT_VERSION,
+                "program": paw_runtime.identity(),
+            },
+        )
+
+    @application.post("/v1/enrich")
+    async def enrich(request: EnrichRequest) -> dict[str, Any]:
+        enrichment = await paw_runtime.enrich_async(
+            request.programId, request.input, request.maxTokens
+        )
+        return {
+            "contractVersion": CONTRACT_VERSION,
+            "programId": paw_runtime.program_id,
+            "enrichment": enrichment,
+        }
+
+    return application
+
+
+app = create_app()
diff --git a/compat/paw_runtime/engine.py b/compat/paw_runtime/engine.py
new file mode 100644
index 00000000..4f176c12
--- /dev/null
+++ b/compat/paw_runtime/engine.py
@@ -0,0 +1,186 @@
+from __future__ import annotations
+
+import asyncio
+import json
+import os
+import re
+import threading
+from concurrent.futures import ThreadPoolExecutor
+from typing import Any
+
+CONTRACT_VERSION = "opencli.paw.runtime.v1"
+PAW_VERSION = "0.4.4"
+DEFAULT_PROGRAM_ID = ""
+MAX_INPUT_CHARS = 8_192
+MAX_TOKENS = 512
+MAX_OUTPUT_BYTES = 65_536
+_PROGRAM_ID_RE = re.compile(r"^[a-f0-9]{16,64}$")
+
+
+class PawRuntimeError(Exception):
+    def __init__(self, code: str, message: str, *, status_code: int = 422) -> None:
+        super().__init__(message)
+        self.code = code
+        self.message = message
+        self.status_code = status_code
+
+
+def _bounded_env_int(name: str, default: int, minimum: int, maximum: int) -> int:
+    try:
+        value = int(os.environ.get(name, str(default)))
+    except ValueError as exc:
+        raise ValueError(f"{name} must be an integer") from exc
+    if not minimum <= value <= maximum:
+        raise ValueError(f"{name} must be between {minimum} and {maximum}")
+    return value
+
+
+def _reject_nonfinite_json(value: str) -> None:
+    raise ValueError(f"Non-finite JSON number: {value}")
+
+
+def _load_sdk() -> Any | None:
+    try:
+        import programasweights
+    except ImportError:
+        return None
+    return programasweights
+
+
+class PawRuntime:
+    """Serial, bounded adapter around the pinned offline PAW SDK."""
+
+    def __init__(
+        self,
+        *,
+        program_id: str | None = None,
+        max_output_bytes: int = MAX_OUTPUT_BYTES,
+        n_ctx: int | None = None,
+        paw_sdk: Any | None = None,
+    ) -> None:
+        self.program_id = program_id or os.environ.get("PAW_PROGRAM_ID", DEFAULT_PROGRAM_ID)
+        self.max_output_bytes = max_output_bytes
+        self.n_ctx = n_ctx or _bounded_env_int("PAW_N_CTX", 2048, 128, 2048)
+        self.execution_timeout_seconds = _bounded_env_int(
+            "PAW_EXECUTION_TIMEOUT_SECONDS", 30, 1, 120
+        )
+        self._paw = paw_sdk if paw_sdk is not None else _load_sdk()
+        self._function: Any | None = None
+        self._inference_lock = threading.Lock()
+        self._dispatch_lock: asyncio.Lock | None = None
+        self._executor = ThreadPoolExecutor(max_workers=1, thread_name_prefix="paw-inference")
+        self._poisoned = False
+
+    def identity(self) -> dict[str, str]:
+        return {
+            "programId": self.program_id,
+            "pawVersion": PAW_VERSION,
+            "contractVersion": CONTRACT_VERSION,
+        }
+
+    def is_ready(self) -> bool:
+        if not _PROGRAM_ID_RE.fullmatch(self.program_id):
+            return False
+        if self._poisoned or self._paw is None:
+            return False
+        try:
+            return bool(self._paw.is_offline_ready(self.program_id))
+        except Exception:
+            return False
+
+    async def enrich_async(
+        self, program_id: str, input_text: str, max_tokens: int
+    ) -> dict[str, Any]:
+        if self._poisoned:
+            raise PawRuntimeError(
+                "runtime.poisoned",
+                "PAW runtime is unavailable.",
+                status_code=503,
+            )
+        if self._dispatch_lock is None:
+            self._dispatch_lock = asyncio.Lock()
+        async with self._dispatch_lock:
+            if self._poisoned:
+                raise PawRuntimeError(
+                    "runtime.poisoned",
+                    "PAW runtime is unavailable.",
+                    status_code=503,
+                )
+            future = asyncio.get_running_loop().run_in_executor(
+                self._executor, self.enrich, program_id, input_text, max_tokens
+            )
+            try:
+                return await asyncio.wait_for(
+                    asyncio.shield(future),
+                    self.execution_timeout_seconds,
+                )
+            except TimeoutError as exc:
+                # SDK inference cannot be safely cancelled. Poison this one-worker
+                # runtime so queued requests fail fast instead of accumulating threads.
+                self._poisoned = True
+                self._executor.shutdown(wait=False, cancel_futures=True)
+                raise PawRuntimeError(
+                    "runtime.timeout",
+                    "PAW inference timed out; runtime is unavailable.",
+                    status_code=503,
+                ) from exc
+
+    def enrich(self, program_id: str, input_text: str, max_tokens: int) -> dict[str, Any]:
+        if not _PROGRAM_ID_RE.fullmatch(self.program_id):
+            raise PawRuntimeError(
+                "runtime.program_not_configured",
+                "PAW_PROGRAM_ID must be an immutable hexadecimal program ID.",
+                status_code=503,
+            )
+        if program_id != self.program_id:
+            raise PawRuntimeError(
+                "request.program_mismatch",
+                "The requested PAW program is not available.",
+            )
+        if not input_text:
+            raise PawRuntimeError("request.input_empty", "PAW input must not be empty.")
+        if len(input_text) > MAX_INPUT_CHARS:
+            raise PawRuntimeError(
+                "request.input_too_large",
+                "The input exceeds the PAW input limit.",
+            )
+        if not 1 <= max_tokens <= MAX_TOKENS:
+            raise PawRuntimeError(
+                "request.max_tokens_invalid",
+                "maxTokens is outside the PAW limit.",
+            )
+        if self._paw is None or not self.is_ready():
+            raise PawRuntimeError(
+                "runtime.not_ready",
+                "The fixed PAW program is unavailable in offline mode.",
+                status_code=503,
+            )
+
+        try:
+            with self._inference_lock:
+                if self._function is None:
+                    self._function = self._paw.function(
+                        self.program_id,
+                        offline=True,
+                        n_ctx=self.n_ctx,
+                        n_gpu_layers=0,
+                        verbose=False,
+                    )
+                output = self._function(input_text, max_tokens=max_tokens, temperature=0.0)
+        except Exception as exc:
+            raise PawRuntimeError(
+                "runtime.inference_failed", "PAW inference failed.", status_code=503
+            ) from exc
+
+        if not isinstance(output, str):
+            raise PawRuntimeError("response.invalid_json", "PAW returned invalid JSON.")
+        encoded = output.encode("utf-8")
+        if len(encoded) > self.max_output_bytes:
+            raise PawRuntimeError("response.too_large", "PAW output exceeds the output limit.")
+        try:
+            enrichment = json.loads(output, parse_constant=_reject_nonfinite_json)
+        except (json.JSONDecodeError, RecursionError, ValueError) as exc:
+            raise PawRuntimeError("response.invalid_json", "PAW returned invalid JSON.") from exc
+        if not isinstance(enrichment, dict):
+            raise PawRuntimeError("response.invalid_enrichment", "PAW must return a JSON object.")
+        return enrichment
diff --git a/compat/paw_runtime/pyproject.toml b/compat/paw_runtime/pyproject.toml
new file mode 100644
index 00000000..a5065a9e
--- /dev/null
+++ b/compat/paw_runtime/pyproject.toml
@@ -0,0 +1,33 @@
+[project]
+name = "opencli-paw-runtime"
+version = "0.4.4"
+description = "Offline fixed-program PAW enrichment sidecar for OpenCLI Admin"
+requires-python = ">=3.13,<3.14"
+dependencies = [
+    "fastapi==0.141.1",
+    "httpx==0.28.1",
+    "llama-cpp-python==0.3.19",
+    "programasweights==0.4.4",
+    "uvicorn==0.52.4",
+]
+
+[dependency-groups]
+dev = [
+    "httpx>=0.28,<1",
+    "pytest>=8.3,<10",
+]
+
+
+[tool.uv]
+package = false
+
+[tool.pytest.ini_options]
+pythonpath = ["."]
+testpaths = ["tests"]
+
+[tool.ruff]
+line-length = 100
+target-version = "py313"
+
+[tool.ruff.lint]
+select = ["E", "F", "I", "UP", "B", "SIM"]
diff --git a/compat/paw_runtime/tests/test_contract.py b/compat/paw_runtime/tests/test_contract.py
new file mode 100644
index 00000000..84d81839
--- /dev/null
+++ b/compat/paw_runtime/tests/test_contract.py
@@ -0,0 +1,214 @@
+import json
+import threading
+from concurrent.futures import ThreadPoolExecutor
+
+from compat.paw_runtime.app import create_app
+from compat.paw_runtime.engine import CONTRACT_VERSION, MAX_INPUT_CHARS, PawRuntime
+from fastapi.testclient import TestClient
+
+
+class FakePaw:
+    def __init__(self, *, ready=True, output='{"summary":"classified"}'):
+        self.ready = ready
+        self.output = output
+        self.function_calls = []
+        self.inference_calls = []
+
+    def is_offline_ready(self, program_id):
+        return self.ready
+
+    def function(self, program_id, **kwargs):
+        self.function_calls.append((program_id, kwargs))
+
+        def infer(input_text, **inference_kwargs):
+            self.inference_calls.append((input_text, inference_kwargs))
+            return self.output
+
+        return infer
+
+
+class BlockingPaw(FakePaw):
+    def __init__(self):
+        super().__init__()
+        self.entered = threading.Event()
+        self.second_entered = threading.Event()
+        self.release = threading.Event()
+        self.active = 0
+        self.active_lock = threading.Lock()
+
+    def function(self, program_id, **kwargs):
+        self.function_calls.append((program_id, kwargs))
+
+        def infer(input_text, **inference_kwargs):
+            with self.active_lock:
+                self.active += 1
+                if self.active == 1:
+                    self.entered.set()
+                else:
+                    self.second_entered.set()
+            self.release.wait(timeout=2)
+            with self.active_lock:
+                self.active -= 1
+            return json.dumps({"summary": input_text})
+
+        return infer
+
+
+def _runtime(paw=None, **kwargs):
+    return PawRuntime(paw_sdk=paw or FakePaw(), program_id="0123456789abcdef", **kwargs)
+
+
+def test_health_exposes_fixed_identity_and_offline_readiness():
+    client = TestClient(create_app(_runtime()))
+
+    response = client.get("/health")
+
+    assert response.status_code == 200
+    assert response.json() == {
+        "status": "ok",
+        "ready": True,
+        "offline": True,
+        "contractVersion": CONTRACT_VERSION,
+        "program": {
+            "programId": "0123456789abcdef",
+            "pawVersion": "0.4.4",
+            "contractVersion": CONTRACT_VERSION,
+        },
+    }
+
+def test_health_rejects_missing_program_identity():
+    client = TestClient(create_app(PawRuntime(paw_sdk=FakePaw(), program_id="")))
+
+    response = client.get("/health")
+
+    assert response.status_code == 503
+    assert response.json()["status"] == "not_ready"
+    assert response.json()["program"]["programId"] == ""
+
+
+
+def test_enrich_calls_official_sdk_offline_with_fixed_program():
+    paw = FakePaw()
+    client = TestClient(create_app(_runtime(paw)))
+
+    response = client.post(
+        "/v1/enrich",
+        json={"programId": "0123456789abcdef", "input": "short prompt", "maxTokens": 32},
+    )
+
+    assert response.status_code == 200
+    assert response.json()["enrichment"] == {"summary": "classified"}
+    assert paw.function_calls == [
+        (
+            "0123456789abcdef",
+            {"offline": True, "n_ctx": 2048, "n_gpu_layers": 0, "verbose": False},
+        )
+    ]
+    assert paw.inference_calls == [("short prompt", {"max_tokens": 32, "temperature": 0.0})]
+
+
+def test_runtime_reuses_successfully_loaded_offline_function():
+    paw = FakePaw()
+    runtime = _runtime(paw)
+
+    assert runtime.enrich(runtime.program_id, "first", 8) == {"summary": "classified"}
+    assert runtime.enrich(runtime.program_id, "second", 8) == {"summary": "classified"}
+
+    assert len(paw.function_calls) == 1
+    assert [call[0] for call in paw.inference_calls] == ["first", "second"]
+
+
+def test_runtime_serializes_concurrent_offline_inference():
+    paw = BlockingPaw()
+    runtime = _runtime(paw)
+
+    with ThreadPoolExecutor(max_workers=2) as executor:
+        first = executor.submit(runtime.enrich, runtime.program_id, "first", 8)
+        assert paw.entered.wait(timeout=1)
+        second = executor.submit(runtime.enrich, runtime.program_id, "second", 8)
+        assert not paw.second_entered.wait(timeout=0.1)
+        paw.release.set()
+        assert first.result() == {"summary": "first"}
+        assert second.result() == {"summary": "second"}
+
+    assert not paw.second_entered.is_set()
+
+
+def test_missing_cache_is_not_ready_and_never_loads_or_prepares_program():
+    paw = FakePaw(ready=False)
+    client = TestClient(create_app(_runtime(paw)))
+
+    health = client.get("/health")
+    assert health.status_code == 503
+    assert health.json()["status"] == "not_ready"
+    response = client.post(
+        "/v1/enrich",
+        json={"programId": "0123456789abcdef", "input": "short", "maxTokens": 8},
+    )
+    assert response.status_code == 503
+    assert response.json()["error"]["code"] == "runtime.not_ready"
+    assert paw.function_calls == []
+
+
+def test_program_mismatch_and_empty_input_are_rejected_without_sdk_call():
+    paw = FakePaw()
+    client = TestClient(create_app(_runtime(paw)))
+
+    mismatch = client.post(
+        "/v1/enrich",
+        json={"programId": "other.program", "input": "short", "maxTokens": 8},
+    )
+    empty = client.post(
+        "/v1/enrich",
+        json={"programId": "0123456789abcdef", "input": "", "maxTokens": 8},
+    )
+
+    assert mismatch.status_code == 422
+    assert mismatch.json()["error"]["code"] == "request.program_mismatch"
+    assert empty.status_code == 422
+    assert empty.json()["error"]["code"] == "request.invalid"
+    assert paw.function_calls == []
+
+
+def test_ascii_escaped_non_bmp_input_at_character_limit_reaches_sdk():
+    paw = FakePaw()
+    client = TestClient(create_app(_runtime(paw)))
+    input_text = "😀" * MAX_INPUT_CHARS
+
+    response = client.post(
+        "/v1/enrich",
+        content=json.dumps(
+            {"programId": "0123456789abcdef", "input": input_text, "maxTokens": 8}
+        ),
+        headers={"content-type": "application/json"},
+    )
+
+    assert response.status_code == 200
+    assert paw.inference_calls == [(input_text, {"max_tokens": 8, "temperature": 0.0})]
+
+
+def test_sdk_output_must_be_bounded_finite_json_object():
+    for output, code in [
+        ('{"score":NaN}', "response.invalid_json"),
+        ("[]", "response.invalid_enrichment"),
+        (json.dumps({"large": "x" * 32}), "response.too_large"),
+    ]:
+        runtime = _runtime(FakePaw(output=output), max_output_bytes=16)
+        try:
+            runtime.enrich(runtime.program_id, "short", 8)
+        except Exception as error:
+            assert error.code == code
+        else:  # pragma: no cover - each malformed value must fail closed
+            raise AssertionError("invalid SDK output must be rejected")
+
+
+def test_request_byte_limit_is_413():
+    client = TestClient(create_app(_runtime(), max_request_bytes=64))
+
+    response = client.post(
+        "/v1/enrich",
+        json={"programId": "0123456789abcdef", "input": "x" * 128, "maxTokens": 8},
+    )
+
+    assert response.status_code == 413
+    assert response.json()["error"]["code"] == "request.too_large"
diff --git a/docker-compose.build.yml b/docker-compose.build.yml
index be7c0836..fe82521d 100644
--- a/docker-compose.build.yml
+++ b/docker-compose.build.yml
@@ -33,10 +33,18 @@ services:
       dockerfile: Dockerfile
       args:
         BACKEND_URL: http://api:8000
+        NEXT_PUBLIC_API_AUTH_TOKEN: ${API_AUTH_TOKEN:?Set API_AUTH_TOKEN in .env or run scripts/install.sh}
+        NEXT_PUBLIC_ENABLE_VERCEL_ANALYTICS: ${NEXT_PUBLIC_ENABLE_VERCEL_ANALYTICS:-}
+        NEXT_PUBLIC_YJS_URL: ${NEXT_PUBLIC_YJS_URL:-ws://127.0.0.1:1234}
 
   api:
     <<: *backend-build
 
+  collaboration:
+    build:
+      context: ./collaboration
+      dockerfile: Dockerfile
+
   worker:
     <<: *backend-build
 
diff --git a/docker-compose.yml b/docker-compose.yml
index 84ef3f63..fc24d86a 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -141,6 +141,42 @@ services:
       start_period: 60s
     restart: unless-stopped
 
+  # ── PAW local enrichment runtime ─────────────────────────────────────────
+  # The fixed PAW 0.4.4 program is opt-in and never receives egress access.
+  paw-runtime:
+    profiles: ["paw"]
+    build:
+      context: ./compat/paw_runtime
+      dockerfile: Dockerfile
+    environment:
+      PAW_OFFLINE: "1"
+      PAW_PROGRAM_ID: ${PAW_PROGRAM_ID:-}
+      PAW_EXECUTION_TIMEOUT_SECONDS: ${PAW_EXECUTION_TIMEOUT_SECONDS:-30}
+      PAW_N_CTX: ${PAW_N_CTX:-2048}
+    volumes:
+      - ${PAW_CACHE_DIR:-./.paw-cache}:/var/cache/paw:ro
+    networks:
+      - paw_runtime
+    read_only: true
+    tmpfs:
+      - /tmp:size=${PAW_RUNTIME_TMPFS_SIZE:-64m},mode=1777
+    cap_drop:
+      - ALL
+    security_opt:
+      - no-new-privileges:true
+    pids_limit: 128
+    healthcheck:
+      test:
+        - CMD
+        - python
+        - -c
+        - "import urllib.request; urllib.request.urlopen('http://localhost:8097/health', timeout=3)"
+      interval: 10s
+      timeout: 5s
+      retries: 6
+      start_period: 20s
+    restart: unless-stopped
+
   # ── Backend API ───────────────────────────────────────────────────────────
   api:
     <<: *backend-common
@@ -155,7 +191,7 @@ services:
       DEBUG: ${DEBUG:-false}
       SECRET_KEY: ${SECRET_KEY:-change-me-in-production}
       API_AUTH_TOKEN: ${API_AUTH_TOKEN:?Set API_AUTH_TOKEN in .env or run scripts/install.sh}
-      BOOTSTRAP_ADMIN_TOKEN: ${BOOTSTRAP_ADMIN_TOKEN:?Set BOOTSTRAP_ADMIN_TOKEN in .env or run scripts/install.sh}
+      BOOTSTRAP_ADMIN_TOKEN: ${BOOTSTRAP_ADMIN_TOKEN:-}
       OPENCLI_MCP_ALLOWED_HOSTS: ${OPENCLI_MCP_ALLOWED_HOSTS:-}
       OPENCLI_MCP_ALLOWED_ORIGINS: ${OPENCLI_MCP_ALLOWED_ORIGINS:-}
       ANTHROPIC_API_KEY: ${ANTHROPIC_API_KEY:-}
@@ -180,6 +216,10 @@ services:
       III_URL: ${III_URL:-}
       DIFY_GRAPHON_RUNTIME_URL: ${DIFY_GRAPHON_RUNTIME_URL:-http://dify-graphon-runtime:8095}
       KATS_RUNTIME_URL: ${KATS_RUNTIME_URL:-http://kats-runtime:8096}
+      PAW_RUNTIME_URL: ${PAW_RUNTIME_URL:-http://paw-runtime:8097}
+      PAW_PROGRAM_ID: ${PAW_PROGRAM_ID:-}
+      PAW_RUNTIME_TIMEOUT_SECONDS: ${PAW_RUNTIME_TIMEOUT_SECONDS:-30}
+      PAW_MAX_TOKENS: ${PAW_MAX_TOKENS:-512}
       INVOKEAI_ENABLED: ${INVOKEAI_ENABLED:-false}
       INVOKEAI_BASE_URL: http://invokeai:9090
       INVOKEAI_API_TOKEN: ${INVOKEAI_API_TOKEN:-}
@@ -194,6 +234,7 @@ services:
       - default
       - dify_runtime
       - kats_runtime
+      - paw_runtime
       - invokeai-private
     healthcheck:
       test: ["CMD", "curl", "-f", "http://localhost:8000/health"]
@@ -205,6 +246,33 @@ services:
       agent-1:
         condition: service_healthy
 
+  # ── Studio Yjs collaboration sidecar ─────────────────────────────────────
+  collaboration:
+    image: ${DOCKER_REGISTRY:-ghcr.io/}${DOCKER_IMAGE_NAMESPACE:-2233admin}/opencli-admin-collaboration:${IMAGE_TAG:-0.4.0}
+    environment:
+      PORT: "1234"
+      COLLABORATION_API_URL: http://api:8000
+      API_AUTH_TOKEN: ${API_AUTH_TOKEN:?Set API_AUTH_TOKEN in .env or run scripts/install.sh}
+      SNAPSHOT_DEBOUNCE_MS: ${SNAPSHOT_DEBOUNCE_MS:-250}
+      MAX_SNAPSHOT_BYTES: ${MAX_SNAPSHOT_BYTES:-1048576}
+      API_REQUEST_TIMEOUT_MS: ${COLLABORATION_API_REQUEST_TIMEOUT_MS:-5000}
+    ports:
+      - "127.0.0.1:1234:1234"
+    depends_on:
+      api:
+        condition: service_healthy
+    healthcheck:
+      test:
+        - CMD
+        - node
+        - -e
+        - "fetch('http://localhost:1234/health').then(r=>{if(!r.ok)process.exit(1)}).catch(()=>process.exit(1))"
+      interval: 10s
+      timeout: 5s
+      retries: 5
+      start_period: 10s
+    restart: unless-stopped
+
   # ── Next.js operator console ─────────────────────────────────────────────
   frontend:
     image: ${DOCKER_REGISTRY:-ghcr.io/}${DOCKER_IMAGE_NAMESPACE:-2233admin}/opencli-admin-frontend:${IMAGE_TAG:-0.4.0}
@@ -244,9 +312,14 @@ services:
       OPENAI_API_KEY: ${OPENAI_API_KEY:-}
       OPENCLI_CDP_ENDPOINT: ${OPENCLI_CDP_ENDPOINT:-http://agent-1:19222}
       AGENT_POOL_ENDPOINTS: ${AGENT_POOL_ENDPOINTS:-}
+      CONTROL_PLANE_URL: ${CONTROL_PLANE_URL:-http://api:8000}
       KATS_RUNTIME_URL: ${KATS_RUNTIME_URL:-http://kats-runtime:8096}
       INVOKEAI_ENABLED: ${INVOKEAI_ENABLED:-false}
       INVOKEAI_BASE_URL: http://invokeai:9090
+      PAW_RUNTIME_URL: ${PAW_RUNTIME_URL:-http://paw-runtime:8097}
+      PAW_PROGRAM_ID: ${PAW_PROGRAM_ID:-}
+      PAW_RUNTIME_TIMEOUT_SECONDS: ${PAW_RUNTIME_TIMEOUT_SECONDS:-30}
+      PAW_MAX_TOKENS: ${PAW_MAX_TOKENS:-512}
       INVOKEAI_API_TOKEN: ${INVOKEAI_API_TOKEN:-}
       IMAGE_ASSET_STORAGE_PATH: /data/image-studio/assets
     depends_on:
@@ -257,6 +330,7 @@ services:
     networks:
       - default
       - kats_runtime
+      - paw_runtime
       - invokeai-private
 
   # ── Private image-generation data plane ─────────────────────────────────
@@ -344,17 +418,19 @@ services:
     profiles: ["agent"]
     image: ${DOCKER_REGISTRY:-ghcr.io/}${DOCKER_IMAGE_NAMESPACE:-2233admin}/opencli-admin-agent:${IMAGE_TAG:-0.4.0}${CHROME_SUFFIX:-}
     environment:
-      CENTRAL_API_URL: ${CENTRAL_API_URL:-}
-      AGENT_ADVERTISE_URL: ${AGENT_ADVERTISE_URL:-}
+      CENTRAL_API_URL: ${CENTRAL_API_URL:-http://api:8000}
+      AGENT_ADVERTISE_URL: ${AGENT_ADVERTISE_URL:-http://agent:19823}
       AGENT_PORT: ${AGENT_PORT:-19823}
       AGENT_MODE: ${AGENT_MODE:-cdp}
-      AGENT_LABEL: ${AGENT_LABEL:-}
-      AGENT_REGISTER: ${AGENT_REGISTER:-http}
+      AGENT_LABEL: ${AGENT_LABEL:-OpenCLI Local Operations Runtime}
+      AGENT_REGISTER: ${AGENT_REGISTER:-ws}
       OPENCLI_TIMEOUT: ${OPENCLI_TIMEOUT:-120}
       API_AUTH_TOKEN: ${API_AUTH_TOKEN:-}
       AGENT_API_TOKEN: ${AGENT_API_TOKEN:-}
+      MINIFLOW_WORKFLOW_ROOT: ${MINIFLOW_WORKFLOW_ROOT:-/app/workflows}
       HTTP_PROXY: ${HTTP_PROXY:-}
       HTTPS_PROXY: ${HTTPS_PROXY:-}
+      NO_PROXY: api,localhost,127.0.0.1,${NO_PROXY:-}
     volumes:
       - agent_profile:/home/agent/.config/chromium
     ports:
@@ -497,6 +573,9 @@ networks:
     internal: true
   kats_runtime:
     internal: true
+  # PAW has no egress; only the API and Celery worker can reach it.
+  paw_runtime:
+    internal: true
   # Dedicated outbound network: the Graphon sidecar can reach server-owned
   # sandbox/model endpoints without becoming reachable from sibling services.
   dify_egress:
diff --git a/docs/adr/0012-pin-automation-deployment-revisions.md b/docs/adr/0012-pin-automation-deployment-revisions.md
index 3e5130b1..b38da703 100644
--- a/docs/adr/0012-pin-automation-deployment-revisions.md
+++ b/docs/adr/0012-pin-automation-deployment-revisions.md
@@ -1,3 +1,35 @@
 # Pin enabled Automations to Deployment Revisions
 
 An enabled Automation pins immutable Agent Deployment Revisions instead of silently following later Runtime or permission changes. Administrators explicitly upgrade and re-enable an Automation, preserving reproducible and auditable behavior; emergency suspension and permission revocation override pinned revisions immediately because safety takes precedence over reproducibility.
+
+## Observable runtime contract
+
+- `schedule` accepts exactly `hourly`, `daily@HH:MM`, `weekdays@HH:MM`, or
+  `weekly@HH:MM`. `weekly` means Monday. Evaluation uses the Automation's IANA
+  `timezone`: nonexistent DST wall-clock minutes are skipped, while the two
+  folds of an ambiguous minute are distinct UTC occurrences.
+- An enabled Automation explicitly pins one enabled, published Operations Agent
+  version. Its permission profile must be `observe_only` or `suggest_changes`
+  and match the Automation approval mode; `low_risk_automatic`, incompatible
+  Agent contracts, and Runtimes not advertised by the bound Fleet node are
+  rejected.
+- **Run Now** uses that persisted Agent/version binding rather than choosing an
+  Agent implicitly. It performs a live Fleet precheck and rejects an offline or
+  disconnected binding before creating a queued Run.
+- Each scheduled occurrence is unique by `(automation_id, scheduled_for)` and
+  persists `trigger_type=scheduled`, a deterministic `trigger_reference`,
+  `scheduled_for`, schedule timezone, the Automation revision and full snapshot,
+  and pinned Agent/profile/version lineage. Repeated scheduler and worker scans
+  reuse the same occurrence.
+- A structurally invalid due binding produces one terminal failed occurrence
+  with the binding error and the same lineage. A structurally valid binding that
+  is temporarily offline remains queued and retryable.
+- Celery Beat surfaces queued scheduled Runs, Celery submits token-authenticated
+  dispatch requests, and the API process that owns the Fleet WebSocket schedules
+  execution. Dispatch is non-blocking and duplicate-safe through the
+  `queued -> running` database compare-and-set. Local scheduler ticks recover
+  the same queued rows, and API restarts preserve queued scheduled claims.
+- Starter Automations are installed paused until a compatible published Agent
+  is explicitly bound. Pausing any Automation prevents future claims without
+  deleting its durable Run history; acceptance/proof Automations remain paused
+  after execution.
diff --git a/docs/local-first-auth-PRD.md b/docs/local-first-auth-PRD.md
new file mode 100644
index 00000000..33c69944
--- /dev/null
+++ b/docs/local-first-auth-PRD.md
@@ -0,0 +1,96 @@
+# 本地优先登录简化 PRD
+
+## 1. 背景
+
+OpenCLI 是部署在 NAS 或自有服务器上的本地数据采集与编排工具，不是需要组织身份提供商才能使用的 SaaS 平台。
+
+现有登录页要求用户填写管理员身份令牌，并可选填写 Fleet API 令牌，同时展示 OIDC、首次部署和紧急恢复概念。这些信息不是本地用户完成首次使用所必需的，直接阻断了“部署后打开即用”的主链路。
+
+## 2. 产品目标
+
+让本地部署用户完成以下最短路径：
+
+```text
+部署 → 打开页面 → admin/admin 登录 → 配置采集 → 持续运行
+```
+
+首次登录后仅通过轻量提示提醒用户修改密码，不要求用户理解令牌、组织登录或 Fleet 网络。
+
+## 3. 范围
+
+### 本期包含
+
+- 本地管理员默认账号：`admin`
+- 本地管理员默认密码：`admin`
+- 登录页只展示用户名和密码
+- 登录成功后创建本地会话
+- 登录页提示默认密码可在设置中修改
+- 设置页提供修改本地管理员密码的入口
+- 修改密码后持久化到部署的 `.env`
+- Fleet API 令牌、OIDC 和组织登录不再出现在本地登录页
+- 保留后端已有 OIDC / Fleet 能力作为后续高级部署能力，但不阻塞本地登录
+
+### 本期不包含
+
+- 组织成员管理
+- 邀请、找回密码、邮件验证
+- 多因素认证
+- 多租户权限模型改造
+- Fleet 网络配置向导
+- 公网 SaaS 登录体验
+
+## 4. 用户流程
+
+### 首次登录
+
+1. 用户打开 `/login`。
+2. 页面预填用户名 `admin`，密码由用户填写；页面说明首次使用可使用 `admin / admin`。
+3. 登录成功后进入控制台。
+4. 控制台账户菜单或设置页显示“建议修改默认密码”提示。
+5. 用户可以稍后处理，不阻塞采集工具使用。
+
+### 修改密码
+
+1. 用户打开账户菜单中的“账户设置”。
+2. 输入当前密码、新密码和确认密码。
+3. 保存成功后当前会话继续有效，下次登录使用新密码。
+
+## 5. 交互要求
+
+- 登录卡片只保留用户名、密码和“登录”按钮。
+- 不出现“首次部署令牌”“管理员身份令牌”“Fleet API 令牌”“OIDC issuer”等部署术语。
+- “本地开发模式”仅在开发环境显示，不作为生产用户入口。
+- 密码修改失败时给出直接错误，不展示后端堆栈或配置细节。
+- 默认密码提示使用普通文案，不使用红色阻断式告警。
+
+## 6. API 契约
+
+### `POST /api/v1/auth/login`
+
+请求：
+
+```json
+{"username":"admin","password":"admin"}
+```
+
+响应：返回 bearer access token、本地管理员身份和 `using_default_password` 标记。
+
+### `POST /api/v1/auth/password`
+
+请求：
+
+```json
+{"current_password":"admin","new_password":"new-password"}
+```
+
+要求当前请求携带本地管理员 bearer token。成功后持久化密码哈希。
+
+## 7. 验收标准
+
+- `/login` 不再渲染任何令牌输入框。
+- 使用 `admin / admin` 可在全新本地部署中登录。
+- 错误密码返回明确失败，不创建会话。
+- 修改密码后，旧密码不能登录，新密码可以登录。
+- 登录成功后的 API 请求不要求用户手工填写 Fleet API 令牌。
+- 登录页 E2E 和后端认证测试覆盖以上行为。
+- OIDC / Fleet 后端代码仍可独立工作，但不进入本地登录主流程。
diff --git a/docs/ptt-acceptance.md b/docs/ptt-acceptance.md
index b66c538a..72d0fe37 100644
--- a/docs/ptt-acceptance.md
+++ b/docs/ptt-acceptance.md
@@ -159,12 +159,101 @@ Pass condition:
 - Docker Agent packaging is part of PTT-0 and must include `backend/agent_runtimes`
   plus `backend/miniflow`; otherwise runtime registration inside the container is
   a false pass.
-- Shell/systemd Python install mode currently downloads only `agent_server.py`.
-  Treat that mode as blocked for runtime PTT until the installer can distribute
-  the runtime adapter package as well.
+- Shell/systemd Python install now downloads and validates the authenticated
+  `agent-runtime.tar.gz` bundle before installing runtime adapters. Native runtime
+  PTT is unblocked at the packaging layer; a real host enrollment is still
+  required for PTT-2 through PTT-6.
 - MiniFlow workflow file distribution is not solved by Fleet itself. PTT uses a
   pre-positioned NAS path until Git sync, file upload, or a managed workflow
   bundle API is implemented.
 - NAS Agent runtime execution is local code execution. PTT must record the
   configured allowlist, workflow directory, bearer token, and audit artifact
   path before the run is accepted.
+
+## 2026-08-24 Local Docker PTT Evidence
+
+This run used isolated local ports `18031` (API), `16080` (noVNC), and
+`19824` (remote Agent) because the default ports were already occupied.
+
+| Gate | Result | Evidence |
+|---|---|---|
+| D0 / PTT-0 | Partial pass | Compose config parsed with isolated credentials; 135 targeted tests passed; `code-intel sentrux check .` rules passed. OpenSpec CLI is unavailable and the Sentrux baseline is missing. |
+| D1 / PTT-1 | Pass | API image and Agent image built; API `/health` returned `{"status":"ok"}`; `/docs` and the built-in browser Agent were healthy. |
+| D2 / PTT-2 persistence | Pass | API and built-in browser Agent restarted; health stayed green, the workspace, published Operations Agent, completed MiniFlow run, and online WS node were readable afterward. |
+| D3 / PTT-2/3 enrollment | Pass | WS Agent `http://host.docker.internal:19824` registered online with `runtimes=["miniflow"]`; Fleet inventory projected `runtime.miniflow`. |
+| D4 / PTT-4 runtime dispatch | Pass | Operations Agent run `9688b25a-26e7-46be-99de-4e6496390202` completed through the real WS path; MiniFlow `ptt-smoke` ran one successful `probe` step and wrote `/tmp/ptt/audit.jsonl`. |
+| D5 / PTT-5 workflow trace | Blocked | The real three-feed Market Situation workflow produced 27 trace events, but all external RSS fetches were blocked because the environment DNS resolved public hosts to `198.18.0.x`, which the SSRF guard correctly rejects; zero records were stored. |
+| D6 backup/restore | Pass | `snailfish_db_data` and Agent profile were archived to `snailfish_ptt_backup`; DB state restored to `snailfish_ptt_restore` and queried successfully (`1` workspace, `1` completed runtime run). |
+| D7 upgrade/rollback | Pass | Rebuilt API image, then launched the locally retained pre-change API image `10f2c4b565a7` against `snailfish_ptt_restore`; `/health` passed and the restored workspace was readable through the old image. |
+PTT-6 OpenTabs smoke remains unproven: this evidence run did not verify the
+OpenTabs `/health`, `/tools` manifest, or a read-only `runtime=opentabs` call.
+Production NAS/edge/NetBird acceptance also remains unproven; this evidence is
+for isolated local Docker only.
+
+The D5 result remains a hard blocker for promoting this deployment profile to
+`supported`; it is not converted into passing evidence.
+## Live Adapter Kernel Smoke
+
+On 2026-08-24, the real RSS adapter/kernel path was exercised against the
+backend at `http://127.0.0.1:18041` using an ephemeral RSS fixture at
+`http://127.0.0.1:18042`. The run used `POST /api/v1/workflows/runs` with
+`runId=live-local-rss-20260824-1110` and returned HTTP `202`,
+`valid=true`, `status=completed`, and `eventCount=22`.
+
+- Source batch: `itemCount=1`.
+- Normalize batch: `recordCount=1`.
+- Trace: `status=completed`, `lastSequence=22`.
+- Failed/blocked events: none.
+
+The public RSS trial remains blocked because the environment's DNS sinkhole
+resolves public hosts to `198.18.*`, which the SSRF guard correctly rejects.
+This local fixture proves the real adapter/kernel execution path only; it does
+not prove internet reachability or public RSS acceptance.
+
+## 2026-08-24 Multi-agent Business Acceptance
+
+The authenticated business path was exercised against the actual backend
+entrypoint (`backend.main:app`) at `127.0.0.1:18042`, using a process-isolated
+temporary SQLite database. `GET /health`, `/docs`, `/api/v1/auth/me`,
+`/api/v1/workflows/capabilities`, and `/api/v1/workflows/fleet/inventory` all
+returned HTTP `200`. Workspace/project bootstrap succeeded, followed by draft
+read/update, validation, and publish.
+
+The final studio run was `36aac778-5a2a-574c-aeec-0d726a0dd73b`, with trace
+`e867db35-e256-498d-ad06-8b285a22a69d`: `status=completed`,
+`valid=true`, and `eventCount=19`. The real local RSS adapter/kernel fetched
+`2` RSS items and produced `2` evidence batches. The project trace was
+readable with events, projection, and checkpoint data. Generic replay run
+`acceptance-generic-rss` was also completed with `19` events, `2` projection
+artifacts, and `2` evidence batches.
+
+The first schedule-only manual run failed correctly with
+`workflow_trigger_kind_mismatch`; adding an explicit manual trigger and
+republishing fixed the operator path. A direct private RSS URL was correctly
+blocked by the SSRF guard; configuring a provider route with local-fixture
+private-network allowance enabled the controlled local RSS acceptance. This
+does not claim public internet success.
+
+Frontend evidence: typecheck and lint passed; the top-right Agent entry opened
+`/operations-agents`, the bottom-right bubble opened the conversation dock,
+and the sidebar Automation entry opened `/schedules`. Workflow-editor browser
+click-through and backend-connected frontend project data remain unverified.
+## 2026-08-25 Full Test Sweep
+
+This run is not a clean full-suite gate; blocked, cancelled, and failing checks
+remain explicitly recorded.
+
+- Backend environment: `uv sync --extra dev` passed.
+- Backend full suite: `uv run pytest` collected `2755` tests, then was bounded
+  and cancelled at
+  `tests/integration/test_chat_api.py::test_viewer_confirmation_is_denied_without_mutation`
+  after `127` passed, `2` skipped, and `0` failed. The standalone viewer test
+  passed, but its overall command exited `1` because coverage was `32.84%`,
+  below the configured `80%` threshold.
+- Backend lint: `ruff check backend tests` found `568` diagnostics in `265`
+  files.
+- Frontend: typecheck, lint, and build passed. Deterministic checks reported
+  `141` passes and `4` stale-contract failures. Playwright smoke was blocked by
+  `EADDRINUSE` on port `3000` and the login-token label.
+
+These results do not constitute a passing full-suite acceptance gate.
diff --git a/frontend/.env.example b/frontend/.env.example
index 58eb0554..c990c134 100644
--- a/frontend/.env.example
+++ b/frontend/.env.example
@@ -1,6 +1,13 @@
 # Experimental local shared-element transitions. Set false for a static fallback build.
 NEXT_PUBLIC_ENABLE_VIEW_TRANSITIONS=true
 
+# Vercel deployments enable Web Analytics automatically. Self-hosted builds stay off unless opted in.
+NEXT_PUBLIC_ENABLE_VERCEL_ANALYTICS=false
+
+# Optional collaboration websocket origin. Unset uses the current browser host
+# with ws:// or wss:// selected from the page protocol.
+NEXT_PUBLIC_YJS_URL=
+
 # Provider-neutral organization login (OpenID Connect Authorization Code + PKCE).
 # Use endpoints published by your OIDC provider; OpenCLI does not infer provider-specific paths.
 NEXT_PUBLIC_OIDC_AUTHORITY=
diff --git a/frontend/Dockerfile b/frontend/Dockerfile
index 052adfc4..3587b903 100644
--- a/frontend/Dockerfile
+++ b/frontend/Dockerfile
@@ -8,7 +8,13 @@ RUN pnpm install --frozen-lockfile
 FROM dependencies AS builder
 
 ARG BACKEND_URL=http://api:8000
+ARG NEXT_PUBLIC_ENABLE_VERCEL_ANALYTICS
+ARG NEXT_PUBLIC_API_AUTH_TOKEN
+ARG NEXT_PUBLIC_YJS_URL
 ENV BACKEND_URL=${BACKEND_URL} \
+    NEXT_PUBLIC_API_AUTH_TOKEN=${NEXT_PUBLIC_API_AUTH_TOKEN} \
+    NEXT_PUBLIC_YJS_URL=${NEXT_PUBLIC_YJS_URL} \
+    NEXT_PUBLIC_ENABLE_VERCEL_ANALYTICS=${NEXT_PUBLIC_ENABLE_VERCEL_ANALYTICS} \
     NEXT_TELEMETRY_DISABLED=1
 COPY . .
 RUN pnpm build
diff --git a/frontend/app/(app)/agents/page.tsx b/frontend/app/(app)/agents/page.tsx
index c401a3ae..7110bb9a 100644
--- a/frontend/app/(app)/agents/page.tsx
+++ b/frontend/app/(app)/agents/page.tsx
@@ -19,6 +19,7 @@ const PROCESSOR_LABEL: Record<string, string> = {
   claude: 'Claude',
   openai: 'OpenAI',
   local: '本地模型',
+  paw: 'PAW 本地富化',
 }
 
 export default function AgentsPage() {
diff --git a/frontend/app/(app)/dashboard/page.tsx b/frontend/app/(app)/dashboard/page.tsx
index f8a71fa6..4147ed75 100644
--- a/frontend/app/(app)/dashboard/page.tsx
+++ b/frontend/app/(app)/dashboard/page.tsx
@@ -36,11 +36,14 @@ import type { FailureItem, StreamTask, ThroughputPoint, WorkerView } from '@/lib
 import { formatNumber, formatRelative } from '@/lib/format'
 import { notificationChannelLabel } from '@/lib/notification-channels'
 import { cn } from '@/lib/utils'
+import { AgentTaskRow, ToolChip } from '@/components/agent-native/agent-primitives'
+import type { AgentTaskStatus } from '@/components/agent-native/agent-primitives'
 import { MatrixClock } from '@/components/monitor/matrix-clock'
 import { FailureFeed, TaskStream } from '@/components/monitor/task-stream'
 import { ThroughputChart } from '@/components/monitor/throughput-chart'
 import { OperationalAnalytics } from '@/components/monitor/operational-analytics'
 import { WorkerAllocation } from '@/components/monitor/worker-allocation'
+import { FancyTestimonialsSlider, type Testimonial } from '@/components/eldoraui/testimonal-slider'
 import { BACKEND_HINT, ErrorState, LoadingState } from '@/components/shell/data-states'
 import { PageContainer } from '@/components/shell/page-container'
 import { Badge } from '@/components/ui/badge'
@@ -200,6 +203,48 @@ function SignalFlow({
   )
 }
 
+function taskStatus(phase: StreamTask['phase']): AgentTaskStatus {
+  if (phase === 'running') return 'running'
+  if (phase === 'failed') return 'failed'
+  if (phase === 'success') return 'completed'
+  return 'queued'
+}
+
+function AgentRunQueue({ tasks }: { tasks: StreamTask[] }) {
+  return (
+    <Card>
+      <CardHeader className="flex flex-row items-start justify-between gap-3">
+        <div>
+          <CardTitle className="flex items-center gap-2 text-base">
+            <Bot className="size-4 text-primary" aria-hidden />
+            Agent 运行队列
+          </CardTitle>
+          <p className="mt-1 text-sm text-muted-foreground">把正在执行、已完成和需要关注的工作集中在一条轻量时间线上。</p>
+        </div>
+        <ToolChip icon={Activity} label="实时任务" detail={`${tasks.length} 条`} tone={tasks.some((task) => task.phase === 'failed') ? 'warning' : 'success'} />
+      </CardHeader>
+      <CardContent className="space-y-2">
+        {tasks.length ? (
+          tasks.slice(0, 4).map((task) => (
+            <AgentTaskRow
+              key={task.id}
+              title={task.title}
+              detail={`${task.endpoint} · ${task.records ? `${formatNumber(task.records)} 条记录` : '等待数据'}`}
+              meta={`${task.workerName || '本地执行器'} · ${formatRelative(new Date(task.startedAt).toISOString())}`}
+              status={taskStatus(task.phase)}
+              progress={task.phase === 'running' ? 62 : task.phase === 'success' ? 100 : undefined}
+            />
+          ))
+        ) : (
+          <div className="rounded-lg border border-dashed p-5 text-sm text-muted-foreground">
+            当前没有最近运行记录。
+          </div>
+        )}
+      </CardContent>
+    </Card>
+  )
+}
+
 function AgentDeliveryPanel({
   agents,
   notificationLogs,
@@ -234,6 +279,11 @@ function AgentDeliveryPanel({
         </Badge>
       </CardHeader>
       <CardContent className="grid gap-4 md:grid-cols-[1.2fr_0.8fr]">
+        <div className="flex flex-wrap gap-1.5 md:col-span-2" aria-label="Agent 交付摘要">
+          <ToolChip icon={Bot} label="活动 Agent" detail={agentsLoading ? '同步中' : `${enabledAgents.length} 个`} tone={enabledAgents.length ? 'success' : 'neutral'} />
+          <ToolChip icon={Send} label="交付渠道" detail={rulesLoading ? '同步中' : `${notificationChannels.length} 个`} />
+          <ToolChip icon={Radio} label="最近送达" detail={logsLoading ? '同步中' : `${delivered} 条`} tone={failed ? 'warning' : 'success'} />
+        </div>
         <div className="space-y-2">
           {agentsLoading ? (
             <p className="text-sm text-muted-foreground">正在同步 Agent…</p>
@@ -538,7 +588,30 @@ export default function DashboardPage() {
   const nextSchedule = [...(schedulesQuery.data?.data ?? [])]
     .filter((schedule) => schedule.enabled && schedule.next_run_at && !Number.isNaN(new Date(schedule.next_run_at).getTime()))
     .sort((left, right) => new Date(left.next_run_at as string).getTime() - new Date(right.next_run_at as string).getTime())[0]
-
+  const insights: Testimonial[] = [
+    {
+      quote: hasAttention
+        ? `${formatNumber(s.tasks.failed)} 个失败任务正在等待处理，建议先检查最近一次运行记录。`
+        : '当前运行链路没有阻塞，可以继续推进工作流和数据源配置。',
+      name: '运行态势',
+      role: hasAttention ? '需要关注' : '运行正常',
+    },
+    {
+      quote: agents.length ? `${agents.length} 个 Agent 已接入当前控制面，结果会按通知规则继续交付。` : '还没有启用 Agent，接入一个执行能力后即可开始自动化。',
+      name: 'Agent 交付',
+      role: `${activeDeliveryChannels.length} 个渠道`,
+    },
+    {
+      quote: `累计采集 ${formatNumber(s.records.total)} 条记录，其中 ${formatNumber(s.records.ai_processed)} 条已经完成 AI 处理。`,
+      name: '数据链路',
+      role: `${normalizedSuccessRate(s.runs.success_rate ?? 0)}% 成功率`,
+    },
+    {
+      quote: nextSchedule ? `下一次调度将在 ${countdownLabel(nextSchedule.next_run_at as string, Date.now())} 后执行。` : '当前没有已启用的下一次调度，可以从自动化与智能体开始配置。',
+      name: '下一步',
+      role: nextSchedule?.name ?? '调度配置',
+    },
+  ]
   return (
     <PageContainer
       eyebrow="Control plane"
@@ -639,7 +712,7 @@ export default function DashboardPage() {
         <div className="grid gap-3 sm:grid-cols-2 xl:grid-cols-4">
           <ActionLink href="/studio" title="编排工作流" description="先选择项目，再设计节点和执行链路" icon={GitBranch} />
           <ActionLink href="/sources" title="接入数据源" description="配置采集来源与凭证" icon={Database} />
-          <ActionLink href="/schedules" title="设置触发调度" description="决定何时自动运行" icon={Clock3} />
+          <ActionLink href="/operations-agents" title="设置自动化与智能体" description="安排任务、选择智能体和配置执行链路" icon={Clock3} />
           <ActionLink href="/tasks" title="检查运行结果" description="查看任务、记录与通知" icon={Activity} />
         </div>
       </section>
@@ -682,6 +755,17 @@ export default function DashboardPage() {
         rulesLoading={notificationRulesQuery.isLoading}
       />
 
+      <AgentRunQueue tasks={stream} />
+
+      <section className="overflow-hidden rounded-xl border border-cyan-500/20 bg-gradient-to-br from-cyan-500/[0.06] via-card to-blue-500/[0.05] py-5" aria-labelledby="agent-insights-title">
+        <div className="mb-2 px-5">
+          <p className="eyebrow-mono">Agent insights / 自动轮播</p>
+          <h2 id="agent-insights-title" className="mt-1 text-lg font-semibold">把系统状态变成下一步建议</h2>
+          <p className="mt-1 text-sm text-muted-foreground">来自 Eldora UI 的 Testimonial Slider 交互，用于浏览运行态势、交付和调度建议。</p>
+        </div>
+        <FancyTestimonialsSlider testimonials={insights} autorotateTiming={6000} />
+      </section>
+
       <section className="grid gap-4" aria-label="运行与异常">
         <FailureFeed failures={failures} />
         <TaskStream tasks={stream} />
diff --git a/frontend/app/(app)/inbox/page.tsx b/frontend/app/(app)/inbox/page.tsx
index d295c8fc..fa8aaa37 100644
--- a/frontend/app/(app)/inbox/page.tsx
+++ b/frontend/app/(app)/inbox/page.tsx
@@ -1,6 +1,5 @@
 'use client'
 
-import Link from 'next/link'
 import { usePathname, useRouter, useSearchParams } from 'next/navigation'
 import {
   Suspense,
@@ -13,7 +12,6 @@ import {
 } from 'react'
 import {
   AlertCircle,
-  ArrowUpRight,
   Bell,
   CheckCircle2,
   Clock3,
@@ -27,19 +25,27 @@ import {
   X,
 } from 'lucide-react'
 
+import { ApprovalQueueDetail, QueueDetail } from '@/components/inbox/queue-detail'
 import { BACKEND_HINT, ErrorState, LoadingState } from '@/components/shell/data-states'
 import { ACTION_CENTER_TABS, RouteTabs } from '@/components/shell/route-tabs'
-import { StatusBadge } from '@/components/shell/status-badge'
-import { Button, buttonVariants } from '@/components/ui/button'
+import { Button } from '@/components/ui/button'
 import { Input } from '@/components/ui/input'
 import { Kbd } from '@/components/ui/kbd'
 import { ScrollArea } from '@/components/ui/scroll-area'
 import {
+  useGovernedWorkspaces,
   useInfiniteControlActions,
   useInfiniteNotificationLogs,
   useInfiniteTasks,
+  useOperationsInbox,
 } from '@/lib/api/hooks'
-import type { CollectionTask, ControlActionRecord, NotificationLog } from '@/lib/api/types'
+import type {
+  CollectionTask,
+  ControlActionRecord,
+  NotificationLog,
+  OperationsWorkItem,
+} from '@/lib/api/types'
+import { resolveApprovalAvailability, shouldIgnoreInboxShortcut } from '@/lib/inbox/workbench-state'
 import { formatRelative } from '@/lib/format'
 import { cn } from '@/lib/utils'
 
@@ -62,6 +68,7 @@ interface QueueItem {
   occurrenceCount: number
   detailLabel: string
   detailValue: string
+  approval?: OperationsWorkItem
 }
 
 const SECTION_META: Record<
@@ -114,6 +121,14 @@ function shortId(value: string) {
   return value.length > 12 ? `${value.slice(0, 8)}…` : value
 }
 
+function evidenceText(evidence: Record<string, unknown>, keys: string[]) {
+  for (const key of keys) {
+    const value = evidence[key]
+    if (typeof value === 'string' && value.trim()) return value.trim()
+  }
+  return null
+}
+
 function taskToQueueItem(task: CollectionTask, waiting = false): QueueItem {
   const title = task.source_name ?? `数据源 ${shortId(task.source_id)}`
   const summary = waiting
@@ -183,6 +198,34 @@ function controlToQueueItem(action: ControlActionRecord): QueueItem {
   }
 }
 
+function approvalToQueueItem(approval: OperationsWorkItem, workspaceName: string): QueueItem {
+  const title =
+    evidenceText(approval.evidence, ['title', 'action', 'operation']) ??
+    'Operations Agent 请求批准'
+  const summary =
+    compact(approval.reason, '') ||
+    evidenceText(approval.evidence, ['summary', 'description', 'reason']) ||
+    '智能体已暂停执行，等待人工决定。'
+
+  return {
+    id: `approval-${approval.id}`,
+    groupKey: `approval:${approval.id}`,
+    section: 'review',
+    eyebrow: `人工审批 · ${workspaceName}`,
+    title,
+    summary,
+    status: approval.status,
+    createdAt: approval.created_at,
+    href: '/operations-agents',
+    hrefLabel: '打开智能体',
+    sourceName: workspaceName,
+    occurrenceCount: 1,
+    detailLabel: '风险级别',
+    detailValue: `${approval.priority} · ${approval.severity}`,
+    approval,
+  }
+}
+
 function groupQueueItems(items: QueueItem[]) {
   const grouped = new Map<string, QueueItem>()
 
@@ -268,104 +311,7 @@ function QueueRow({
   )
 }
 
-function QueueDetail({ item }: { item: QueueItem }) {
-  const meta = SECTION_META[item.section]
-  const nextStep = {
-    blocked: '先检查错误和运行参数，再决定是否重新触发采集。',
-    waiting: '确认执行容量或通知目标状态，避免事项长期停留在队列。',
-    review: '观察后续运行是否恢复，并在控制证据中完成结果判断。',
-  }[item.section]
 
-  return (
-    <div className="flex min-h-0 flex-1 flex-col">
-      <div className="flex min-h-16 items-start justify-between gap-4 border-b px-5 py-3.5">
-        <div className="flex min-w-0 items-start gap-3">
-          <span className={cn('grid size-8 shrink-0 place-items-center rounded-md', meta.iconTone)}>
-            {meta.icon}
-          </span>
-          <div className="min-w-0">
-            <p className="text-xs font-medium text-muted-foreground">{item.eyebrow}</p>
-            <h2 className="mt-0.5 truncate text-base font-semibold leading-tight">{item.title}</h2>
-          </div>
-        </div>
-        <StatusBadge status={item.status} className="shrink-0" />
-      </div>
-
-      <ScrollArea data-testid="inbox-detail-scroll" className="min-h-0 flex-1">
-        <div className="mx-auto w-full max-w-3xl space-y-7 p-5 lg:p-7">
-          <section aria-labelledby="signal-context-heading">
-            <h3 id="signal-context-heading" className="text-xs font-medium text-muted-foreground">
-              信号上下文
-            </h3>
-            <p className="mt-3 whitespace-pre-wrap break-words text-sm leading-6 text-foreground/90">
-              {item.summary}
-            </p>
-            {item.occurrenceCount > 1 ? (
-              <p className="mt-4 border-l-2 border-border pl-3 text-xs leading-5 text-muted-foreground">
-                已将 {item.occurrenceCount} 条同一对象、同一处理阶段的信号合并为一个主题，当前展示最近一次上下文。
-              </p>
-            ) : null}
-          </section>
-
-          <section aria-labelledby="signal-facts-heading">
-            <h3 id="signal-facts-heading" className="text-xs font-medium text-muted-foreground">
-              关键信息
-            </h3>
-            <dl className="mt-3 divide-y border-y text-sm">
-              <div className="grid grid-cols-[7rem_minmax(0,1fr)] gap-3 py-2.5">
-                <dt className="text-muted-foreground">队列</dt>
-                <dd>{meta.label}</dd>
-              </div>
-              <div className="grid grid-cols-[7rem_minmax(0,1fr)] gap-3 py-2.5">
-                <dt className="text-muted-foreground">{item.detailLabel}</dt>
-                <dd className="break-words">{item.detailValue}</dd>
-              </div>
-              {item.sourceId ? (
-                <div className="grid grid-cols-[7rem_minmax(0,1fr)] gap-3 py-2.5">
-                  <dt className="text-muted-foreground">数据源</dt>
-                  <dd>
-                    <Link
-                      href={`/sources/${item.sourceId}`}
-                      className="inline-flex items-center gap-1 font-medium hover:underline focus-visible:rounded-sm focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring"
-                    >
-                      {item.sourceName ?? shortId(item.sourceId)}
-                      <ArrowUpRight aria-hidden="true" className="size-3.5 text-muted-foreground" />
-                    </Link>
-                  </dd>
-                </div>
-              ) : null}
-              <div className="grid grid-cols-[7rem_minmax(0,1fr)] gap-3 py-2.5">
-                <dt className="text-muted-foreground">最近发生</dt>
-                <dd>{formatRelative(item.createdAt)}</dd>
-              </div>
-            </dl>
-          </section>
-
-          <section aria-labelledby="signal-next-step-heading">
-            <h3 id="signal-next-step-heading" className="text-xs font-medium text-muted-foreground">
-              建议下一步
-            </h3>
-            <p className="mt-3 text-sm leading-6 text-muted-foreground">{nextStep}</p>
-          </section>
-        </div>
-      </ScrollArea>
-
-      <div className="flex min-h-14 flex-wrap items-center justify-between gap-3 border-t px-5 py-2.5">
-        <span className="hidden text-xs text-muted-foreground sm:inline">
-          <Kbd>J</Kbd>
-          <Kbd className="ml-1">K</Kbd>
-          <span className="ml-2">切换</span>
-          <Kbd className="ml-3">Enter</Kbd>
-          <span className="ml-2">打开</span>
-        </span>
-        <Link href={item.href} className={buttonVariants({ size: 'sm' })}>
-          {item.hrefLabel}
-          <ArrowUpRight aria-hidden="true" className="size-3.5" />
-        </Link>
-      </div>
-    </div>
-  )
-}
 
 function InboxLoadingFallback() {
   return (
@@ -394,6 +340,24 @@ function InboxContent() {
   const [search, setSearch] = useState(searchParams.get('q') ?? '')
   const [selectedId, setSelectedId] = useState<string | null>(null)
 
+  const workspaces = useGovernedWorkspaces()
+  const requestedWorkspaceId = searchParams.get('workspace')
+  const workspaceId =
+    workspaces.data?.find((workspace) => workspace.id === requestedWorkspaceId)?.id ??
+    workspaces.data?.[0]?.id ??
+    null
+  const workspaceName =
+    workspaces.data?.find((workspace) => workspace.id === workspaceId)?.name ?? 'Workspace'
+  const operationsInbox = useOperationsInbox(workspaceId, 'open')
+  const approvalAvailability = resolveApprovalAvailability({
+    workspaceLoading: workspaces.isLoading,
+    workspaceError: workspaces.isError,
+    workspaceCount: workspaces.data?.length ?? 0,
+    workspaceId,
+    inboxLoading: operationsInbox.isLoading,
+    inboxError: operationsInbox.isError,
+  })
+
   const failedTasks = useInfiniteTasks({ status: 'failed', limit: 100 })
   const pendingTasks = useInfiniteTasks({ status: 'pending', limit: 100 })
   const notificationLogs = useInfiniteNotificationLogs({ limit: 100 })
@@ -418,12 +382,16 @@ function InboxContent() {
     () => pendingControlActions.data?.pages.flatMap((page) => page.data) ?? [],
     [pendingControlActions.data?.pages],
   )
+  const approvals = useMemo(
+    () => operationsInbox.data?.data?.filter((item) => item.type === 'approval') ?? [],
+    [operationsInbox.data],
+  )
 
   const rawCounts: Record<QueueFilter, number> = {
-    all: failed.length + pending.length + notifications.length + controls.length,
+    all: failed.length + pending.length + notifications.length + controls.length + approvals.length,
     blocked: failed.length + notifications.filter((log) => /fail|error/i.test(log.status)).length,
     waiting: pending.length + notifications.filter((log) => !/fail|error/i.test(log.status)).length,
-    review: controls.length,
+    review: controls.length + approvals.length,
   }
 
   const queueItems = useMemo(
@@ -433,8 +401,9 @@ function InboxContent() {
         ...pending.map((task) => taskToQueueItem(task, true)),
         ...notifications.map(notificationToQueueItem),
         ...controls.map(controlToQueueItem),
+        ...approvals.map((approval) => approvalToQueueItem(approval, workspaceName)),
       ]),
-    [controls, failed, notifications, pending],
+    [approvals, controls, failed, notifications, pending, workspaceName],
   )
 
   const filteredItems = useMemo(() => {
@@ -466,6 +435,16 @@ function InboxContent() {
     [pathname, router, searchParamsKey],
   )
 
+  const selectWorkspace = useCallback(
+    (nextWorkspaceId: string) => {
+      const params = new URLSearchParams(searchParamsKey)
+      params.set('workspace', nextWorkspaceId)
+      const nextQuery = params.toString()
+      router.replace(nextQuery ? `${pathname}?${nextQuery}` : pathname, { scroll: false })
+    },
+    [pathname, router, searchParamsKey],
+  )
+
   useEffect(() => {
     const params = new URLSearchParams(searchParamsKey)
     const view = params.get('view')
@@ -494,11 +473,25 @@ function InboxContent() {
 
   useEffect(() => {
     const onKeyDown = (event: KeyboardEvent) => {
-      const target = event.target as HTMLElement | null
-      const isEditing =
-        target?.tagName === 'INPUT' ||
-        target?.tagName === 'TEXTAREA' ||
-        target?.isContentEditable
+      const target = event.target instanceof Element ? event.target : null
+      const targetElement = target instanceof HTMLElement ? target : null
+      const isInteractive = shouldIgnoreInboxShortcut({
+        tagName: target?.tagName,
+        isContentEditable: targetElement?.isContentEditable,
+        withinInteractive: Boolean(
+          target?.closest(
+            'a,button,input,select,textarea,summary,[contenteditable="true"],[role="button"],[role="link"],[role="menuitem"],[role="option"],[role="tab"],[tabindex]:not([tabindex="-1"])',
+          ),
+        ),
+      })
+
+      if (isInteractive) {
+        if (event.key === 'Escape' && target === searchRef.current) {
+          setSearch('')
+          searchRef.current?.blur()
+        }
+        return
+      }
 
       if ((event.ctrlKey || event.metaKey) && event.key.toLowerCase() === 'f') {
         event.preventDefault()
@@ -507,14 +500,6 @@ function InboxContent() {
         return
       }
 
-      if (isEditing) {
-        if (event.key === 'Escape') {
-          setSearch('')
-          searchRef.current?.blur()
-        }
-        return
-      }
-
       const currentIndex = selectedItem
         ? filteredItems.findIndex((item) => item.id === selectedItem.id)
         : -1
@@ -528,7 +513,7 @@ function InboxContent() {
         setSelectedId(filteredItems[nextIndex].id)
       }
 
-      if (event.key === 'Enter' && selectedItem) {
+      if (event.key === 'Enter' && selectedItem && !selectedItem.approval) {
         event.preventDefault()
         router.push(selectedItem.href)
       }
@@ -539,19 +524,31 @@ function InboxContent() {
   }, [filteredItems, router, selectedItem])
 
   const queries = [failedTasks, pendingTasks, notificationLogs, pendingControlActions]
-  const isInitialLoading = queries.every((query) => query.isLoading)
-  const isTotalFailure = queries.every((query) => query.isError)
+  const isInitialLoading =
+    queries.every((query) => query.isLoading) &&
+    (approvalAvailability !== 'ready' || approvals.length === 0)
+  const isTotalFailure =
+    queries.every((query) => query.isError) &&
+    (approvalAvailability === 'workspace_error' ||
+      approvalAvailability === 'inbox_error' ||
+      approvalAvailability === 'no_workspace')
   const partialFailures = [
     failedTasks.isError ? '失败任务' : null,
     pendingTasks.isError ? '等待任务' : null,
     notificationLogs.isError ? '通知记录' : null,
     pendingControlActions.isError ? '控制结果' : null,
+    workspaces.isError ? 'Workspace' : null,
+    operationsInbox.isError ? '人工审批' : null,
   ].filter(Boolean)
   const hasMoreSignals = queries.some((query) => query.hasNextPage)
   const isFetchingNextPage = queries.some((query) => query.isFetchingNextPage)
 
   const refetchAll = () => {
-    void Promise.all(queries.map((query) => query.refetch()))
+    void Promise.all([
+      ...queries.map((query) => query.refetch()),
+      workspaces.refetch(),
+      operationsInbox.refetch(),
+    ])
   }
 
   const loadMoreSignals = () => {
@@ -569,6 +566,12 @@ function InboxContent() {
     { key: 'waiting', label: '等待' },
     { key: 'review', label: '复核' },
   ]
+  const approvalNotice =
+    approvalAvailability === 'loading'
+      ? '正在读取 Workspace 人工审批…'
+      : approvalAvailability === 'no_workspace'
+        ? '尚未加入 Workspace，人工审批目前不可用。'
+        : null
 
   return (
     <div
@@ -589,6 +592,20 @@ function InboxContent() {
         />
 
         <div className="ml-auto flex w-full items-center gap-1.5 md:w-auto">
+          {workspaceId ? (
+            <select
+              value={workspaceId}
+              onChange={(event) => selectWorkspace(event.target.value)}
+              aria-label="选择人工审批 Workspace"
+              className="h-9 max-w-40 shrink-0 rounded-md border bg-background px-2 text-xs md:h-8"
+            >
+              {workspaces.data?.map((workspace) => (
+                <option key={workspace.id} value={workspace.id}>
+                  {workspace.name}
+                </option>
+              ))}
+            </select>
+          ) : null}
           <div className="relative min-w-0 flex-1 md:w-64 md:flex-none">
             <Search
               aria-hidden="true"
@@ -642,6 +659,12 @@ function InboxContent() {
         </div>
       ) : null}
 
+      {approvalNotice ? (
+        <div role="status" className="shrink-0 border-b bg-muted/25 px-4 py-2 text-xs text-muted-foreground">
+          {approvalNotice}
+        </div>
+      ) : null}
+
       {isInitialLoading ? (
         <div className="min-h-0 flex-1 p-4">
           <LoadingState rows={5} />
@@ -776,7 +799,15 @@ function InboxContent() {
             className="flex min-h-[32rem] min-w-0 flex-col bg-muted/10 lg:min-h-0"
           >
             {selectedItem ? (
-              <QueueDetail item={selectedItem} />
+              selectedItem.approval ? (
+                <ApprovalQueueDetail key={selectedItem.id} item={selectedItem} />
+              ) : (
+                <QueueDetail
+                  item={selectedItem}
+                  meta={SECTION_META[selectedItem.section]}
+                  createdAtLabel={formatRelative(selectedItem.createdAt)}
+                />
+              )
             ) : (
               <div className="grid min-h-80 flex-1 place-items-center px-6 text-center">
                 <div>
diff --git a/frontend/app/(app)/operations-agents/page.tsx b/frontend/app/(app)/operations-agents/page.tsx
index 2476c970..b4e4ed07 100644
--- a/frontend/app/(app)/operations-agents/page.tsx
+++ b/frontend/app/(app)/operations-agents/page.tsx
@@ -1,16 +1,21 @@
 'use client'
 
 import { useEffect, useMemo, useState } from 'react'
-import Link from 'next/link'
-import { ArrowUp, Bell, Bot, CalendarClock, ChevronDown, CircleDot, Cloud, Code2, FileSearch, Pause, Play, Plus, Repeat2, Sparkles, Terminal } from 'lucide-react'
+import { ArrowUp, Bell, Bot, CalendarClock, ChevronDown, CircleDot, FileSearch, Pause, Play, Plus, Repeat2, Terminal } from 'lucide-react'
 import { toast } from 'sonner'
 
 import AgentAvatar from '@/components/smoothui/agent-avatar'
-import { useAutomations, useCreateAutomation, useMyWorkspaces, useOperationsAgentActivity, useOperationsAgentDraft, useOperationsAgents, useOperationsAgentVersions, usePatchAutomation, usePublishOperationsAgentVersion, useStartOperationsAgentRun, useUpdateOperationsAgentDraft } from '@/lib/api/hooks'
-import type { Automation, OperationsAgent, OperationsAgentMode } from '@/lib/api/types'
+import SwitchboardCard from '@/components/smoothui/switchboard-card'
+import { useAutomations, useCreateAutomation, useCreateOperationsAgent, useGovernedWorkspaces, useInstallAutomationStarters, useNodes, useOperationsAgentActivity, useOperationsAgentDraft, useOperationsAgents, useOperationsAgentTeams, useOperationsAgentVersions, usePatchAutomation, usePublishOperationsAgentVersion, useStartAutomationRun, useStartOperationsAgentRun, useUpdateOperationsAgentDraft } from '@/lib/api/hooks'
+import type { AgentRuntimeBindingV2, Automation, OperationsAgent, OperationsAgentMode } from '@/lib/api/types'
+import { latestOperationsAgentRuns } from '@/lib/automations/activity'
+import { AUTOMATION_APPROVALS as APPROVALS } from '@/lib/automations/approval'
+import { compatibleOperationsAgents, runnableOperationsAgents } from '@/lib/automations/binding'
+import { automationExecutorMeta as executorMeta } from '@/lib/automations/executors'
+import { automationScheduleText, automationScheduleValue } from '@/lib/automations/schedule'
 import { cn } from '@/lib/utils'
 import { BACKEND_HINT, EmptyState, ErrorState, LoadingState } from '@/components/shell/data-states'
-import { Button, buttonVariants } from '@/components/ui/button'
+import { Button } from '@/components/ui/button'
 import { Dialog, DialogContent, DialogDescription, DialogFooter, DialogHeader, DialogTitle } from '@/components/ui/dialog'
 import { Input } from '@/components/ui/input'
 import { Textarea } from '@/components/ui/textarea'
@@ -21,27 +26,34 @@ const SUGGESTIONS = [
   { name: '异常跟进监控', prompt: '检查最近的异常活动，并将有证据的问题整理为待处理建议。', icon: FileSearch, color: 'text-emerald-400', schedule: 'weekdays@09:00' },
 ] as const
 
-const EXECUTORS = [
-  { id: 'codex', name: 'Codex', icon: Code2, color: 'text-sky-400' },
-  { id: 'claude', name: 'Claude', icon: Sparkles, color: 'text-orange-400' },
-  { id: 'chatcloud', name: 'ChatCloud', icon: Cloud, color: 'text-violet-400' },
-  { id: 'custom', name: '自定义', icon: Terminal, color: 'text-emerald-400' },
+const AGENT_STARTERS = [
+  {
+    ...SUGGESTIONS[0],
+    name: '运行简报 Agent',
+    subtitle: '每天汇总运行、失败与待批准事项',
+    pattern: [0, 1, 2, 18, 19, 20, 36, 37, 38, 54, 55, 56, 72, 73, 74],
+  },
+  {
+    ...SUGGESTIONS[1],
+    name: '系统回顾 Agent',
+    subtitle: '每周整理变化、风险与待处理建议',
+    pattern: [4, 5, 6, 22, 23, 24, 40, 41, 42, 58, 59, 60, 76, 77, 78],
+  },
+  {
+    ...SUGGESTIONS[2],
+    name: '异常跟进 Agent',
+    subtitle: '工作日检查异常并生成证据化建议',
+    pattern: [8, 9, 10, 26, 27, 28, 44, 45, 46, 62, 63, 64, 80, 81, 82],
+  },
 ] as const
 
-const APPROVALS: Array<{ id: OperationsAgentMode; label: string; detail: string }> = [
-  { id: 'observe_only', label: '仅观察', detail: '不提出或执行变更' },
-  { id: 'suggest_changes', label: '建议需批准', detail: '送入 Inbox 后由人决定' },
-  { id: 'low_risk_automatic', label: '低风险自动', detail: '白名单外仍需批准' },
-]
-
-function executorMeta(id: string) {
-  return EXECUTORS.find((item) => item.id === id) ?? EXECUTORS[3]
+type AgentStarterInput = {
+  name: string
+  prompt: string
+  schedule: string
 }
 
-function scheduleText(value: string) {
-  const [kind, time] = value.split('@')
-  return `${kind === 'daily' ? '每天' : kind === 'weekdays' ? '工作日' : kind === 'weekly' ? '每周' : kind}${time ? ` ${time}` : ''}`
-}
+
 
 const EMPTY_SCHEMA = { type: 'object', properties: {} }
 
@@ -51,37 +63,84 @@ function parseJsonObject(value: string, label: string) {
   return parsed as Record<string, unknown>
 }
 
+function parseJsonArray(value: string, label: string) {
+  const parsed = JSON.parse(value)
+  if (!Array.isArray(parsed) || parsed.some((item) => !item || Array.isArray(item) || typeof item !== 'object')) {
+    throw new Error(`${label} 必须是 JSON 对象数组`)
+  }
+  return parsed as Array<Record<string, unknown>>
+}
+
+function parseIdentifierList(value: string) {
+  return [...new Set(value.split(',').map((item) => item.trim()).filter(Boolean))]
+}
+
 function ContractEditor({ workspaceId, agent }: { workspaceId: string; agent: OperationsAgent }) {
   const draft = useOperationsAgentDraft(workspaceId, agent.id)
   const versions = useOperationsAgentVersions(workspaceId, agent.id)
   const updateDraft = useUpdateOperationsAgentDraft()
+  const nodes = useNodes()
   const publishVersion = usePublishOperationsAgentVersion()
   const [instructions, setInstructions] = useState('')
+  const [role, setRole] = useState('operations_agent')
+  const [requiredCapabilities, setRequiredCapabilities] = useState('streaming')
+  const [toolPolicy, setToolPolicy] = useState('{}')
+  const [budget, setBudget] = useState('{}')
+  const [qualityGates, setQualityGates] = useState('[]')
+  const [evidenceRequirements, setEvidenceRequirements] = useState('')
   const [inputSchema, setInputSchema] = useState('')
   const [outputSchema, setOutputSchema] = useState('')
   const [stateSchema, setStateSchema] = useState('')
   const [agentUrl, setAgentUrl] = useState('')
+  const [runtime, setRuntime] = useState<AgentRuntimeBindingV2['preferred_runtimes'][number]>('')
   const [workflow, setWorkflow] = useState('')
-  const [dispatchTimeout, setDispatchTimeout] = useState(600)
+  const [dispatchTimeout, setDispatchTimeout] = useState(1800)
   const [runtimeConfig, setRuntimeConfig] = useState('')
+  const [provider, setProvider] = useState('')
+  const [model, setModel] = useState('')
+  const [authProfile, setAuthProfile] = useState('')
   const [reason, setReason] = useState('')
-
+  const runtimeNodes = useMemo(
+    () => [...(nodes.data?.data ?? [])].reverse().filter(
+      (node) => node.protocol === 'ws' && node.status === 'online' && Object.keys(node.runtime_capabilities ?? {}).length,
+    ),
+    [nodes.data],
+  )
+  const selectedRuntimeNode = runtimeNodes.find((node) => node.url === agentUrl)
+  const runtimeOptions = selectedRuntimeNode
+    ? Object.keys(selectedRuntimeNode.runtime_capabilities ?? {})
+    : [...new Set(runtimeNodes.flatMap((node) => Object.keys(node.runtime_capabilities ?? {})))]
   useEffect(() => {
     if (!draft.data) return
     const contract = draft.data.model_configuration.agent_contract
     const binding = draft.data.model_configuration.runtime_binding
     setInstructions(draft.data.instructions)
+    setRole(contract?.role ?? 'operations_agent')
+    setRequiredCapabilities((contract?.required_capabilities ?? ['streaming']).join(', '))
+    setToolPolicy(JSON.stringify(contract?.tool_policy ?? {}, null, 2))
+    setBudget(JSON.stringify(contract?.budget ?? {}, null, 2))
+    setQualityGates(JSON.stringify(contract?.quality_gates ?? [], null, 2))
+    setEvidenceRequirements((contract?.evidence_requirements ?? []).join(', '))
     setInputSchema(JSON.stringify(contract?.input_schema ?? EMPTY_SCHEMA, null, 2))
     setOutputSchema(JSON.stringify(contract?.output_schema ?? EMPTY_SCHEMA, null, 2))
     setStateSchema(JSON.stringify(contract?.state_schema ?? EMPTY_SCHEMA, null, 2))
-    setAgentUrl(binding?.agent_url ?? '')
+    setAgentUrl(binding?.preferred_agent_urls[0] ?? '')
+    setRuntime(binding?.preferred_runtimes[0] ?? '')
     setWorkflow(binding?.workflow ?? '')
-    setDispatchTimeout(binding?.dispatch_timeout_seconds ?? 600)
-    setRuntimeConfig(JSON.stringify(binding?.config ?? {}, null, 2))
+    setDispatchTimeout(binding?.dispatch_timeout_seconds ?? 1800)
+    setRuntimeConfig(JSON.stringify(binding?.config ?? { timeout_seconds: 1800 }, null, 2))
+    setProvider(binding?.model_binding?.provider ?? '')
+    setModel(binding?.model_binding?.model ?? '')
+    setAuthProfile(binding?.model_binding?.auth_profile ?? '')
   }, [draft.data])
 
+
   async function saveDraft() {
     if (!draft.data) return
+    if (Boolean(provider.trim()) !== Boolean(model.trim())) {
+      toast.error('Provider 与 Model 必须同时填写或同时留空')
+      return
+    }
     try {
       await updateDraft.mutateAsync({
         workspaceId,
@@ -92,16 +151,28 @@ function ContractEditor({ workspaceId, agent }: { workspaceId: string; agent: Op
           model_configuration: {
             ...draft.data.model_configuration,
             agent_contract: {
-              schema_version: 'agent.contract.v1',
+              schema_version: 'agent.contract.v2',
+              role: role.trim(),
               input_schema: parseJsonObject(inputSchema, 'Input schema'),
               output_schema: parseJsonObject(outputSchema, 'Output schema'),
               state_schema: parseJsonObject(stateSchema, 'State schema'),
+              required_capabilities: parseIdentifierList(requiredCapabilities),
+              tool_policy: parseJsonObject(toolPolicy, 'Tool policy'),
+              budget: parseJsonObject(budget, 'Budget'),
+              quality_gates: parseJsonArray(qualityGates, 'Quality gates'),
+              evidence_requirements: parseIdentifierList(evidenceRequirements),
             },
             runtime_binding: {
-              schema_version: 'agent.runtime-binding.v1',
-              agent_url: agentUrl.trim(),
-              runtime: 'pi',
+              schema_version: 'agent.runtime-binding.v2',
               workflow: workflow.trim(),
+              preferred_agent_urls: agentUrl ? [agentUrl.trim()] : [],
+              preferred_runtimes: runtime ? [runtime] : [],
+              model_binding: provider.trim() && model.trim() ? {
+                schema_version: 'agent.model-binding.v1',
+                provider: provider.trim(),
+                model: model.trim(),
+                auth_profile: authProfile.trim() || null,
+              } : null,
               dispatch_timeout_seconds: dispatchTimeout,
               config: parseJsonObject(runtimeConfig, 'Runtime config'),
             },
@@ -140,24 +211,44 @@ function ContractEditor({ workspaceId, agent }: { workspaceId: string; agent: Op
       <div className="min-h-0 overflow-y-auto bg-[#090a0b] p-5">
         <div className="mb-5 flex flex-wrap items-center justify-between gap-3">
           <div><h3 className="text-sm font-medium">Agent Contract</h3><p className="mt-1 text-xs text-muted-foreground">Draft r{draft.data?.revision ?? '—'} · 当前发布 v{currentPublishedVersion || '—'}</p></div>
-          <Button size="sm" onClick={() => void saveDraft()} disabled={!instructions.trim() || !agentUrl.trim() || !workflow.trim() || updateDraft.isPending}>保存草稿</Button>
+          <Button size="sm" onClick={() => void saveDraft()} disabled={!instructions.trim() || !role.trim() || !workflow.trim() || updateDraft.isPending}>保存草稿</Button>
         </div>
         <div className="space-y-5">
           <label className="block space-y-1.5 text-xs text-muted-foreground">Instructions<Textarea value={instructions} onChange={(event) => setInstructions(event.target.value)} className="min-h-36 resize-y text-sm text-foreground" placeholder="描述智能体的职责、边界和执行要求" /></label>
+          <div className="grid gap-4 lg:grid-cols-3">
+            <label className="block space-y-1.5 text-xs text-muted-foreground">业务角色<Input value={role} onChange={(event) => setRole(event.target.value)} placeholder="sales_researcher" /></label>
+            <label className="block space-y-1.5 text-xs text-muted-foreground">必需 Capability（逗号分隔）<Input value={requiredCapabilities} onChange={(event) => setRequiredCapabilities(event.target.value)} placeholder="streaming, browser" /></label>
+            <label className="block space-y-1.5 text-xs text-muted-foreground">证据要求（逗号分隔）<Input value={evidenceRequirements} onChange={(event) => setEvidenceRequirements(event.target.value)} placeholder="citations, lineage" /></label>
+          </div>
           <div className="grid gap-4 lg:grid-cols-3">
             <label className="block space-y-1.5 text-xs text-muted-foreground">Input schema<Textarea value={inputSchema} onChange={(event) => setInputSchema(event.target.value)} spellCheck={false} className="min-h-64 resize-y font-mono text-xs text-foreground" /></label>
             <label className="block space-y-1.5 text-xs text-muted-foreground">Output schema<Textarea value={outputSchema} onChange={(event) => setOutputSchema(event.target.value)} spellCheck={false} className="min-h-64 resize-y font-mono text-xs text-foreground" /></label>
             <label className="block space-y-1.5 text-xs text-muted-foreground">State schema<Textarea value={stateSchema} onChange={(event) => setStateSchema(event.target.value)} spellCheck={false} className="min-h-64 resize-y font-mono text-xs text-foreground" /></label>
           </div>
           <details className="rounded-lg border border-white/[0.08] px-4 py-3">
-            <summary className="cursor-pointer text-sm text-muted-foreground">Runtime Binding 高级配置</summary>
+            <summary className="cursor-pointer text-sm text-muted-foreground">Contract 策略与质量门禁</summary>
+            <div className="mt-4 grid gap-4 lg:grid-cols-3">
+              <label className="block space-y-1.5 text-xs text-muted-foreground">Tool policy<Textarea value={toolPolicy} onChange={(event) => setToolPolicy(event.target.value)} spellCheck={false} className="min-h-32 resize-y font-mono text-xs text-foreground" /></label>
+              <label className="block space-y-1.5 text-xs text-muted-foreground">Budget<Textarea value={budget} onChange={(event) => setBudget(event.target.value)} spellCheck={false} className="min-h-32 resize-y font-mono text-xs text-foreground" /></label>
+              <label className="block space-y-1.5 text-xs text-muted-foreground">Quality gates<Textarea value={qualityGates} onChange={(event) => setQualityGates(event.target.value)} spellCheck={false} className="min-h-32 resize-y font-mono text-xs text-foreground" /></label>
+            </div>
+          </details>
+          <details className="rounded-lg border border-white/[0.08] px-4 py-3">
+            <summary className="cursor-pointer text-sm text-muted-foreground">Runtime / Provider / Model 高级配置</summary>
+            <p className="mt-2 text-xs leading-5 text-muted-foreground">节点和 Runtime 仅作为调度偏好；最终执行器按 Capability 匹配。Provider 凭据保留在 Fleet 节点，控制面只保存 auth profile 引用。</p>
             <div className="mt-4 grid gap-4 sm:grid-cols-4">
-              <label className="space-y-1.5 text-xs text-muted-foreground">Agent URL<Input type="url" value={agentUrl} onChange={(event) => setAgentUrl(event.target.value)} placeholder="https://agent.example.com" /></label>
-              <label className="space-y-1.5 text-xs text-muted-foreground">Runtime<Input value="pi" readOnly aria-readonly="true" /></label>
-              <label className="space-y-1.5 text-xs text-muted-foreground">Workflow<Input value={workflow} onChange={(event) => setWorkflow(event.target.value)} placeholder="default" /></label>
-              <label className="space-y-1.5 text-xs text-muted-foreground">超时（秒）<Input type="number" min={1} max={3600} value={dispatchTimeout} onChange={(event) => setDispatchTimeout(Number(event.target.value))} /></label>
+              <label className="space-y-1.5 text-xs text-muted-foreground">Fleet 节点偏好<select value={agentUrl} onChange={(event) => { const nextUrl = event.target.value; const node = runtimeNodes.find((candidate) => candidate.url === nextUrl); setAgentUrl(nextUrl); if (runtime && node && !Object.hasOwn(node.runtime_capabilities ?? {}, runtime)) setRuntime('') }} className="h-9 w-full rounded-lg border bg-background px-3 text-sm text-foreground"><option value="">无节点偏好</option>{runtimeNodes.map((node) => <option key={node.id} value={node.url}>{node.label}</option>)}</select></label>
+              <label className="space-y-1.5 text-xs text-muted-foreground">Runtime 偏好<select value={runtime} onChange={(event) => setRuntime(event.target.value)} className="h-9 w-full rounded-lg border bg-background px-3 text-sm text-foreground"><option value="">无 Runtime 偏好</option>{runtimeOptions.map((runtimeName) => <option key={runtimeName} value={runtimeName}>{runtimeName}</option>)}</select></label>
+              <label className="space-y-1.5 text-xs text-muted-foreground">Workflow<Input value={workflow} onChange={(event) => setWorkflow(event.target.value)} placeholder="operations-agent" /></label>
+              <label className="space-y-1.5 text-xs text-muted-foreground">深度执行超时（秒）<Input type="number" min={1} max={3600} value={dispatchTimeout} onChange={(event) => setDispatchTimeout(Number(event.target.value))} /></label>
             </div>
-            <label className="mt-4 block space-y-1.5 text-xs text-muted-foreground">Task config（仅 timeout_seconds）<Textarea value={runtimeConfig} onChange={(event) => setRuntimeConfig(event.target.value)} spellCheck={false} className="min-h-32 resize-y font-mono text-xs text-foreground" /></label>
+            <div className="mt-4 grid gap-4 sm:grid-cols-4">
+              <label className="space-y-1.5 text-xs text-muted-foreground">Provider<Input value={provider} onChange={(event) => setProvider(event.target.value)} placeholder="openrouter" /></label>
+              <label className="space-y-1.5 text-xs text-muted-foreground">Model<Input value={model} onChange={(event) => setModel(event.target.value)} placeholder="anthropic/claude-sonnet" /></label>
+              <label className="space-y-1.5 text-xs text-muted-foreground">Auth profile 引用<Input value={authProfile} onChange={(event) => setAuthProfile(event.target.value)} placeholder="sales-production" /></label>
+            </div>
+            {!runtimeNodes.length ? <p className="mt-3 text-xs text-amber-300">没有已连接并报告 Capability manifest 的 Fleet 节点；草稿可以保存，但运行会按 Contract 要求失败关闭。</p> : null}
+            <label className="mt-4 block space-y-1.5 text-xs text-muted-foreground">Task config（仅 timeout_seconds）<Textarea value={runtimeConfig} onChange={(event) => setRuntimeConfig(event.target.value)} spellCheck={false} className="min-h-24 resize-y font-mono text-xs text-foreground" /></label>
           </details>
         </div>
       </div>
@@ -175,14 +266,18 @@ function ContractEditor({ workspaceId, agent }: { workspaceId: string; agent: Op
 }
 
 export default function OperationsAgentsPage() {
-  const workspaces = useMyWorkspaces()
+  const workspaces = useGovernedWorkspaces()
   const [workspaceId, setWorkspaceId] = useState<string | null>(null)
   const [view, setView] = useState<'automations' | 'agents'>('automations')
   const automations = useAutomations(workspaceId)
   const agents = useOperationsAgents(workspaceId)
+  const teams = useOperationsAgentTeams(workspaceId)
   const activity = useOperationsAgentActivity(workspaceId)
+  const installStarterPack = useInstallAutomationStarters()
+  const createOperationsAgent = useCreateOperationsAgent()
   const createAutomation = useCreateAutomation()
   const patchAutomation = usePatchAutomation()
+  const startAutomationRun = useStartAutomationRun()
   const startRunMutation = useStartOperationsAgentRun()
   const [open, setOpen] = useState(false)
   const [selectedAgent, setSelectedAgent] = useState<OperationsAgent | null>(null)
@@ -191,27 +286,58 @@ export default function OperationsAgentsPage() {
   const [name, setName] = useState('')
   const [prompt, setPrompt] = useState('')
   const [precheck, setPrecheck] = useState('')
-  const [executor, setExecutor] = useState('codex')
+  const [automationAgentId, setAutomationAgentId] = useState('')
   const [projectPath, setProjectPath] = useState('')
   const [branch, setBranch] = useState('main')
   const [scheduleKind, setScheduleKind] = useState('weekdays')
   const [time, setTime] = useState('09:00')
   const [sessionMode, setSessionMode] = useState<'fresh' | 'reuse'>('fresh')
   const [approvalMode, setApprovalMode] = useState<OperationsAgentMode>('suggest_changes')
-  const [runTargetType, setRunTargetType] = useState('manual')
   const [runTargetId, setRunTargetId] = useState('')
   const [runInput, setRunInput] = useState('{}')
   const [runState, setRunState] = useState('{}')
+  const [runTargetType, setRunTargetType] = useState('manual')
+  const [agentCreateOpen, setAgentCreateOpen] = useState(false)
+  const [agentName, setAgentName] = useState('')
+  const [agentDescription, setAgentDescription] = useState('')
+  const [owningTeamId, setOwningTeamId] = useState('')
+  const [automationToBind, setAutomationToBind] = useState<Automation | null>(null)
+  const [automationToRun, setAutomationToRun] = useState<Automation | null>(null)
+  const [bindingAgentId, setBindingAgentId] = useState('')
+  const latestRun = useMemo(
+    () => latestOperationsAgentRuns(activity.data ?? []),
+    [activity.data],
+  )
+  const runnableAgents = useMemo(
+    () => runnableOperationsAgents(agents.data ?? []),
+    [agents.data],
+  )
+  const compatibleAgents = useMemo(
+    () => compatibleOperationsAgents(agents.data ?? [], approvalMode),
+    [agents.data, approvalMode],
+  )
 
   useEffect(() => {
-    if (!workspaceId && workspaces.data?.length) setWorkspaceId(workspaces.data[0].id)
+    if (!workspaceId && workspaces.data?.length) {
+      setWorkspaceId(workspaces.data[0].id)
+    }
   }, [workspaceId, workspaces.data])
 
-  const latestRun = useMemo(() => new Map(activity.data?.map((run) => [run.operations_agent_id, run]) ?? []), [activity.data])
+  useEffect(() => {
+    if (agentCreateOpen && teams.data?.length === 1 && !owningTeamId) {
+      setOwningTeamId(teams.data[0].id)
+    }
+  }, [agentCreateOpen, owningTeamId, teams.data])
 
-  function startCreate(preset?: (typeof SUGGESTIONS)[number]) {
+  useEffect(() => {
+    if (!open) return
+    const selected = compatibleAgents.find((agent) => agent.id === automationAgentId)
+    if (!selected) setAutomationAgentId(compatibleAgents[0]?.id ?? '')
+  }, [automationAgentId, compatibleAgents, open])
+  function startCreate(preset?: AgentStarterInput) {
     setName(preset?.name ?? '')
     setPrompt(preset?.prompt ?? '')
+    setAutomationAgentId('')
     if (preset) {
       const [kind, presetTime] = preset.schedule.split('@')
       setScheduleKind(kind)
@@ -220,6 +346,16 @@ export default function OperationsAgentsPage() {
     setOpen(true)
   }
 
+  async function installAgentStarters() {
+    if (!workspaceId || installStarterPack.isPending) return
+    try {
+      const result = await installStarterPack.mutateAsync({ workspaceId })
+      toast.success(`已安装 ${result.created_count} 个自动化模板，跳过 ${result.skipped_count} 个`)
+    } catch (error) {
+      toast.error(error instanceof Error ? error.message : 'Agent Starter 创建失败')
+    }
+  }
+
   function configureDraft() {
     const draft = automationDraft.trim()
     if (!draft) return
@@ -228,17 +364,49 @@ export default function OperationsAgentsPage() {
     setOpen(true)
   }
 
+  function startAgentCreate(starter?: AgentStarterInput) {
+    setAgentName(starter?.name ?? '')
+    setAgentDescription(starter?.prompt ?? '')
+    setOwningTeamId(teams.data?.length === 1 ? teams.data[0].id : '')
+    setAgentCreateOpen(true)
+  }
+
+  async function submitAgentCreate() {
+    if (!workspaceId || !agentName.trim() || !owningTeamId) return
+    try {
+      const agent = await createOperationsAgent.mutateAsync({
+        workspaceId,
+        data: {
+          name: agentName.trim(),
+          description: agentDescription.trim() || null,
+          owning_team_id: owningTeamId,
+        },
+      })
+      setAgentCreateOpen(false)
+      setView('agents')
+      setSelectedAgent(agent)
+      setAgentDetailView('contract')
+      toast.success('Operations Agent 已创建；请确认 Runtime Contract 后发布')
+    } catch (error) {
+      toast.error(error instanceof Error ? error.message : 'Operations Agent 创建失败')
+    }
+  }
+
   async function submitCreate() {
     if (!workspaceId) return
+    const boundAgent = compatibleAgents.find((agent) => agent.id === automationAgentId)
+    if (!boundAgent?.current_published_version) return
     try {
       await createAutomation.mutateAsync({ workspaceId, data: {
+        operations_agent_id: boundAgent.id,
+        operations_agent_version: boundAgent.current_published_version,
         name: name.trim(), prompt: prompt.trim(), precheck: precheck.trim() || null,
-        executor, schedule: `${scheduleKind}@${time}`, timezone: Intl.DateTimeFormat().resolvedOptions().timeZone,
-        session_mode: sessionMode, approval_mode: approvalMode,
+        executor: 'operations-agent', schedule: automationScheduleValue(scheduleKind, time), timezone: Intl.DateTimeFormat().resolvedOptions().timeZone,
+        session_mode: sessionMode, approval_mode: boundAgent.current_profile.mode,
         project: { path: projectPath.trim() || null, branch: branch.trim() || null }, enabled: true,
       } })
       setOpen(false)
-      toast.success('自动化已创建')
+      toast.success('自动化已创建并绑定已发布智能体')
     } catch (error) {
       toast.error(error instanceof Error ? error.message : '创建失败')
     }
@@ -254,6 +422,60 @@ export default function OperationsAgentsPage() {
     }
   }
 
+  function openAutomationRun(automation: Automation) {
+    setAutomationToRun(automation)
+  }
+
+  async function runAutomationNow() {
+    if (!workspaceId || !automationToRun) return
+    try {
+      const run = await startAutomationRun.mutateAsync({
+        workspaceId,
+        automationId: automationToRun.id,
+      })
+      const agent = agents.data?.find(
+        (candidate) => candidate.id === automationToRun.operations_agent_id,
+      )
+      setAutomationToRun(null)
+      if (agent) {
+        setSelectedAgent(agent)
+        setAgentDetailView('activity')
+        setView('agents')
+      }
+      toast.success(`Automation Run ${run.id} 已提交`)
+    } catch (error) {
+      toast.error(error instanceof Error ? error.message : 'Automation Run 启动失败')
+    }
+  }
+
+  function openAutomationBinding(automation: Automation) {
+    setAutomationToBind(automation)
+    setBindingAgentId(automation.operations_agent_id ?? runnableAgents[0]?.id ?? '')
+  }
+
+  async function bindAutomation() {
+    if (!workspaceId || !automationToBind) return
+    const agent = runnableAgents.find((candidate) => candidate.id === bindingAgentId)
+    if (!agent?.current_published_version) return
+    try {
+      await patchAutomation.mutateAsync({
+        workspaceId,
+        automationId: automationToBind.id,
+        data: {
+          operations_agent_id: agent.id,
+          operations_agent_version: agent.current_published_version,
+          approval_mode: agent.current_profile.mode,
+          executor: 'operations-agent',
+          enabled: true,
+        },
+      })
+      setAutomationToBind(null)
+      toast.success(`已绑定 ${agent.name} v${agent.current_published_version}`)
+    } catch (error) {
+      toast.error(error instanceof Error ? error.message : '智能体绑定失败')
+    }
+  }
+
   async function startRun() {
     if (!workspaceId || !selectedAgent || !runTargetType.trim() || !runTargetId.trim()) return
     try {
@@ -288,11 +510,38 @@ export default function OperationsAgentsPage() {
           <label className="relative"><select value={workspaceId ?? ''} onChange={(event) => setWorkspaceId(event.target.value)} className="h-9 appearance-none rounded-lg border bg-background py-1 pl-3 pr-8 text-xs" aria-label="选择 Workspace">{workspaces.data.map((workspace) => <option key={workspace.id} value={workspace.id}>{workspace.name}</option>)}</select><ChevronDown className="pointer-events-none absolute right-2.5 top-2.5 size-4 text-muted-foreground" /></label>
         </header>
 
-        <div className="mt-10 flex items-center gap-1 border-b border-white/[0.08]">
-          <button type="button" onClick={() => setView('automations')} className={cn('border-b-2 px-4 py-3 text-sm transition-colors', view === 'automations' ? 'border-foreground text-foreground' : 'border-transparent text-muted-foreground hover:text-foreground')}><CalendarClock className="mr-2 inline size-4" />自动化</button>
-          <button type="button" onClick={() => setView('agents')} className={cn('border-b-2 px-4 py-3 text-sm transition-colors', view === 'agents' ? 'border-foreground text-foreground' : 'border-transparent text-muted-foreground hover:text-foreground')}><Bot className="mr-2 inline size-4" />智能体</button>
+        <div className="mt-10 flex items-center gap-1 border-b border-white/[0.08]" role="tablist" aria-label="自动化与智能体视图">
+          <button type="button" role="tab" aria-selected={view === 'automations'} onClick={() => setView('automations')} className={cn('border-b-2 px-4 py-3 text-sm transition-colors', view === 'automations' ? 'border-foreground text-foreground' : 'border-transparent text-muted-foreground hover:text-foreground')}><CalendarClock className="mr-2 inline size-4" />自动化</button>
+          <button type="button" role="tab" aria-selected={view === 'agents'} onClick={() => setView('agents')} className={cn('border-b-2 px-4 py-3 text-sm transition-colors', view === 'agents' ? 'border-foreground text-foreground' : 'border-transparent text-muted-foreground hover:text-foreground')}><Bot className="mr-2 inline size-4" />智能体</button>
         </div>
 
+        <section className="mt-8" aria-labelledby="agent-starters-title">
+          <div className="mb-4 flex flex-wrap items-end justify-between gap-3">
+            <div>
+              <p className="font-mono text-[10px] uppercase tracking-[0.16em] text-muted-foreground">SmoothUI / Agent starters</p>
+              <h2 id="agent-starters-title" className="mt-1 text-xl font-medium tracking-[-0.015em]">添加 Operations Agent</h2>
+              <p className="mt-1 text-sm text-muted-foreground">Starter 卡片创建真实 Agent 身份；随后确认 Runtime Contract 并发布即可运行。</p>
+            </div>
+            <Button size="sm" variant="outline" onClick={() => void installAgentStarters()} disabled={installStarterPack.isPending}>
+              {installStarterPack.isPending ? '正在安装…' : '安装三个自动化模板'}
+            </Button>
+          </div>
+          <div className="grid gap-3 lg:grid-cols-3">
+            {AGENT_STARTERS.map((starter) => (
+              <SwitchboardCard
+                key={starter.name}
+                title={starter.name}
+                subtitle={starter.subtitle}
+                columns={18}
+                rows={5}
+                gridPattern={[...starter.pattern]}
+                className="h-[260px] p-4"
+                onButtonClick={() => startAgentCreate(starter)}
+              />
+            ))}
+          </div>
+        </section>
+
         {view === 'automations' ? (
           <div>
             <section className="mt-10">
@@ -303,17 +552,53 @@ export default function OperationsAgentsPage() {
                   <Textarea value={automationDraft} onChange={(event) => setAutomationDraft(event.target.value)} onKeyDown={(event) => { if (event.key === 'Enter' && !event.shiftKey) { event.preventDefault(); configureDraft() } }} placeholder="例如：每个工作日检查失败任务，把需要处理的项目送到 Inbox" aria-label="描述自动化任务" className="min-h-12 resize-none border-0 bg-transparent px-0 py-3 shadow-none focus-visible:ring-0" />
                   <Button size="icon" className="mb-0.5 rounded-full" aria-label="配置自动化" disabled={!automationDraft.trim()} onClick={configureDraft}><ArrowUp /></Button>
                 </div>
-                <div className="mt-7 space-y-1">{SUGGESTIONS.map((item) => { const Icon = item.icon; return <button key={item.name} type="button" onClick={() => startCreate(item)} className="group flex w-full items-start gap-4 rounded-xl px-3 py-3 text-left transition-colors hover:bg-white/[0.04]"><Icon className={`mt-0.5 size-5 ${item.color}`} /><span className="flex-1"><span className="block text-sm font-medium">{item.name} <span className="ml-2 font-normal text-muted-foreground">{scheduleText(item.schedule)}</span></span><span className="mt-1 block text-sm text-muted-foreground">{item.prompt}</span></span><Plus className="mt-2 size-4 opacity-0 transition-opacity group-hover:opacity-100" /></button> })}</div>
+                <div className="mt-7 space-y-1">{SUGGESTIONS.map((item) => { const Icon = item.icon; return <button key={item.name} type="button" onClick={() => startCreate(item)} className="group flex w-full items-start gap-4 rounded-xl px-3 py-3 text-left transition-colors hover:bg-white/[0.04]"><Icon className={`mt-0.5 size-5 ${item.color}`} /><span className="flex-1"><span className="block text-sm font-medium">{item.name} <span className="ml-2 font-normal text-muted-foreground">{automationScheduleText(item.schedule)}</span></span><span className="mt-1 block text-sm text-muted-foreground">{item.prompt}</span></span><Plus className="mt-2 size-4 opacity-0 transition-opacity group-hover:opacity-100" /></button> })}</div>
               </div>
             </section>
             <section className="mt-12">
               <div className="mb-4 flex items-center justify-between"><h2 className="text-sm font-medium text-muted-foreground">我的自动化</h2><Button variant="ghost" size="sm" onClick={() => startCreate()}><Plus />手动配置</Button></div>
-              {automations.isLoading ? <LoadingState rows={3} /> : automations.isError ? <ErrorState message={(automations.error as Error)?.message} hint={BACKEND_HINT} /> : !automations.data?.length ? <EmptyState title="还没有自动化" description="使用模板或创建一个新的定时任务。" /> : <div className="divide-y divide-white/[0.06]">{automations.data.map((automation) => { const meta = executorMeta(automation.executor); const Icon = meta.icon; return <div key={automation.id} className="flex items-start gap-4 rounded-lg px-3 py-4 hover:bg-white/[0.03]"><span className={cn('mt-0.5 flex size-7 items-center justify-center rounded-md bg-white/[0.05]', meta.color)}><Icon className="size-4" /></span><div className="min-w-0 flex-1"><div className="flex flex-wrap items-center gap-3"><span className="text-sm font-medium">{automation.name}</span><span className={cn('rounded-full px-2 py-0.5 text-[11px]', automation.enabled ? 'bg-white/[0.06] text-muted-foreground' : 'bg-amber-400/10 text-amber-300')}>{automation.enabled ? '等待首次运行' : '已暂停'}</span><span className="text-xs text-muted-foreground">{scheduleText(automation.schedule)}</span><span className="text-xs text-muted-foreground">{meta.name}</span></div><p className="mt-1 truncate text-sm text-muted-foreground">{automation.prompt}</p></div><Button size="icon-sm" variant="ghost" aria-label={automation.enabled ? `暂停 ${automation.name}` : `恢复 ${automation.name}`} onClick={() => void toggleAutomation(automation)}>{automation.enabled ? <Pause /> : <Play />}</Button></div> })}</div>}
+              {automations.isLoading ? (
+                <LoadingState rows={3} />
+              ) : automations.isError ? (
+                <ErrorState message={(automations.error as Error)?.message} hint={BACKEND_HINT} />
+              ) : !automations.data?.length ? (
+                <EmptyState title="还没有自动化" description="使用模板或创建一个新的定时任务。" />
+              ) : (
+                <div className="divide-y divide-white/[0.06]">
+                  {automations.data.map((automation) => {
+                    const meta = executorMeta(automation.executor)
+                    const Icon = meta.icon
+                    const boundAgent = agents.data?.find(
+                      (agent) => agent.id === automation.operations_agent_id,
+                    )
+                    return (
+                      <div key={automation.id} className="flex items-start gap-4 rounded-lg px-3 py-4 hover:bg-white/[0.03]">
+                        <span className={cn('mt-0.5 flex size-7 items-center justify-center rounded-md bg-white/[0.05]', meta.color)}><Icon className="size-4" /></span>
+                        <div className="min-w-0 flex-1">
+                          <div className="flex flex-wrap items-center gap-3">
+                            <span className="text-sm font-medium">{automation.name}</span>
+                            <span className={cn('rounded-full px-2 py-0.5 text-[11px]', automation.enabled ? 'bg-white/[0.06] text-muted-foreground' : 'bg-amber-400/10 text-amber-300')}>{automation.enabled ? '已启用' : '已暂停'}</span>
+                            <span className="text-xs text-muted-foreground">{automationScheduleText(automation.schedule)}</span>
+                            <span className="text-xs text-muted-foreground">{meta.name}</span>
+                          </div>
+                          <p className="mt-1 truncate text-sm text-muted-foreground">{automation.prompt}</p>
+                          <p className="mt-1 text-xs text-muted-foreground">{boundAgent ? `绑定 ${boundAgent.name} · Agent v${automation.operations_agent_version} · Automation r${automation.revision}` : '尚未绑定已发布 Operations Agent；调度保持暂停'}</p>
+                        </div>
+                        <div className="flex items-center gap-1">
+                          <Button size="sm" variant="ghost" onClick={() => openAutomationBinding(automation)}>{boundAgent ? '更换绑定' : '绑定智能体'}</Button>
+                          <Button size="sm" variant="outline" disabled={!automation.enabled || !boundAgent} onClick={() => openAutomationRun(automation)}><Play />立即运行</Button>
+                          <Button size="icon-sm" variant="ghost" disabled={!automation.enabled && !boundAgent} aria-label={automation.enabled ? `暂停 ${automation.name}` : `恢复 ${automation.name}`} onClick={() => void toggleAutomation(automation)}>{automation.enabled ? <Pause /> : <Play />}</Button>
+                        </div>
+                      </div>
+                    )
+                  })}
+                </div>
+              )}
             </section>
           </div>
         ) : (
           <section className="mt-10">
-            <div className="mb-7 flex items-start justify-between gap-4"><div><h2 className="text-xl font-medium tracking-[-0.015em]">智能体正在做什么</h2><p className="mt-1 text-sm text-muted-foreground">状态每 5 秒更新；需要你决定的动作会进入 Inbox。</p></div><Link href="/agents" className={buttonVariants()}><Plus />添加智能体</Link></div>
+            <div className="mb-7 flex items-start justify-between gap-4"><div><h2 className="text-xl font-medium tracking-[-0.015em]">智能体正在做什么</h2><p className="mt-1 text-sm text-muted-foreground">状态每 5 秒更新；需要你决定的动作会进入 Inbox。</p></div><Button onClick={() => startAgentCreate()}><Plus />添加智能体</Button></div>
             {agents.isLoading ? <LoadingState rows={3} /> : agents.isError ? <ErrorState message={(agents.error as Error)?.message} hint={BACKEND_HINT} /> : !agents.data?.length ? <EmptyState title="还没有智能体" description="添加一个智能体后，它的当前任务和最近活动会显示在这里。" /> : <div className="divide-y divide-white/[0.07] border-y border-white/[0.07]">{agents.data.map((agent) => { const run = latestRun.get(agent.id); const working = run?.status === 'running' || run?.status === 'queued'; return <article key={agent.id}><button type="button" onClick={() => { setSelectedAgent(agent); setAgentDetailView('activity') }} className="group flex w-full items-start gap-4 px-2 py-5 text-left transition-colors hover:bg-white/[0.025] focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-inset focus-visible:ring-white/30"><div className="relative size-10 shrink-0"><AgentAvatar seed={agent.id} size={40} />{working ? <span className="absolute -right-0.5 -top-0.5 size-2.5 animate-pulse rounded-full border-2 border-[#0f1012] bg-emerald-400" /> : null}</div><div className="min-w-0 flex-1"><div className="flex flex-wrap items-center gap-x-3 gap-y-1"><h3 className="text-sm font-medium">{agent.name}</h3><span className={cn('text-xs', working ? 'text-emerald-400' : agent.disabled ? 'text-amber-300' : 'text-muted-foreground')}>{agent.disabled ? '已停用' : working ? '工作中' : '空闲'}</span></div><p className="mt-1 text-sm text-muted-foreground">{run ? `${run.target_resource_type} · ${run.target_resource_id}` : agent.description || '等待任务'}</p>{run ? <div className="mt-3 flex items-center gap-2 text-xs text-muted-foreground"><CircleDot className="size-3" />{run.trigger_type} · {run.status}</div> : <div className="mt-3 text-xs text-muted-foreground">暂无进行中的任务</div>}</div><ChevronDown className="mt-2 size-4 -rotate-90 text-muted-foreground opacity-0 transition-opacity group-hover:opacity-100" /></button></article> })}</div>}
           </section>
         )}
@@ -325,13 +610,66 @@ export default function OperationsAgentsPage() {
           <div className="grid gap-5 py-2">
             <Input value={name} onChange={(event) => setName(event.target.value)} placeholder="自动化名称" className="text-base font-medium" autoFocus />
             <label className="space-y-1.5 text-xs text-muted-foreground">提示词<Textarea value={prompt} onChange={(event) => setPrompt(event.target.value)} placeholder="每次运行要完成什么？" className="min-h-40 resize-y text-sm" /></label>
-            <fieldset><legend className="mb-2 text-xs text-muted-foreground">选择智能体</legend><div className="grid grid-cols-2 gap-2 sm:grid-cols-4">{EXECUTORS.map((item) => { const Icon = item.icon; return <button key={item.id} type="button" aria-pressed={executor === item.id} onClick={() => setExecutor(item.id)} className={cn('flex items-center gap-2 rounded-lg border px-3 py-3 text-left text-sm', executor === item.id ? 'border-foreground bg-white/[0.06]' : 'border-white/[0.08] hover:bg-white/[0.03]')}><Icon className={cn('size-4', item.color)} />{item.name}</button> })}</div></fieldset>
+            <label className="space-y-1.5 text-xs text-muted-foreground">绑定已发布智能体<select value={automationAgentId} onChange={(event) => setAutomationAgentId(event.target.value)} className="h-10 w-full rounded-lg border bg-background px-3 text-sm text-foreground"><option value="">选择与审批模式兼容的智能体</option>{compatibleAgents.map((agent) => <option key={agent.id} value={agent.id}>{agent.name} · v{agent.current_published_version} · {APPROVALS.find((item) => item.id === agent.current_profile.mode)?.label}</option>)}</select></label>
             <div className="grid gap-4 sm:grid-cols-2"><label className="space-y-1.5 text-xs text-muted-foreground">项目路径<Input value={projectPath} onChange={(event) => setProjectPath(event.target.value)} placeholder="/workspace/project" /></label><label className="space-y-1.5 text-xs text-muted-foreground">基础分支<Input value={branch} onChange={(event) => setBranch(event.target.value)} /></label></div>
             <div className="grid gap-4 sm:grid-cols-3"><label className="space-y-1.5 text-xs text-muted-foreground">日程<select value={scheduleKind} onChange={(event) => setScheduleKind(event.target.value)} className="h-9 w-full rounded-lg border bg-background px-3 text-sm"><option value="hourly">每小时</option><option value="daily">每天</option><option value="weekdays">工作日</option><option value="weekly">每周</option></select></label><label className="space-y-1.5 text-xs text-muted-foreground">时间<Input type="time" value={time} onChange={(event) => setTime(event.target.value)} disabled={scheduleKind === 'hourly'} /></label><label className="space-y-1.5 text-xs text-muted-foreground">会话<select value={sessionMode} onChange={(event) => setSessionMode(event.target.value as 'fresh' | 'reuse')} className="h-9 w-full rounded-lg border bg-background px-3 text-sm"><option value="fresh">每次新会话</option><option value="reuse">重复利用会话</option></select></label></div>
-            <fieldset><legend className="mb-2 text-xs text-muted-foreground">审批方式</legend><div className="grid gap-2 sm:grid-cols-3">{APPROVALS.map((item) => <button key={item.id} type="button" onClick={() => setApprovalMode(item.id)} className={cn('rounded-lg border p-3 text-left', approvalMode === item.id ? 'border-foreground bg-white/[0.06]' : 'border-white/[0.08]')}><span className="block text-sm font-medium">{item.label}</span><span className="mt-1 block text-xs text-muted-foreground">{item.detail}</span></button>)}</div></fieldset>
+            <fieldset><legend className="mb-2 text-xs text-muted-foreground">审批方式</legend><div className="grid gap-2 sm:grid-cols-2">{APPROVALS.filter((item) => item.id !== 'low_risk_automatic').map((item) => <button key={item.id} type="button" aria-pressed={approvalMode === item.id} onClick={() => setApprovalMode(item.id)} className={cn('rounded-lg border p-3 text-left', approvalMode === item.id ? 'border-foreground bg-white/[0.06]' : 'border-white/[0.08]')}><span className="block text-sm font-medium">{item.label}</span><span className="mt-1 block text-xs text-muted-foreground">{item.detail}</span></button>)}</div></fieldset>
             <details className="rounded-lg border border-white/[0.08] px-4 py-3"><summary className="cursor-pointer text-sm text-muted-foreground">高级设置</summary><label className="mt-4 block space-y-1.5 text-xs text-muted-foreground"><Terminal className="mr-1 inline size-3" />预检查<Input value={precheck} onChange={(event) => setPrecheck(event.target.value)} placeholder="可选：运行前检查命令" className="font-mono" /></label></details>
           </div>
-          <DialogFooter><Button variant="outline" onClick={() => setOpen(false)}>取消</Button><Button onClick={() => void submitCreate()} disabled={!name.trim() || !prompt.trim() || createAutomation.isPending}><Plus />创建</Button></DialogFooter>
+          <DialogFooter><Button variant="outline" onClick={() => setOpen(false)}>取消</Button><Button onClick={() => void submitCreate()} disabled={!name.trim() || !prompt.trim() || !automationAgentId || createAutomation.isPending}><Plus />创建并启用</Button></DialogFooter>
+        </DialogContent>
+      </Dialog>
+
+      <Dialog open={agentCreateOpen} onOpenChange={setAgentCreateOpen}>
+        <DialogContent className="sm:max-w-xl">
+          <DialogHeader>
+            <DialogTitle>添加 Operations Agent</DialogTitle>
+            <DialogDescription>创建 Observe Only 身份后，继续配置已连接 Fleet 节点的 Runtime Contract 并发布。</DialogDescription>
+          </DialogHeader>
+          <div className="grid gap-4 py-2">
+            <Input value={agentName} onChange={(event) => setAgentName(event.target.value)} placeholder="智能体名称" autoFocus />
+            <Textarea value={agentDescription} onChange={(event) => setAgentDescription(event.target.value)} placeholder="职责与只读边界" className="min-h-32" />
+            <label className="space-y-1.5 text-xs text-muted-foreground">所属 Team<select value={owningTeamId} onChange={(event) => setOwningTeamId(event.target.value)} className="h-10 w-full rounded-lg border bg-background px-3 text-sm text-foreground"><option value="">选择 Team</option>{(teams.data ?? []).map((team) => <option key={team.id} value={team.id}>{team.name}</option>)}</select></label>
+          </div>
+          <DialogFooter>
+            <Button variant="outline" onClick={() => setAgentCreateOpen(false)}>取消</Button>
+            <Button onClick={() => void submitAgentCreate()} disabled={!agentName.trim() || !owningTeamId || createOperationsAgent.isPending}><Plus />创建并配置</Button>
+          </DialogFooter>
+        </DialogContent>
+      </Dialog>
+
+      <Dialog open={Boolean(automationToRun)} onOpenChange={(nextOpen) => { if (!nextOpen) setAutomationToRun(null) }}>
+        <DialogContent className="sm:max-w-xl">
+          <DialogHeader>
+            <DialogTitle>立即运行自动化</DialogTitle>
+            <DialogDescription>{automationToRun?.name} 将使用已持久化的 Agent/version 绑定执行；Run 会固定 Automation revision 与完整快照。</DialogDescription>
+          </DialogHeader>
+          <p className="rounded-lg border border-white/[0.08] p-3 text-sm text-muted-foreground">
+            Agent {automationToRun?.operations_agent_id} · v{automationToRun?.operations_agent_version} · Automation r{automationToRun?.revision}
+          </p>
+          <DialogFooter>
+            <Button variant="outline" onClick={() => setAutomationToRun(null)}>取消</Button>
+            <Button onClick={() => void runAutomationNow()} disabled={startAutomationRun.isPending}><Play />提交 Run</Button>
+          </DialogFooter>
+        </DialogContent>
+      </Dialog>
+
+      <Dialog open={Boolean(automationToBind)} onOpenChange={(nextOpen) => { if (!nextOpen) setAutomationToBind(null) }}>
+        <DialogContent className="sm:max-w-xl">
+          <DialogHeader>
+            <DialogTitle>绑定 Operations Agent</DialogTitle>
+            <DialogDescription>绑定会固定当前已发布 Agent 版本；Low-Risk Automatic Agent 不可用于 Automation。</DialogDescription>
+          </DialogHeader>
+          <label className="grid gap-2 text-sm text-muted-foreground">已发布智能体
+            <select value={bindingAgentId} onChange={(event) => setBindingAgentId(event.target.value)} className="h-10 rounded-lg border bg-background px-3 text-foreground">
+              <option value="">选择智能体</option>
+              {runnableAgents.map((agent) => <option key={agent.id} value={agent.id}>{agent.name} · v{agent.current_published_version} · {APPROVALS.find((item) => item.id === agent.current_profile.mode)?.label}</option>)}
+            </select>
+          </label>
+          <DialogFooter>
+            <Button variant="outline" onClick={() => setAutomationToBind(null)}>取消</Button>
+            <Button onClick={() => void bindAutomation()} disabled={!bindingAgentId || patchAutomation.isPending}>绑定并启用</Button>
+          </DialogFooter>
         </DialogContent>
       </Dialog>
 
@@ -340,7 +678,7 @@ export default function OperationsAgentsPage() {
           {selectedAgent ? <>
             <DialogHeader className="border-b border-white/[0.08] px-6 py-4">
               <div className="flex items-center gap-3 pr-8"><AgentAvatar seed={selectedAgent.id} size={36} /><div className="min-w-0"><DialogTitle className="flex items-center gap-2 text-base"><span className="truncate">{selectedAgent.name}</span><span className={cn('text-xs font-normal', selectedAgent.disabled ? 'text-amber-300' : latestRun.get(selectedAgent.id)?.status === 'running' ? 'text-emerald-400' : 'text-muted-foreground')}>{selectedAgent.disabled ? '已停用' : latestRun.get(selectedAgent.id)?.status === 'running' ? '工作中' : '空闲'}</span></DialogTitle><DialogDescription className="mt-0.5">CLI Agent 会话 · {APPROVALS.find((item) => item.id === selectedAgent.current_profile.mode)?.label ?? selectedAgent.current_profile.mode}</DialogDescription></div></div>
-              <div className="mt-3 flex gap-1"><button type="button" onClick={() => setAgentDetailView('activity')} className={cn('rounded-md px-3 py-1.5 text-xs', agentDetailView === 'activity' ? 'bg-white/[0.08] text-foreground' : 'text-muted-foreground hover:text-foreground')}>活动</button><button type="button" onClick={() => setAgentDetailView('contract')} className={cn('rounded-md px-3 py-1.5 text-xs', agentDetailView === 'contract' ? 'bg-white/[0.08] text-foreground' : 'text-muted-foreground hover:text-foreground')}>Contract</button></div>
+              <div className="mt-3 flex gap-1" role="tablist" aria-label="智能体详情"><button type="button" role="tab" aria-selected={agentDetailView === 'activity'} onClick={() => setAgentDetailView('activity')} className={cn('rounded-md px-3 py-1.5 text-xs', agentDetailView === 'activity' ? 'bg-white/[0.08] text-foreground' : 'text-muted-foreground hover:text-foreground')}>活动</button><button type="button" role="tab" aria-selected={agentDetailView === 'contract'} onClick={() => setAgentDetailView('contract')} className={cn('rounded-md px-3 py-1.5 text-xs', agentDetailView === 'contract' ? 'bg-white/[0.08] text-foreground' : 'text-muted-foreground hover:text-foreground')}>Contract</button></div>
             </DialogHeader>
             {agentDetailView === 'activity' ? <div className="grid min-h-0 flex-1 md:grid-cols-[minmax(0,1fr)_250px]">
               <div className="flex min-h-0 flex-col bg-[#090a0b]">
@@ -361,7 +699,7 @@ export default function OperationsAgentsPage() {
                   </details>
                 </div>
               </div>
-              <aside className="border-l border-white/[0.08] bg-white/[0.015] p-5 text-xs"><div className="text-muted-foreground">职责</div><p className="mt-2 text-sm leading-6 text-foreground">{selectedAgent.description || '尚未填写职责说明'}</p><div className="mt-6 text-muted-foreground">权限模式</div><p className="mt-2 text-sm text-foreground">{APPROVALS.find((item) => item.id === selectedAgent.current_profile.mode)?.label}</p><p className="mt-1 leading-5 text-muted-foreground">Profile v{selectedAgent.current_profile.version} · {selectedAgent.current_profile.reason}</p><Link href="/agents" className={cn(buttonVariants({ variant: 'outline', size: 'sm' }), 'mt-6 w-full')}>配置智能体</Link></aside>
+              <aside className="border-l border-white/[0.08] bg-white/[0.015] p-5 text-xs"><div className="text-muted-foreground">职责</div><p className="mt-2 text-sm leading-6 text-foreground">{selectedAgent.description || '尚未填写职责说明'}</p><div className="mt-6 text-muted-foreground">权限模式</div><p className="mt-2 text-sm text-foreground">{APPROVALS.find((item) => item.id === selectedAgent.current_profile.mode)?.label}</p><p className="mt-1 leading-5 text-muted-foreground">Profile v{selectedAgent.current_profile.version} · {selectedAgent.current_profile.reason}</p><Button variant="outline" size="sm" className="mt-6 w-full" onClick={() => setAgentDetailView('contract')}>配置智能体</Button></aside>
             </div> : workspaceId ? <ContractEditor workspaceId={workspaceId} agent={selectedAgent} /> : null}
           </> : null}
         </DialogContent>
diff --git a/frontend/app/(app)/records/graph/page.tsx b/frontend/app/(app)/records/graph/page.tsx
index 39de08e5..c33ee4cd 100644
--- a/frontend/app/(app)/records/graph/page.tsx
+++ b/frontend/app/(app)/records/graph/page.tsx
@@ -13,6 +13,7 @@ import {
   Waypoints,
 } from 'lucide-react'
 
+import { GpuSurface } from '@/components/gpu/gpu-surface'
 import { BACKEND_HINT, EmptyState, ErrorState, LoadingState } from '@/components/shell/data-states'
 import { PageContainer } from '@/components/shell/page-container'
 import { DATA_EXPLORER_TABS, RouteTabs } from '@/components/shell/route-tabs'
@@ -26,7 +27,7 @@ import {
   SelectValue,
 } from '@/components/ui/select'
 import { useMyWorkspaces, useProjectRecordGraph, useWorkspaceProjects } from '@/lib/api/hooks'
-import type { RecordGraphNode } from '@/lib/api/types'
+import type { ProjectRecordGraphPreview, RecordGraphNode } from '@/lib/api/types'
 import {
   RECORD_GRAPH_KIND_COLOR,
   RECORD_GRAPH_KIND_LABEL,
@@ -58,6 +59,61 @@ function selectedNodeDescription(node: RecordGraphNode) {
   return node.subtitle ?? RECORD_GRAPH_KIND_LABEL[node.kind]
 }
 
+function RecordGraphFallback({
+  preview,
+  selectedNodeId,
+  onSelectNode,
+}: {
+  preview: ProjectRecordGraphPreview
+  selectedNodeId: string | null
+  onSelectNode: (nodeId: string | null) => void
+}) {
+  const initialNodes = preview.nodes.slice(0, 200)
+
+  return (
+    <div
+      className="h-full min-h-[38rem] overflow-y-auto bg-muted/20 p-4"
+      data-record-graph-fallback
+    >
+      <div className="rounded-lg border bg-background p-4">
+        <h2 className="text-sm font-semibold">WebGL 图谱不可用</h2>
+        <p className="mt-1 text-sm text-muted-foreground">
+          当前显示前 200 个节点；可使用上方搜索定位其余节点，并继续查看双向邻居。
+        </p>
+      </div>
+      <div className="mt-4 space-y-2">
+        {initialNodes.map((node) => (
+          <button
+            key={node.id}
+            type="button"
+            aria-pressed={selectedNodeId === node.id}
+            onClick={() => onSelectNode(node.id)}
+            className={`w-full rounded-lg border bg-background p-3 text-left transition-colors hover:bg-muted focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring ${
+              selectedNodeId === node.id ? 'border-primary bg-primary/5' : ''
+            }`}
+          >
+            <span className="flex items-center justify-between gap-3">
+              <span className="flex min-w-0 items-center gap-2">
+                <span
+                  className="size-2 shrink-0 rounded-full"
+                  style={{ backgroundColor: RECORD_GRAPH_KIND_COLOR[node.kind] }}
+                />
+                <span className="truncate text-sm font-medium">{node.label}</span>
+              </span>
+              <span className="shrink-0 font-mono text-xs text-muted-foreground">
+                {node.count.toLocaleString('zh-CN')}
+              </span>
+            </span>
+            <span className="mt-1 block text-xs text-muted-foreground">
+              {RECORD_GRAPH_KIND_LABEL[node.kind]}
+            </span>
+          </button>
+        ))}
+      </div>
+    </div>
+  )
+}
+
 export default function RecordRelationshipGraphPage() {
   const [selectedProjectId, setSelectedProjectId] = useState<string | null>(null)
   const [selectedNodeId, setSelectedNodeId] = useState<string | null>(null)
@@ -258,15 +314,29 @@ export default function RecordRelationshipGraphPage() {
         ) : (
           <div className="grid h-[70rem] min-h-0 grid-rows-[38rem_32rem] lg:h-[42rem] lg:grid-cols-[minmax(0,1fr)_20rem] lg:grid-rows-1">
             <div className="relative min-h-0 overflow-hidden bg-[#09090b]">
-              <ProjectRecordGraphCanvas
-                preview={preview}
-                selectedNodeId={selectedNodeId}
-                onSelectNode={setSelectedNodeId}
-              />
-              <div className="pointer-events-none absolute left-3 top-3 flex items-center gap-2 rounded-md border border-white/10 bg-black/60 px-2.5 py-1.5 text-xs text-zinc-400 backdrop-blur-sm">
-                <Waypoints className="size-3.5" />
-                滚轮缩放 · 双击聚焦 · 点击查看双向邻居
-              </div>
+              <GpuSurface
+                surface="record-relationship-graph"
+                className="h-full"
+                fallback={(
+                  <RecordGraphFallback
+                    preview={preview}
+                    selectedNodeId={selectedNodeId}
+                    onSelectNode={setSelectedNodeId}
+                  />
+                )}
+              >
+                <>
+                  <ProjectRecordGraphCanvas
+                    preview={preview}
+                    selectedNodeId={selectedNodeId}
+                    onSelectNode={setSelectedNodeId}
+                  />
+                  <div className="pointer-events-none absolute left-3 top-3 flex items-center gap-2 rounded-md border border-white/10 bg-black/60 px-2.5 py-1.5 text-xs text-zinc-400 backdrop-blur-sm">
+                    <Waypoints className="size-3.5" />
+                    滚轮缩放 · 双击聚焦 · 点击查看双向邻居
+                  </div>
+                </>
+              </GpuSurface>
             </div>
 
             <aside
diff --git a/frontend/app/(app)/schedules/page.tsx b/frontend/app/(app)/schedules/page.tsx
index 495f06b7..ccc3dc7c 100644
--- a/frontend/app/(app)/schedules/page.tsx
+++ b/frontend/app/(app)/schedules/page.tsx
@@ -1,18 +1,5 @@
-'use client'
-
-import { ScheduleListPanel } from '@/components/schedules/schedule-list-panel'
-import { PageContainer } from '@/components/shell/page-container'
-import { AUTOMATION_TABS, RouteTabs } from '@/components/shell/route-tabs'
+import { redirect } from 'next/navigation'
 
 export default function SchedulesPage() {
-  return (
-    <PageContainer
-      eyebrow="Automation"
-      title="自动化与 Agent"
-      description="管理自动化链路的定时触发计划。"
-      tabs={<RouteTabs tabs={AUTOMATION_TABS} />}
-    >
-      <ScheduleListPanel />
-    </PageContainer>
-  )
+  redirect('/operations-agents')
 }
diff --git a/frontend/app/(app)/settings/page.tsx b/frontend/app/(app)/settings/page.tsx
index be4e27a3..71dc5cc8 100644
--- a/frontend/app/(app)/settings/page.tsx
+++ b/frontend/app/(app)/settings/page.tsx
@@ -1,432 +1,5 @@
-'use client'
-
-import { useEffect, useState } from 'react'
-import { AlertTriangle, Loader2, RotateCcw, Save } from 'lucide-react'
-import { toast } from 'sonner'
-
-import { useResetWorkspaceSettings, useUpdateWorkspaceSettings, useWorkspaceSettings } from '@/lib/api/hooks'
-import type { WorkspaceSettingsRead, WorkspaceSettingsValues } from '@/lib/api/types'
-import { BACKEND_HINT, ErrorState, LoadingState } from '@/components/shell/data-states'
-import { PageContainer } from '@/components/shell/page-container'
-import { Badge } from '@/components/ui/badge'
-import { Button } from '@/components/ui/button'
-import { Card, CardContent, CardDescription, CardHeader, CardTitle } from '@/components/ui/card'
-import {
-  Dialog,
-  DialogContent,
-  DialogDescription,
-  DialogFooter,
-  DialogHeader,
-  DialogTitle,
-} from '@/components/ui/dialog'
-import { Field, FieldDescription, FieldGroup, FieldLabel } from '@/components/ui/field'
-import { Input } from '@/components/ui/input'
-import { Select, SelectContent, SelectItem, SelectTrigger, SelectValue } from '@/components/ui/select'
-import { Switch } from '@/components/ui/switch'
-
-const THEME_OPTIONS = [
-  { value: 'system', label: '跟随系统' },
-  { value: 'dark', label: '深色' },
-  { value: 'light', label: '浅色' },
-] as const
-
-const SIDEBAR_OPTIONS = [
-  { value: 'expanded', label: '展开' },
-  { value: 'icon', label: '仅图标' },
-  { value: 'collapsed', label: '收起' },
-] as const
-
-const LANDING_OPTIONS = [
-  { value: '/dashboard', label: '概览' },
-  { value: '/canvas', label: '节点画布' },
-  { value: '/inbox', label: '任务与通知' },
-] as const
-
-const RETENTION_OPTIONS = [7, 30, 90, 365] as const
-
-/** Small provenance hint next to a field's label — "已自定义" when the value
- *  has been overridden from the system default, "下次运行生效" when this
- *  particular field only takes effect on the next scheduled run rather than
- *  immediately (both come from WorkspaceSettingsRead, not guessed client-side). */
-function FieldMeta({ read, field }: { read: WorkspaceSettingsRead; field: keyof WorkspaceSettingsValues }) {
-  const isOverride = read.sources[field] === 'override'
-  const isNextRun = read.apply_modes[field] === 'next_run'
-  if (!isOverride && !isNextRun) return null
-  return (
-    <span className="inline-flex items-center gap-1.5">
-      {isOverride ? (
-        <Badge variant="outline" className="h-4 px-1.5 text-[10px] font-normal">
-          已自定义
-        </Badge>
-      ) : null}
-      {isNextRun ? <span className="text-[10px] text-muted-foreground">下次运行生效</span> : null}
-    </span>
-  )
-}
+import { redirect } from 'next/navigation'
 
 export default function SettingsPage() {
-  const { data, isLoading, isError, error } = useWorkspaceSettings()
-  const updateMutation = useUpdateWorkspaceSettings()
-  const resetMutation = useResetWorkspaceSettings()
-  const [form, setForm] = useState<WorkspaceSettingsValues | null>(null)
-  const [confirmReset, setConfirmReset] = useState(false)
-
-  // Sync local form state whenever the server value changes — on first load,
-  // and again after our own save/reset mutations invalidate the query. Safe
-  // to re-run: refetchOnWindowFocus is off (components/providers.tsx), so
-  // this doesn't clobber an in-progress edit from a background refetch.
-  useEffect(() => {
-    if (data?.values) setForm(data.values)
-  }, [data?.values])
-
-  const set = <K extends keyof WorkspaceSettingsValues>(key: K, value: WorkspaceSettingsValues[K]) => {
-    setForm((current) => (current ? { ...current, [key]: value } : current))
-  }
-
-  const hasChanges = Boolean(
-    form && data?.values && Object.keys(form).some((key) => {
-      const k = key as keyof WorkspaceSettingsValues
-      return form[k] !== data.values[k]
-    }),
-  )
-
-  const handleSave = (event: React.FormEvent) => {
-    event.preventDefault()
-    if (!form || !data?.values) return
-    const changed: Partial<WorkspaceSettingsValues> = {}
-    for (const key of Object.keys(form) as (keyof WorkspaceSettingsValues)[]) {
-      if (form[key] !== data.values[key]) {
-        Object.assign(changed, { [key]: form[key] })
-      }
-    }
-    if (Object.keys(changed).length === 0) return
-    updateMutation.mutate(changed, {
-      onSuccess: () => toast.success('设置已保存'),
-      onError: (cause) => toast.error((cause as Error).message),
-    })
-  }
-
-  const handleReset = () => {
-    resetMutation.mutate(undefined, {
-      onSuccess: () => {
-        toast.success('已重置为默认设置')
-        setConfirmReset(false)
-      },
-      onError: (cause) => toast.error((cause as Error).message),
-    })
-  }
-
-  return (
-    <PageContainer
-      eyebrow="Workspace"
-      title="工作区设置"
-      description="配置外观、执行、数据保留与通知偏好；随时可重置为默认值。"
-    >
-      {isLoading ? (
-        <LoadingState />
-      ) : isError || !data ? (
-        <ErrorState message={(error as Error)?.message} hint={BACKEND_HINT} />
-      ) : !form ? (
-        <LoadingState />
-      ) : (
-        <form onSubmit={handleSave} className="flex flex-col gap-6">
-          <div className="flex flex-wrap items-center gap-2 text-xs text-muted-foreground">
-            <Badge variant="secondary">第 {data.revision} 版</Badge>
-            <span>
-              {data.updated_at ? `最近更新于 ${new Date(data.updated_at).toLocaleString()}` : '尚未自定义过设置'}
-            </span>
-          </div>
-
-          <Card>
-            <CardHeader>
-              <CardTitle className="text-base">外观与导航</CardTitle>
-              <CardDescription>控制台的显示方式和登录后默认落地页。</CardDescription>
-            </CardHeader>
-            <CardContent>
-              <FieldGroup className="gap-5">
-                <div className="grid gap-4 sm:grid-cols-2">
-                  <Field>
-                    <FieldLabel htmlFor="settings-theme">
-                      主题
-                      <FieldMeta read={data} field="theme" />
-                    </FieldLabel>
-                    <Select value={form.theme} onValueChange={(value) => set('theme', value as WorkspaceSettingsValues['theme'])}>
-                      <SelectTrigger id="settings-theme" className="w-full">
-                        <SelectValue>
-                          {(value: WorkspaceSettingsValues['theme'] | null) =>
-                            value ? (THEME_OPTIONS.find((option) => option.value === value)?.label ?? value) : '选择主题'
-                          }
-                        </SelectValue>
-                      </SelectTrigger>
-                      <SelectContent>
-                        {THEME_OPTIONS.map((option) => (
-                          <SelectItem key={option.value} value={option.value}>
-                            {option.label}
-                          </SelectItem>
-                        ))}
-                      </SelectContent>
-                    </Select>
-                  </Field>
-
-                  <Field>
-                    <FieldLabel htmlFor="settings-sidebar-mode">
-                      侧边栏
-                      <FieldMeta read={data} field="sidebar_mode" />
-                    </FieldLabel>
-                    <Select
-                      value={form.sidebar_mode}
-                      onValueChange={(value) => set('sidebar_mode', value as WorkspaceSettingsValues['sidebar_mode'])}
-                    >
-                      <SelectTrigger id="settings-sidebar-mode" className="w-full">
-                        <SelectValue>
-                          {(value: WorkspaceSettingsValues['sidebar_mode'] | null) =>
-                            value ? (SIDEBAR_OPTIONS.find((option) => option.value === value)?.label ?? value) : '选择侧边栏'
-                          }
-                        </SelectValue>
-                      </SelectTrigger>
-                      <SelectContent>
-                        {SIDEBAR_OPTIONS.map((option) => (
-                          <SelectItem key={option.value} value={option.value}>
-                            {option.label}
-                          </SelectItem>
-                        ))}
-                      </SelectContent>
-                    </Select>
-                  </Field>
-                </div>
-
-                <div className="grid gap-4 sm:grid-cols-2">
-                  <Field>
-                    <FieldLabel htmlFor="settings-landing-page">
-                      默认落地页
-                      <FieldMeta read={data} field="landing_page" />
-                    </FieldLabel>
-                    <Select
-                      value={form.landing_page}
-                      onValueChange={(value) => set('landing_page', value as WorkspaceSettingsValues['landing_page'])}
-                    >
-                      <SelectTrigger id="settings-landing-page" className="w-full">
-                        <SelectValue>
-                          {(value: WorkspaceSettingsValues['landing_page'] | null) =>
-                            value ? (LANDING_OPTIONS.find((option) => option.value === value)?.label ?? value) : '选择默认落地页'
-                          }
-                        </SelectValue>
-                      </SelectTrigger>
-                      <SelectContent>
-                        {LANDING_OPTIONS.map((option) => (
-                          <SelectItem key={option.value} value={option.value}>
-                            {option.label}
-                          </SelectItem>
-                        ))}
-                      </SelectContent>
-                    </Select>
-                  </Field>
-
-                  <Field>
-                    <FieldLabel htmlFor="settings-timezone">
-                      时区
-                      <FieldMeta read={data} field="timezone" />
-                    </FieldLabel>
-                    <Input
-                      id="settings-timezone"
-                      value={form.timezone}
-                      onChange={(event) => set('timezone', event.target.value)}
-                      placeholder="例如 Asia/Shanghai"
-                    />
-                  </Field>
-                </div>
-
-                <Field orientation="horizontal" className="items-center justify-between">
-                  <FieldLabel htmlFor="settings-motion-enabled">
-                    启用界面动效
-                    <FieldMeta read={data} field="motion_enabled" />
-                  </FieldLabel>
-                  <Switch
-                    id="settings-motion-enabled"
-                    checked={form.motion_enabled}
-                    onCheckedChange={(value) => set('motion_enabled', value)}
-                  />
-                </Field>
-              </FieldGroup>
-            </CardContent>
-          </Card>
-
-          <Card>
-            <CardHeader>
-              <CardTitle className="text-base">执行</CardTitle>
-              <CardDescription>调度和工作流运行时的默认行为。</CardDescription>
-            </CardHeader>
-            <CardContent>
-              <FieldGroup className="gap-5">
-                <Field>
-                  <FieldLabel htmlFor="settings-default-concurrency">
-                    默认并发数
-                    <FieldMeta read={data} field="default_concurrency" />
-                  </FieldLabel>
-                  <Input
-                    id="settings-default-concurrency"
-                    type="number"
-                    min={1}
-                    value={form.default_concurrency}
-                    onChange={(event) => {
-                      const next = Number(event.target.value)
-                      set('default_concurrency', Number.isFinite(next) ? next : form.default_concurrency)
-                    }}
-                  />
-                  <FieldDescription>新建调度未单独指定并发数时使用的默认值。</FieldDescription>
-                </Field>
-
-                <Field orientation="horizontal" className="items-center justify-between">
-                  <FieldLabel htmlFor="settings-automatic-retries">
-                    失败自动重试
-                    <FieldMeta read={data} field="automatic_retries" />
-                  </FieldLabel>
-                  <Switch
-                    id="settings-automatic-retries"
-                    checked={form.automatic_retries}
-                    onCheckedChange={(value) => set('automatic_retries', value)}
-                  />
-                </Field>
-              </FieldGroup>
-            </CardContent>
-          </Card>
-
-          <Card>
-            <CardHeader>
-              <CardTitle className="text-base">数据保留</CardTitle>
-              <CardDescription>采集到的原始数据在系统中保存多久。</CardDescription>
-            </CardHeader>
-            <CardContent>
-              <FieldGroup className="gap-5">
-                <Field orientation="horizontal" className="items-center justify-between">
-                  <FieldLabel htmlFor="settings-retain-raw-data">
-                    保留原始数据
-                    <FieldMeta read={data} field="retain_raw_data" />
-                  </FieldLabel>
-                  <Switch
-                    id="settings-retain-raw-data"
-                    checked={form.retain_raw_data}
-                    onCheckedChange={(value) => set('retain_raw_data', value)}
-                  />
-                </Field>
-
-                <Field>
-                  <FieldLabel htmlFor="settings-retention-days">
-                    保留天数
-                    <FieldMeta read={data} field="retention_days" />
-                  </FieldLabel>
-                  <Select
-                    value={String(form.retention_days)}
-                    onValueChange={(value) =>
-                      set('retention_days', Number(value) as WorkspaceSettingsValues['retention_days'])
-                    }
-                    disabled={!form.retain_raw_data}
-                  >
-                    <SelectTrigger id="settings-retention-days" className="w-full">
-                      <SelectValue>
-                        {(value: string | null) => (value ? `${value} 天` : '选择保留天数')}
-                      </SelectValue>
-                    </SelectTrigger>
-                    <SelectContent>
-                      {RETENTION_OPTIONS.map((days) => (
-                        <SelectItem key={days} value={String(days)}>
-                          {days} 天
-                        </SelectItem>
-                      ))}
-                    </SelectContent>
-                  </Select>
-                  <FieldDescription>
-                    {form.retain_raw_data ? '超过此天数的原始数据将被清理。' : '关闭保留后，此项不生效。'}
-                  </FieldDescription>
-                </Field>
-              </FieldGroup>
-            </CardContent>
-          </Card>
-
-          <Card>
-            <CardHeader>
-              <CardTitle className="text-base">通知</CardTitle>
-              <CardDescription>选择哪些事件需要提醒。</CardDescription>
-            </CardHeader>
-            <CardContent>
-              <FieldGroup className="gap-4">
-                <Field orientation="horizontal" className="items-center justify-between">
-                  <FieldLabel htmlFor="settings-inbox-alerts">
-                    收件箱提醒
-                    <FieldMeta read={data} field="inbox_alerts" />
-                  </FieldLabel>
-                  <Switch
-                    id="settings-inbox-alerts"
-                    checked={form.inbox_alerts}
-                    onCheckedChange={(value) => set('inbox_alerts', value)}
-                  />
-                </Field>
-                <Field orientation="horizontal" className="items-center justify-between">
-                  <FieldLabel htmlFor="settings-failure-alerts">
-                    失败提醒
-                    <FieldMeta read={data} field="failure_alerts" />
-                  </FieldLabel>
-                  <Switch
-                    id="settings-failure-alerts"
-                    checked={form.failure_alerts}
-                    onCheckedChange={(value) => set('failure_alerts', value)}
-                  />
-                </Field>
-                <Field orientation="horizontal" className="items-center justify-between">
-                  <FieldLabel htmlFor="settings-agent-alerts">
-                    Agent 提醒
-                    <FieldMeta read={data} field="agent_alerts" />
-                  </FieldLabel>
-                  <Switch
-                    id="settings-agent-alerts"
-                    checked={form.agent_alerts}
-                    onCheckedChange={(value) => set('agent_alerts', value)}
-                  />
-                </Field>
-              </FieldGroup>
-            </CardContent>
-          </Card>
-
-          <div className="flex flex-wrap items-center justify-end gap-2">
-            <Button
-              type="button"
-              variant="outline"
-              onClick={() => setConfirmReset(true)}
-              disabled={resetMutation.isPending}
-              className="gap-1.5"
-            >
-              <RotateCcw className="size-4" />
-              重置为默认值
-            </Button>
-            <Button type="submit" disabled={!hasChanges || updateMutation.isPending} className="gap-1.5">
-              {updateMutation.isPending ? <Loader2 className="size-4 animate-spin" /> : <Save className="size-4" />}
-              {updateMutation.isPending ? '保存中…' : '保存更改'}
-            </Button>
-          </div>
-        </form>
-      )}
-
-      <Dialog open={confirmReset} onOpenChange={setConfirmReset}>
-        <DialogContent className="sm:max-w-md">
-          <DialogHeader>
-            <DialogTitle className="flex items-center gap-2">
-              <AlertTriangle className="size-4 text-destructive" aria-hidden />
-              确认重置为默认值
-            </DialogTitle>
-            <DialogDescription>
-              这会清除所有已自定义的工作区设置，恢复为系统默认值。此操作无法撤销。
-            </DialogDescription>
-          </DialogHeader>
-          <DialogFooter>
-            <Button variant="outline" onClick={() => setConfirmReset(false)} disabled={resetMutation.isPending}>
-              取消
-            </Button>
-            <Button variant="destructive" onClick={handleReset} disabled={resetMutation.isPending}>
-              {resetMutation.isPending ? '正在重置…' : '确认重置'}
-            </Button>
-          </DialogFooter>
-        </DialogContent>
-      </Dialog>
-    </PageContainer>
-  )
+  redirect('/system')
 }
diff --git a/frontend/app/(app)/studio/projects/[projectId]/operations/page.tsx b/frontend/app/(app)/studio/projects/[projectId]/operations/page.tsx
index bed8066a..47874f0a 100644
--- a/frontend/app/(app)/studio/projects/[projectId]/operations/page.tsx
+++ b/frontend/app/(app)/studio/projects/[projectId]/operations/page.tsx
@@ -5,6 +5,7 @@ import Link from 'next/link'
 import { useSearchParams } from 'next/navigation'
 import { use, useDeferredValue, useEffect, useState, type ReactNode } from 'react'
 
+import AITaskList from '@/components/smoothui/ai-task-list'
 import { EmptyState, ErrorState, LoadingState } from '@/components/shell/data-states'
 import { PageContainer } from '@/components/shell/page-container'
 import { ProjectNavigation } from '@/components/studio/project-navigation'
@@ -18,6 +19,7 @@ import { Table, TableBody, TableCell, TableHead, TableHeader, TableRow } from '@
 import { useProjectRuntimeLogs, useProjectRuntimeSummary, useProjectRuntimeTrace, useProjectWorkflows, useWorkspaceProjects } from '@/lib/api/hooks'
 import type { ProjectRuntimeLog } from '@/lib/api/types'
 import { formatDateTime, formatRelative } from '@/lib/format'
+import { buildOperationsNodeTasks } from '@/lib/studio/operations-task-model'
 import { cn } from '@/lib/utils'
 
 const PAGE_SIZE = 20
@@ -75,6 +77,10 @@ export default function ProjectOperationsPage({
   const traceNextCursor = traceQuery.data?.trace.nextAfterSequence ?? traceCursor
   const traceTotalEvents = traceQuery.data?.trace.projection.eventCount ?? selectedLog?.event_count ?? 0
   const traceHasNextPage = traceNextCursor > traceCursor && traceNextCursor < traceTotalEvents
+  const traceStatus = traceQuery.data?.trace.projection.status ?? selectedLog?.status ?? 'queued'
+  const traceNodeTasks = buildOperationsNodeTasks(
+    traceQuery.data?.trace.projection.nodeStates ?? [],
+  )
 
   useEffect(() => {
     setPage(1)
@@ -187,7 +193,7 @@ export default function ProjectOperationsPage({
                 <div className="grid gap-2 sm:grid-cols-4">
                   <TraceMetric label="状态"><StatusBadge status={traceQuery.data?.trace.projection.status ?? selectedLog.status} /></TraceMetric>
                   <TraceMetric label="版本" value={traceQuery.data?.workflow_version ? `Published v${traceQuery.data.workflow_version}` : 'Draft'} />
-                  <TraceMetric label="用户" value={traceQuery.data?.user ?? 'operator'} />
+                  <TraceMetric label="用户" value={traceQuery.data?.user?.trim() || '未提供'} />
                   <TraceMetric label="事件" value={String(traceQuery.data?.trace.projection.eventCount ?? selectedLog.event_count)} />
                 </div>
 
@@ -197,35 +203,68 @@ export default function ProjectOperationsPage({
                 </div>
 
                 <div>
-                  <div className="mb-2 flex items-center gap-2 text-sm font-medium"><Workflow className="size-4" />事件时间线</div>
+                  <div className="mb-2 flex items-center gap-2 text-sm font-medium"><Workflow className="size-4" />节点状态与事件</div>
                   {traceQuery.isLoading ? <LoadingState rows={5} /> : traceQuery.isError ? (
                     <ErrorState message={traceQuery.error?.message ?? 'Trace 加载失败'} hint="确认该运行仍属于当前项目和工作流。" />
-                  ) : traceEvents.length === 0 ? (
-                    <div className="rounded-lg border border-dashed p-4 text-sm text-muted-foreground">这个运行还没有持久化事件。</div>
                   ) : (
-                    <div className="space-y-2">
-                      {traceEvents.map((event) => (
-                        <div key={event.id} className="grid gap-3 rounded-lg border p-3 sm:grid-cols-[2.5rem_minmax(0,1fr)_8rem]">
-                          <div className="font-mono text-xs text-muted-foreground">#{event.sequence}</div>
-                          <div className="min-w-0">
-                            <div className="flex flex-wrap items-center gap-2"><StatusBadge status={event.eventType} /><span className="truncate font-mono text-xs">{event.nodeId}</span></div>
-                            {event.message ? <p className="mt-2 text-xs leading-5 text-muted-foreground">{event.message}</p> : null}
-                            {Object.keys(event.details ?? {}).length ? <pre className="mt-2 max-h-32 overflow-auto rounded border bg-muted/25 p-2 font-mono text-[10px] leading-4">{JSON.stringify(event.details, null, 2)}</pre> : null}
+                    <div className="space-y-3">
+                      {traceNodeTasks.length ? (
+                        <AITaskList
+                          label={`节点执行状态 · ${traceStatus}`}
+                          tasks={traceNodeTasks}
+                          className="bg-card"
+                        />
+                      ) : (
+                        <div className="rounded-lg border border-dashed p-4 text-sm text-muted-foreground">这个运行还没有节点状态投影。</div>
+                      )}
+
+                      {traceEvents.length ? (
+                        <details className="rounded-xl border bg-card">
+                          <summary className="cursor-pointer px-3 py-2.5 text-sm font-medium focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring">
+                            当前批次事件详情 · {traceEvents.length}
+                          </summary>
+                          <div className="space-y-2 border-t p-3">
+                            {traceEvents.map((event) => (
+                              <article key={event.id} className="grid gap-3 rounded-lg border p-3 sm:grid-cols-[2.5rem_minmax(0,1fr)_8rem]">
+                                <div className="font-mono text-xs text-muted-foreground">#{event.sequence}</div>
+                                <div className="min-w-0">
+                                  <div className="flex flex-wrap items-start gap-2">
+                                    <StatusBadge status={event.eventType} />
+                                    <span className="min-w-0 font-mono text-xs [overflow-wrap:anywhere]">{event.nodeId}</span>
+                                  </div>
+                                  {event.message ? <p className="mt-2 text-xs leading-5 text-muted-foreground">{event.message}</p> : null}
+                                  {event.blockReason ? (
+                                    <div className="mt-2 rounded border border-warning/30 bg-warning/5 p-2 text-xs text-warning">
+                                      {event.blockReason.code} · {event.blockReason.message}
+                                    </div>
+                                  ) : null}
+                                  {Object.keys(event.details ?? {}).length ? (
+                                    <details className="mt-2 rounded border bg-muted/25">
+                                      <summary className="cursor-pointer px-2 py-1.5 text-xs font-medium focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring">原始事件详情</summary>
+                                      <pre className="max-h-40 overflow-auto border-t p-2 font-mono text-[10px] leading-4">{JSON.stringify(event.details, null, 2)}</pre>
+                                    </details>
+                                  ) : null}
+                                </div>
+                                <time className="text-left text-[10px] text-muted-foreground sm:text-right" dateTime={event.createdAt}>{formatDateTime(event.createdAt)}</time>
+                              </article>
+                            ))}
+                          </div>
+                        </details>
+                      ) : (
+                        <div className="rounded-lg border border-dashed p-4 text-sm text-muted-foreground">当前批次没有持久化事件。</div>
+                      )}
+
+                      {traceQuery.data ? (
+                        <div className="flex items-center justify-between gap-3 text-xs text-muted-foreground">
+                          <span>{traceEvents.length ? `事件 #${traceEvents[0]?.sequence}–#${traceEvents.at(-1)?.sequence}` : '当前批次无事件'} · 共 {traceTotalEvents} 条</span>
+                          <div className="flex gap-2">
+                            <Button type="button" size="sm" variant="outline" disabled={traceCursorHistory.length === 1} onClick={() => setTraceCursorHistory((history) => history.slice(0, -1))}>上一批</Button>
+                            <Button type="button" size="sm" variant="outline" disabled={!traceHasNextPage} onClick={() => setTraceCursorHistory((history) => [...history, traceNextCursor])}>下一批</Button>
                           </div>
-                          <div className="text-right text-[10px] text-muted-foreground">{formatDateTime(event.createdAt)}</div>
                         </div>
-                      ))}
+                      ) : null}
                     </div>
                   )}
-                  {!traceQuery.isLoading && !traceQuery.isError && traceQuery.data ? (
-                    <div className="mt-3 flex items-center justify-between gap-3 text-xs text-muted-foreground">
-                      <span>{traceEvents.length ? `事件 #${traceEvents[0]?.sequence}–#${traceEvents.at(-1)?.sequence}` : '当前批次无事件'} · 共 {traceTotalEvents} 条</span>
-                      <div className="flex gap-2">
-                        <Button type="button" size="sm" variant="outline" disabled={traceCursorHistory.length === 1} onClick={() => setTraceCursorHistory((history) => history.slice(0, -1))}>上一批</Button>
-                        <Button type="button" size="sm" variant="outline" disabled={!traceHasNextPage} onClick={() => setTraceCursorHistory((history) => [...history, traceNextCursor])}>下一批</Button>
-                      </div>
-                    </div>
-                  ) : null}
                 </div>
               </div>
             </>
diff --git a/frontend/app/(app)/system/page.tsx b/frontend/app/(app)/system/page.tsx
index d1cd028a..a6f17f70 100644
--- a/frontend/app/(app)/system/page.tsx
+++ b/frontend/app/(app)/system/page.tsx
@@ -1,23 +1,793 @@
 'use client'
 
+import { useState } from 'react'
+import { toast } from 'sonner'
+
+import { useAuth } from '@/components/auth/auth-provider'
+import { BACKEND_HINT, ErrorState, LoadingState } from '@/components/shell/data-states'
 import { PageContainer } from '@/components/shell/page-container'
-import { CeleryStatsCard } from '@/components/system/celery-stats-card'
-import { ChromePoolCard } from '@/components/system/chrome-pool-card'
-import { RestartApiCard } from '@/components/system/restart-api-card'
-import { SystemConfigCard } from '@/components/system/system-config-card'
+import { Button } from '@/components/ui/button'
+import { Card, CardContent, CardDescription, CardFooter, CardHeader, CardTitle } from '@/components/ui/card'
+import { Field, FieldDescription, FieldGroup, FieldLabel } from '@/components/ui/field'
+import { Input } from '@/components/ui/input'
+import { Select, SelectContent, SelectItem, SelectTrigger, SelectValue } from '@/components/ui/select'
+import { Switch } from '@/components/ui/switch'
+import { type SystemConfig, type SystemConfigPatch, useSystemConfig, useUpdateSystemConfig } from '@/lib/api/hooks'
+
+type StatusBadge = {
+  label: string
+  critical?: boolean
+}
+
+function ConfigStatusRow({
+  label,
+  value,
+  badge,
+}: {
+  label: string
+  value: React.ReactNode
+  badge?: StatusBadge
+}) {
+  return (
+    <div className="flex flex-col gap-2 rounded-lg border px-3 py-2.5 sm:flex-row sm:items-start sm:justify-between sm:gap-4">
+      <span className="text-sm text-muted-foreground">{label}</span>
+      <span className="flex min-w-0 flex-wrap items-center gap-2 text-sm font-medium sm:justify-end sm:text-right">
+        <span className="min-w-0 break-all">{value}</span>
+        {badge ? (
+          <span className={badge.critical ? 'text-xs text-destructive' : 'text-xs text-muted-foreground'}>
+            {badge.label}
+          </span>
+        ) : null}
+      </span>
+    </div>
+  )
+}
+
+function configuredBadge(configured: boolean): StatusBadge {
+  return { label: configured ? '已配置' : '未配置' }
+}
+
+function SystemConfigOverview({ config }: { config: SystemConfig }) {
+  const configuredIntegrations = [
+    config.api_auth_configured,
+    config.credential_encryption_configured,
+    config.oidc_configured,
+    config.smtp_configured,
+  ].filter(Boolean).length
+
+  return (
+    <div className="space-y-5">
+      <Card>
+        <CardHeader>
+          <CardTitle><h2>部署概览</h2></CardTitle>
+          <CardDescription>来自当前服务端配置；这里不把配置值描述为运行健康状态。</CardDescription>
+        </CardHeader>
+        <CardContent className="grid gap-3 sm:grid-cols-2 lg:grid-cols-4">
+          <div className="rounded-lg border p-3">
+            <p className="text-xs text-muted-foreground">应用环境</p>
+            <p className="mt-1 font-medium">{config.app_env}</p>
+          </div>
+          <div className="rounded-lg border p-3">
+            <p className="text-xs text-muted-foreground">任务执行器</p>
+            <p className="mt-1 font-medium">{config.task_executor}</p>
+          </div>
+          <div className="rounded-lg border p-3">
+            <p className="text-xs text-muted-foreground">数据库</p>
+            <p className="mt-1 font-medium">{config.database_kind}</p>
+          </div>
+          <div className="rounded-lg border p-3">
+            <p className="text-xs text-muted-foreground">镜像版本</p>
+            <p className="mt-1 break-all font-mono text-xs">{config.image_tag}</p>
+          </div>
+        </CardContent>
+      </Card>
+
+      <div className="grid gap-5 lg:grid-cols-2">
+        <Card>
+          <CardHeader>
+            <CardTitle><h2>当前执行策略</h2></CardTitle>
+            <CardDescription>展示服务端正在使用的配置值。</CardDescription>
+          </CardHeader>
+          <CardContent className="space-y-3">
+            <ConfigStatusRow
+              label="采集执行位置"
+              value={config.collection_mode === 'local' ? '本机执行' : '远程 Agent 执行'}
+            />
+            <ConfigStatusRow label="调度编排器" value={config.collection_orchestrator} />
+            <ConfigStatusRow
+              label="最大并发任务"
+              value={`${config.local_max_concurrent_pipelines} 个`}
+            />
+            <ConfigStatusRow label="OpenCLI 超时" value={`${config.opencli_timeout} 秒`} />
+            <ConfigStatusRow label="默认时区" value={config.default_timezone} />
+          </CardContent>
+        </Card>
+
+        <Card>
+          <CardHeader>
+            <CardTitle><h2>浏览器与 Agent 配置</h2></CardTitle>
+            <CardDescription>仅展示地址和组网配置，不代表端点在线或可连接。</CardDescription>
+          </CardHeader>
+          <CardContent className="space-y-3">
+            <ConfigStatusRow label="默认 CDP" value={config.opencli_cdp_endpoint} />
+            <ConfigStatusRow
+              label="已配置的 Agent Pool"
+              value={
+                config.agent_pool_endpoints.length ? (
+                  <span className="flex flex-col gap-1 font-mono text-xs">
+                    {config.agent_pool_endpoints.map((endpoint) => (
+                      <span key={endpoint}>{endpoint}</span>
+                    ))}
+                  </span>
+                ) : (
+                  '未配置，将使用默认 CDP'
+                )
+              }
+              badge={{ label: `${config.agent_pool_endpoints.length} 个地址` }}
+            />
+            <ConfigStatusRow
+              label="当前有效 CDP 地址"
+              value={
+                <span className="flex flex-col gap-1 font-mono text-xs">
+                  {config.effective_cdp_endpoints.map((endpoint) => (
+                    <span key={endpoint}>{endpoint}</span>
+                  ))}
+                </span>
+              }
+            />
+            <ConfigStatusRow label="Fleet 网络" value={config.fleet_network_provider} />
+            <ConfigStatusRow label="NetBird 模式" value={config.netbird_mode} />
+            <ConfigStatusRow
+              label="对外访问地址"
+              value={config.public_url || '未配置'}
+              badge={configuredBadge(Boolean(config.public_url))}
+            />
+            <Button variant="ghost" render={<a href="/nodes" />}>
+              查看节点
+            </Button>
+          </CardContent>
+        </Card>
+
+        <Card>
+          <CardHeader>
+            <CardTitle><h2>AI 与自动执行</h2></CardTitle>
+            <CardDescription>当前模型请求边界和全局控制策略。</CardDescription>
+          </CardHeader>
+          <CardContent className="space-y-3">
+            <ConfigStatusRow
+              label="模型请求超时"
+              value={`${config.llm_request_timeout_seconds} 秒`}
+            />
+            <ConfigStatusRow label="模型并发数" value={String(config.llm_max_concurrency)} />
+            <ConfigStatusRow
+              label="控制策略"
+              value={config.control_mode === 'advisory' ? '建议模式' : '自动模式'}
+            />
+            <ConfigStatusRow
+              label="全局暂停自动执行"
+              value={config.control_kill_switch ? '已开启' : '已关闭'}
+              badge={
+                config.control_kill_switch
+                  ? { label: '拦截自动动作', critical: true }
+                  : { label: '未拦截' }
+              }
+            />
+          </CardContent>
+        </Card>
+
+        <Card>
+          <CardHeader>
+            <CardTitle><h2>集成与密钥状态</h2></CardTitle>
+            <CardDescription>
+              {configuredIntegrations}/4 项已配置；只显示是否存在，不在浏览器中暴露密钥。
+            </CardDescription>
+          </CardHeader>
+          <CardContent className="space-y-3">
+            <ConfigStatusRow
+              label="API / Fleet 访问令牌"
+              value={config.api_auth_configured ? '服务端已保存' : '服务端未保存'}
+              badge={configuredBadge(config.api_auth_configured)}
+            />
+            <ConfigStatusRow
+              label="凭证加密密钥"
+              value={config.credential_encryption_configured ? '服务端已保存' : '服务端未保存'}
+              badge={configuredBadge(config.credential_encryption_configured)}
+            />
+            <ConfigStatusRow
+              label="OIDC 组织登录"
+              value={config.oidc_configured ? '环境变量完整' : '环境变量不完整'}
+              badge={configuredBadge(config.oidc_configured)}
+            />
+            <ConfigStatusRow
+              label="SMTP 通知"
+              value={config.smtp_configured ? '主机与发件人已配置' : '主机或发件人未配置'}
+              badge={configuredBadge(config.smtp_configured)}
+            />
+          </CardContent>
+        </Card>
+      </div>
+    </div>
+  )
+}
+type RuntimeSettingsValues = {
+  collectionMode: SystemConfig['collection_mode']
+  collectionOrchestrator: SystemConfig['collection_orchestrator']
+  localMaxConcurrentPipelines: string
+  opencliTimeout: string
+  defaultTimezone: string
+  publicUrl: string
+  fleetNetworkProvider: SystemConfig['fleet_network_provider']
+  netbirdMode: SystemConfig['netbird_mode']
+  opencliCdpEndpoint: string
+  agentPoolEndpoints: string
+  llmRequestTimeoutSeconds: string
+  llmMaxConcurrency: string
+  controlMode: SystemConfig['control_mode']
+  controlKillSwitch: boolean
+}
+
+type RuntimeSettingsChange = <Key extends keyof RuntimeSettingsValues>(
+  key: Key,
+  value: RuntimeSettingsValues[Key],
+) => void
+
+type SettingsSectionProps = {
+  values: RuntimeSettingsValues
+  onChange: RuntimeSettingsChange
+}
+
+function ExecutionSettingsSection({ values, onChange }: SettingsSectionProps) {
+  return (
+    <section className="space-y-4" aria-labelledby="execution-settings-title">
+      <div>
+        <h3 id="execution-settings-title" className="text-sm font-medium">
+          任务与执行
+        </h3>
+        <p className="mt-1 text-xs text-muted-foreground">
+          决定采集任务在哪里运行、由谁编排，以及本机执行的资源边界。
+        </p>
+      </div>
+      <div className="grid gap-4 sm:grid-cols-2">
+        <Field>
+          <FieldLabel htmlFor="collection-mode">采集执行位置</FieldLabel>
+          <Select
+            value={values.collectionMode}
+            onValueChange={(value) =>
+              onChange('collectionMode', value as SystemConfig['collection_mode'])
+            }
+          >
+            <SelectTrigger id="collection-mode" className="w-full">
+              <SelectValue />
+            </SelectTrigger>
+            <SelectContent>
+              <SelectItem value="local">本机执行</SelectItem>
+              <SelectItem value="agent">远程 Agent 执行</SelectItem>
+            </SelectContent>
+          </Select>
+        </Field>
+
+        <Field>
+          <FieldLabel htmlFor="collection-orchestrator">调度编排器</FieldLabel>
+          <Select
+            value={values.collectionOrchestrator}
+            onValueChange={(value) =>
+              onChange(
+                'collectionOrchestrator',
+                value as SystemConfig['collection_orchestrator'],
+              )
+            }
+          >
+            <SelectTrigger id="collection-orchestrator" className="w-full">
+              <SelectValue />
+            </SelectTrigger>
+            <SelectContent>
+              <SelectItem value="admin">Admin</SelectItem>
+              <SelectItem value="iii">III</SelectItem>
+            </SelectContent>
+          </Select>
+        </Field>
+
+        <Field>
+          <FieldLabel htmlFor="max-concurrent-pipelines">本机最大并发任务</FieldLabel>
+          <Input
+            id="max-concurrent-pipelines"
+            type="number"
+            min={1}
+            max={64}
+            value={values.localMaxConcurrentPipelines}
+            onChange={(event) => onChange('localMaxConcurrentPipelines', event.target.value)}
+            aria-describedby="max-concurrent-pipelines-description"
+            required
+          />
+          <FieldDescription id="max-concurrent-pipelines-description">允许范围 1–64；只约束本机执行管线。</FieldDescription>
+        </Field>
+
+        <Field>
+          <FieldLabel htmlFor="opencli-timeout">OpenCLI 超时（秒）</FieldLabel>
+          <Input
+            id="opencli-timeout"
+            type="number"
+            min={1}
+            max={3600}
+            value={values.opencliTimeout}
+            onChange={(event) => onChange('opencliTimeout', event.target.value)}
+            aria-describedby="opencli-timeout-description"
+            required
+          />
+          <FieldDescription id="opencli-timeout-description">允许范围 1–3600 秒。</FieldDescription>
+        </Field>
+
+        <Field>
+          <FieldLabel htmlFor="default-timezone">默认时区</FieldLabel>
+          <Input
+            id="default-timezone"
+            value={values.defaultTimezone}
+            onChange={(event) => onChange('defaultTimezone', event.target.value)}
+            placeholder="Asia/Shanghai"
+            maxLength={64}
+            pattern=".*\S.*"
+            aria-describedby="default-timezone-description"
+            required
+          />
+          <FieldDescription id="default-timezone-description">填写非空的 IANA 时区名称，例如 Asia/Shanghai。</FieldDescription>
+        </Field>
+
+        <Field>
+          <FieldLabel htmlFor="public-url">对外访问地址</FieldLabel>
+          <Input
+            id="public-url"
+            value={values.publicUrl}
+            onChange={(event) => onChange('publicUrl', event.target.value)}
+            placeholder="未配置"
+            maxLength={2048}
+          />
+          <FieldDescription>留空表示未配置；不会在此处探测连通性。</FieldDescription>
+        </Field>
+      </div>
+    </section>
+  )
+}
+
+function BrowserSettingsSection({ values, onChange }: SettingsSectionProps) {
+  return (
+    <section className="space-y-4 border-t pt-6" aria-labelledby="browser-settings-title">
+      <div>
+        <h3 id="browser-settings-title" className="text-sm font-medium">
+          浏览器与 Agent 网络
+        </h3>
+        <p className="mt-1 text-xs text-muted-foreground">
+          这些是执行地址和组网方式的配置值，不是实时健康检查。
+        </p>
+      </div>
+      <div className="grid gap-4 sm:grid-cols-2">
+        <Field>
+          <FieldLabel htmlFor="fleet-network-provider">Fleet 网络</FieldLabel>
+          <Select
+            value={values.fleetNetworkProvider}
+            onValueChange={(value) =>
+              onChange(
+                'fleetNetworkProvider',
+                value as SystemConfig['fleet_network_provider'],
+              )
+            }
+          >
+            <SelectTrigger id="fleet-network-provider" className="w-full">
+              <SelectValue />
+            </SelectTrigger>
+            <SelectContent>
+              <SelectItem value="lan">LAN</SelectItem>
+              <SelectItem value="netbird">NetBird</SelectItem>
+              <SelectItem value="wireguard">WireGuard</SelectItem>
+              <SelectItem value="ssh">SSH</SelectItem>
+              <SelectItem value="custom">自定义</SelectItem>
+            </SelectContent>
+          </Select>
+        </Field>
+
+        <Field>
+          <FieldLabel htmlFor="netbird-mode">NetBird 模式</FieldLabel>
+          <Select
+            value={values.netbirdMode}
+            onValueChange={(value) =>
+              onChange('netbirdMode', value as SystemConfig['netbird_mode'])
+            }
+          >
+            <SelectTrigger id="netbird-mode" className="w-full">
+              <SelectValue />
+            </SelectTrigger>
+            <SelectContent>
+              <SelectItem value="off">关闭</SelectItem>
+              <SelectItem value="host">宿主机</SelectItem>
+              <SelectItem value="docker">Docker</SelectItem>
+            </SelectContent>
+          </Select>
+        </Field>
+
+        <Field className="sm:col-span-2">
+          <FieldLabel htmlFor="opencli-cdp-endpoint">默认 CDP 地址</FieldLabel>
+          <Input
+            id="opencli-cdp-endpoint"
+            value={values.opencliCdpEndpoint}
+            onChange={(event) => onChange('opencliCdpEndpoint', event.target.value)}
+            maxLength={2048}
+            pattern=".*\S.*"
+            aria-describedby="opencli-cdp-endpoint-description"
+            required
+          />
+          <FieldDescription id="opencli-cdp-endpoint-description">不能为空；Agent Pool 未配置时使用此地址。</FieldDescription>
+        </Field>
+
+        <Field className="sm:col-span-2">
+          <FieldLabel htmlFor="agent-pool-endpoints">Agent Pool 地址</FieldLabel>
+          <Input
+            id="agent-pool-endpoints"
+            value={values.agentPoolEndpoints}
+            onChange={(event) => onChange('agentPoolEndpoints', event.target.value)}
+            placeholder="使用逗号分隔多个地址"
+            maxLength={8192}
+            className="font-mono text-xs"
+            aria-describedby="agent-pool-endpoints-description"
+          />
+          <FieldDescription id="agent-pool-endpoints-description">多个地址使用逗号分隔；留空表示不配置 Agent Pool，届时使用默认 CDP。</FieldDescription>
+        </Field>
+      </div>
+    </section>
+  )
+}
+
+function AutomationSettingsSection({ values, onChange }: SettingsSectionProps) {
+  return (
+    <section className="space-y-4 border-t pt-6" aria-labelledby="automation-settings-title">
+      <div>
+        <h3 id="automation-settings-title" className="text-sm font-medium">
+          AI 与自动执行
+        </h3>
+        <p className="mt-1 text-xs text-muted-foreground">
+          设置模型请求资源上限和全局控制策略。
+        </p>
+      </div>
+      <div className="grid gap-4 sm:grid-cols-2">
+        <Field>
+          <FieldLabel htmlFor="llm-request-timeout">模型请求超时（秒）</FieldLabel>
+          <Input
+            id="llm-request-timeout"
+            type="number"
+            min={1}
+            max={3600}
+            value={values.llmRequestTimeoutSeconds}
+            onChange={(event) => onChange('llmRequestTimeoutSeconds', event.target.value)}
+            aria-describedby="llm-request-timeout-description"
+            required
+          />
+          <FieldDescription id="llm-request-timeout-description">允许范围 1–3600 秒。</FieldDescription>
+        </Field>
+
+        <Field>
+          <FieldLabel htmlFor="llm-max-concurrency">模型最大并发数</FieldLabel>
+          <Input
+            id="llm-max-concurrency"
+            type="number"
+            min={1}
+            max={64}
+            value={values.llmMaxConcurrency}
+            onChange={(event) => onChange('llmMaxConcurrency', event.target.value)}
+            aria-describedby="llm-max-concurrency-description"
+            required
+          />
+          <FieldDescription id="llm-max-concurrency-description">允许范围 1–64。</FieldDescription>
+        </Field>
+
+        <Field>
+          <FieldLabel htmlFor="control-mode">控制策略</FieldLabel>
+          <Select
+            value={values.controlMode}
+            onValueChange={(value) =>
+              onChange('controlMode', value as SystemConfig['control_mode'])
+            }
+          >
+            <SelectTrigger id="control-mode" className="w-full">
+              <SelectValue />
+            </SelectTrigger>
+            <SelectContent>
+              <SelectItem value="advisory">建议模式</SelectItem>
+              <SelectItem value="automatic">自动模式</SelectItem>
+            </SelectContent>
+          </Select>
+        </Field>
+
+        <Field orientation="horizontal" className="items-center justify-between rounded-lg border p-3">
+          <div className="space-y-1">
+            <FieldLabel htmlFor="control-kill-switch">暂停自动执行</FieldLabel>
+            <FieldDescription id="control-kill-switch-description">开启后，全局阻止自动控制动作；建议模式不受影响。</FieldDescription>
+          </div>
+          <Switch
+            id="control-kill-switch"
+            checked={values.controlKillSwitch}
+            onCheckedChange={(checked) => onChange('controlKillSwitch', checked)}
+            aria-describedby="control-kill-switch-description"
+          />
+        </Field>
+      </div>
+    </section>
+  )
+}
+
+type RuntimeSettingsFormProps = {
+  config: SystemConfig
+}
+
+
+
+function valuesFromConfig(config: SystemConfig): RuntimeSettingsValues {
+  return {
+    collectionMode: config.collection_mode,
+    collectionOrchestrator: config.collection_orchestrator,
+    localMaxConcurrentPipelines: String(config.local_max_concurrent_pipelines),
+    opencliTimeout: String(config.opencli_timeout),
+    defaultTimezone: config.default_timezone,
+    publicUrl: config.public_url,
+    fleetNetworkProvider: config.fleet_network_provider,
+    netbirdMode: config.netbird_mode,
+    opencliCdpEndpoint: config.opencli_cdp_endpoint,
+    agentPoolEndpoints: config.agent_pool_endpoints.join('\n'),
+    llmRequestTimeoutSeconds: String(config.llm_request_timeout_seconds),
+    llmMaxConcurrency: String(config.llm_max_concurrency),
+    controlMode: config.control_mode,
+    controlKillSwitch: config.control_kill_switch,
+  }
+}
+
+function normalizedEndpoints(value: string): string[] {
+  return value
+    .split(/[\n,]/)
+    .map((endpoint) => endpoint.trim())
+    .filter(Boolean)
+}
+
+function patchFromValues(
+  values: RuntimeSettingsValues,
+  config: SystemConfig,
+): SystemConfigPatch {
+  const patch: SystemConfigPatch = {}
+  const localMaxConcurrentPipelines = Number(values.localMaxConcurrentPipelines)
+  const opencliTimeout = Number(values.opencliTimeout)
+  const llmRequestTimeoutSeconds = Number(values.llmRequestTimeoutSeconds)
+  const llmMaxConcurrency = Number(values.llmMaxConcurrency)
+  const defaultTimezone = values.defaultTimezone.trim()
+  const publicUrl = values.publicUrl.trim()
+  const opencliCdpEndpoint = values.opencliCdpEndpoint.trim()
+  const agentPoolEndpoints = normalizedEndpoints(values.agentPoolEndpoints)
+
+  if (values.collectionMode !== config.collection_mode) {
+    patch.collection_mode = values.collectionMode
+  }
+  if (values.collectionOrchestrator !== config.collection_orchestrator) {
+    patch.collection_orchestrator = values.collectionOrchestrator
+  }
+  if (localMaxConcurrentPipelines !== config.local_max_concurrent_pipelines) {
+    patch.local_max_concurrent_pipelines = localMaxConcurrentPipelines
+  }
+  if (opencliTimeout !== config.opencli_timeout) patch.opencli_timeout = opencliTimeout
+  if (defaultTimezone !== config.default_timezone) patch.default_timezone = defaultTimezone
+  if (publicUrl !== config.public_url) patch.public_url = publicUrl
+  if (values.fleetNetworkProvider !== config.fleet_network_provider) {
+    patch.fleet_network_provider = values.fleetNetworkProvider
+  }
+  if (values.netbirdMode !== config.netbird_mode) patch.netbird_mode = values.netbirdMode
+  if (opencliCdpEndpoint !== config.opencli_cdp_endpoint) {
+    patch.opencli_cdp_endpoint = opencliCdpEndpoint
+  }
+  if (agentPoolEndpoints.join(',') !== config.agent_pool_endpoints.join(',')) {
+    patch.agent_pool_endpoints = agentPoolEndpoints.join(',')
+  }
+  if (llmRequestTimeoutSeconds !== config.llm_request_timeout_seconds) {
+    patch.llm_request_timeout_seconds = llmRequestTimeoutSeconds
+  }
+  if (llmMaxConcurrency !== config.llm_max_concurrency) {
+    patch.llm_max_concurrency = llmMaxConcurrency
+  }
+  if (values.controlMode !== config.control_mode) patch.control_mode = values.controlMode
+  if (values.controlKillSwitch !== config.control_kill_switch) {
+    patch.control_kill_switch = values.controlKillSwitch
+  }
+
+  return patch
+}
+
+function RuntimeSettingsForm({ config }: RuntimeSettingsFormProps) {
+  const [values, setValues] = useState(() => valuesFromConfig(config))
+  const updateConfig = useUpdateSystemConfig()
+  const patch = patchFromValues(values, config)
+  const hasChanges = Object.keys(patch).length > 0
+
+  function setValue<Key extends keyof RuntimeSettingsValues>(
+    key: Key,
+    value: RuntimeSettingsValues[Key],
+  ) {
+    setValues((current) => ({ ...current, [key]: value }))
+  }
+
+  async function saveSettings(event: React.FormEvent<HTMLFormElement>) {
+    event.preventDefault()
+    if (!hasChanges) return
+    if (!values.defaultTimezone.trim() || !values.opencliCdpEndpoint.trim()) {
+      toast.error('默认时区和默认 CDP 地址不能为空')
+      return
+    }
+
+    try {
+      const updated = await updateConfig.mutateAsync(patch)
+      setValues(valuesFromConfig(updated))
+      toast.success('运行配置已保存', {
+        description: '当前 API 进程已刷新配置；独立 Worker 或 Agent 可能需要按部署方式重启。',
+      })
+    } catch (error) {
+      toast.error(error instanceof Error ? error.message : '运行配置保存失败')
+    }
+  }
+
+  return (
+    <form onSubmit={saveSettings}>
+      <Card>
+        <CardHeader>
+          <CardTitle><h2 id="runtime-settings-title">运行配置</h2></CardTitle>
+          <CardDescription>
+            这里修改服务端已开放的安全配置字段。保存会写入部署环境文件并刷新当前 API 进程，不代表外部服务已连通。
+          </CardDescription>
+        </CardHeader>
+
+        <CardContent>
+          <fieldset
+            className="space-y-7 border-0 p-0"
+            disabled={updateConfig.isPending}
+            aria-labelledby="runtime-settings-title"
+          >
+            <legend className="sr-only">运行配置字段</legend>
+            <ExecutionSettingsSection values={values} onChange={setValue} />
+            <BrowserSettingsSection values={values} onChange={setValue} />
+            <AutomationSettingsSection values={values} onChange={setValue} />
+          </fieldset>
+        </CardContent>
+
+        <CardFooter className="flex flex-wrap justify-between gap-3">
+          <p className="text-xs text-muted-foreground">
+            {hasChanges ? '有尚未保存的更改。' : '当前表单与服务端配置一致。'}
+          </p>
+          <div className="flex gap-2">
+            <Button
+              type="button"
+              variant="outline"
+              disabled={!hasChanges || updateConfig.isPending}
+              onClick={() => setValues(valuesFromConfig(config))}
+            >
+              放弃更改
+            </Button>
+            <Button type="submit" disabled={!hasChanges || updateConfig.isPending}>
+              {updateConfig.isPending ? '保存中…' : '保存运行配置'}
+            </Button>
+          </div>
+        </CardFooter>
+      </Card>
+    </form>
+  )
+}
+function AccountSecurityCard() {
+  const { changePassword } = useAuth()
+  const [currentPassword, setCurrentPassword] = useState('')
+  const [newPassword, setNewPassword] = useState('')
+  const [confirmPassword, setConfirmPassword] = useState('')
+  const [savingPassword, setSavingPassword] = useState(false)
+
+  async function savePassword(event: React.FormEvent<HTMLFormElement>) {
+    event.preventDefault()
+    if (newPassword !== confirmPassword) {
+      toast.error('两次输入的新密码不一致')
+      return
+    }
+
+    setSavingPassword(true)
+    try {
+      await changePassword(currentPassword, newPassword)
+      setCurrentPassword('')
+      setNewPassword('')
+      setConfirmPassword('')
+      toast.success('密码已更新')
+    } catch (error) {
+      toast.error(error instanceof Error ? error.message : '密码更新失败')
+    } finally {
+      setSavingPassword(false)
+    }
+  }
+
+  return (
+    <Card>
+      <CardHeader>
+        <CardTitle><h2>管理员账户</h2></CardTitle>
+        <CardDescription>
+          修改当前本地管理员密码。此操作不依赖运行配置是否成功加载。
+        </CardDescription>
+      </CardHeader>
+      <CardContent>
+        <form className="max-w-xl space-y-5" onSubmit={savePassword}>
+          <FieldGroup>
+            <Field>
+              <FieldLabel htmlFor="current-password">当前密码</FieldLabel>
+              <Input
+                id="current-password"
+                type="password"
+                autoComplete="current-password"
+                value={currentPassword}
+                onChange={(event) => setCurrentPassword(event.target.value)}
+                required
+              />
+            </Field>
+            <Field>
+              <FieldLabel htmlFor="new-password">新密码</FieldLabel>
+              <Input
+                id="new-password"
+                type="password"
+                autoComplete="new-password"
+                minLength={6}
+                value={newPassword}
+                onChange={(event) => setNewPassword(event.target.value)}
+                required
+                aria-describedby="new-password-description"
+              />
+              <FieldDescription id="new-password-description">至少 6 个字符。</FieldDescription>
+            </Field>
+            <Field>
+              <FieldLabel htmlFor="confirm-password">确认新密码</FieldLabel>
+              <Input
+                id="confirm-password"
+                type="password"
+                autoComplete="new-password"
+                minLength={6}
+                value={confirmPassword}
+                onChange={(event) => setConfirmPassword(event.target.value)}
+                required
+              />
+            </Field>
+          </FieldGroup>
+          <Button type="submit" disabled={savingPassword}>
+            {savingPassword ? '保存中…' : '保存密码'}
+          </Button>
+        </form>
+      </CardContent>
+    </Card>
+  )
+}
+
+export default function SystemSettingsPage() {
+  const config = useSystemConfig()
 
-export default function SystemPage() {
   return (
     <PageContainer
-      title="系统与运维"
-      description="部署配置与计算池的运行状态一览；重启操作会影响全部用户，请谨慎执行。"
+      eyebrow="SYSTEM"
+      title="系统设置"
+      description="查看真实部署配置，调整服务端开放的运行参数，并管理管理员账户。"
     >
-      <SystemConfigCard />
-      <div className="grid gap-4 lg:grid-cols-2">
-        <ChromePoolCard />
-        <CeleryStatsCard />
-      </div>
-      <RestartApiCard />
+      {config.isLoading ? (
+        <Card>
+          <CardHeader>
+            <CardTitle><h2>部署与运行配置</h2></CardTitle>
+            <CardDescription>正在从服务端读取当前配置。</CardDescription>
+          </CardHeader>
+          <CardContent>
+            <LoadingState rows={3} />
+          </CardContent>
+        </Card>
+      ) : config.isError || !config.data ? (
+        <section aria-label="部署与运行配置加载失败">
+          <ErrorState message={(config.error as Error)?.message} hint={BACKEND_HINT} />
+        </section>
+      ) : (
+        <>
+          <SystemConfigOverview config={config.data} />
+          <RuntimeSettingsForm key={JSON.stringify(config.data)} config={config.data} />
+        </>
+      )}
+
+      <AccountSecurityCard />
     </PageContainer>
   )
 }
diff --git a/frontend/app/layout.tsx b/frontend/app/layout.tsx
index 035324e7..353a476c 100644
--- a/frontend/app/layout.tsx
+++ b/frontend/app/layout.tsx
@@ -20,6 +20,11 @@ const ibmPlexMono = IBM_Plex_Mono({
   display: 'swap',
 })
 
+const analyticsFlag = process.env.NEXT_PUBLIC_ENABLE_VERCEL_ANALYTICS?.trim().toLowerCase()
+const analyticsEnabled = analyticsFlag
+  ? analyticsFlag === 'true'
+  : process.env.VERCEL === '1'
+
 export const metadata: Metadata = {
   title: 'OpenCLI Admin',
   description: '采集编排控制台 — 以节点工作流为核心的数据采集管理平台',
@@ -44,7 +49,7 @@ export default function RootLayout({
       <body className="font-sans antialiased bg-background text-foreground">
         <Providers>{children}</Providers>
         <Toaster />
-        {process.env.NODE_ENV === 'production' && <Analytics />}
+        {analyticsEnabled ? <Analytics /> : null}
       </body>
     </html>
   )
diff --git a/frontend/app/login/page.tsx b/frontend/app/login/page.tsx
index 16447936..c47c8a43 100644
--- a/frontend/app/login/page.tsx
+++ b/frontend/app/login/page.tsx
@@ -1,6 +1,6 @@
 'use client'
 
-import { Droplets, Grid3X3, KeyRound, LoaderCircle, ShieldCheck, SquareTerminal } from 'lucide-react'
+import { Droplets, Grid3X3, LoaderCircle, SquareTerminal } from 'lucide-react'
 import { AnimatePresence, motion } from 'motion/react'
 import { useRouter, useSearchParams } from 'next/navigation'
 import { Suspense, useEffect, useState } from 'react'
@@ -11,6 +11,7 @@ import Dither from '@/components/Dither'
 import { useAuth } from '@/components/auth/auth-provider'
 import { PixelLiquidBg } from '@/components/unlumen-ui/pixel-liquid-bg'
 import { Button } from '@/components/ui/button'
+import { GpuSurface } from '@/components/gpu/gpu-surface'
 import {
   Card,
   CardContent,
@@ -21,7 +22,6 @@ import {
 } from '@/components/ui/card'
 import { Field, FieldDescription, FieldGroup, FieldLabel } from '@/components/ui/field'
 import { Input } from '@/components/ui/input'
-import { Separator } from '@/components/ui/separator'
 import RevealText from '@/components/ui/smoothui/reveal-text'
 import { sanitizeReturnTo } from '@/lib/auth/oidc'
 
@@ -61,88 +61,126 @@ const LIQUID_LIGHT_PALETTE = [
   '#8f2112',
 ]
 
+function StaticLoginBackground({ theme }: { theme: LoginBackdrop }) {
+  const themeClassName = theme === 'liquid'
+    ? 'bg-[radial-gradient(circle_at_20%_20%,#ff8a2b_0%,transparent_34%),radial-gradient(circle_at_80%_75%,#bd2d17_0%,transparent_38%),#070604]'
+    : theme === 'terminal'
+      ? 'bg-[linear-gradient(135deg,#0d0805_0%,#2a1008_48%,#070604_100%)]'
+      : 'bg-[radial-gradient(circle_at_30%_20%,#f0441f_0%,transparent_25%),linear-gradient(145deg,#130604_0%,#020202_72%)]'
+
+  return (
+    <div
+      aria-hidden="true"
+      className={`relative size-full overflow-hidden ${themeClassName}`}
+      data-login-static-background
+      data-login-theme={theme}
+    >
+      {theme === 'terminal' ? (
+        <div className="absolute inset-0 bg-[linear-gradient(rgba(249,115,22,0.13)_1px,transparent_1px),linear-gradient(90deg,rgba(249,115,22,0.13)_1px,transparent_1px)] bg-[size:28px_28px]" />
+      ) : null}
+      <div className="absolute inset-0 bg-[radial-gradient(circle_at_center,transparent_0%,rgba(0,0,0,0.42)_100%)]" />
+    </div>
+  )
+}
+
+function LoginPendingShell() {
+  return (
+    <main className="relative min-h-screen overflow-hidden bg-[#070604] text-white">
+      <div className="absolute inset-0 opacity-90" aria-hidden="true">
+        <div
+          className="absolute inset-0"
+          data-gpu-surface="login-background"
+          data-gpu-backend="pending"
+          data-gpu-fallback-reason="probing"
+        >
+          <StaticLoginBackground theme="liquid" />
+        </div>
+      </div>
+    </main>
+  )
+}
+
 function LoginBackground({ theme, reduceMotion }: { theme: LoginBackdrop; reduceMotion: boolean }) {
   return (
-    <AnimatePresence mode="wait" initial={false}>
-      <motion.div
-        key={theme}
-        className="absolute inset-0"
-        initial={reduceMotion ? false : { opacity: 0, filter: 'blur(6px)', scale: 1.01 }}
-        animate={{ opacity: 1, filter: 'blur(0px)', scale: 1 }}
-        exit={reduceMotion ? { opacity: 0 } : { opacity: 0, filter: 'blur(3px)', scale: 0.995 }}
-        transition={{ type: 'spring', duration: 0.45, bounce: 0 }}
-      >
-        {theme === 'liquid' ? (
-          <div className="relative size-full overflow-hidden bg-[#070604]">
-            <PixelLiquidBg
-              className="absolute inset-0"
-              darkPalette={LIQUID_DARK_PALETTE}
-              lightPalette={LIQUID_LIGHT_PALETTE}
-              pixelSize={11}
-              resolution={0.34}
-              mouseForce={5.5}
-              cursorSize={150}
-              viscosity={14}
-              surfaceStrength={0.76}
-              autoDemo={!reduceMotion}
-            />
-          </div>
-        ) : theme === 'terminal' ? (
-          <FaultyTerminal
-            tint="#F97316"
-            scale={1.35}
-            gridMul={[2, 1]}
-            digitSize={1.2}
-            timeScale={0.22}
-            pause={reduceMotion}
-            scanlineIntensity={0.35}
-            glitchAmount={0.65}
-            flickerAmount={0.25}
-            noiseAmp={0.85}
-            chromaticAberration={0.5}
-            curvature={0.08}
-            mouseReact={!reduceMotion}
-            mouseStrength={0.12}
-            dpr={1}
-            pageLoadAnimation={!reduceMotion}
-            brightness={0.9}
-          />
-        ) : (
-          <div className="relative size-full overflow-hidden bg-black">
-            <div className="absolute inset-0 opacity-85">
-              <Dither
-                waveSpeed={0.1}
-                waveFrequency={2.6}
-                waveAmplitude={0.38}
-                waveColor={[1, 0.27, 0.08]}
-                colorNum={7}
-                pixelSize={3}
-                disableAnimation={reduceMotion}
-                enableMouseInteraction={!reduceMotion}
-                mouseRadius={0.72}
+    <GpuSurface
+      surface="login-background"
+      className="absolute inset-0"
+      fallback={<StaticLoginBackground theme={theme} />}
+    >
+      <AnimatePresence mode="wait" initial={false}>
+        <motion.div
+          key={theme}
+          className="absolute inset-0"
+          initial={reduceMotion ? false : { opacity: 0, filter: 'blur(6px)', scale: 1.01 }}
+          animate={{ opacity: 1, filter: 'blur(0px)', scale: 1 }}
+          exit={reduceMotion ? { opacity: 0 } : { opacity: 0, filter: 'blur(3px)', scale: 0.995 }}
+          transition={{ type: 'spring', duration: 0.45, bounce: 0 }}
+        >
+          {theme === 'liquid' ? (
+            <div className="relative size-full overflow-hidden bg-[#070604]">
+              <PixelLiquidBg
+                className="absolute inset-0"
+                darkPalette={LIQUID_DARK_PALETTE}
+                lightPalette={LIQUID_LIGHT_PALETTE}
+                pixelSize={11}
+                resolution={0.34}
+                mouseForce={5.5}
+                cursorSize={150}
+                viscosity={14}
+                surfaceStrength={0.76}
+                autoDemo={!reduceMotion}
               />
             </div>
-          </div>
-        )}
-      </motion.div>
-    </AnimatePresence>
+          ) : theme === 'terminal' ? (
+            <FaultyTerminal
+              tint="#F97316"
+              scale={1.35}
+              gridMul={[2, 1]}
+              digitSize={1.2}
+              timeScale={0.22}
+              pause={reduceMotion}
+              scanlineIntensity={0.35}
+              glitchAmount={0.65}
+              flickerAmount={0.25}
+              noiseAmp={0.85}
+              chromaticAberration={0.5}
+              curvature={0.08}
+              mouseReact={!reduceMotion}
+              mouseStrength={0.12}
+              dpr={1}
+              pageLoadAnimation={!reduceMotion}
+              brightness={0.9}
+            />
+          ) : (
+            <div className="relative size-full overflow-hidden bg-black">
+              <div className="absolute inset-0 opacity-85">
+                <Dither
+                  waveSpeed={0.1}
+                  waveFrequency={2.6}
+                  waveAmplitude={0.38}
+                  waveColor={[1, 0.27, 0.08]}
+                  colorNum={7}
+                  pixelSize={3}
+                  disableAnimation={reduceMotion}
+                  enableMouseInteraction={!reduceMotion}
+                  mouseRadius={0.72}
+                />
+              </div>
+            </div>
+          )}
+        </motion.div>
+      </AnimatePresence>
+    </GpuSurface>
   )
 }
 
 function LoginForm() {
   const router = useRouter()
   const searchParams = useSearchParams()
-  const {
-    status,
-    oidcEnabled,
-    developmentLoginEnabled,
-    signInWithOidc,
-    signInWithBootstrap,
-    enterDevelopmentMode,
-  } = useAuth()
-  const [identityToken, setIdentityToken] = useState('')
-  const [fleetToken, setFleetToken] = useState('')
-  const [submitting, setSubmitting] = useState<'oidc' | 'bootstrap' | 'development' | null>(null)
+  const { status, developmentLoginEnabled, signInWithPassword, enterDevelopmentMode } = useAuth()
+  const [username, setUsername] = useState('admin')
+  const [password, setPassword] = useState('')
+  const [submitting, setSubmitting] = useState<'password' | 'development' | null>(null)
   const [reduceMotion, setReduceMotion] = useState(true)
   const [backdrop, setBackdrop] = useState<LoginBackdrop>('liquid')
   const [headlineWord, setHeadlineWord] = useState(0)
@@ -173,27 +211,19 @@ function LoginForm() {
     return () => window.clearInterval(interval)
   }, [reduceMotion])
 
-  const optionalFleetToken = fleetToken.trim() || undefined
-
-  async function startOidcLogin() {
-    setSubmitting('oidc')
-    try {
-      await signInWithOidc(returnTo, optionalFleetToken)
-    } catch (error) {
-      toast.error(error instanceof Error ? error.message : '无法启动 OIDC 登录')
-      setSubmitting(null)
-    }
-  }
-
-  async function handleBootstrapLogin(event: React.FormEvent) {
+  async function handlePasswordLogin(event: React.FormEvent) {
     event.preventDefault()
-    setSubmitting('bootstrap')
+    setSubmitting('password')
     try {
-      await signInWithBootstrap(identityToken, optionalFleetToken)
-      toast.success('管理员身份验证成功')
+      const usingDefaultPassword = await signInWithPassword(username, password)
+      toast.success(
+        usingDefaultPassword
+          ? '登录成功。你可以在账户设置中修改默认密码。'
+          : '登录成功',
+      )
       router.replace(returnTo)
     } catch (error) {
-      toast.error(error instanceof Error ? error.message : '身份验证失败')
+      toast.error(error instanceof Error ? error.message : '用户名或密码错误')
       setSubmitting(null)
     }
   }
@@ -201,7 +231,7 @@ function LoginForm() {
   function handleDevelopmentLogin() {
     setSubmitting('development')
     try {
-      enterDevelopmentMode(optionalFleetToken)
+      enterDevelopmentMode()
       window.setTimeout(() => router.replace(returnTo), reduceMotion ? 0 : 285)
     } catch (error) {
       toast.error(error instanceof Error ? error.message : '无法进入本地开发模式')
@@ -213,9 +243,7 @@ function LoginForm() {
     <motion.main
       className="relative min-h-screen overflow-hidden bg-[#070604] text-white"
       animate={
-        submitting === 'development' && !reduceMotion
-          ? { x: '-100%' }
-          : { x: 0 }
+        submitting === 'development' && !reduceMotion ? { x: '-100%' } : { x: 0 }
       }
       transition={{ duration: 0.28, ease: [0.32, 0.72, 0, 1] }}
     >
@@ -257,11 +285,7 @@ function LoginForm() {
             <div className="min-w-0 pt-0.5">
               <div className="flex overflow-hidden text-[clamp(2.8rem,4.5vw,4.4rem)] font-bold leading-[0.86] tracking-[-0.075em] text-white/95">
                 {Array.from('OPENCLI').map((character, index) => (
-                  <RevealText
-                    key={`${character}-${index}`}
-                    delay={80 + index * 45}
-                    direction="up"
-                  >
+                  <RevealText key={`${character}-${index}`} delay={80 + index * 45} direction="up">
                     {character}
                   </RevealText>
                 ))}
@@ -329,68 +353,33 @@ function LoginForm() {
             <CardHeader>
               <CardTitle>登录控制台</CardTitle>
               <CardDescription>
-                使用组织账号登录；Bootstrap Admin 仅用于首次部署和紧急恢复。
+                本地部署直接登录。首次使用可使用默认账号 admin / admin。
               </CardDescription>
             </CardHeader>
-            <CardContent className="space-y-5">
-              {oidcEnabled ? (
-                <Button
-                  className={`w-full overflow-hidden transition-[height,border-radius,background-color] duration-300 ${submitting === 'oidc' ? 'h-16 rounded-2xl' : 'h-10'}`}
-                  data-triggered={submitting === 'oidc'}
-                  disabled={submitting !== null}
-                  onClick={startOidcLogin}
-                >
-                  {submitting === 'oidc' ? (
-                    <span className="flex items-center gap-3 text-left">
-                      <LoaderCircle className="size-5 animate-spin" />
-                      <span className="grid">
-                        <span>正在连接组织账号</span>
-                        <span className="text-xs font-normal opacity-65">等待身份提供方响应</span>
-                      </span>
-                    </span>
-                  ) : (
-                    <><ShieldCheck />使用组织账号登录</>
-                  )}
-                </Button>
-              ) : (
-                <div className="rounded-lg border bg-muted/40 p-3 text-sm text-muted-foreground">
-                  当前未配置组织登录。请配置 OIDC issuer、client ID 和授权端点。
-                </div>
-              )}
-
-              <div className="flex items-center gap-3">
-                <Separator className="flex-1" />
-                <span className="text-xs text-muted-foreground">紧急管理员访问</span>
-                <Separator className="flex-1" />
-              </div>
-
-              <form id="bootstrap-login" onSubmit={handleBootstrapLogin}>
+            <CardContent>
+              <form id="local-login" className="space-y-4" onSubmit={handlePasswordLogin}>
                 <FieldGroup>
                   <Field>
-                    <FieldLabel htmlFor="identity-token">管理员身份令牌</FieldLabel>
+                    <FieldLabel htmlFor="local-username">用户名</FieldLabel>
                     <Input
-                      id="identity-token"
-                      type="password"
-                      placeholder="BOOTSTRAP_ADMIN_TOKEN"
-                      value={identityToken}
-                      onChange={(event) => setIdentityToken(event.target.value)}
-                      autoComplete="off"
+                      id="local-username"
+                      autoComplete="username"
+                      value={username}
+                      onChange={(event) => setUsername(event.target.value)}
+                      required
                     />
-                    <FieldDescription>验证成功后仅保存在当前标签页会话中。</FieldDescription>
                   </Field>
                   <Field>
-                    <FieldLabel htmlFor="fleet-token">Fleet API 令牌（可选）</FieldLabel>
+                    <FieldLabel htmlFor="local-password">密码</FieldLabel>
                     <Input
-                      id="fleet-token"
+                      id="local-password"
                       type="password"
-                      placeholder="API_AUTH_TOKEN"
-                      value={fleetToken}
-                      onChange={(event) => setFleetToken(event.target.value)}
-                      autoComplete="off"
+                      autoComplete="current-password"
+                      value={password}
+                      onChange={(event) => setPassword(event.target.value)}
+                      required
                     />
-                    <FieldDescription>
-                      后端启用 Fleet Auth 时填写；留空沿用部署配置或浏览器中已有值。
-                    </FieldDescription>
+                    <FieldDescription>登录后可以在账户设置中修改密码。</FieldDescription>
                   </Field>
                 </FieldGroup>
               </form>
@@ -398,22 +387,21 @@ function LoginForm() {
             <CardFooter className="flex-col gap-2">
               <Button
                 type="submit"
-                form="bootstrap-login"
-                variant={oidcEnabled ? 'outline' : 'default'}
-                className={`w-full overflow-hidden transition-[height,border-radius,background-color] duration-300 ${submitting === 'bootstrap' ? 'h-16 rounded-2xl' : 'h-10'}`}
-                data-triggered={submitting === 'bootstrap'}
+                form="local-login"
+                className={`w-full overflow-hidden transition-[height,border-radius,background-color] duration-300 ${submitting === 'password' ? 'h-16 rounded-2xl' : 'h-10'}`}
+                data-triggered={submitting === 'password'}
                 disabled={submitting !== null}
               >
-                {submitting === 'bootstrap' ? (
+                {submitting === 'password' ? (
                   <span className="flex items-center gap-3 text-left">
                     <LoaderCircle className="size-5 animate-spin" />
                     <span className="grid">
-                      <span>正在验证管理员令牌</span>
-                      <span className="text-xs font-normal opacity-65">验证通过后建立本地会话</span>
+                      <span>正在登录</span>
+                      <span className="text-xs font-normal opacity-65">建立本地会话</span>
                     </span>
                   </span>
                 ) : (
-                  <><KeyRound />使用管理员令牌登录</>
+                  '登录'
                 )}
               </Button>
               {developmentLoginEnabled ? (
@@ -451,7 +439,7 @@ function LoginForm() {
 
 export default function LoginPage() {
   return (
-    <Suspense>
+    <Suspense fallback={<LoginPendingShell />}>
       <LoginForm />
     </Suspense>
   )
diff --git a/frontend/components/agent-native/agent-primitives.tsx b/frontend/components/agent-native/agent-primitives.tsx
new file mode 100644
index 00000000..a1ba0a70
--- /dev/null
+++ b/frontend/components/agent-native/agent-primitives.tsx
@@ -0,0 +1,230 @@
+'use client'
+
+import type { FormEvent, KeyboardEventHandler, ReactNode } from 'react'
+import { Check, CheckCircle2, ChevronDown, CircleAlert, Clock3, Loader2, ShieldCheck, Terminal, X } from 'lucide-react'
+
+import { Badge } from '@/components/ui/badge'
+import { Button } from '@/components/ui/button'
+import { Textarea } from '@/components/ui/textarea'
+import { cn } from '@/lib/utils'
+
+export type AgentTaskStatus = 'queued' | 'running' | 'completed' | 'failed'
+
+type StatusConfig = {
+  label: string
+  icon: typeof CheckCircle2
+  className: string
+}
+
+const statusConfig: Record<AgentTaskStatus, StatusConfig> = {
+  queued: { label: '排队中', icon: Clock3, className: 'text-muted-foreground' },
+  running: { label: '运行中', icon: Loader2, className: 'text-info' },
+  completed: { label: '已完成', icon: CheckCircle2, className: 'text-success' },
+  failed: { label: '失败', icon: CircleAlert, className: 'text-destructive' },
+}
+
+export function AgentStatusBadge({ status }: { status: AgentTaskStatus }) {
+  const config = statusConfig[status]
+  const Icon = config.icon
+
+  return (
+    <Badge variant="outline" className={cn('gap-1.5 font-normal', config.className)}>
+      <Icon className={cn('size-3.5', status === 'running' && 'animate-spin')} aria-hidden />
+      {config.label}
+    </Badge>
+  )
+}
+
+export function ToolChip({
+  label,
+  detail,
+  icon: Icon = Terminal,
+  tone = 'neutral',
+}: {
+  label: string
+  detail?: string
+  icon?: typeof Terminal
+  tone?: 'neutral' | 'success' | 'warning'
+}) {
+  return (
+    <span
+      className={cn(
+        'inline-flex max-w-full items-center gap-1.5 rounded-full border px-2 py-1 text-[10px] leading-none',
+        tone === 'success' && 'border-success/30 bg-success/8 text-success',
+        tone === 'warning' && 'border-warning/35 bg-warning/10 text-warning-foreground dark:text-warning',
+        tone === 'neutral' && 'border-border/80 bg-muted/40 text-muted-foreground',
+      )}
+    >
+      <Icon className="size-3 shrink-0" aria-hidden />
+      <span className="truncate font-medium text-foreground/80">{label}</span>
+      {detail ? <span className="truncate text-muted-foreground">{detail}</span> : null}
+    </span>
+  )
+}
+
+export function ThinkingTrace({
+  label = 'Agent 正在处理',
+  steps,
+}: {
+  label?: string
+  steps?: string[]
+}) {
+  return (
+    <details open className="group rounded-lg border border-border/70 bg-muted/20">
+      <summary className="flex cursor-pointer list-none items-center gap-2 px-3 py-2.5 text-xs text-muted-foreground [&::-webkit-details-marker]:hidden">
+        <Loader2 className="size-3.5 animate-spin text-info" aria-hidden />
+        <span className="font-medium text-foreground/80">{label}</span>
+        {steps?.length ? <span className="text-muted-foreground">· {steps.length} 个步骤</span> : null}
+        <ChevronDown className="ml-auto size-3.5 transition-transform group-open:rotate-180" aria-hidden />
+      </summary>
+      {steps?.length ? (
+        <div className="space-y-1.5 border-t border-border/60 px-3 py-2.5">
+          {steps.map((step, index) => (
+            <div key={step} className="flex items-center gap-2 text-[11px] text-muted-foreground">
+              <span className="grid size-4 place-items-center rounded-full bg-background font-mono text-[9px]">{index + 1}</span>
+              {step}
+            </div>
+          ))}
+        </div>
+      ) : null}
+    </details>
+  )
+}
+
+export function ApprovalCard({
+  summary,
+  diff,
+  metadata,
+  confirming = false,
+  disabled = false,
+  onReject,
+  onConfirm,
+}: {
+  summary: string
+  diff: string
+  metadata?: Array<{ label: string; value: string }>
+  confirming?: boolean
+  disabled?: boolean
+  onReject: () => void
+  onConfirm: () => void
+}) {
+  return (
+    <section className="overflow-hidden rounded-xl border border-warning/35 bg-warning/8" aria-label="待确认操作">
+      <div className="flex items-start gap-2.5 border-b border-warning/20 px-3 py-3">
+        <span className="grid size-7 shrink-0 place-items-center rounded-full bg-warning/15 text-warning">
+          <ShieldCheck className="size-3.5" aria-hidden />
+        </span>
+        <div className="min-w-0">
+          <div className="text-sm font-medium">待确认操作</div>
+          <p className="mt-0.5 text-xs leading-5 text-muted-foreground">写入变更先展示提案，确认后才会执行。</p>
+        </div>
+      </div>
+      <div className="space-y-3 p-3">
+        <p className="text-sm leading-5">{summary}</p>
+        <div className="rounded-lg border border-border/70 bg-background/75 p-2.5 font-mono text-2xs leading-5 whitespace-pre-wrap">
+          {diff}
+        </div>
+        {metadata?.length ? (
+          <div className="grid gap-1 font-mono text-3xs text-muted-foreground">
+            {metadata.map((item) => (
+              <div key={item.label} className="flex justify-between gap-3">
+                <span>{item.label}</span>
+                <span className="truncate text-right text-foreground/70">{item.value}</span>
+              </div>
+            ))}
+          </div>
+        ) : null}
+        <div className="flex justify-end gap-2">
+          <Button variant="ghost" size="sm" disabled={confirming} onClick={onReject}>
+            <X aria-hidden />
+            拒绝
+          </Button>
+          <Button size="sm" disabled={disabled || confirming} onClick={onConfirm}>
+            {confirming ? <Loader2 className="animate-spin" aria-hidden /> : <Check aria-hidden />}
+            确认执行
+          </Button>
+        </div>
+      </div>
+    </section>
+  )
+}
+
+export function AgentTaskRow({
+  title,
+  detail,
+  meta,
+  status,
+  progress,
+}: {
+  title: string
+  detail?: string
+  meta?: string
+  status: AgentTaskStatus
+  progress?: number
+}) {
+  return (
+    <div className="group rounded-lg border border-border/70 bg-background/55 p-3 transition-colors hover:border-primary/25 hover:bg-muted/30">
+      <div className="flex items-start gap-3">
+        <span className="mt-0.5 grid size-7 shrink-0 place-items-center rounded-md bg-muted/70 text-muted-foreground">
+          {status === 'running' ? <Loader2 className="size-3.5 animate-spin text-info" aria-hidden /> : <Terminal className="size-3.5" aria-hidden />}
+        </span>
+        <div className="min-w-0 flex-1">
+          <div className="flex flex-wrap items-center justify-between gap-2">
+            <p className="truncate text-sm font-medium">{title}</p>
+            <AgentStatusBadge status={status} />
+          </div>
+          {detail ? <p className="mt-1 text-xs text-muted-foreground">{detail}</p> : null}
+          {meta ? <p className="mt-2 font-mono text-3xs text-muted-foreground">{meta}</p> : null}
+          {typeof progress === 'number' ? (
+            <div className="mt-2 h-1 overflow-hidden rounded-full bg-muted" aria-label={`进度 ${progress}%`}>
+              <div className="h-full rounded-full bg-primary/70 transition-[width]" style={{ width: `${Math.max(0, Math.min(100, progress))}%` }} />
+            </div>
+          ) : null}
+        </div>
+      </div>
+    </div>
+  )
+}
+
+export function AgentPromptBar({
+  value,
+  onChange,
+  onSubmit,
+  onKeyDown,
+  disabled = false,
+  submitting = false,
+  placeholder,
+  footer,
+}: {
+  value: string
+  onChange: (value: string) => void
+  onSubmit: (event: FormEvent<HTMLFormElement>) => void
+  onKeyDown: KeyboardEventHandler<HTMLTextAreaElement>
+  disabled?: boolean
+  submitting?: boolean
+  placeholder: string
+  footer?: ReactNode
+}) {
+  return (
+    <form className="border-t bg-background/85 p-3 backdrop-blur" onSubmit={onSubmit}>
+      <div className="rounded-xl border border-border/80 bg-muted/20 p-2 transition-colors focus-within:border-primary/40 focus-within:bg-background">
+        <Textarea
+          value={value}
+          onChange={(event) => onChange(event.target.value)}
+          onKeyDown={onKeyDown}
+          placeholder={placeholder}
+          aria-label="给全局 Agent 的消息"
+          className="max-h-32 min-h-16 resize-none border-0 bg-transparent p-1.5 shadow-none focus-visible:ring-0"
+          disabled={disabled}
+        />
+        <div className="flex items-center justify-between gap-2 px-1.5 pt-2">
+          <div className="min-w-0 truncate text-3xs text-muted-foreground">{footer}</div>
+          <Button type="submit" size="sm" disabled={!value.trim() || disabled}>
+            {submitting ? <Loader2 className="animate-spin" aria-hidden /> : <Terminal aria-hidden />}
+            发送
+          </Button>
+        </div>
+      </div>
+    </form>
+  )
+}
diff --git a/frontend/components/agents/agent-form-dialog.tsx b/frontend/components/agents/agent-form-dialog.tsx
index c26b4d93..587da92d 100644
--- a/frontend/components/agents/agent-form-dialog.tsx
+++ b/frontend/components/agents/agent-form-dialog.tsx
@@ -22,10 +22,10 @@ import { Select, SelectContent, SelectItem, SelectTrigger, SelectValue } from '@
 import { Switch } from '@/components/ui/switch'
 import { Textarea } from '@/components/ui/textarea'
 
-// Keep in sync with backend/models/agent.py's "claude | openai | local" comment
-// and AIAgentCreate.processor_type — the schema stores a plain str, so this is
-// a UX constraint, not an enum enforced server-side.
-const PROCESSOR_TYPES = ['claude', 'openai', 'local'] as const
+// Keep in sync with backend/models/agent.py's supported processor types. The
+// schema stores a plain str, so this is a UX constraint, not an enum enforced
+// server-side.
+const PROCESSOR_TYPES = ['claude', 'openai', 'local', 'paw'] as const
 type ProcessorType = (typeof PROCESSOR_TYPES)[number]
 
 // Mirrors the label map in app/(app)/agents/page.tsx (kept local/duplicated
@@ -34,6 +34,7 @@ const PROCESSOR_LABEL: Record<ProcessorType, string> = {
   claude: 'Claude',
   openai: 'OpenAI',
   local: '本地模型',
+  paw: 'PAW 本地富化',
 }
 
 // Sentinel for "no linked provider" — Select items need a non-empty value.
@@ -122,6 +123,10 @@ export function AgentFormDialog({
       toast.error('请填写 Agent 名称')
       return
     }
+    if (form.processor_type === 'paw' && !form.prompt_template.trim()) {
+      toast.error('PAW Agent 必须填写提示词模板')
+      return
+    }
 
     let processorConfig: Record<string, unknown> | undefined
     if (form.processor_config.trim()) {
@@ -216,6 +221,11 @@ export function AgentFormDialog({
                     ))}
                   </SelectContent>
                 </Select>
+                {form.processor_type === 'paw' ? (
+                  <FieldDescription>
+                    使用固定、离线的 PAW sidecar。仅发送渲染后的短提示词；程序身份与就绪状态由运行时健康检查管理。
+                  </FieldDescription>
+                ) : null}
               </Field>
 
               <Field>
@@ -280,10 +290,14 @@ export function AgentFormDialog({
                 onChange={(event) =>
                   setForm((current) => ({ ...current, prompt_template: event.target.value }))
                 }
+                required={form.processor_type === 'paw'}
                 placeholder="发送给模型的系统提示词"
                 rows={6}
                 className="font-mono text-xs"
               />
+              {form.processor_type === 'paw' ? (
+                <FieldDescription>PAW 必须使用非空的短提示词模板。</FieldDescription>
+              ) : null}
             </Field>
 
             <Field orientation="horizontal" className="items-center justify-between">
@@ -310,11 +324,15 @@ export function AgentFormDialog({
                   onChange={(event) =>
                     setForm((current) => ({ ...current, processor_config: event.target.value }))
                   }
-                  placeholder='{"temperature": 0.3}'
+                  placeholder={form.processor_type === 'paw' ? '{"max_tokens": 256, "output_schema": {"type": "object"}}' : '{"temperature": 0.3}'}
                   rows={4}
                   className="font-mono text-xs"
                 />
-                <FieldDescription>传给处理器的额外参数，留空表示使用默认配置。</FieldDescription>
+                <FieldDescription>
+                  {form.processor_type === 'paw'
+                    ? '仅可设置 max_tokens（1–512）和 output_schema；程序、缓存、URL 与离线模式由部署配置固定。'
+                    : '传给处理器的额外参数，留空表示使用默认配置。'}
+                </FieldDescription>
               </Field>
             </div>
           </details>
diff --git a/frontend/components/auth/auth-provider.tsx b/frontend/components/auth/auth-provider.tsx
index d00ee1b9..74cd01a2 100644
--- a/frontend/components/auth/auth-provider.tsx
+++ b/frontend/components/auth/auth-provider.tsx
@@ -2,9 +2,8 @@
 
 import { createContext, useCallback, useContext, useEffect, useMemo, useState } from 'react'
 
-import { getCurrentIdentity } from '@/lib/api/endpoints'
+import { changeLocalPassword, getCurrentIdentity, loginWithPassword } from '@/lib/api/endpoints'
 import { AUTH_REQUIRED_EVENT } from '@/lib/api/auth-events'
-import { setApiAuthToken } from '@/lib/api/auth-token'
 import { getOidcManager, isOidcConfigured, oidcReturnTo, sanitizeReturnTo } from '@/lib/auth/oidc'
 import {
   clearIdentityToken,
@@ -22,10 +21,11 @@ type AuthContextValue = {
   identity: AuthIdentity | null
   oidcEnabled: boolean
   developmentLoginEnabled: boolean
-  signInWithOidc: (returnTo?: string, fleetToken?: string) => Promise<void>
+  signInWithOidc: (returnTo?: string) => Promise<void>
+  signInWithPassword: (username: string, password: string) => Promise<boolean>
+  changePassword: (currentPassword: string, newPassword: string) => Promise<void>
   completeOidcSignIn: () => Promise<string>
-  signInWithBootstrap: (identityToken: string, fleetToken?: string) => Promise<void>
-  enterDevelopmentMode: (fleetToken?: string) => void
+  enterDevelopmentMode: () => void
   signOut: () => Promise<void>
 }
 
@@ -135,13 +135,26 @@ export function AuthProvider({ children }: { children: React.ReactNode }) {
     return () => window.removeEventListener(AUTH_REQUIRED_EVENT, onAuthRequired)
   }, [becomeAnonymous, developmentLoginEnabled])
 
-  const signInWithOidc = useCallback(async (returnTo = '/studio', fleetToken?: string) => {
+  const signInWithOidc = useCallback(async (returnTo = '/studio') => {
     const manager = getOidcManager()
     if (!manager) throw new Error('OIDC 登录尚未配置')
-    if (fleetToken !== undefined) setApiAuthToken(fleetToken)
     await manager.signinRedirect({ state: { returnTo: sanitizeReturnTo(returnTo) } })
   }, [])
 
+  const signInWithPassword = useCallback(
+    async (username: string, password: string) => {
+      const result = await loginWithPassword(username, password)
+      await acceptIdentityToken(result.access_token)
+      persistBootstrapIdentityToken(result.access_token)
+      return result.using_default_password
+    },
+    [acceptIdentityToken],
+  )
+
+  const changePassword = useCallback(async (currentPassword: string, newPassword: string) => {
+    await changeLocalPassword(currentPassword, newPassword)
+  }, [])
+
   const completeOidcSignIn = useCallback(async () => {
     const manager = getOidcManager()
     if (!manager) throw new Error('OIDC 登录尚未配置')
@@ -151,28 +164,13 @@ export function AuthProvider({ children }: { children: React.ReactNode }) {
     return oidcReturnTo(user)
   }, [acceptIdentityToken])
 
-  const signInWithBootstrap = useCallback(
-    async (identityToken: string, fleetToken?: string) => {
-      const trimmed = identityToken.trim()
-      if (!trimmed) throw new Error('请输入管理员身份令牌')
-      if (fleetToken !== undefined) setApiAuthToken(fleetToken)
-      await acceptIdentityToken(trimmed)
-      persistBootstrapIdentityToken(trimmed)
-    },
-    [acceptIdentityToken],
-  )
-
-  const enterDevelopmentMode = useCallback(
-    (fleetToken?: string) => {
-      if (!developmentLoginEnabled) throw new Error('本地开发模式不可用')
-      if (fleetToken !== undefined) setApiAuthToken(fleetToken)
-      clearIdentityToken()
-      setDevelopmentSession(true)
-      setIdentity(DEVELOPMENT_IDENTITY)
-      setStatus('authenticated')
-    },
-    [developmentLoginEnabled],
-  )
+  const enterDevelopmentMode = useCallback(() => {
+    if (!developmentLoginEnabled) throw new Error('本地开发模式不可用')
+    clearIdentityToken()
+    setDevelopmentSession(true)
+    setIdentity(DEVELOPMENT_IDENTITY)
+    setStatus('authenticated')
+  }, [developmentLoginEnabled])
 
   const signOut = useCallback(async () => {
     const manager = getOidcManager()
@@ -193,19 +191,21 @@ export function AuthProvider({ children }: { children: React.ReactNode }) {
       oidcEnabled,
       developmentLoginEnabled,
       signInWithOidc,
+      signInWithPassword,
+      changePassword,
       completeOidcSignIn,
-      signInWithBootstrap,
       enterDevelopmentMode,
       signOut,
     }),
     [
+      changePassword,
       completeOidcSignIn,
       developmentLoginEnabled,
       enterDevelopmentMode,
       identity,
       oidcEnabled,
-      signInWithBootstrap,
       signInWithOidc,
+      signInWithPassword,
       signOut,
       status,
     ],
diff --git a/frontend/components/eldoraui/testimonal-slider.tsx b/frontend/components/eldoraui/testimonal-slider.tsx
new file mode 100644
index 00000000..5dc54f4d
--- /dev/null
+++ b/frontend/components/eldoraui/testimonal-slider.tsx
@@ -0,0 +1,147 @@
+'use client'
+
+import Image from 'next/image'
+import { useEffect, useRef, useState } from 'react'
+
+/*
+ * Adapted from EldoraUI's MIT-style copy-paste component:
+ * https://www.eldoraui.site/docs/components/testimonal-slider
+ * The original uses Headless UI Transition and next/image; this version keeps
+ * the same rotating avatar / quote / pill navigation interaction while using
+ * the host app's image and transition primitives.
+ */
+
+export interface Testimonial {
+  img?: string
+  quote: string
+  name: string
+  role: string
+}
+
+export function FancyTestimonialsSlider({
+  testimonials,
+  autorotateTiming = 7000,
+}: {
+  testimonials: Testimonial[]
+  autorotateTiming?: number
+}) {
+  const [active, setActive] = useState(0)
+  const [autorotate, setAutorotate] = useState(true)
+  const [hovered, setHovered] = useState(false)
+  const sliderRef = useRef<HTMLDivElement>(null)
+
+  useEffect(() => {
+    if (!autorotate || hovered || testimonials.length < 2) return
+    const interval = window.setInterval(() => {
+      setActive((current) => (current + 1) % testimonials.length)
+    }, autorotateTiming)
+    return () => window.clearInterval(interval)
+  }, [autorotate, autorotateTiming, hovered, testimonials.length])
+
+  if (!testimonials.length) return null
+
+  function select(index: number) {
+    setActive(index)
+    setAutorotate(false)
+  }
+
+  function handleKeyDown(event: React.KeyboardEvent<HTMLDivElement>) {
+    if (event.key !== 'ArrowLeft' && event.key !== 'ArrowRight') return
+    event.preventDefault()
+    const delta = event.key === 'ArrowRight' ? 1 : -1
+    select((active + delta + testimonials.length) % testimonials.length)
+  }
+
+
+  return (
+    <div
+      ref={sliderRef}
+      tabIndex={0}
+      onKeyDown={handleKeyDown}
+      onMouseEnter={() => setHovered(true)}
+      onMouseLeave={() => setHovered(false)}
+      onFocus={() => setHovered(true)}
+      onBlur={() => setHovered(false)}
+      className="mx-auto w-full max-w-3xl px-4 text-center outline-none sm:px-6"
+      aria-label="Agent 洞察轮播"
+    >
+      <div className="relative h-24 sm:h-28">
+        <div className="pointer-events-none absolute top-0 left-1/2 h-[280px] w-[280px] -translate-x-1/2 before:absolute before:inset-0 before:-z-10 before:rounded-full before:bg-gradient-to-b before:from-cyan-500/25 before:via-cyan-500/5 before:via-25% before:to-cyan-500/0 before:to-75% sm:h-[360px] sm:w-[360px]">
+          <div className="h-24 [mask-image:linear-gradient(0deg,transparent,white_20%,white)] sm:h-28">
+            {testimonials.map((testimonial, index) => {
+              const visible = active === index
+              const initialsForItem = testimonial.name
+                .split(/\s+/)
+                .map((part) => part[0])
+                .join('')
+                .slice(0, 2)
+                .toUpperCase()
+              return (
+                <div
+                  key={`${testimonial.name}-${index}`}
+                  className="absolute inset-0 h-full transition-[opacity,transform] duration-700"
+                  style={{
+                    opacity: visible ? 1 : 0,
+                    transform: visible ? 'rotate(0deg)' : `rotate(${index < active ? '-60deg' : '60deg'})`,
+                    pointerEvents: visible ? 'auto' : 'none',
+                  }}
+                  aria-hidden={!visible}
+                >
+                  {testimonial.img ? (
+                    <Image className="relative top-8 left-1/2 size-12 -translate-x-1/2 rounded-full object-cover ring-4 ring-cyan-500/10 sm:top-10" src={testimonial.img} alt="" width={48} height={48} unoptimized />
+                  ) : (
+                    <span className="relative top-8 left-1/2 flex size-12 -translate-x-1/2 items-center justify-center rounded-full bg-gradient-to-br from-cyan-400 to-blue-600 text-sm font-semibold text-white ring-4 ring-cyan-500/10 sm:top-10" aria-hidden>
+                      {initialsForItem}
+                    </span>
+                  )}
+                </div>
+              )
+            })}
+          </div>
+        </div>
+      </div>
+
+      <div className="mb-6 min-h-[76px] transition-all delay-300 duration-150 ease-in-out sm:mb-8">
+        <div className="relative flex flex-col">
+          {testimonials.map((testimonial, index) => (
+            <div
+              key={`${testimonial.role}-${index}`}
+              className="px-2 text-lg font-bold text-cyan-900 transition-[opacity,transform] duration-500 dark:text-cyan-100 sm:px-0 sm:text-xl"
+              style={{
+                opacity: active === index ? 1 : 0,
+                transform: active === index ? 'translateX(0)' : `translateX(${index < active ? '-16px' : '16px'})`,
+                position: active === index ? 'relative' : 'absolute',
+                inset: active === index ? undefined : 0,
+                pointerEvents: active === index ? 'auto' : 'none',
+              }}
+              aria-hidden={active !== index}
+            >
+              <span aria-hidden>“</span>
+              {testimonial.quote}
+              <span aria-hidden>”</span>
+            </div>
+          ))}
+        </div>
+      </div>
+
+      <div className="-m-1 flex flex-wrap justify-center gap-1 sm:gap-1.5">
+        {testimonials.map((testimonial, index) => (
+          <button
+            key={`${testimonial.name}-${testimonial.role}`}
+            type="button"
+            aria-pressed={active === index}
+            aria-label={`查看 ${testimonial.name} 的洞察`}
+            onClick={() => select(index)}
+            className={`m-1 inline-flex max-w-full justify-center rounded-full px-2 py-1 text-xs whitespace-nowrap shadow-sm transition-colors duration-150 focus-visible:ring-2 focus-visible:ring-cyan-300 focus-visible:outline-none sm:px-3 sm:py-1.5 dark:focus-visible:ring-cyan-600 ${active === index ? 'bg-cyan-500 text-white shadow-cyan-950/10' : 'bg-white text-cyan-900 hover:bg-cyan-100 dark:bg-slate-800 dark:text-cyan-100 dark:hover:bg-slate-700'}`}
+          >
+            <span className="truncate">{testimonial.name}</span>
+            <span className={active === index ? 'px-1 text-cyan-200' : 'px-1 text-cyan-300'} aria-hidden>-</span>
+            <span className="truncate">{testimonial.role}</span>
+          </button>
+        ))}
+      </div>
+    </div>
+  )
+}
+
+export default FancyTestimonialsSlider
diff --git a/frontend/components/flow/collaboration.tsx b/frontend/components/flow/collaboration.tsx
index b4464f34..8fca568f 100644
--- a/frontend/components/flow/collaboration.tsx
+++ b/frontend/components/flow/collaboration.tsx
@@ -5,54 +5,70 @@ import { useReactFlow, useStore } from "@xyflow/react"
 import { useFlowStore } from "@/lib/flow/store"
 import { useSettingsStore } from "@/lib/flow/settings-store"
 import { useYjs } from "@/lib/flow/yjs-provider"
+import { persistableWorkflowProject } from "@/lib/workflow/persistence"
 
 export function Collaboration() {
-  const provider = useSettingsStore((s) => s.collabProvider)
-  const yjsUrl = useSettingsStore((s) => s.yjsUrl)
-  const yjsRoom = useSettingsStore((s) => s.yjsRoom)
-  const enabled = provider === "yjs"
-
+  const provider = useSettingsStore((state) => state.collabProvider)
+  const yjsUrl = useSettingsStore((state) => state.yjsUrl)
+  const yjsRoom = useSettingsStore((state) => state.yjsRoom)
+  const yjsEnabled = useSettingsStore((state) => state.yjsEnabled)
+  const configured = yjsEnabled && provider === "yjs"
+  const enabled = configured && Boolean(yjsUrl) && Boolean(yjsRoom)
   const binding = useYjs(enabled, { url: yjsUrl, room: yjsRoom })
+  const transportEnabled = enabled && binding.authorized
   const { screenToFlowPosition } = useReactFlow()
-  const transform = useStore((s) => s.transform)
+  const transform = useStore((state) => state.transform)
   const applyingRemote = useRef(false)
 
-  // outgoing: whenever local nodes/edges change, publish to yjs
   useEffect(() => {
-    if (!enabled) return
-    const unsub = useFlowStore.subscribe((state, prev) => {
-      if (applyingRemote.current) return
-      if (state.nodes === prev.nodes && state.edges === prev.edges) return
-      binding.publish(state.nodes, state.edges)
-    })
-    // initial push
-    const { nodes, edges } = useFlowStore.getState()
-    binding.publish(nodes, edges)
-    return unsub
-  }, [enabled, binding])
+    const ready = binding.connected && binding.synced
+    const settings = useSettingsStore.getState()
+    if (settings.yjsConnected !== ready) settings.patch({ yjsConnected: ready })
+  }, [binding.connected, binding.synced])
 
-  // incoming: apply remote to local store
+  // Incoming updates are canonical WorkflowProject records. Projecting them
+  // here keeps the canvas and the full-PUT fallback on the same source graph.
   useEffect(() => {
-    if (!enabled) return
+    if (!transportEnabled) return
     return binding.onRemote((nodes, edges) => {
       applyingRemote.current = true
-      useFlowStore.setState({ nodes, edges })
+      const current = useFlowStore.getState()
+      const { drawings, workflowProject } = current
+      current.importWorkflowProject({
+        ...workflowProject,
+        nodes: nodes as typeof workflowProject.nodes,
+        edges: edges as typeof workflowProject.edges,
+      })
+      useFlowStore.setState({ drawings })
       queueMicrotask(() => {
         applyingRemote.current = false
       })
     })
-  }, [enabled, binding])
+  }, [binding, transportEnabled])
+
+  // A synced, non-empty room is authoritative. Only seed once for a room that
+  // contains no graph yet, then publish subsequent local canonical changes.
+  useEffect(() => {
+    if (!transportEnabled) return
+    const unsubscribe = useFlowStore.subscribe((state, previous) => {
+      if (applyingRemote.current || state.workflowProject === previous.workflowProject) return
+      const persistable = persistableWorkflowProject(state.workflowProject)
+      binding.publish(persistable.nodes, persistable.edges)
+    })
+    const persistable = persistableWorkflowProject(useFlowStore.getState().workflowProject)
+    binding.initialize(persistable.nodes, persistable.edges)
+    return unsubscribe
+  }, [binding, transportEnabled])
 
-  // cursor broadcast
   useEffect(() => {
-    if (!enabled) return
+    if (!transportEnabled) return
     let raf = 0
-    const onMove = (e: MouseEvent) => {
+    const onMove = (event: MouseEvent) => {
       if (raf) return
       raf = requestAnimationFrame(() => {
         raf = 0
-        const p = screenToFlowPosition({ x: e.clientX, y: e.clientY })
-        binding.publishCursor(p.x, p.y)
+        const position = screenToFlowPosition({ x: event.clientX, y: event.clientY })
+        binding.publishCursor(position.x, position.y)
       })
     }
     window.addEventListener("mousemove", onMove)
@@ -60,51 +76,51 @@ export function Collaboration() {
       window.removeEventListener("mousemove", onMove)
       if (raf) cancelAnimationFrame(raf)
     }
-  }, [enabled, binding, screenToFlowPosition])
+  }, [binding, screenToFlowPosition, transportEnabled])
 
-  if (!enabled) return null
+  if (!configured) return null
   const [tx, ty, zoom] = transform
+  const stateLabel = !binding.authorized ? "○ TOKEN REQUIRED" : binding.connected ? binding.synced ? "● LIVE" : "○ SYNCING" : "○ OFFLINE"
+  const stateClassName = binding.connected && binding.synced ? "text-[#4ade80]" : "text-[#ff7a17]"
 
   return (
     <>
       <div className="pointer-events-none absolute inset-0 z-20 overflow-hidden">
-        {binding.cursors.map((c) => (
+        {binding.cursors.map((cursor) => (
           <div
-            key={c.id}
+            key={cursor.id}
             className="absolute flex items-center gap-1"
-            style={{ transform: `translate(${tx + c.x * zoom}px, ${ty + c.y * zoom}px)` }}
+            style={{ transform: `translate(${tx + cursor.x * zoom}px, ${ty + cursor.y * zoom}px)` }}
           >
-            <svg width="18" height="18" viewBox="0 0 24 24" style={{ color: c.color }}>
+            <svg width="18" height="18" viewBox="0 0 24 24" style={{ color: cursor.color }}>
               <path fill="currentColor" d="M4 2l6 16 2.5-6.5L19 9z" />
             </svg>
             <span
               className="rounded px-1.5 py-0.5 text-[10px] font-medium text-white shadow"
-              style={{ backgroundColor: c.color }}
+              style={{ backgroundColor: cursor.color }}
             >
-              {c.name}
+              {cursor.name}
             </span>
           </div>
         ))}
       </div>
 
       <div className="pointer-events-none absolute right-3 top-3 z-30 flex items-center gap-2 rounded-md border bg-card/90 px-2 py-1 font-mono text-[10px] shadow backdrop-blur">
-        <span className={binding.connected ? "text-[#4ade80]" : "text-[#ff7a17]"}>
-          {binding.connected ? "● LIVE" : "○ CONNECTING"}
-        </span>
+        <span className={stateClassName}>{stateLabel}</span>
         <span className="text-muted-foreground">·</span>
         <span className="text-muted-foreground">room {yjsRoom}</span>
         {binding.users.length > 0 ? (
           <>
             <span className="text-muted-foreground">·</span>
             <div className="flex -space-x-1">
-              {binding.users.slice(0, 6).map((u) => (
+              {binding.users.slice(0, 6).map((user) => (
                 <span
-                  key={u.id}
+                  key={user.id}
                   className="flex size-4 items-center justify-center rounded-full border border-background text-[8px] font-semibold text-white"
-                  style={{ backgroundColor: u.color }}
-                  title={u.name}
+                  style={{ backgroundColor: user.color }}
+                  title={user.name}
                 >
-                  {u.name[0]?.toUpperCase()}
+                  {user.name[0]?.toUpperCase()}
                 </span>
               ))}
             </div>
diff --git a/frontend/components/flow/interaction-settings-panel.tsx b/frontend/components/flow/interaction-settings-panel.tsx
index ef996723..f2ec87e0 100644
--- a/frontend/components/flow/interaction-settings-panel.tsx
+++ b/frontend/components/flow/interaction-settings-panel.tsx
@@ -201,13 +201,13 @@ export function InteractionSettingsPanel() {
             </div>
             <div className="space-y-1.5">
               <Label className="font-mono text-[10px] uppercase tracking-wider">Yjs Room</Label>
-              <Input value={s.yjsRoom} onChange={(e) => s.set("yjsRoom", e.target.value)} />
+              <Input readOnly value={s.yjsRoom} />
             </div>
             <div className="space-y-1.5">
               <Label className="font-mono text-[10px] uppercase tracking-wider">Yjs WebSocket URL</Label>
-              <Input value={s.yjsUrl} onChange={(e) => s.set("yjsUrl", e.target.value)} />
+              <Input readOnly value={s.yjsUrl} />
               <p className="text-[10px] text-muted-foreground">
-                默认 wss://demos.yjs.dev — 或自建 y-websocket-server。
+                由 Studio 工作流地址生成；可通过 NEXT_PUBLIC_YJS_URL 配置服务地址。
               </p>
             </div>
           </div>
diff --git a/frontend/components/flow/workflow-editor-session.tsx b/frontend/components/flow/workflow-editor-session.tsx
index dbe07ea6..afe69159 100644
--- a/frontend/components/flow/workflow-editor-session.tsx
+++ b/frontend/components/flow/workflow-editor-session.tsx
@@ -15,34 +15,18 @@ import { getProjectWorkflowDraft, publishProjectWorkflow, updateProjectWorkflowD
 import { useCreateProjectWorkflow, useWorkspaceProjects } from '@/lib/api/hooks'
 import type { WorkflowAssetSummary } from '@/lib/api/types'
 import { useFlowStore } from '@/lib/flow/store'
+import { resolveYjsUrl, useSettingsStore } from '@/lib/flow/settings-store'
+import { IMAGE_ASSET_CATALOG_ID } from '@/lib/workflow/node-catalog'
+import { persistableWorkflowProject } from '@/lib/workflow/persistence'
+import { parseWorkflowProject, type WorkflowProject } from '@/lib/workflow/schema'
 import { studioGraphForTemplate } from '@/lib/workflow/studio-templates'
 import { useWorkflowCapabilities } from '@/lib/workflow/use-workflow-capabilities'
-import { parseWorkflowProject, type WorkflowProject, type WorkflowProjectNode } from '@/lib/workflow/schema'
-import { IMAGE_ASSET_CATALOG_ID } from '@/lib/workflow/node-catalog'
 
 import { WorkflowEditor } from './workflow-editor'
 
-function persistableProject(project: WorkflowProject): WorkflowProject {
-  const persistableNode = (node: WorkflowProjectNode): WorkflowProjectNode => {
-    const ui = { ...(node.ui ?? {}) }
-    delete ui.runtimeCapability
-    return {
-      ...node,
-      ...(node.ui ? { ui } : {}),
-      ...(node.internals
-        ? {
-            internals: {
-              ...node.internals,
-              nodes: node.internals.nodes.map((item) => persistableNode(item as WorkflowProjectNode)),
-            },
-          }
-        : {}),
-    }
-  }
-  return { ...project, nodes: project.nodes.map(persistableNode) }
-}
 
-const projectFingerprint = (project: WorkflowProject) => JSON.stringify(persistableProject(project))
+const projectFingerprint = (project: WorkflowProject) =>
+  JSON.stringify(persistableWorkflowProject(project))
 
 type WorkflowEditorSessionProps = {
   forceStandalone?: boolean
@@ -64,6 +48,8 @@ export function WorkflowEditorSession({ forceStandalone = false }: WorkflowEdito
   const workflowProject = useFlowStore((state) => state.workflowProject)
   const importWorkflowProject = useFlowStore((state) => state.importWorkflowProject)
   const updateWorkflowNodeParams = useFlowStore((state) => state.updateWorkflowNodeParams)
+  const yjsConnected = useSettingsStore((state) => state.yjsConnected)
+  const yjsEnabled = useSettingsStore((state) => state.yjsEnabled)
   const [workflowId, setWorkflowId] = useState<string | null>(null)
   const [loadState, setLoadState] = useState<WorkflowLoadState>('loading')
   const [loadError, setLoadError] = useState<string | null>(null)
@@ -88,11 +74,22 @@ export function WorkflowEditorSession({ forceStandalone = false }: WorkflowEdito
   const missingProjectContext = !standalone && (!workspaceId || !projectId)
   const projectHref = workspaceId && projectId ? `/studio/projects/${projectId}?workspace=${workspaceId}` : '/studio'
 
+  useEffect(() => {
+    return () => {
+      useSettingsStore.getState().patch({ collabProvider: 'off', yjsEnabled: false, yjsConnected: false })
+    }
+  }, [])
+
   const saveDraft = useCallback(
-    (graph: typeof workflowProject) => {
-      if (!workspaceId || !projectId || !workflowId || revision.current === null || saveBlocked.current) return Promise.reject(new Error('草稿尚未就绪'))
-      const persistableGraph = persistableProject(graph)
-      if (lastSavedFingerprint.current === projectFingerprint(persistableGraph)) return Promise.resolve()
+    async (graph: typeof workflowProject) => {
+      if (!workspaceId || !projectId || !workflowId || revision.current === null || saveBlocked.current) throw new Error('草稿尚未就绪')
+      const persistableGraph = persistableWorkflowProject(graph)
+      if (lastSavedFingerprint.current === projectFingerprint(persistableGraph)) return
+      if (yjsEnabled) {
+        const latestDraft = await getProjectWorkflowDraft(workspaceId, projectId, workflowId)
+        revision.current = latestDraft.revision
+        setSavedRevision(latestDraft.revision)
+      }
       pendingGraph.current = persistableGraph
       setDocumentState('saving')
       if (!saveQueuePromise.current) {
@@ -120,12 +117,13 @@ export function WorkflowEditorSession({ forceStandalone = false }: WorkflowEdito
       }
       return saveQueuePromise.current
     },
-    [projectId, workflowId, workspaceId],
+    [projectId, workflowId, workspaceId, yjsEnabled],
   )
 
   useEffect(() => {
     if (!workspaceId || !projectId) return
     if (primaryWorkflowPending) return
+    useSettingsStore.getState().patch({ collabProvider: 'off', yjsEnabled: false, yjsConnected: false })
     let active = true
     loaded.current = false
     revision.current = null
@@ -160,6 +158,13 @@ export function WorkflowEditorSession({ forceStandalone = false }: WorkflowEdito
         setSavedRevision(draft.revision)
         lastSavedFingerprint.current = projectFingerprint(graph)
         setWorkflowId(resolvedWorkflowId)
+        useSettingsStore.getState().patch({
+          collabProvider: 'yjs',
+          yjsEnabled: true,
+          yjsConnected: false,
+          yjsRoom: `workspace:${workspaceId}:project:${projectId}:workflow:${resolvedWorkflowId}`,
+          yjsUrl: resolveYjsUrl(),
+        })
         loaded.current = true
         setDocumentState('saved')
         setLoadState('ready')
@@ -184,12 +189,47 @@ export function WorkflowEditorSession({ forceStandalone = false }: WorkflowEdito
   }, [importWorkflowProject, primaryWorkflowPending, project, projectId, requestedWorkflowId, resolvedWorkflowId, workspaceId, workspaceProjects.error, workspaceProjects.isError])
 
   useEffect(() => {
-    if (!loaded.current || !workspaceId || !projectId || !workflowId) return
+    if (!loaded.current || !workspaceId || !projectId || !workflowId || yjsConnected) return
     const timer = window.setTimeout(() => {
       saveDraft(workflowProject).catch((reason: Error) => toast.error(`自动保存失败：${reason.message}`))
     }, 800)
     return () => window.clearTimeout(timer)
-  }, [projectId, saveDraft, workflowId, workflowProject, workspaceId])
+  }, [projectId, saveDraft, workflowId, workflowProject, workspaceId, yjsConnected])
+
+  useEffect(() => {
+    if (!loaded.current || !workspaceId || !projectId || !workflowId || !yjsConnected) return
+    const fingerprint = projectFingerprint(workflowProject)
+    if (lastSavedFingerprint.current === fingerprint) return
+    const expectedRevision = revision.current ?? 0
+    setDocumentState('saving')
+    const timer = window.setTimeout(() => {
+      void (async () => {
+        try {
+          const latest = await getProjectWorkflowDraft(workspaceId, projectId, workflowId)
+          if (latest.revision <= expectedRevision) {
+            await saveDraft(workflowProject)
+            return
+          }
+          revision.current = latest.revision
+          setSavedRevision(latest.revision)
+          lastSavedFingerprint.current = fingerprint
+          setDocumentState('saved')
+        } catch (reason) {
+          try {
+            await saveDraft(workflowProject)
+          } catch {
+            setDocumentState('error')
+            toast.error(
+              reason instanceof Error
+                ? `协同快照确认失败：${reason.message}`
+                : '协同快照确认失败',
+            )
+          }
+        }
+      })()
+    }, 1_000)
+    return () => window.clearTimeout(timer)
+  }, [projectId, saveDraft, workflowId, workflowProject, workspaceId, yjsConnected])
 
   useEffect(() => {
     if (!loaded.current) return
@@ -284,6 +324,7 @@ export function WorkflowEditorSession({ forceStandalone = false }: WorkflowEdito
     }
     setReleaseState('publishing')
     try {
+      await saveDraft(workflowProject)
       const version = await publishProjectWorkflow(workspaceId, projectId, workflowId, {
         reason: '通过工作区画布发布',
         expectedRevision: revision.current,
diff --git a/frontend/components/gpu/gpu-surface.tsx b/frontend/components/gpu/gpu-surface.tsx
new file mode 100644
index 00000000..3861f451
--- /dev/null
+++ b/frontend/components/gpu/gpu-surface.tsx
@@ -0,0 +1,220 @@
+'use client'
+
+import {
+  Component,
+  type ReactNode,
+  useEffect,
+  useRef,
+  useState,
+} from 'react'
+
+import {
+  invalidateGpuCapabilitiesAfterContextLoss,
+  probeGpuCapabilities,
+  type GpuRendererBackend,
+} from '@/lib/rendering/gpu-capabilities'
+
+type GpuFallbackReason =
+  | 'probing'
+  | 'unavailable'
+  | 'reduced-motion'
+  | 'page-hidden'
+  | 'context-lost'
+  | 'renderer-error'
+
+type GpuSurfaceState = {
+  backend: GpuRendererBackend
+  fallbackReason: GpuFallbackReason | null
+  webgpuApiPresent: boolean
+}
+
+type GpuSurfaceProps = {
+  children: ReactNode
+  fallback: ReactNode
+  surface: string
+  className?: string
+}
+
+type RendererErrorBoundaryProps = {
+  children: ReactNode
+  onError: () => void
+}
+
+type RendererErrorBoundaryState = {
+  failed: boolean
+}
+
+class RendererErrorBoundary extends Component<
+  RendererErrorBoundaryProps,
+  RendererErrorBoundaryState
+> {
+  state: RendererErrorBoundaryState = { failed: false }
+
+  static getDerivedStateFromError(): RendererErrorBoundaryState {
+    return { failed: true }
+  }
+
+  componentDidCatch() {
+    this.props.onError()
+  }
+
+  render() {
+    return this.state.failed ? null : this.props.children
+  }
+}
+
+export function GpuSurface({
+  children,
+  fallback,
+  surface,
+  className,
+}: GpuSurfaceProps) {
+  const hostRef = useRef<HTMLDivElement>(null)
+  const retryTimerRef = useRef<number | null>(null)
+  const refreshSurfaceRef = useRef<() => void>(() => {})
+  const [state, setState] = useState<GpuSurfaceState>({
+    backend: 'pending',
+    fallbackReason: 'probing',
+    webgpuApiPresent: false,
+  })
+
+  useEffect(() => {
+    let disposed = false
+    const motionQuery = typeof window.matchMedia === 'function'
+      ? window.matchMedia('(prefers-reduced-motion: reduce)')
+      : null
+
+    const updateSurface = () => {
+      if (disposed) return
+
+      const capabilities = probeGpuCapabilities()
+      if (document.hidden) {
+        setState({
+          backend: 'fallback',
+          fallbackReason: 'page-hidden',
+          webgpuApiPresent: capabilities.webgpuApiPresent,
+        })
+        return
+      }
+      if (motionQuery?.matches) {
+        setState({
+          backend: 'fallback',
+          fallbackReason: 'reduced-motion',
+          webgpuApiPresent: capabilities.webgpuApiPresent,
+        })
+        return
+      }
+      if (!capabilities.webgl2) {
+        setState({
+          backend: 'fallback',
+          fallbackReason: 'unavailable',
+          webgpuApiPresent: capabilities.webgpuApiPresent,
+        })
+        return
+      }
+      setState({
+        backend: 'webgl2',
+        fallbackReason: null,
+        webgpuApiPresent: capabilities.webgpuApiPresent,
+      })
+    }
+
+    refreshSurfaceRef.current = updateSurface
+    updateSurface()
+    document.addEventListener('visibilitychange', updateSurface)
+    motionQuery?.addEventListener('change', updateSurface)
+    return () => {
+      disposed = true
+      refreshSurfaceRef.current = () => {}
+      document.removeEventListener('visibilitychange', updateSurface)
+      motionQuery?.removeEventListener('change', updateSurface)
+    }
+  }, [])
+
+  useEffect(() => () => {
+    if (retryTimerRef.current !== null) {
+      window.clearTimeout(retryTimerRef.current)
+    }
+  }, [])
+
+  useEffect(() => {
+    if (state.backend !== 'webgl2') return
+
+    const host = hostRef.current
+    if (!host) return
+
+    const observedCanvases = new Set<HTMLCanvasElement>()
+    const scheduleRecovery = () => {
+      invalidateGpuCapabilitiesAfterContextLoss()
+      if (retryTimerRef.current !== null) {
+        window.clearTimeout(retryTimerRef.current)
+      }
+      retryTimerRef.current = window.setTimeout(() => {
+        retryTimerRef.current = null
+        refreshSurfaceRef.current()
+      }, 250)
+    }
+    const handleContextLost = (event: Event) => {
+      event.preventDefault()
+      scheduleRecovery()
+      setState((current) => current.backend === 'webgl2'
+        ? {
+            ...current,
+            backend: 'fallback',
+            fallbackReason: 'context-lost',
+          }
+        : current)
+    }
+    const observeCanvases = () => {
+      observedCanvases.forEach((canvas) => {
+        if (host.contains(canvas)) return
+        canvas.removeEventListener('webglcontextlost', handleContextLost)
+        observedCanvases.delete(canvas)
+      })
+      host.querySelectorAll('canvas').forEach((canvas) => {
+        if (observedCanvases.has(canvas)) return
+        canvas.addEventListener('webglcontextlost', handleContextLost)
+        observedCanvases.add(canvas)
+      })
+    }
+
+    observeCanvases()
+    const observer = new MutationObserver(observeCanvases)
+    observer.observe(host, { childList: true, subtree: true })
+    return () => {
+      observer.disconnect()
+      observedCanvases.forEach((canvas) => {
+        canvas.removeEventListener('webglcontextlost', handleContextLost)
+      })
+    }
+  }, [state.backend])
+
+  return (
+    <div
+      ref={hostRef}
+      className={className}
+      data-gpu-surface={surface}
+      data-gpu-backend={state.backend}
+      data-gpu-fallback-reason={state.fallbackReason ?? undefined}
+      data-gpu-webgpu-api={state.webgpuApiPresent ? 'present' : 'absent'}
+    >
+      {state.backend === 'webgl2' ? (
+        <RendererErrorBoundary
+          onError={() => {
+            setState((current) => ({
+              ...current,
+              backend: 'fallback',
+              fallbackReason: 'renderer-error',
+            }))
+          }}
+        >
+          {children}
+        </RendererErrorBoundary>
+      ) : (
+        <div className="h-full">
+          {fallback}
+        </div>
+      )}
+    </div>
+  )
+}
diff --git a/frontend/components/inbox/queue-detail.tsx b/frontend/components/inbox/queue-detail.tsx
new file mode 100644
index 00000000..3466d613
--- /dev/null
+++ b/frontend/components/inbox/queue-detail.tsx
@@ -0,0 +1,247 @@
+'use client'
+
+import { ArrowUpRight } from 'lucide-react'
+import Link from 'next/link'
+import { useState, type ReactNode } from 'react'
+
+import AIApproval, { type AIApprovalOption } from '@/components/smoothui/ai-approval'
+import { StatusBadge } from '@/components/shell/status-badge'
+import { buttonVariants } from '@/components/ui/button'
+import { ScrollArea } from '@/components/ui/scroll-area'
+import { useDecideOperationsApproval } from '@/lib/api/hooks'
+import type { ApprovalDecision, OperationsWorkItem } from '@/lib/api/types'
+
+export type QueueDetailItem = {
+  id: string
+  section: 'blocked' | 'waiting' | 'review'
+  eyebrow: string
+  title: string
+  summary: string
+  status: string
+  createdAt: string
+  href: string
+  hrefLabel: string
+  sourceId?: string
+  sourceName?: string
+  occurrenceCount: number
+  detailLabel: string
+  detailValue: string
+  approval?: OperationsWorkItem
+}
+
+export type QueueSectionMeta = {
+  label: string
+  icon: ReactNode
+  iconTone: string
+}
+
+function shortId(value: string) {
+  return value.length > 12 ? `${value.slice(0, 8)}…` : value
+}
+
+export function QueueDetail({
+  createdAtLabel,
+  item,
+  meta,
+}: {
+  createdAtLabel: string
+  item: QueueDetailItem
+  meta: QueueSectionMeta
+}) {
+  const nextStep = {
+    blocked: '先检查错误和运行参数，再决定是否重新触发采集。',
+    waiting: '确认执行容量或通知目标状态，避免事项长期停留在队列。',
+    review: '观察后续运行是否恢复，并在控制证据中完成结果判断。',
+  }[item.section]
+
+  return (
+    <div className="flex min-h-0 flex-1 flex-col">
+      <div className="flex min-h-16 items-start justify-between gap-4 border-b px-5 py-3.5">
+        <div className="flex min-w-0 items-start gap-3">
+          <span className={`grid size-8 shrink-0 place-items-center rounded-md ${meta.iconTone}`}>
+            {meta.icon}
+          </span>
+          <div className="min-w-0">
+            <p className="text-xs font-medium text-muted-foreground">{item.eyebrow}</p>
+            <h2 className="mt-0.5 truncate text-base font-semibold leading-tight">{item.title}</h2>
+          </div>
+        </div>
+        <StatusBadge status={item.status} className="shrink-0" />
+      </div>
+
+      <ScrollArea data-testid="inbox-detail-scroll" className="min-h-0 flex-1">
+        <div className="mx-auto w-full max-w-3xl space-y-7 p-5 lg:p-7">
+          <section aria-labelledby="signal-context-heading">
+            <h3 id="signal-context-heading" className="text-xs font-medium text-muted-foreground">
+              信号上下文
+            </h3>
+            <p className="mt-3 whitespace-pre-wrap break-words text-sm leading-6 text-foreground/90">
+              {item.summary}
+            </p>
+            {item.occurrenceCount > 1 ? (
+              <p className="mt-4 border-l-2 border-border pl-3 text-xs leading-5 text-muted-foreground">
+                已将 {item.occurrenceCount} 条同一对象、同一处理阶段的信号合并为一个主题，当前展示最近一次上下文。
+              </p>
+            ) : null}
+          </section>
+
+          <section aria-labelledby="signal-facts-heading">
+            <h3 id="signal-facts-heading" className="text-xs font-medium text-muted-foreground">
+              关键信息
+            </h3>
+            <dl className="mt-3 divide-y border-y text-sm">
+              <div className="grid grid-cols-[7rem_minmax(0,1fr)] gap-3 py-2.5">
+                <dt className="text-muted-foreground">队列</dt>
+                <dd>{meta.label}</dd>
+              </div>
+              <div className="grid grid-cols-[7rem_minmax(0,1fr)] gap-3 py-2.5">
+                <dt className="text-muted-foreground">{item.detailLabel}</dt>
+                <dd className="break-words">{item.detailValue}</dd>
+              </div>
+              {item.sourceId ? (
+                <div className="grid grid-cols-[7rem_minmax(0,1fr)] gap-3 py-2.5">
+                  <dt className="text-muted-foreground">数据源</dt>
+                  <dd>
+                    <Link
+                      href={`/sources/${item.sourceId}`}
+                      className="inline-flex items-center gap-1 font-medium hover:underline focus-visible:rounded-sm focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring"
+                    >
+                      {item.sourceName ?? shortId(item.sourceId)}
+                      <ArrowUpRight aria-hidden="true" className="size-3.5 text-muted-foreground" />
+                    </Link>
+                  </dd>
+                </div>
+              ) : null}
+              <div className="grid grid-cols-[7rem_minmax(0,1fr)] gap-3 py-2.5">
+                <dt className="text-muted-foreground">最近发生</dt>
+                <dd>{createdAtLabel}</dd>
+              </div>
+            </dl>
+          </section>
+
+          <section aria-labelledby="signal-next-step-heading">
+            <h3 id="signal-next-step-heading" className="text-xs font-medium text-muted-foreground">
+              建议下一步
+            </h3>
+            <p className="mt-3 text-sm leading-6 text-muted-foreground">{nextStep}</p>
+          </section>
+        </div>
+      </ScrollArea>
+
+      <div className="flex min-h-14 flex-wrap items-center justify-between gap-3 border-t px-5 py-2.5">
+        <span className="hidden text-xs text-muted-foreground sm:inline">
+          <kbd className="rounded border bg-muted px-1.5 py-0.5 font-mono text-[10px]">J</kbd>
+          <kbd className="ml-1 rounded border bg-muted px-1.5 py-0.5 font-mono text-[10px]">K</kbd>
+          <span className="ml-2">切换</span>
+          <kbd className="ml-3 rounded border bg-muted px-1.5 py-0.5 font-mono text-[10px]">Enter</kbd>
+          <span className="ml-2">打开</span>
+        </span>
+        <Link href={item.href} className={buttonVariants({ size: 'sm' })}>
+          {item.hrefLabel}
+          <ArrowUpRight aria-hidden="true" className="size-3.5" />
+        </Link>
+      </div>
+    </div>
+  )
+}
+
+const APPROVAL_OPTIONS: AIApprovalOption[] = [
+  { id: 'approve', label: '批准执行', detail: '进入执行队列' },
+  { id: 'request_changes', label: '要求修改', detail: '返回提案作者' },
+  { id: 'reject', label: '拒绝', detail: '不执行此提案', destructive: true },
+]
+
+export function ApprovalQueueDetail({ item }: { item: QueueDetailItem }) {
+  const decision = useDecideOperationsApproval()
+  const [reason, setReason] = useState('')
+  const [decisionError, setDecisionError] = useState<string | null>(null)
+  const [cardRevision, setCardRevision] = useState(0)
+  const approval = item.approval
+
+  async function decideApproval(option: AIApprovalOption) {
+    if (!approval || !reason.trim() || decision.isPending) return
+    setDecisionError(null)
+    try {
+      await decision.mutateAsync({
+        workspaceId: approval.workspace_id,
+        approvalId: approval.id,
+        decision: option.id as ApprovalDecision,
+        reason: reason.trim(),
+      })
+    } catch (error) {
+      setDecisionError(error instanceof Error ? error.message : '审批决定提交失败')
+      setCardRevision((revision) => revision + 1)
+    }
+  }
+
+  if (!approval) return null
+
+  return (
+    <div className="flex min-h-0 flex-1 flex-col">
+      <div className="flex min-h-16 items-start justify-between gap-4 border-b px-5 py-3.5">
+        <div className="min-w-0">
+          <p className="text-xs font-medium text-muted-foreground">{item.eyebrow}</p>
+          <h2 className="mt-0.5 truncate text-base font-semibold leading-tight">{item.title}</h2>
+        </div>
+        <StatusBadge status={item.status} className="shrink-0" />
+      </div>
+
+      <ScrollArea data-testid="inbox-detail-scroll" className="min-h-0 flex-1">
+        <div className="mx-auto w-full max-w-3xl space-y-5 p-5 lg:p-7">
+          <label className="block space-y-2">
+            <span className="text-xs font-medium text-muted-foreground">审批理由（必填）</span>
+            <input
+              value={reason}
+              onChange={(event) => setReason(event.target.value)}
+              placeholder="说明批准、拒绝或要求修改的依据"
+              aria-describedby="approval-reason-hint"
+              disabled={decision.isPending}
+              className="h-9 w-full rounded-md border bg-transparent px-3 text-sm outline-none transition-shadow placeholder:text-muted-foreground focus-visible:ring-2 focus-visible:ring-ring disabled:cursor-not-allowed disabled:opacity-50"
+            />
+            <span id="approval-reason-hint" className="block text-[11px] text-muted-foreground">
+              理由会随决定写入审计记录。
+            </span>
+          </label>
+
+          <AIApproval
+            key={`${approval.id}-${cardRevision}`}
+            question={item.title}
+            options={APPROVAL_OPTIONS}
+            disabled={!reason.trim()}
+            pending={decision.isPending}
+            onDecide={(option) => void decideApproval(option)}
+          >
+            <p>{item.summary}</p>
+            <p className="mt-1">
+              {item.detailLabel}：{item.detailValue} · Workspace：{item.sourceName}
+            </p>
+          </AIApproval>
+
+          {decisionError ? (
+            <p role="alert" className="rounded-lg border border-destructive/30 bg-destructive/5 p-3 text-sm text-destructive">
+              {decisionError}。决定尚未生效，请检查连接后重试。
+            </p>
+          ) : null}
+
+          {Object.keys(approval.evidence).length ? (
+            <details className="rounded-lg border bg-background">
+              <summary className="cursor-pointer px-3 py-2 text-xs font-medium focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring">
+                查看原始审批证据
+              </summary>
+              <pre className="max-h-72 overflow-auto border-t p-3 font-mono text-[10px] leading-4">
+                {JSON.stringify(approval.evidence, null, 2)}
+              </pre>
+            </details>
+          ) : null}
+        </div>
+      </ScrollArea>
+
+      <div className="flex min-h-14 items-center justify-end border-t px-5 py-2.5">
+        <Link href="/operations-agents" className={buttonVariants({ variant: 'outline', size: 'sm' })}>
+          打开智能体
+          <ArrowUpRight aria-hidden="true" className="size-3.5" />
+        </Link>
+      </div>
+    </div>
+  )
+}
diff --git a/frontend/components/motion/app-route-transition.tsx b/frontend/components/motion/app-route-transition.tsx
index e04b5145..c5cc121b 100644
--- a/frontend/components/motion/app-route-transition.tsx
+++ b/frontend/components/motion/app-route-transition.tsx
@@ -16,6 +16,7 @@ const APP_ROUTES = [
   '/schedules',
   '/notifications',
   '/agents',
+  '/operations-agents',
   '/skills',
   '/providers',
   '/nodes',
diff --git a/frontend/components/providers/primary-model-card.tsx b/frontend/components/providers/primary-model-card.tsx
index 88bb7990..69888a49 100644
--- a/frontend/components/providers/primary-model-card.tsx
+++ b/frontend/components/providers/primary-model-card.tsx
@@ -2,7 +2,7 @@
 
 import { useEffect, useMemo, useState } from 'react'
 import Link from 'next/link'
-import { CheckCircle2, Loader2, RefreshCw, Route, Sparkles } from 'lucide-react'
+import { CheckCircle2, CircleHelp, Loader2, RefreshCw, Route, Sparkles } from 'lucide-react'
 import { toast } from 'sonner'
 
 import {
@@ -27,6 +27,42 @@ function sameCandidate(a?: ModelDefaultCandidate, b?: ModelDefaultCandidate) {
   return a?.provider_id === b?.provider_id && a?.model_id === b?.model_id
 }
 
+function UsageVisibilityCard({ providers }: { providers: ModelProvider[] }) {
+  const enabled = providers.filter((provider) => provider.enabled).length
+
+  return (
+    <Card className="border-dashed">
+      <CardHeader className="flex flex-row items-start justify-between gap-3">
+        <div>
+          <CardTitle className="flex items-center gap-2 text-base">
+            <CircleHelp className="size-4 text-muted-foreground" />
+            用量与额度
+          </CardTitle>
+          <CardDescription>不把运行时长误报成额度，只有供应商返回 usage 数据时才展示数字。</CardDescription>
+        </div>
+        <Badge variant="outline">诚实状态</Badge>
+      </CardHeader>
+      <CardContent className="grid gap-2 text-sm sm:grid-cols-3">
+        <div className="rounded-lg border bg-muted/20 p-3">
+          <div className="text-xs text-muted-foreground">本地 Claude Code</div>
+          <div className="mt-1 font-medium">额度不可读取</div>
+          <p className="mt-1 text-xs leading-5 text-muted-foreground">CLI 没有稳定的 5 小时额度查询接口。</p>
+        </div>
+        <div className="rounded-lg border bg-muted/20 p-3">
+          <div className="text-xs text-muted-foreground">已启用 API</div>
+          <div className="mt-1 font-mono text-lg">{enabled}</div>
+          <p className="mt-1 text-xs leading-5 text-muted-foreground">当前没有接入 provider usage endpoint。</p>
+        </div>
+        <div className="rounded-lg border border-primary/20 bg-primary/[0.04] p-3">
+          <div className="text-xs text-muted-foreground">执行策略</div>
+          <div className="mt-1 font-medium text-primary">允许深度执行</div>
+          <p className="mt-1 text-xs leading-5 text-muted-foreground">任务按完成度和超时策略结束，不因猜测额度提前截断。</p>
+        </div>
+      </CardContent>
+    </Card>
+  )
+}
+
 export function PrimaryModelCard({ providers }: { providers: ModelProvider[] }) {
   const { data: defaultsData, isLoading: defaultsLoading } = useModelDefaults()
   const defaultMap = useMemo(
@@ -283,6 +319,8 @@ export function PrimaryModelCard({ providers }: { providers: ModelProvider[] })
         </p>
       )}
 
+      <UsageVisibilityCard providers={providers} />
+
       <ProviderFormDialog
         mode="create"
         triggerLabel="添加供应商"
diff --git a/frontend/components/records/project-graph-explorer.tsx b/frontend/components/records/project-graph-explorer.tsx
index 2e3a5788..6939f986 100644
--- a/frontend/components/records/project-graph-explorer.tsx
+++ b/frontend/components/records/project-graph-explorer.tsx
@@ -13,6 +13,7 @@ import Link from 'next/link'
 import { useSearchParams } from 'next/navigation'
 import { useMemo, useState } from 'react'
 
+import { GpuSurface } from '@/components/gpu/gpu-surface'
 import { EmptyState, ErrorState, LoadingState } from '@/components/shell/data-states'
 import { PageContainer } from '@/components/shell/page-container'
 import { ProjectNavigation } from '@/components/studio/project-navigation'
@@ -31,7 +32,7 @@ import {
   useProjectWorkflows,
   useWorkspaceProjects,
 } from '@/lib/api/hooks'
-import type { RecordGraphNode } from '@/lib/api/types'
+import type { ProjectRecordGraphPreview, RecordGraphNode } from '@/lib/api/types'
 import {
   RECORD_GRAPH_KIND_COLOR,
   RECORD_GRAPH_KIND_LABEL,
@@ -246,21 +247,37 @@ export function ProjectGraphExplorer({
           />
         ) : (
           <div className="relative min-h-[44rem]">
-            <div className="min-h-[44rem]">
-              {isGalaxy ? (
+            {isGalaxy ? (
+              <GpuSurface
+                surface="project-galaxy-graph"
+                className="min-h-[44rem]"
+                fallback={(
+                  <ProjectGraphFallback
+                    preview={preview}
+                    selectedNodeId={selectedNodeId}
+                    onSelectNode={setSelectedNodeId}
+                  />
+                )}
+              >
                 <ProjectGalaxyForceGraph
                   preview={preview}
                   selectedNodeId={selectedNodeId}
                   onSelectNode={setSelectedNodeId}
                 />
-              ) : (
+              </GpuSurface>
+            ) : (
+              <div
+                className="min-h-[44rem]"
+                data-gpu-surface="project-relationships-graph"
+                data-gpu-backend="canvas2d"
+              >
                 <ProjectRelationshipForceGraph
                   preview={preview}
                   selectedNodeId={selectedNodeId}
                   onSelectNode={setSelectedNodeId}
                 />
-              )}
-            </div>
+              </div>
+            )}
             {selectedNode ? (
               <GraphInspector
                 node={selectedNode}
@@ -274,6 +291,58 @@ export function ProjectGraphExplorer({
   )
 }
 
+function ProjectGraphFallback({
+  preview,
+  selectedNodeId,
+  onSelectNode,
+}: {
+  preview: ProjectRecordGraphPreview
+  selectedNodeId: string | null
+  onSelectNode: (nodeId: string | null) => void
+}) {
+  const initialNodes = preview.nodes.slice(0, 200)
+
+  return (
+    <div
+      className="min-h-[44rem] overflow-y-auto bg-muted/20 p-4"
+      data-project-graph-fallback
+    >
+      <div className="rounded-lg border bg-background p-4">
+        <h2 className="text-sm font-semibold">GPU 图谱不可用</h2>
+        <p className="mt-1 text-sm text-muted-foreground">
+          当前显示前 200 个节点；可使用上方搜索定位其余节点，并查看关联详情。
+        </p>
+      </div>
+      <div className="mt-4 grid gap-2 sm:grid-cols-2 xl:grid-cols-3">
+        {initialNodes.map((node) => (
+          <button
+            key={node.id}
+            type="button"
+            aria-pressed={selectedNodeId === node.id}
+            onClick={() => onSelectNode(node.id)}
+            className={cn(
+              'rounded-lg border bg-background p-3 text-left transition-colors hover:bg-muted focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring',
+              selectedNodeId === node.id && 'border-primary bg-primary/5',
+            )}
+          >
+            <span className="flex items-center gap-2 text-xs text-muted-foreground">
+              <span
+                className="size-2 rounded-full"
+                style={{ backgroundColor: RECORD_GRAPH_KIND_COLOR[node.kind] }}
+              />
+              {RECORD_GRAPH_KIND_LABEL[node.kind]}
+            </span>
+            <span className="mt-2 block break-words text-sm font-medium">{node.label}</span>
+            <span className="mt-1 block text-xs text-muted-foreground">
+              {node.count.toLocaleString('zh-CN')} 条关联数据
+            </span>
+          </button>
+        ))}
+      </div>
+    </div>
+  )
+}
+
 function GraphInspector({
   node,
   relatedCount,
diff --git a/frontend/components/shell/app-header.tsx b/frontend/components/shell/app-header.tsx
index 2f2b5455..04db630d 100644
--- a/frontend/components/shell/app-header.tsx
+++ b/frontend/components/shell/app-header.tsx
@@ -1,7 +1,7 @@
 'use client'
 
 import Link from 'next/link'
-import { Bot, LogOut, Search } from 'lucide-react'
+import { Bot, LogOut, Search, Settings } from 'lucide-react'
 import { usePathname, useRouter } from 'next/navigation'
 import { Fragment } from 'react'
 
@@ -39,10 +39,8 @@ function resolveLabels(pathname: string): string[] {
   return match ? [ROUTE_LABELS[match]] : []
 }
 export function AppHeader({
-  onOpenAgent,
   onOpenCommand,
 }: {
-  onOpenAgent?: () => void
   onOpenCommand?: () => void
 }) {
   const pathname = usePathname()
@@ -91,17 +89,19 @@ export function AppHeader({
           variant="outline"
           size="sm"
           className="hidden gap-2 sm:flex"
-          onClick={onOpenAgent}
+          nativeButton={false}
+          render={<Link href="/operations-agents" />}
         >
           <Bot />
-          <span>Agent</span>
+          <span>自动化与智能体</span>
         </Button>
         <Button
           variant="ghost"
           size="icon"
           className="sm:hidden"
-          aria-label="打开全局 Agent"
-          onClick={onOpenAgent}
+          aria-label="打开自动化与智能体"
+          nativeButton={false}
+          render={<Link href="/operations-agents" />}
         >
           <Bot />
         </Button>
@@ -140,6 +140,10 @@ export function AppHeader({
                 <span className="block truncate text-xs">{accountLabel}</span>
               </DropdownMenuLabel>
             </DropdownMenuGroup>
+            <DropdownMenuItem onClick={() => router.push('/system')}>
+              <Settings />
+              系统设置
+            </DropdownMenuItem>
             <DropdownMenuSeparator />
             <DropdownMenuItem onClick={handleSignOut}>
               <LogOut />
diff --git a/frontend/components/shell/app-shell.tsx b/frontend/components/shell/app-shell.tsx
index 5d5c7730..d56885b0 100644
--- a/frontend/components/shell/app-shell.tsx
+++ b/frontend/components/shell/app-shell.tsx
@@ -1,24 +1,34 @@
 'use client'
 
-import { useState } from 'react'
+import { useEffect, useState } from 'react'
 
 import { AppRouteTransition } from '@/components/motion/app-route-transition'
 import { AppHeader } from '@/components/shell/app-header'
 import { AppSidebar } from '@/components/shell/app-sidebar'
 import { CommandPalette } from '@/components/shell/command-palette'
+import { GlobalAgentBubble } from '@/components/shell/global-agent-bubble'
 import { GlobalAgentDock } from '@/components/shell/global-agent-dock'
 import { SidebarInset, SidebarProvider } from '@/components/ui/sidebar'
 
 export function AppShell({ children }: { children: React.ReactNode }) {
   const [commandOpen, setCommandOpen] = useState(false)
   const [agentOpen, setAgentOpen] = useState(false)
+  const [agentPrompt, setAgentPrompt] = useState('')
 
+  useEffect(() => {
+    function openAgent(event: Event) {
+      const prompt = (event as CustomEvent<{ prompt?: string }>).detail?.prompt?.trim() ?? ''
+      setAgentPrompt(prompt)
+      setAgentOpen(true)
+    }
+    window.addEventListener('open-global-agent', openAgent)
+    return () => window.removeEventListener('open-global-agent', openAgent)
+  }, [])
   return (
     <SidebarProvider>
       <AppSidebar />
       <SidebarInset className="min-w-0">
         <AppHeader
-          onOpenAgent={() => setAgentOpen(true)}
           onOpenCommand={() => setCommandOpen(true)}
         />
         <div className="relative z-0 flex-1 overflow-auto overflow-x-clip bg-background [scrollbar-gutter:stable]">
@@ -26,7 +36,8 @@ export function AppShell({ children }: { children: React.ReactNode }) {
         </div>
       </SidebarInset>
       <CommandPalette open={commandOpen} onOpenChange={setCommandOpen} />
-      <GlobalAgentDock open={agentOpen} onOpenChange={setAgentOpen} />
+      <GlobalAgentBubble onClick={() => { setAgentPrompt(''); setAgentOpen(true) }} />
+      <GlobalAgentDock open={agentOpen} onOpenChange={setAgentOpen} initialPrompt={agentPrompt} />
     </SidebarProvider>
   )
 }
diff --git a/frontend/components/shell/global-agent-bubble.tsx b/frontend/components/shell/global-agent-bubble.tsx
new file mode 100644
index 00000000..dc27ee6f
--- /dev/null
+++ b/frontend/components/shell/global-agent-bubble.tsx
@@ -0,0 +1,20 @@
+'use client'
+
+import { Bot } from 'lucide-react'
+
+import { Button } from '@/components/ui/button'
+
+export function GlobalAgentBubble({ onClick }: { onClick: () => void }) {
+  return (
+    <Button
+      type="button"
+      size="icon"
+      className="fixed bottom-4 right-4 z-40 size-12 rounded-full shadow-lg"
+      aria-label="打开全局 Agent"
+      onClick={onClick}
+    >
+      <Bot className="size-5" aria-hidden />
+      <span className="sr-only">打开全局 Agent</span>
+    </Button>
+  )
+}
diff --git a/frontend/components/shell/global-agent-dock.tsx b/frontend/components/shell/global-agent-dock.tsx
index 148599db..60070e06 100644
--- a/frontend/components/shell/global-agent-dock.tsx
+++ b/frontend/components/shell/global-agent-dock.tsx
@@ -1,24 +1,30 @@
 'use client'
 
+
 import { useQueryClient } from '@tanstack/react-query'
-import { Bot, Check, Loader2, Send, ShieldCheck, X } from 'lucide-react'
+import { Bot, ShieldCheck } from 'lucide-react'
 import { usePathname } from 'next/navigation'
-import { FormEvent, KeyboardEvent, useState } from 'react'
+import { FormEvent, KeyboardEvent, useEffect, useState } from 'react'
 
-import { Button } from '@/components/ui/button'
+import {
+  AgentPromptBar,
+  ApprovalCard,
+  ThinkingTrace,
+  ToolChip,
+} from '@/components/agent-native/agent-primitives'
 import { ScrollArea } from '@/components/ui/scroll-area'
 import {
-  Sheet,
-  SheetContent,
-  SheetDescription,
-  SheetHeader,
-  SheetTitle,
-} from '@/components/ui/sheet'
-import { Textarea } from '@/components/ui/textarea'
+  Dialog,
+  DialogContent,
+  DialogDescription,
+  DialogHeader,
+  DialogTitle,
+} from '@/components/ui/dialog'
 import { apiClient } from '@/lib/api/client'
 import type { ApiResponse } from '@/lib/api/types'
 import { ROUTE_LABELS } from '@/lib/navigation'
 
+
 type AgentMessage = {
   role: 'user' | 'assistant'
   content: string
@@ -43,9 +49,11 @@ type AgentReply = {
 export function GlobalAgentDock({
   open,
   onOpenChange,
+  initialPrompt = '',
 }: {
   open: boolean
   onOpenChange: (open: boolean) => void
+  initialPrompt?: string
 }) {
   const pathname = usePathname()
   const queryClient = useQueryClient()
@@ -55,6 +63,11 @@ export function GlobalAgentDock({
   const [error, setError] = useState<string | null>(null)
   const [sending, setSending] = useState(false)
   const [confirming, setConfirming] = useState(false)
+  useEffect(() => {
+    if (open && initialPrompt) {
+      setInput(initialPrompt)
+    }
+  }, [initialPrompt, open])
 
   async function sendMessage(event?: FormEvent) {
     event?.preventDefault()
@@ -136,23 +149,30 @@ export function GlobalAgentDock({
   }
 
   return (
-    <Sheet open={open} onOpenChange={onOpenChange}>
-      <SheetContent className="w-full gap-0 sm:max-w-md" aria-label="全局 Agent">
-        <SheetHeader className="border-b">
-          <SheetTitle className="flex items-center gap-2">
+    <Dialog open={open} onOpenChange={onOpenChange}>
+      <DialogContent
+        className="fixed bottom-4 right-4 top-auto left-auto flex h-[min(680px,calc(100vh-2rem))] w-[min(420px,calc(100vw-2rem))] max-w-none translate-x-0 translate-y-0 flex-col gap-0 overflow-hidden p-0 shadow-2xl"
+        aria-label="全局 Agent"
+      >
+        <DialogHeader className="border-b px-4 py-3">
+          <DialogTitle className="flex items-center gap-2">
             <Bot className="size-4 text-primary" aria-hidden />
             全局 Agent
-          </SheetTitle>
-          <SheetDescription>
+          </DialogTitle>
+          <DialogDescription>
             当前上下文：{ROUTE_LABELS[pathname] ?? pathname}。读取可直接执行，写入操作先生成确认提案。
             未明确指定 Workspace 时，仅在后端能解析出唯一授权范围时允许确认写操作。
-          </SheetDescription>
-        </SheetHeader>
+          </DialogDescription>
+          <div className="flex flex-wrap gap-1.5 pt-1" aria-label="Agent 能力状态">
+            <ToolChip icon={Bot} label="当前页面" detail={ROUTE_LABELS[pathname] ?? pathname} />
+            <ToolChip icon={ShieldCheck} label="写入保护" detail="需确认" tone="warning" />
+          </div>
+        </DialogHeader>
 
         <ScrollArea className="min-h-0 flex-1">
           <div className="space-y-3 p-4" aria-live="polite">
             {messages.length === 0 ? (
-              <div className="rounded-md border border-dashed p-4">
+              <div className="rounded-lg border border-dashed p-4">
                 <div className="flex items-center gap-2 text-sm font-medium">
                   <ShieldCheck className="size-4 text-success" aria-hidden />
                   所有页面共用一个操作入口
@@ -166,79 +186,52 @@ export function GlobalAgentDock({
               <div
                 key={`${message.role}-${index}`}
                 className={message.role === 'user'
-                  ? 'ml-8 rounded-md bg-primary px-3 py-2 text-sm text-primary-foreground'
-                  : 'mr-8 rounded-md border bg-muted/30 px-3 py-2 text-sm'}
+                  ? 'ml-8 rounded-lg bg-primary px-3 py-2 text-sm text-primary-foreground'
+                  : 'mr-8 rounded-lg border bg-muted/30 px-3 py-2 text-sm'}
               >
                 {message.content}
               </div>
             ))}
             {sending ? (
-              <div className="flex items-center gap-2 text-xs text-muted-foreground" role="status">
-                <Loader2 className="size-3.5 animate-spin" aria-hidden />
-                Agent 正在处理
-              </div>
+              <ThinkingTrace
+                steps={['读取当前页面上下文', '整理可执行建议', '等待 Agent 回复']}
+              />
             ) : null}
             {proposal ? (
-              <div className="rounded-md border border-warning/40 bg-warning/10 p-3">
-                <div className="text-sm font-medium">待确认操作</div>
-                <p className="mt-1 text-xs text-muted-foreground">{proposal.summary}</p>
-                <div className="mt-3 rounded-xs border bg-background/70 p-2 font-mono text-2xs">
-                  {proposal.diff}
-                </div>
-                <div className="mt-2 space-y-1 font-mono text-3xs text-muted-foreground">
-                  <div>工作项：{proposal.work_item_id ?? '未生成'}</div>
-                  <div>工作区：{proposal.workspace_id ?? '未绑定'}</div>
-                  <div>提案版本：{proposal.proposal_version ?? '未生成'}</div>
-                </div>
-                <div className="mt-3 flex justify-end gap-2">
-                  <Button
-                    variant="ghost"
-                    size="sm"
-                    disabled={confirming}
-                    onClick={() => setProposal(null)}
-                  >
-                    <X aria-hidden />
-                    拒绝
-                  </Button>
-                  <Button
-                    size="sm"
-                    disabled={
-                      confirming
-                      || !proposal.work_item_id
-                      || !proposal.workspace_id
-                      || !proposal.proposal_version
-                    }
-                    onClick={() => void confirmProposal()}
-                  >
-                    {confirming ? <Loader2 className="animate-spin" aria-hidden /> : <Check aria-hidden />}
-                    确认执行
-                  </Button>
-                </div>
-              </div>
+              <ApprovalCard
+                summary={proposal.summary}
+                diff={proposal.diff}
+                metadata={[
+                  { label: '工具', value: proposal.tool },
+                  { label: '工作项', value: proposal.work_item_id ?? '未生成' },
+                  { label: '工作区', value: proposal.workspace_id ?? '未绑定' },
+                  { label: '提案版本', value: proposal.proposal_version ?? '未生成' },
+                ]}
+                confirming={confirming}
+                disabled={
+                  !proposal.work_item_id
+                  || !proposal.workspace_id
+                  || !proposal.proposal_version
+                }
+                onReject={() => setProposal(null)}
+                onConfirm={() => void confirmProposal()}
+              />
             ) : null}
             {error ? <p className="text-xs text-destructive" role="alert">{error}</p> : null}
           </div>
         </ScrollArea>
 
-        <form className="border-t p-4" onSubmit={(event) => void sendMessage(event)}>
-          <Textarea
-            value={input}
-            onChange={(event) => setInput(event.target.value)}
-            onKeyDown={handleKeyDown}
-            placeholder="告诉 Agent 你要查询或执行什么…"
-            aria-label="给全局 Agent 的消息"
-            className="max-h-36 min-h-20 resize-none rounded-xs"
-            disabled={sending || confirming}
-          />
-          <div className="mt-2 flex items-center justify-between gap-3">
-            <span className="text-3xs text-muted-foreground">Enter 发送 · Shift+Enter 换行</span>
-            <Button type="submit" size="sm" disabled={!input.trim() || sending || confirming || Boolean(proposal)}>
-              {sending ? <Loader2 className="animate-spin" aria-hidden /> : <Send aria-hidden />}
-              发送
-            </Button>
-          </div>
-        </form>
-      </SheetContent>
-    </Sheet>
+        <AgentPromptBar
+          value={input}
+          onChange={setInput}
+          onKeyDown={handleKeyDown}
+          onSubmit={(event) => void sendMessage(event)}
+          placeholder="描述你要查询或执行的事情…"
+          disabled={sending || confirming || Boolean(proposal)}
+          submitting={sending}
+          footer={<>Enter 发送 · Shift+Enter 换行</>}
+        />
+      </DialogContent>
+    </Dialog>
   )
 }
diff --git a/frontend/components/shell/route-tabs.tsx b/frontend/components/shell/route-tabs.tsx
index def4ddf4..258a8fa1 100644
--- a/frontend/components/shell/route-tabs.tsx
+++ b/frontend/components/shell/route-tabs.tsx
@@ -62,8 +62,7 @@ export const ACTION_CENTER_TABS: RouteTab[] = [
 ]
 
 export const AUTOMATION_TABS: RouteTab[] = [
-  { href: '/schedules', label: '调度' },
-  { href: '/plans', label: '计划' },
+  { href: '/operations-agents', label: '自动化与智能体' },
   { href: '/agents', label: 'Agent' },
   { href: '/skills', label: '技能' },
 ]
diff --git a/frontend/components/smoothui/LICENSE b/frontend/components/smoothui/LICENSE
new file mode 100644
index 00000000..dee522da
--- /dev/null
+++ b/frontend/components/smoothui/LICENSE
@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2024 Eduardo Calvo
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
diff --git a/frontend/components/smoothui/ai-approval/index.tsx b/frontend/components/smoothui/ai-approval/index.tsx
new file mode 100644
index 00000000..442d4447
--- /dev/null
+++ b/frontend/components/smoothui/ai-approval/index.tsx
@@ -0,0 +1,186 @@
+"use client";
+
+/*
+ * SmoothUI AI Approval
+ * Source: https://smoothui.dev/docs/components/ai-approval
+ * Copyright (c) 2024 Eduardo Calvo — MIT; see ../LICENSE.
+ * Adapted with disabled/pending states so server-backed decisions stay truthful.
+ */
+
+import { cn } from "@/lib/utils";
+import { AnimatePresence, motion, useReducedMotion } from "motion/react";
+import { type ReactNode, useState } from "react";
+
+const SPRING_DEFAULT = {
+  bounce: 0.1,
+  duration: 0.25,
+  type: "spring" as const,
+};
+const EASE_OUT = [0.23, 1, 0.32, 1] as const;
+const CHOICE_STAGGER = 0.03;
+
+export type AIApprovalOption = {
+  /** Marks the option as the destructive one, e.g. "Delete everything". */
+  destructive?: boolean;
+  /** Secondary line under the label. */
+  detail?: string;
+  id: string;
+  label: string;
+};
+
+export type AIApprovalProps = {
+  className?: string;
+  /** Extra context under the question. */
+  children?: ReactNode;
+  /** Prevent decisions until surrounding validation is complete. */
+  disabled?: boolean;
+  /** Called once, with the chosen option. */
+  onDecide?: (option: AIApprovalOption) => void;
+  options: AIApprovalOption[];
+  /** Marks the chosen decision as awaiting server confirmation. */
+  pending?: boolean;
+  /** What the agent needs a human to settle. */
+  question: string;
+  /** Render already-resolved, e.g. when replaying a transcript. */
+  resolvedId?: string;
+};
+
+/**
+ * The card an agent puts up before it acts.
+ *
+ * Choosing does not tick a radio button — the chosen option expands to fill the
+ * card and the alternatives collapse out of existence. The decision should look
+ * as irreversible as it is; leaving the rejected options sitting there greyed out
+ * invites a second look at something that already happened.
+ */
+const AIApproval = ({
+  className,
+  children,
+  disabled = false,
+  onDecide,
+  options,
+  pending = false,
+  question,
+  resolvedId,
+}: AIApprovalProps) => {
+  const shouldReduceMotion = useReducedMotion();
+  const [chosenId, setChosenId] = useState<string | null>(resolvedId ?? null);
+
+  const chosen = options.find((option) => option.id === chosenId) ?? null;
+
+  const decide = (option: AIApprovalOption) => {
+    if (chosenId || disabled || pending) {
+      return;
+    }
+    setChosenId(option.id);
+    onDecide?.(option);
+  };
+
+  return (
+    <motion.div
+      aria-busy={pending}
+      className={cn(
+        "w-full rounded-xl border border-border bg-background p-3.5",
+        className
+      )}
+      layout={!shouldReduceMotion}
+      transition={shouldReduceMotion ? { duration: 0 } : SPRING_DEFAULT}
+    >
+      <motion.div layout={!shouldReduceMotion}>
+        <p className="font-medium text-foreground text-sm">{question}</p>
+        {children ? (
+          <div className="mt-1 text-muted-foreground text-xs leading-relaxed">
+            {children}
+          </div>
+        ) : null}
+      </motion.div>
+
+      <div className="mt-3">
+        <AnimatePresence initial={false} mode="popLayout">
+          {chosen ? (
+            <motion.div
+              animate={{ opacity: 1, scale: 1 }}
+              className={cn(
+                "flex items-center gap-2 rounded-lg px-3 py-2 text-sm",
+                chosen.destructive
+                  ? "bg-destructive/10 text-destructive"
+                  : "bg-foreground text-background"
+              )}
+              initial={
+                shouldReduceMotion
+                  ? { opacity: 1, scale: 1 }
+                  : { opacity: 0, scale: 0.97 }
+              }
+              key="resolved"
+              role="status"
+              transition={shouldReduceMotion ? { duration: 0 } : SPRING_DEFAULT}
+            >
+              <span className="flex size-4 items-center justify-center">
+                <svg aria-hidden="true" viewBox="0 0 14 14" className="size-3.5 fill-none stroke-current">
+                  <path d="m2.5 7.5 3 3 6-7" strokeLinecap="round" strokeLinejoin="round" strokeWidth="1.8" />
+                </svg>
+              </span>
+              <span className="font-medium">{chosen.label}</span>
+              {chosen.detail ? (
+                <span className="ml-auto text-xs opacity-70">
+                  {pending ? "正在提交…" : chosen.detail}
+                </span>
+              ) : null}
+            </motion.div>
+          ) : (
+            <motion.div
+              className="flex flex-col gap-1.5"
+              exit={
+                shouldReduceMotion
+                  ? { opacity: 0, transition: { duration: 0 } }
+                  : { opacity: 0, scale: 0.98 }
+              }
+              key="options"
+              transition={
+                shouldReduceMotion
+                  ? { duration: 0 }
+                  : { duration: 0.16, ease: EASE_OUT }
+              }
+            >
+              {options.map((option, index) => (
+                <motion.button
+                  animate={{ opacity: 1, y: 0 }}
+                  className={cn(
+                    "flex w-full cursor-pointer items-center gap-2 rounded-lg border px-3 py-2 text-left text-sm transition-colors disabled:cursor-not-allowed disabled:opacity-50",
+                    option.destructive
+                      ? "border-destructive/30 text-destructive hover:bg-destructive/10"
+                      : "border-border text-foreground hover:bg-muted"
+                  )}
+                  disabled={disabled || pending}
+                  initial={
+                    shouldReduceMotion
+                      ? { opacity: 1, y: 0 }
+                      : { opacity: 0, y: 4 }
+                  }
+                  key={option.id}
+                  onClick={() => decide(option)}
+                  transition={
+                    shouldReduceMotion
+                      ? { duration: 0 }
+                      : { ...SPRING_DEFAULT, delay: index * CHOICE_STAGGER }
+                  }
+                  type="button"
+                  whileTap={shouldReduceMotion || disabled ? undefined : { scale: 0.99 }}
+                >
+                  <span className="font-medium">{option.label}</span>
+                  {option.detail ? (
+                    <span className="ml-auto text-muted-foreground text-xs">
+                      {option.detail}
+                    </span>
+                  ) : null}
+                </motion.button>
+              ))}
+            </motion.div>
+          )}
+        </AnimatePresence>
+      </div>
+    </motion.div>
+  );
+};
+
+export default AIApproval;
diff --git a/frontend/components/smoothui/ai-task-list/index.tsx b/frontend/components/smoothui/ai-task-list/index.tsx
new file mode 100644
index 00000000..8da27e46
--- /dev/null
+++ b/frontend/components/smoothui/ai-task-list/index.tsx
@@ -0,0 +1,304 @@
+"use client";
+
+/*
+ * SmoothUI AI Task List
+ * Source: https://smoothui.dev/docs/components/ai-task-list
+ * Copyright (c) 2024 Eduardo Calvo — MIT; see ../LICENSE.
+ * Adapted with blocked/partial states and raw-detail slots for workflow traces.
+ */
+
+import { cn } from "@/lib/utils";
+import { motion, useReducedMotion } from "motion/react";
+import { Fragment, type ReactNode } from "react";
+
+const SPRING_DEFAULT = {
+  bounce: 0.1,
+  duration: 0.25,
+  type: "spring" as const,
+};
+const EASE_IN_OUT = [0.645, 0.045, 0.355, 1] as const;
+const CHECK_PATH = "M 3.5 7.5 L 6 10 L 10.5 4.5";
+const BOX_SIZE = 14;
+const UNDERLINE_SECONDS = 1.4;
+
+export type AITaskStatus =
+  | "pending"
+  | "running"
+  | "done"
+  | "failed"
+  | "blocked"
+  | "partial";
+
+export type AITask = {
+  /** Nested sub-steps, one level. */
+  children?: AITask[];
+  /** Optional operational detail retained under the summary row. */
+  details?: ReactNode;
+  id: string;
+  label: string;
+  /** Short right-aligned note, e.g. "12/12" or "3 files". */
+  note?: string;
+  status: AITaskStatus;
+};
+
+export type AITaskListProps = {
+  className?: string;
+  /** Heading text. The counts are derived, never passed in. */
+  label?: string;
+  tasks: AITask[];
+};
+
+const flatten = (tasks: AITask[]): AITask[] =>
+  tasks.flatMap((task) => [task, ...flatten(task.children ?? [])]);
+
+const SUCCESS_COLOR = "oklch(72% 0.17 150)";
+const DANGER_COLOR = "oklch(63% 0.21 25)";
+const WARNING_COLOR = "oklch(75% 0.16 75)";
+const INFO_COLOR = "oklch(68% 0.14 240)";
+
+const STATUS_LABELS: Record<AITaskStatus, string> = {
+  pending: "已记录",
+  running: "运行中",
+  done: "已完成",
+  failed: "失败",
+  blocked: "阻塞",
+  partial: "部分完成",
+};
+
+const boxStroke = (status: AITaskStatus): string => {
+  if (status === "failed") {
+    return DANGER_COLOR;
+  }
+  if (status === "done") {
+    return SUCCESS_COLOR;
+  }
+  if (status === "blocked") {
+    return WARNING_COLOR;
+  }
+  if (status === "partial") {
+    return INFO_COLOR;
+  }
+  return "currentColor";
+};
+
+const TaskBox = ({
+  status,
+  shouldReduceMotion,
+}: {
+  shouldReduceMotion: boolean;
+  status: AITaskStatus;
+}) => {
+  const isDone = status === "done";
+  const isFailed = status === "failed";
+  const isBlocked = status === "blocked";
+  const isPartial = status === "partial";
+
+  return (
+    <span className="mt-0.5 flex size-3.5 shrink-0 items-center justify-center">
+      <svg
+        aria-hidden="true"
+        className="size-3.5"
+        viewBox={`0 0 ${BOX_SIZE} ${BOX_SIZE}`}
+      >
+        <motion.rect
+          animate={{
+            fillOpacity: isDone || isBlocked || isPartial ? 0.12 : 0,
+            stroke: boxStroke(status),
+          }}
+          fill={
+            isFailed
+              ? DANGER_COLOR
+              : isBlocked
+                ? WARNING_COLOR
+                : isPartial
+                  ? INFO_COLOR
+                  : SUCCESS_COLOR
+          }
+          height={12}
+          rx={3.5}
+          strokeWidth={1.4}
+          transition={shouldReduceMotion ? { duration: 0 } : { duration: 0.2 }}
+          width={12}
+          x={1}
+          y={1}
+        />
+        {isDone && (
+          // Drawn, not faded: a check that draws itself reads as the act of
+          // ticking the box rather than a state that was always there.
+          <path
+            className="ai-task-draw"
+            d={CHECK_PATH}
+            fill="none"
+            pathLength={1}
+            stroke={SUCCESS_COLOR}
+            strokeDasharray="1 1"
+            strokeLinecap="round"
+            strokeLinejoin="round"
+            strokeWidth={1.8}
+          />
+        )}
+        {isFailed && (
+          <path
+            className="ai-task-draw"
+            d="M 5 5 L 9 9 M 9 5 L 5 9"
+            fill="none"
+            pathLength={1}
+            stroke={DANGER_COLOR}
+            strokeDasharray="1 1"
+            strokeLinecap="round"
+            strokeWidth={1.8}
+          />
+        )}
+        {isBlocked && (
+          <path
+            d="M 5 4.5 L 5 9.5 M 9 4.5 L 9 9.5"
+            fill="none"
+            stroke={WARNING_COLOR}
+            strokeLinecap="round"
+            strokeWidth={1.6}
+          />
+        )}
+        {isPartial && (
+          <path
+            d="M 4.5 7 L 9.5 7"
+            fill="none"
+            stroke={INFO_COLOR}
+            strokeLinecap="round"
+            strokeWidth={1.8}
+          />
+        )}
+      </svg>
+    </span>
+  );
+};
+
+const TaskRow = ({
+  depth,
+  shouldReduceMotion,
+  task,
+}: {
+  depth: number;
+  shouldReduceMotion: boolean;
+  task: AITask;
+}) => {
+  const isRunning = task.status === "running";
+  const isDone = task.status === "done";
+
+  return (
+    <motion.li
+      // Completed rows settle down a pixel and lose a little contrast: they go
+      // quiet so whatever is running is the only thing asking for attention.
+      animate={{
+        opacity: isDone ? 0.65 : 1,
+        y: isDone && !shouldReduceMotion ? 1 : 0,
+      }}
+      className="list-none"
+      style={{ paddingLeft: depth * 20 }}
+      transition={shouldReduceMotion ? { duration: 0 } : SPRING_DEFAULT}
+    >
+      <span className="relative flex items-start gap-2 py-1">
+        <TaskBox shouldReduceMotion={shouldReduceMotion} status={task.status} />
+        <span className="sr-only">状态：{STATUS_LABELS[task.status]}。</span>
+
+        <span className="min-w-0 flex-1 text-foreground text-sm leading-snug [overflow-wrap:anywhere]">
+          {task.label}
+        </span>
+
+        {task.note ? (
+          <span className="shrink-0 text-muted-foreground text-xs tabular-nums">
+            {task.note}
+          </span>
+        ) : null}
+
+        {isRunning && !shouldReduceMotion && (
+          // A travelling underline under the active row only. One moving thing
+          // at a time is what makes "which step is live" readable at a glance.
+          <motion.span
+            animate={{ backgroundPositionX: ["0%", "200%"] }}
+            className="pointer-events-none absolute inset-x-0 bottom-0 h-px"
+            style={{
+              backgroundImage:
+                "linear-gradient(90deg, transparent 0%, currentColor 50%, transparent 100%)",
+              backgroundSize: "50% 100%",
+              opacity: 0.5,
+            }}
+            transition={{
+              duration: UNDERLINE_SECONDS,
+              ease: EASE_IN_OUT,
+              repeat: Number.POSITIVE_INFINITY,
+            }}
+          />
+        )}
+      </span>
+      {task.details ? (
+        <div className="ml-[22px] pb-1 text-xs text-muted-foreground">
+          {task.details}
+        </div>
+      ) : null}
+    </motion.li>
+  );
+};
+
+/**
+ * A plan an agent works through.
+ *
+ * Header counts are derived from the tasks, so the summary can never disagree
+ * with the rows — a "3/7" that has drifted from what is on screen destroys trust
+ * in the whole panel.
+ */
+const AITaskList = ({ className, label = "Plan", tasks }: AITaskListProps) => {
+  const shouldReduceMotion = Boolean(useReducedMotion());
+  const all = flatten(tasks);
+  const done = all.filter((task) => task.status === "done").length;
+
+  return (
+    <div
+      className={cn(
+        "w-full rounded-xl border border-border bg-background p-3",
+        className
+      )}
+    >
+      <div className="mb-1.5 flex items-baseline justify-between">
+        <p className="font-medium text-foreground text-sm">{label}</p>
+        <p className="text-muted-foreground text-xs tabular-nums">
+          {done}/{all.length}
+        </p>
+      </div>
+
+      <style>{`
+        .ai-task-draw { stroke-dashoffset: 0; }
+        @media (prefers-reduced-motion: no-preference) {
+          .ai-task-draw {
+            animation: ai-task-draw 200ms cubic-bezier(0.23, 1, 0.32, 1) both;
+          }
+        }
+        @keyframes ai-task-draw {
+          from { stroke-dashoffset: 1; }
+          to { stroke-dashoffset: 0; }
+        }
+      `}</style>
+
+      <ul className="list-none">
+        {tasks.map((task) => (
+          <Fragment key={task.id}>
+            <TaskRow
+              depth={0}
+              shouldReduceMotion={shouldReduceMotion}
+              task={task}
+            />
+            {task.children?.map((child) => (
+              <TaskRow
+                depth={1}
+                key={child.id}
+                shouldReduceMotion={shouldReduceMotion}
+                task={child}
+              />
+            ))}
+          </Fragment>
+        ))}
+      </ul>
+    </div>
+  );
+};
+
+export default AITaskList;
diff --git a/frontend/e2e/gpu-graph.spec.mjs b/frontend/e2e/gpu-graph.spec.mjs
new file mode 100644
index 00000000..f78f54bc
--- /dev/null
+++ b/frontend/e2e/gpu-graph.spec.mjs
@@ -0,0 +1,205 @@
+import { expect, test } from '@playwright/test'
+
+const api = (path) => `**/api/v1${path}`
+
+const workspace = {
+  id: 'workspace-gpu',
+  name: 'GPU fixture workspace',
+  slug: 'gpu-fixture-workspace',
+  active: true,
+  created_at: '2026-08-28T00:00:00Z',
+  updated_at: '2026-08-28T00:00:00Z',
+}
+
+const project = {
+  id: 'project-gpu',
+  workspace_id: workspace.id,
+  name: 'GPU fixture project',
+  slug: 'gpu-fixture-project',
+  description: null,
+  app_type: 'workflow',
+  primary_workflow_id: 'workflow-gpu',
+  created_by_user_id: 'e2e-admin',
+  archived: false,
+  created_at: '2026-08-28T00:00:00Z',
+  updated_at: '2026-08-28T00:00:00Z',
+}
+
+const workflow = {
+  id: 'workflow-gpu',
+  project_id: project.id,
+  name: 'GPU fixture workflow',
+  description: null,
+  current_published_version: 1,
+  archived: false,
+  created_at: '2026-08-28T00:00:00Z',
+  updated_at: '2026-08-28T00:00:00Z',
+}
+
+const graphPreview = {
+  workspace_id: workspace.id,
+  project_id: project.id,
+  project_name: project.name,
+  strategy: 'server-aggregated-sample',
+  truncated: false,
+  max_nodes: 700,
+  nodes: [
+    {
+      id: 'project-node',
+      kind: 'project',
+      label: project.name,
+      subtitle: 'Fixture project root',
+      count: 1,
+      record_id: null,
+      source_id: null,
+      workflow_id: null,
+      workflow_run_id: null,
+      url: null,
+      preview: null,
+      status: 'ready',
+      source_published_at: null,
+      created_at: '2026-08-28T00:00:00Z',
+    },
+    {
+      id: 'source-node',
+      kind: 'source',
+      label: 'Fixture source',
+      subtitle: 'Fixture source detail',
+      count: 4,
+      record_id: null,
+      source_id: 'source-gpu',
+      workflow_id: null,
+      workflow_run_id: null,
+      url: 'https://example.test/source',
+      preview: null,
+      status: 'ready',
+      source_published_at: '2026-08-28T00:00:00Z',
+      created_at: '2026-08-28T00:00:00Z',
+    },
+  ],
+  edges: [
+    {
+      id: 'project-source',
+      source: 'project-node',
+      target: 'source-node',
+      kind: 'contains',
+      label: 'contains',
+      weight: 1,
+      bidirectional: true,
+    },
+  ],
+  stats: {
+    total_records: 4,
+    sampled_records: 4,
+    hidden_records: 0,
+    total_sources: 1,
+    total_workflows: 1,
+    total_runs: 0,
+    visible_nodes: 2,
+    visible_edges: 1,
+  },
+  generated_at: '2026-08-28T00:00:00Z',
+}
+
+async function disableWebGl2(page) {
+  await page.addInitScript(() => {
+    const getContext = HTMLCanvasElement.prototype.getContext
+    HTMLCanvasElement.prototype.getContext = function getContextWithoutWebGl2(
+      contextId,
+      ...argumentsAfterContextId
+    ) {
+      if (contextId === 'webgl2') return null
+      return getContext.call(this, contextId, ...argumentsAfterContextId)
+    }
+  })
+}
+
+async function installGraphFixtures(page) {
+  await page.route(api('/auth/login'), (route) => route.fulfill({
+    contentType: 'application/json',
+    body: JSON.stringify({
+      success: true,
+      data: {
+        access_token: 'e2e-token',
+        token_type: 'bearer',
+        using_default_password: false,
+      },
+    }),
+  }))
+  await page.route(api('/auth/me'), (route) => route.fulfill({
+    contentType: 'application/json',
+    body: JSON.stringify({
+      success: true,
+      data: {
+        subject: 'e2e-admin',
+        name: 'E2E Admin',
+        is_platform_admin: true,
+        auth_method: 'password',
+      },
+    }),
+  }))
+  await page.route(api('/workspaces'), (route) => route.fulfill({
+    contentType: 'application/json',
+    body: JSON.stringify({ success: true, data: [workspace] }),
+  }))
+  await page.route(api(`/workspaces/${workspace.id}/projects`), (route) => route.fulfill({
+    contentType: 'application/json',
+    body: JSON.stringify({ success: true, data: [project] }),
+  }))
+  await page.route(api(`/workspaces/${workspace.id}/projects/${project.id}/workflows`), (route) => route.fulfill({
+    contentType: 'application/json',
+    body: JSON.stringify({ success: true, data: [workflow] }),
+  }))
+  await page.route(`**/api/v1/workspaces/${workspace.id}/projects/${project.id}/record-graph**`, (route) => route.fulfill({
+    contentType: 'application/json',
+    body: JSON.stringify({ success: true, data: graphPreview }),
+  }))
+}
+
+async function goAuthed(page, path) {
+  await page.goto(`/login?returnTo=${encodeURIComponent(path)}`)
+  await page.getByLabel('用户名').fill('admin')
+  await page.getByLabel('密码').fill('admin')
+  await page.getByRole('button', { name: '登录' }).click()
+  await expect(page).toHaveURL(new RegExp(`${path.split('?')[0].replaceAll('/', '\\/')}(?:\\?.*)?$`))
+}
+
+test('records graph mounts its interactive WebGL2 renderer when available', async ({ page }) => {
+  await installGraphFixtures(page)
+  await goAuthed(page, '/records/graph')
+
+  const surface = page.locator('[data-gpu-surface="record-relationship-graph"]')
+  await expect(surface).toHaveAttribute('data-gpu-backend', 'webgl2')
+  await expect(surface.locator('canvas').first()).toBeVisible()
+})
+
+test('Galaxy mounts its interactive WebGL2 renderer when available', async ({ page }) => {
+  await installGraphFixtures(page)
+  await goAuthed(page, `/studio/projects/${project.id}/galaxy?workspace=${workspace.id}`)
+
+  const surface = page.locator('[data-gpu-surface="project-galaxy-graph"]')
+  await expect(surface).toHaveAttribute('data-gpu-backend', 'webgl2')
+  await expect(surface.locator('canvas').first()).toBeVisible()
+})
+
+test('records graph keeps fallback nodes selectable without WebGL2', async ({ page }) => {
+  await disableWebGl2(page)
+  await installGraphFixtures(page)
+  await goAuthed(page, '/records/graph')
+
+  await expect(page.locator('[data-gpu-surface="record-relationship-graph"]'))
+    .toHaveAttribute('data-gpu-backend', 'fallback')
+  await page.locator('[data-record-graph-fallback]').getByRole('button', { name: /Fixture source/ }).click()
+  await expect(page.getByRole('heading', { name: 'Fixture source', exact: true })).toBeVisible()
+})
+
+test('Galaxy fallback nodes open the existing inspector without WebGL2', async ({ page }) => {
+  await disableWebGl2(page)
+  await installGraphFixtures(page)
+  await goAuthed(page, `/studio/projects/${project.id}/galaxy?workspace=${workspace.id}`)
+
+  await expect(page.locator('[data-gpu-surface="project-galaxy-graph"]'))
+    .toHaveAttribute('data-gpu-backend', 'fallback')
+  await page.locator('[data-project-graph-fallback]').getByRole('button', { name: /Fixture source/ }).click()
+  await expect(page.getByRole('heading', { name: 'Fixture source', exact: true })).toBeVisible()
+})
diff --git a/frontend/e2e/login.spec.mjs b/frontend/e2e/login.spec.mjs
index f44344b7..d8036fac 100644
--- a/frontend/e2e/login.spec.mjs
+++ b/frontend/e2e/login.spec.mjs
@@ -1,7 +1,79 @@
 import { expect, test } from '@playwright/test'
 
-test('login page renders its administrator credentials form', async ({ page }) => {
+test('login page mounts its WebGL2 backdrop and renders local administrator credentials', async ({ page }) => {
   await page.goto('/login')
   await expect(page.getByText('登录控制台')).toBeVisible()
-  await expect(page.getByLabel('管理员身份令牌')).toBeVisible()
+  await expect(page.getByLabel('用户名')).toHaveValue('admin')
+  await expect(page.getByLabel('密码')).toBeVisible()
+  await expect(page.getByLabel('管理员身份令牌')).toHaveCount(0)
+  await expect(page.getByLabel('Fleet API 令牌（可选）')).toHaveCount(0)
+  await expect(page.locator('[data-gpu-surface="login-background"]'))
+    .toHaveAttribute('data-gpu-backend', 'webgl2')
+})
+
+test('login page keeps its static backdrop and form usable without WebGL2', async ({ page }) => {
+  await page.addInitScript(() => {
+    const getContext = HTMLCanvasElement.prototype.getContext
+    HTMLCanvasElement.prototype.getContext = function getContextWithoutWebGl2(
+      contextId,
+      ...argumentsAfterContextId
+    ) {
+      if (contextId === 'webgl2') return null
+      return getContext.call(this, contextId, ...argumentsAfterContextId)
+    }
+  })
+
+  await page.goto('/login')
+  await expect(page.locator('[data-gpu-surface="login-background"]'))
+    .toHaveAttribute('data-gpu-backend', 'fallback')
+  await expect(page.locator('[data-login-static-background]')).toBeVisible()
+  await expect(page.getByLabel('用户名')).toHaveValue('admin')
+  await expect(page.getByLabel('密码')).toBeVisible()
+})
+
+test('login SSR output retains the pending fallback without JavaScript', async ({ browser }) => {
+  const context = await browser.newContext({ javaScriptEnabled: false })
+  const page = await context.newPage()
+  const pageErrors = []
+  page.on('pageerror', (error) => pageErrors.push(error.message))
+
+  try {
+    await page.goto('/login')
+    const pendingSurface = page.locator('[data-gpu-surface="login-background"]').first()
+    await expect(pendingSurface).toHaveAttribute('data-gpu-backend', 'pending')
+    await expect(pendingSurface.locator('[data-login-static-background]')).toHaveCount(1)
+    expect(pageErrors).toEqual([])
+  } finally {
+    await context.close()
+  }
+})
+
+test('login releases its GPU backdrop for reduced motion while keeping the form usable', async ({ page }) => {
+  await page.goto('/login')
+  await expect(page.locator('[data-gpu-surface="login-background"]'))
+    .toHaveAttribute('data-gpu-backend', 'webgl2')
+
+  await page.emulateMedia({ reducedMotion: 'reduce' })
+
+  await expect(page.locator('[data-gpu-surface="login-background"]'))
+    .toHaveAttribute('data-gpu-backend', 'fallback')
+  await expect(page.locator('[data-login-static-background]')).toBeVisible()
+  await expect(page.getByLabel('用户名')).toHaveValue('admin')
+  await expect(page.getByLabel('密码')).toBeVisible()
+})
+
+test('login recreates its GPU backdrop after a WebGL context loss', async ({ page }) => {
+  await page.goto('/login')
+  const surface = page.locator('[data-gpu-surface="login-background"]')
+  await expect(surface).toHaveAttribute('data-gpu-backend', 'webgl2')
+
+  const canvas = surface.locator('canvas').first()
+  await expect(canvas).toBeAttached()
+  await page.waitForTimeout(50)
+  await canvas.dispatchEvent('webglcontextlost', { cancelable: true })
+
+  await expect(surface).toHaveAttribute('data-gpu-fallback-reason', 'context-lost')
+  await expect(page.locator('[data-login-static-background]')).toBeVisible()
+  await expect(surface).toHaveAttribute('data-gpu-backend', 'webgl2')
+  await expect(surface.locator('canvas').first()).toBeAttached()
 })
diff --git a/frontend/e2e/p0-regressions.spec.mjs b/frontend/e2e/p0-regressions.spec.mjs
new file mode 100644
index 00000000..d2a92701
--- /dev/null
+++ b/frontend/e2e/p0-regressions.spec.mjs
@@ -0,0 +1,90 @@
+import { expect, test } from '@playwright/test'
+
+const api = (path) => `**/api/v1${path}`
+
+async function installPluginFixtures(page) {
+  await page.route(api('/plugins'), (route) => route.fulfill({
+    contentType: 'application/json',
+    body: JSON.stringify({ success: true, data: [
+      { id: 'e2e-provider', providerKey: 'e2e/test-plugin', name: 'E2E Provider', author: 'E2E', version: '1.0.0', sourceKind: 'bundled', sourceDigest: 'e2e', manifestSpecVersion: '1.0', signatureState: 'bundled', labels: { zh_Hans: 'E2E Provider' }, descriptions: { zh_Hans: 'Deterministic test provider' }, icon: 'globe', pluginTypes: ['adapter'], runtimeStatus: 'READY', capabilities: [{ id: 'browser', family: 'tool', key: 'browser', label: 'Browser', blockers: [], flowCapability: true, status: 'READY' }], nodeDefinitions: [], manifest: {}, permissions: {}, blockers: [], bundled: false, installedAt: '2026-01-01T00:00:00Z', updatedAt: '2026-01-01T00:00:00Z' },
+    ] }),
+  }))
+  await page.route('**/api/workflow/capabilities', (route) => route.fulfill({
+    contentType: 'application/json',
+    body: JSON.stringify({ success: true, data: { version: 'test', catalog: [], primitives: [], channels: [], notifiers: [], triggers: [], resources: [] } }),
+  }))
+  await page.route('**/api/v1/nodes/capabilities', (route) => route.fulfill({
+    contentType: 'application/json',
+    body: JSON.stringify({ success: true, data: { version: 'test', nodes: [] } }),
+  }))
+}
+
+test.beforeEach(async ({ page }) => {
+  await page.emulateMedia({ reducedMotion: 'reduce' })
+  await page.addInitScript(() => {
+    const style = document.createElement('style')
+    style.textContent = '*, *::before, *::after { animation-duration: 0s !important; transition-duration: 0s !important; }'
+    document.documentElement.appendChild(style)
+  })
+  await page.route('**/api/v1/auth/login', (route) => route.fulfill({ contentType: 'application/json', body: JSON.stringify({ success: true, data: { access_token: 'e2e-token', token_type: 'bearer', using_default_password: false } }) }))
+  await page.route('**/api/v1/auth/me', (route) => {
+    if (route.request().headers().authorization !== 'Bearer e2e-token') throw new Error('auth/me missing Bearer e2e-token')
+    return route.fulfill({ contentType: 'application/json', body: JSON.stringify({ success: true, data: { subject: 'e2e-admin', name: 'E2E Admin', is_platform_admin: true, auth_method: 'password' } }) })
+  })
+})
+async function goAuthed(page, path) {
+  await page.goto(`/login?returnTo=${encodeURIComponent(path)}`)
+  await page.getByLabel('用户名').fill('admin')
+  await page.getByLabel('密码').fill('admin')
+  await page.getByRole('button', { name: '登录' }).click()
+  await expect(page).toHaveURL(new RegExp(`${path.split('?')[0].replaceAll('/', '\\/')}(?:\\?.*)?$`), { timeout: 10000 })
+}
+
+test('plugins tabs expose semantic selection and installed capability detail', async ({ page }) => {
+  await installPluginFixtures(page)
+  await goAuthed(page, '/plugins')
+  const tabs = page.getByRole('navigation', { name: '插件页面' })
+  await expect(tabs.getByRole('button', { name: '已安装' })).toHaveAttribute('aria-current', 'page')
+  await tabs.getByRole('button', { name: '节点能力' }).dispatchEvent('click')
+  await expect(page.getByRole('navigation', { name: '插件页面' }).getByRole('button', { name: '节点能力' })).toHaveAttribute('aria-current', 'page')
+  await page.getByRole('navigation', { name: '插件页面' }).getByRole('button', { name: '探索市场' }).dispatchEvent('click')
+  await expect(page.getByRole('navigation', { name: '插件页面' }).getByRole('button', { name: '探索市场' })).toHaveAttribute('aria-current', 'page')
+  await page.getByRole('navigation', { name: '插件页面' }).getByRole('button', { name: '已安装' }).dispatchEvent('click')
+  await page.getByRole('button', { name: '查看 E2E Provider 插件' }).click()
+  await expect(page.getByText('声明的能力')).toBeVisible()
+  await expect(page.getByText('Browser', { exact: true })).toBeVisible()
+})
+
+
+test('skill correction proposal can be dismissed and rollback is not stale after rollback', async ({ page }) => {
+  let rolledBack = false
+  let dismissed = false
+  await page.route(api('/skills/skill-1'), (route) => {
+    const evidence = rolledBack
+      ? [{ event: 'corrected', from_version: 1, to_version: 2 }, { event: 'rolled_back', from_version: 2, to_version: 1 }]
+      : dismissed
+        ? [{ event: 'corrected', from_version: 1, to_version: 2 }, { event: 'correction_dismissed', from_version: 1, to_version: 2 }]
+        : [{ event: 'corrected', from_version: 1, to_version: 2 }, { event: 'correction_proposed', trace_ids: ['trace-1'], prior_redistill_count: 0 }]
+    return route.fulfill({
+      contentType: 'application/json',
+      body: JSON.stringify({
+        success: true,
+        data: { id: 'skill-1', name: 'Browser skill', domain: 'web', capability: 'browser', version: 2, status: 'active', enabled: true, evidence_count: 2, evidence },
+      }),
+    })
+  })
+  await page.route(api('/skills/skill-1/dismiss-correction'), async (route) => { dismissed = true; await route.fulfill({ contentType: 'application/json', body: JSON.stringify({ success: true, data: {} }) }) })
+  await page.route(api('/skills/skill-1/rollback'), async (route) => { rolledBack = true; await route.fulfill({ contentType: 'application/json', body: JSON.stringify({ success: true, data: {} }) }) })
+  await goAuthed(page, '/skills/skill-1')
+  await expect(page).toHaveURL(/\/skills\/skill-1(?:\?.*)?$/, { timeout: 10000 })
+  await expect(page.getByText('待复核的纠正建议')).toBeVisible()
+  await page.getByRole('button', { name: '忽略纠正建议' }).click()
+  await page.getByRole('button', { name: '确认忽略' }).click()
+  await expect(page.getByText('待复核的纠正建议')).toHaveCount(0)
+  const rollback = page.getByRole('button', { name: /回滚/ })
+  await expect(rollback).toBeVisible()
+  await rollback.click()
+  await page.getByRole('button', { name: '确认回滚' }).click()
+  await page.reload()
+  await expect(page.getByRole('button', { name: /回滚/ })).toHaveCount(0)
+})
diff --git a/frontend/e2e/paw-agent.spec.mjs b/frontend/e2e/paw-agent.spec.mjs
new file mode 100644
index 00000000..3f1a3028
--- /dev/null
+++ b/frontend/e2e/paw-agent.spec.mjs
@@ -0,0 +1,42 @@
+import { expect, test } from '@playwright/test'
+
+const api = (path) => `**/api/v1${path}`
+
+async function login(page) {
+  await page.goto('/login?returnTo=%2Fagents')
+  await page.getByLabel('用户名').fill('admin')
+  await page.getByLabel('密码').fill('admin')
+  await page.getByRole('button', { name: '登录' }).click()
+  await expect(page).toHaveURL(/\/agents(?:\?.*)?$/)
+}
+
+test('PAW agent form exposes governed config and blocks blank prompt submission', async ({ page }) => {
+  let submitted = null
+  await page.route(api('/auth/login'), (route) => route.fulfill({ contentType: 'application/json', body: JSON.stringify({ success: true, data: { access_token: 'e2e-token', token_type: 'bearer' } }) }))
+  await page.route(api('/auth/me'), (route) => route.fulfill({ contentType: 'application/json', body: JSON.stringify({ success: true, data: { subject: 'e2e', name: 'E2E', is_platform_admin: true } }) }))
+  await page.route(api('/agents'), async (route) => {
+    if (route.request().method() === 'GET') return route.fulfill({ contentType: 'application/json', body: JSON.stringify({ success: true, data: [] }) })
+    submitted = route.request().postDataJSON()
+    return route.fulfill({ contentType: 'application/json', body: JSON.stringify({ success: true, data: { id: 'paw-e2e', ...submitted, created_at: '2026-08-28T00:00:00Z', updated_at: '2026-08-28T00:00:00Z' } }) })
+  })
+  await page.route(api('/providers'), (route) => route.fulfill({ contentType: 'application/json', body: JSON.stringify({ success: true, data: [] }) }))
+
+  await login(page)
+  await page.getByRole('button', { name: '添加 Agent' }).click()
+  await page.getByLabel('处理器类型').click()
+  await page.getByRole('option', { name: 'PAW 本地富化' }).click()
+  await expect(page.getByText('使用固定、离线的 PAW sidecar')).toBeVisible()
+  await expect(page.getByText('PAW 必须使用非空的短提示词模板。')).toBeVisible()
+  await page.getByLabel('名称').fill('PAW fixture')
+  await page.getByRole('button', { name: '创建 Agent' }).click()
+  await expect.poll(() => submitted).toBeNull()
+  await page.getByLabel('提示词模板').fill('Classify {{title}}')
+  await page.getByText('高级设置').click()
+  await page.getByLabel('处理器配置（JSON，可选）').fill('{"max_tokens": 8, "output_schema": {"type": "object"}}')
+  await page.getByRole('button', { name: '创建 Agent' }).click()
+  await expect.poll(() => submitted).toMatchObject({
+    processor_type: 'paw',
+    prompt_template: 'Classify {{title}}',
+    processor_config: { max_tokens: 8, output_schema: { type: 'object' } },
+  })
+})
diff --git a/frontend/lib/api/endpoints.ts b/frontend/lib/api/endpoints.ts
index fe0b8dd0..e03a9d13 100644
--- a/frontend/lib/api/endpoints.ts
+++ b/frontend/lib/api/endpoints.ts
@@ -73,6 +73,7 @@ import type {
   OperationsAgentMode,
   OperationsAgentProfile,
   OperationsAgentRun,
+  OperationsAgentTeam,
   PublishedOperationsAgentVersion,
   Automation,
   WorkspaceSettingsRead,
@@ -92,6 +93,25 @@ export const resetWorkspaceSettings = () =>
 export const getCurrentIdentity = () =>
   apiClient.get<ApiResponse<AuthIdentity>>('/auth/me').then((r) => r.data.data)
 
+export const loginWithPassword = (username: string, password: string) =>
+  apiClient
+    .post<
+      ApiResponse<{
+        access_token: string
+        token_type: 'bearer'
+        using_default_password: boolean
+      }>
+    >('/auth/login', { username, password })
+    .then((r) => r.data.data)
+
+export const changeLocalPassword = (currentPassword: string, newPassword: string) =>
+  apiClient
+    .post<ApiResponse<{ message: string }>>('/auth/password', {
+      current_password: currentPassword,
+      new_password: newPassword,
+    })
+    .then((r) => r.data.data)
+
 export const listMyWorkspaces = () =>
   apiClient.get<ApiResponse<WorkspaceSummary[]>>('/workspaces').then((r) => r.data.data)
 
@@ -257,6 +277,19 @@ export const listOperationsAgents = (workspaceId: string) =>
     .get<ApiResponse<OperationsAgent[]>>(`/workspaces/${workspaceId}/operations-agents`)
     .then((r) => r.data.data)
 
+export const listOperationsAgentTeams = (workspaceId: string) =>
+  apiClient
+    .get<ApiResponse<OperationsAgentTeam[]>>(`/workspaces/${workspaceId}/operations-agents/teams`)
+    .then((r) => r.data.data)
+
+export const createOperationsAgent = (
+  workspaceId: string,
+  data: { name: string; description?: string | null; owning_team_id?: string | null },
+) =>
+  apiClient
+    .post<ApiResponse<OperationsAgent>>(`/workspaces/${workspaceId}/operations-agents`, data)
+    .then((r) => r.data.data)
+
 export const listOperationsAgentActivity = (workspaceId: string) =>
   apiClient.get<ApiResponse<OperationsAgentRun[]>>(`/workspaces/${workspaceId}/operations-agents/activity`).then((r) => r.data.data)
 
@@ -299,12 +332,33 @@ export const startOperationsAgentRun = (
 export const listAutomations = (workspaceId: string) =>
   apiClient.get<ApiResponse<Automation[]>>(`/workspaces/${workspaceId}/automations`).then((r) => r.data.data)
 
-export const createAutomation = (workspaceId: string, data: Omit<Automation, 'id' | 'workspace_id' | 'created_by_user_id' | 'created_at' | 'updated_at'>) =>
+export const installAutomationStarters = (workspaceId: string) =>
+  apiClient
+    .post<ApiResponse<{ created_count: number; skipped_count: number }>>(
+      `/workspaces/${workspaceId}/automations/starters/install`,
+    )
+    .then((r) => r.data.data)
+
+export const createAutomation = (
+  workspaceId: string,
+  data: Omit<Automation, 'id' | 'workspace_id' | 'revision' | 'created_by_user_id' | 'created_at' | 'updated_at'>,
+) =>
   apiClient.post<ApiResponse<Automation>>(`/workspaces/${workspaceId}/automations`, data).then((r) => r.data.data)
 
-export const patchAutomation = (workspaceId: string, automationId: string, data: Partial<Automation>) =>
+export const patchAutomation = (
+  workspaceId: string,
+  automationId: string,
+  data: Partial<Omit<Automation, 'id' | 'workspace_id' | 'revision' | 'created_by_user_id' | 'created_at' | 'updated_at'>>,
+) =>
   apiClient.patch<ApiResponse<Automation>>(`/workspaces/${workspaceId}/automations/${automationId}`, data).then((r) => r.data.data)
 
+export const startAutomationRun = (workspaceId: string, automationId: string) =>
+  apiClient
+    .post<ApiResponse<OperationsAgentRun>>(
+      `/workspaces/${workspaceId}/automations/${automationId}/runs`,
+    )
+    .then((r) => r.data.data)
+
 export const patchOperationsAgent = (workspaceId: string, agentId: string, disabled: boolean) =>
   apiClient
     .patch<ApiResponse<OperationsAgent>>(`/workspaces/${workspaceId}/operations-agents/${agentId}`, { disabled })
@@ -717,8 +771,28 @@ export const getHealth = () =>
 
 export const getSystemConfig = () =>
   apiClient.get<ApiResponse<SystemConfig>>('/system/config').then((r) => r.data.data)
+export type SystemConfigPatch = Partial<
+  Pick<
+    SystemConfig,
+    | 'collection_mode'
+    | 'collection_orchestrator'
+    | 'local_max_concurrent_pipelines'
+    | 'opencli_timeout'
+    | 'default_timezone'
+    | 'public_url'
+    | 'fleet_network_provider'
+    | 'netbird_mode'
+    | 'opencli_cdp_endpoint'
+    | 'llm_request_timeout_seconds'
+    | 'llm_max_concurrency'
+    | 'control_mode'
+    | 'control_kill_switch'
+  >
+> & {
+  agent_pool_endpoints?: string
+}
 
-export const updateSystemConfig = (data: Partial<SystemConfig>) =>
+export const updateSystemConfig = (data: SystemConfigPatch) =>
   apiClient.patch<ApiResponse<SystemConfig>>('/system/config', data).then((r) => r.data.data)
 
 export const getWsAgentStatus = () =>
diff --git a/frontend/lib/api/hooks.ts b/frontend/lib/api/hooks.ts
index fbaede06..41ead0af 100644
--- a/frontend/lib/api/hooks.ts
+++ b/frontend/lib/api/hooks.ts
@@ -15,13 +15,16 @@ import type {
   OperationsAgentMode,
   ProviderModelDiscoveryInput,
   SystemConfig,
-  WorkspaceSettingsValues,
 } from './types'
 
+export type { SystemConfig }
+export type SystemConfigPatch = api.SystemConfigPatch
+
 export function useMyWorkspaces() {
   return useQuery({ queryKey: ['workspaces'], queryFn: api.listMyWorkspaces })
 }
 
+
 export function useWorkspaceProjects(workspaceId: string | null) {
   return useQuery({
     queryKey: ['workspace-projects', workspaceId],
@@ -243,6 +246,29 @@ export function useOperationsAgents(workspaceId: string | null) {
   })
 }
 
+export function useOperationsAgentTeams(workspaceId: string | null) {
+  return useQuery({
+    queryKey: ['operations-agent-teams', workspaceId],
+    queryFn: () => api.listOperationsAgentTeams(workspaceId as string),
+    enabled: !!workspaceId,
+  })
+}
+
+export function useCreateOperationsAgent() {
+  const queryClient = useQueryClient()
+  return useMutation({
+    mutationFn: ({
+      workspaceId,
+      data,
+    }: {
+      workspaceId: string
+      data: Parameters<typeof api.createOperationsAgent>[1]
+    }) => api.createOperationsAgent(workspaceId, data),
+    onSuccess: (_agent, { workspaceId }) =>
+      queryClient.invalidateQueries({ queryKey: ['operations-agents', workspaceId] }),
+  })
+}
+
 export function useOperationsAgentActivity(workspaceId: string | null) {
   return useQuery({
     queryKey: ['operations-agent-activity', workspaceId],
@@ -311,10 +337,19 @@ export function useAutomations(workspaceId: string | null) {
   return useQuery({ queryKey: ['automations', workspaceId], queryFn: () => api.listAutomations(workspaceId as string), enabled: !!workspaceId, refetchInterval: 15_000 })
 }
 
+export function useInstallAutomationStarters() {
+  const queryClient = useQueryClient()
+  return useMutation({
+    mutationFn: ({ workspaceId }: { workspaceId: string }) => api.installAutomationStarters(workspaceId),
+    onSuccess: (_result, { workspaceId }) =>
+      queryClient.invalidateQueries({ queryKey: ['automations', workspaceId] }),
+  })
+}
+
 export function useCreateAutomation() {
   const queryClient = useQueryClient()
   return useMutation({
-    mutationFn: ({ workspaceId, data }: { workspaceId: string; data: Omit<Automation, 'id' | 'workspace_id' | 'created_by_user_id' | 'created_at' | 'updated_at'> }) => api.createAutomation(workspaceId, data),
+    mutationFn: ({ workspaceId, data }: { workspaceId: string; data: Parameters<typeof api.createAutomation>[1] }) => api.createAutomation(workspaceId, data),
     onSuccess: (_result, { workspaceId }) => queryClient.invalidateQueries({ queryKey: ['automations', workspaceId] }),
   })
 }
@@ -322,11 +357,21 @@ export function useCreateAutomation() {
 export function usePatchAutomation() {
   const queryClient = useQueryClient()
   return useMutation({
-    mutationFn: ({ workspaceId, automationId, data }: { workspaceId: string; automationId: string; data: Partial<Automation> }) => api.patchAutomation(workspaceId, automationId, data),
+    mutationFn: ({ workspaceId, automationId, data }: { workspaceId: string; automationId: string; data: Parameters<typeof api.patchAutomation>[2] }) => api.patchAutomation(workspaceId, automationId, data),
     onSuccess: (_result, { workspaceId }) => queryClient.invalidateQueries({ queryKey: ['automations', workspaceId] }),
   })
 }
 
+export function useStartAutomationRun() {
+  const queryClient = useQueryClient()
+  return useMutation({
+    mutationFn: ({ workspaceId, automationId }: { workspaceId: string; automationId: string }) =>
+      api.startAutomationRun(workspaceId, automationId),
+    onSuccess: (_run, { workspaceId }) =>
+      queryClient.invalidateQueries({ queryKey: ['operations-agent-activity', workspaceId] }),
+  })
+}
+
 export function usePatchOperationsAgent() {
   const queryClient = useQueryClient()
   return useMutation({
@@ -709,7 +754,8 @@ export function useWorkspaceSettings() {
 export function useUpdateWorkspaceSettings() {
   const queryClient = useQueryClient()
   return useMutation({
-    mutationFn: (data: Partial<WorkspaceSettingsValues>) => api.updateWorkspaceSettings(data),
+    mutationFn: (data: Parameters<typeof api.updateWorkspaceSettings>[0]) =>
+      api.updateWorkspaceSettings(data),
     onSuccess: () => queryClient.invalidateQueries({ queryKey: ['workspace-settings'] }),
   })
 }
@@ -1016,7 +1062,8 @@ export function useSystemConfig() {
 export function useUpdateSystemConfig() {
   const queryClient = useQueryClient()
   return useMutation({
-    mutationFn: (data: Partial<SystemConfig>) => api.updateSystemConfig(data),
+    mutationFn: (data: Parameters<typeof api.updateSystemConfig>[0]) =>
+      api.updateSystemConfig(data),
     onSuccess: (result) => queryClient.setQueryData(['system-config'], result),
   })
 }
diff --git a/frontend/lib/api/types.ts b/frontend/lib/api/types.ts
index d9160ff6..7bd763c8 100644
--- a/frontend/lib/api/types.ts
+++ b/frontend/lib/api/types.ts
@@ -187,7 +187,7 @@ export interface AIAgent {
   id: string
   name: string
   description?: string
-  processor_type: 'claude' | 'openai' | 'local'
+  processor_type: 'claude' | 'openai' | 'local' | 'paw'
   model?: string
   prompt_template: string
   processor_config: Record<string, unknown>
@@ -439,6 +439,8 @@ export interface EdgeNode {
   status: 'online' | 'offline'
   last_seen_at?: string | null
   ip?: string | null
+  runtimes?: string[]
+  runtime_capabilities?: Record<string, string[]>
   created_at: string
   updated_at: string
 }
@@ -453,9 +455,31 @@ export interface EdgeNodeEvent {
 }
 
 export interface SystemConfig {
+  app_name: string
+  app_env: string
+  debug: boolean
   collection_mode: 'local' | 'agent'
+  collection_orchestrator: 'admin' | 'iii'
   task_executor: 'local' | 'celery'
+  local_max_concurrent_pipelines: number
+  opencli_timeout: number
+  default_timezone: string
+  public_url: string
+  fleet_network_provider: 'lan' | 'netbird' | 'wireguard' | 'ssh' | 'custom'
+  netbird_mode: 'off' | 'host' | 'docker'
+  opencli_cdp_endpoint: string
+  agent_pool_endpoints: string[]
+  effective_cdp_endpoints: string[]
+  llm_request_timeout_seconds: number
+  llm_max_concurrency: number
+  control_mode: 'advisory' | 'automatic'
+  control_kill_switch: boolean
   image_tag: string
+  database_kind: string
+  api_auth_configured: boolean
+  oidc_configured: boolean
+  smtp_configured: boolean
+  credential_encryption_configured: boolean
 }
 
 export interface NodeStats {
@@ -1244,18 +1268,40 @@ export interface OperationsAgent {
   updated_at: string
 }
 
-export interface AgentContractV1 {
-  schema_version: 'agent.contract.v1'
+export interface OperationsAgentTeam {
+  id: string
+  workspace_id: string
+  name: string
+  slug: string
+  created_at: string
+}
+
+export interface AgentContractV2 {
+  schema_version: 'agent.contract.v2'
+  role: string
   input_schema: Record<string, unknown>
   output_schema: Record<string, unknown>
   state_schema: Record<string, unknown>
+  required_capabilities: string[]
+  tool_policy: Record<string, unknown>
+  budget: Record<string, unknown>
+  quality_gates: Array<Record<string, unknown>>
+  evidence_requirements: string[]
 }
 
-export interface AgentRuntimeBindingV1 {
-  schema_version: 'agent.runtime-binding.v1'
-  agent_url: string
-  runtime: 'pi'
+export interface AgentModelBindingV1 {
+  schema_version: 'agent.model-binding.v1'
+  provider: string
+  model: string
+  auth_profile: string | null
+}
+
+export interface AgentRuntimeBindingV2 {
+  schema_version: 'agent.runtime-binding.v2'
   workflow: string
+  preferred_agent_urls: string[]
+  preferred_runtimes: string[]
+  model_binding: AgentModelBindingV1 | null
   config: Record<string, unknown>
   dispatch_timeout_seconds: number
 }
@@ -1264,8 +1310,8 @@ export interface OperationsAgentDraft {
   revision: number
   instructions: string
   model_configuration: Record<string, unknown> & {
-    agent_contract?: AgentContractV1
-    runtime_binding?: AgentRuntimeBindingV1
+    agent_contract?: AgentContractV2
+    runtime_binding?: AgentRuntimeBindingV2
   }
   tool_configuration: Record<string, unknown>
   updated_by_user_id: string
@@ -1291,11 +1337,26 @@ export interface OperationsAgentRun {
   profile_version: number
   trigger_type: string
   trigger_reference: string | null
+  automation_id: string | null
+  automation_revision: number | null
+  automation_snapshot: Record<string, unknown> | null
+  scheduled_for: string | null
+  schedule_timezone: string | null
   target_resource_type: string
   target_resource_id: string
   input_payload: Record<string, unknown>
   state_payload: Record<string, unknown>
   output_payload: Record<string, unknown> | null
+  execution_binding: Record<string, unknown> | null
+  evidence_payload: {
+    schema_version: 'agent.run-evidence.v1'
+    runtime: Record<string, unknown>
+    events: Array<Record<string, unknown>>
+    artifacts: Array<Record<string, unknown>>
+    evidence: Array<Record<string, unknown>>
+    lineage: Array<Record<string, unknown>>
+    audit: Array<Record<string, unknown>>
+  } | null
   error_message: string | null
   status: 'queued' | 'running' | 'paused' | 'completed' | 'failed' | 'cancelled'
   started_by_user_id: string
@@ -1306,6 +1367,9 @@ export interface OperationsAgentRun {
 export interface Automation {
   id: string
   workspace_id: string
+  revision: number
+  operations_agent_id: string | null
+  operations_agent_version: number | null
   name: string
   prompt: string
   precheck: string | null
diff --git a/frontend/lib/automations/activity.ts b/frontend/lib/automations/activity.ts
new file mode 100644
index 00000000..409ab4f1
--- /dev/null
+++ b/frontend/lib/automations/activity.ts
@@ -0,0 +1,5 @@
+import type { OperationsAgentRun } from '@/lib/api/types'
+
+export function latestOperationsAgentRuns(runs: OperationsAgentRun[]) {
+  return new Map(runs.map((run) => [run.operations_agent_id, run]))
+}
diff --git a/frontend/lib/automations/approval.ts b/frontend/lib/automations/approval.ts
new file mode 100644
index 00000000..48f3832f
--- /dev/null
+++ b/frontend/lib/automations/approval.ts
@@ -0,0 +1,11 @@
+import type { OperationsAgentMode } from '@/lib/api/types'
+
+export const AUTOMATION_APPROVALS: Array<{
+  id: OperationsAgentMode
+  label: string
+  detail: string
+}> = [
+  { id: 'observe_only', label: '仅观察', detail: '不提出或执行变更' },
+  { id: 'suggest_changes', label: '建议需批准', detail: '送入 Inbox 后由人决定' },
+  { id: 'low_risk_automatic', label: '低风险自动', detail: '白名单外仍需批准' },
+]
diff --git a/frontend/lib/automations/binding.ts b/frontend/lib/automations/binding.ts
new file mode 100644
index 00000000..968c095f
--- /dev/null
+++ b/frontend/lib/automations/binding.ts
@@ -0,0 +1,19 @@
+import type { OperationsAgent, OperationsAgentMode } from '@/lib/api/types'
+
+export function runnableOperationsAgents(agents: OperationsAgent[]) {
+  return agents.filter(
+    (agent) =>
+      !agent.disabled
+      && agent.current_published_version !== null
+      && agent.current_profile.mode !== 'low_risk_automatic',
+  )
+}
+
+export function compatibleOperationsAgents(
+  agents: OperationsAgent[],
+  mode: OperationsAgentMode,
+) {
+  return runnableOperationsAgents(agents).filter(
+    (agent) => agent.current_profile.mode === mode,
+  )
+}
diff --git a/frontend/lib/automations/executors.ts b/frontend/lib/automations/executors.ts
new file mode 100644
index 00000000..83c1da23
--- /dev/null
+++ b/frontend/lib/automations/executors.ts
@@ -0,0 +1,12 @@
+import { Cloud, Code2, Sparkles, Terminal } from 'lucide-react'
+
+export const AUTOMATION_EXECUTORS = [
+  { id: 'codex', name: 'Codex', icon: Code2, color: 'text-sky-400' },
+  { id: 'claude', name: 'Claude', icon: Sparkles, color: 'text-orange-400' },
+  { id: 'chatcloud', name: 'ChatCloud', icon: Cloud, color: 'text-violet-400' },
+  { id: 'custom', name: '自定义', icon: Terminal, color: 'text-emerald-400' },
+] as const
+
+export function automationExecutorMeta(id: string) {
+  return AUTOMATION_EXECUTORS.find((item) => item.id === id) ?? AUTOMATION_EXECUTORS[3]
+}
diff --git a/frontend/lib/automations/schedule.ts b/frontend/lib/automations/schedule.ts
new file mode 100644
index 00000000..d5751a61
--- /dev/null
+++ b/frontend/lib/automations/schedule.ts
@@ -0,0 +1,9 @@
+export function automationScheduleText(value: string) {
+  const [kind, time] = value.split('@')
+  const label = kind === 'daily' ? '每天' : kind === 'weekdays' ? '工作日' : kind === 'weekly' ? '每周' : kind === 'hourly' ? '每小时' : kind
+  return `${label}${time ? ` ${time}` : ''}`
+}
+
+export function automationScheduleValue(kind: string, time: string) {
+  return kind === 'hourly' ? 'hourly' : `${kind}@${time}`
+}
diff --git a/frontend/lib/flow/settings-store.ts b/frontend/lib/flow/settings-store.ts
index b40d7b67..942369c0 100644
--- a/frontend/lib/flow/settings-store.ts
+++ b/frontend/lib/flow/settings-store.ts
@@ -40,9 +40,20 @@ export interface CanvasSettings {
   collabProvider: "off" | "broadcast" | "yjs"
   yjsRoom: string
   yjsUrl: string
+  yjsEnabled: boolean
+  yjsConnected: boolean
 }
 
-interface SettingsState extends CanvasSettings {
+export function resolveYjsUrl(): string {
+  const configured = process.env.NEXT_PUBLIC_YJS_URL?.trim()
+  if (configured) return configured
+  if (typeof window === "undefined") return ""
+  const protocol = window.location.protocol === "https:" ? "wss:" : "ws:"
+  return `${protocol}//${window.location.host}`
+}
+
+export interface SettingsState extends CanvasSettings {
+
   set: <K extends keyof CanvasSettings>(key: K, value: CanvasSettings[K]) => void
   patch: (partial: Partial<CanvasSettings>) => void
   reset: () => void
@@ -73,8 +84,10 @@ export const DEFAULT_SETTINGS: CanvasSettings = {
   showControls: true,
   showBackground: true,
   collabProvider: "off",
-  yjsRoom: "workflow-demo",
-  yjsUrl: "wss://demos.yjs.dev",
+  yjsRoom: "",
+  yjsUrl: "",
+  yjsEnabled: false,
+  yjsConnected: false,
 }
 
 export const useSettingsStore = create<SettingsState>((set) => ({
diff --git a/frontend/lib/flow/yjs-provider.ts b/frontend/lib/flow/yjs-provider.ts
index 13475caa..9e225ac6 100644
--- a/frontend/lib/flow/yjs-provider.ts
+++ b/frontend/lib/flow/yjs-provider.ts
@@ -1,17 +1,19 @@
 "use client"
 
-// Thin Yjs wrapper for real-time collaboration on nodes / edges + awareness
-// cursors. Mirrors the recipe in https://reactflow.dev/examples/interaction/collaborative
-// but reused as a hook so the canvas is decoupled from transport.
+// Thin Yjs wrapper for real-time collaboration on canonical workflow graph
+// records plus awareness cursors. The transport remains decoupled from canvas UI.
 
-import { useEffect, useMemo, useRef, useState } from "react"
+import { useCallback, useEffect, useMemo, useRef, useState } from "react"
 import * as Y from "yjs"
 import { WebsocketProvider } from "y-websocket"
 import { nanoid } from "nanoid"
-import type { WorkflowNode, WorkflowEdge } from "./types"
+import { getApiAuthToken } from "@/lib/api/auth-token"
 
 const COLORS = ["#ff7a17", "#4ade80", "#a0c3ec", "#f87171", "#c084fc", "#facc15"]
 const NAMES = ["Ada", "Alan", "Grace", "Linus", "Rita", "Ken", "Barbara", "Dennis"]
+const LOCAL_ORIGIN = Symbol("workflow-yjs-local")
+
+type CanonicalWorkflowRecord = { id: string }
 
 export interface RemoteCursor {
   id: number
@@ -28,34 +30,63 @@ export interface RemoteUser {
 }
 
 export interface YjsBinding {
+  authorized: boolean
   connected: boolean
+  synced: boolean
   users: RemoteUser[]
   cursors: RemoteCursor[]
-  /** publish local graph state — call from a store subscription. */
-  publish(nodes: WorkflowNode[], edges: WorkflowEdge[]): void
-  /** publish local cursor position (flow coords). */
+  /**
+   * Seed an empty room once synchronization completes. A non-empty room wins
+   * over this client's loaded draft.
+   */
+  initialize(nodes: CanonicalWorkflowRecord[], edges: CanonicalWorkflowRecord[]): void
+  /** Publish local canonical graph state after the initial room decision. */
+  publish(nodes: CanonicalWorkflowRecord[], edges: CanonicalWorkflowRecord[]): void
+  /** Publish local cursor position (flow coords). */
   publishCursor(x: number, y: number): void
-  /** subscribe to remote graph updates. */
-  onRemote(handler: (nodes: WorkflowNode[], edges: WorkflowEdge[]) => void): () => void
+  /** Subscribe to remote canonical graph updates. */
+  onRemote(handler: (nodes: CanonicalWorkflowRecord[], edges: CanonicalWorkflowRecord[]) => void): () => void
 }
 
 const NULL_BINDING: YjsBinding = {
+  authorized: false,
   connected: false,
+  synced: false,
   users: [],
   cursors: [],
+  initialize: () => {},
   publish: () => {},
   publishCursor: () => {},
   onRemote: () => () => {},
 }
 
+type ProviderBinding = {
+  doc: Y.Doc
+  provider: WebsocketProvider
+  nodesMap: Y.Map<CanonicalWorkflowRecord>
+  edgesMap: Y.Map<CanonicalWorkflowRecord>
+  initialized: boolean
+}
+
+function setRecordIfChanged(
+  records: Y.Map<CanonicalWorkflowRecord>,
+  record: CanonicalWorkflowRecord,
+): void {
+  const current = records.get(record.id)
+  if (current && JSON.stringify(current) === JSON.stringify(record)) return
+  records.set(record.id, record)
+}
+
 export function useYjs(
   enabled: boolean,
   { url, room }: { url: string; room: string },
 ): YjsBinding {
+  const token = getApiAuthToken()
   const [connected, setConnected] = useState(false)
+  const [synced, setSynced] = useState(false)
   const [users, setUsers] = useState<RemoteUser[]>([])
   const [cursors, setCursors] = useState<RemoteCursor[]>([])
-  const remoteHandlerRef = useRef<((nodes: WorkflowNode[], edges: WorkflowEdge[]) => void) | null>(null)
+  const remoteHandlerRef = useRef<((nodes: CanonicalWorkflowRecord[], edges: CanonicalWorkflowRecord[]) => void) | null>(null)
   const applyingRemote = useRef(false)
 
   const identity = useMemo(
@@ -66,106 +97,129 @@ export function useYjs(
     [],
   )
 
-  const bindingRef = useRef<{
-    doc: Y.Doc
-    provider: WebsocketProvider
-    nodesMap: Y.Map<WorkflowNode>
-    edgesMap: Y.Map<WorkflowEdge>
-  } | null>(null)
+  const bindingRef = useRef<ProviderBinding | null>(null)
+
+  const emitRemote = useCallback(() => {
+    const binding = bindingRef.current
+    const handler = remoteHandlerRef.current
+    if (!binding || !handler) return
+    applyingRemote.current = true
+    handler(Array.from(binding.nodesMap.values()), Array.from(binding.edgesMap.values()))
+    queueMicrotask(() => {
+      applyingRemote.current = false
+    })
+  }, [])
 
   useEffect(() => {
-    if (!enabled) return
-    if (typeof window === "undefined") return
+    if (!enabled || !token || typeof window === "undefined") return
 
     const doc = new Y.Doc()
-    const nodesMap = doc.getMap<WorkflowNode>("nodes")
-    const edgesMap = doc.getMap<WorkflowEdge>("edges")
-    const provider = new WebsocketProvider(url, room, doc)
+    const nodesMap = doc.getMap<CanonicalWorkflowRecord>("nodes")
+    const edgesMap = doc.getMap<CanonicalWorkflowRecord>("edges")
+    const provider = new WebsocketProvider(url, room, doc, { params: { token } })
 
     provider.awareness.setLocalStateField("user", identity)
+    bindingRef.current = { doc, provider, nodesMap, edgesMap, initialized: false }
 
-    bindingRef.current = { doc, provider, nodesMap, edgesMap }
-
-    const onStatus = ({ status }: { status: string }) => setConnected(status === "connected")
-    provider.on("status", onStatus)
-
-    const onGraphUpdate = () => {
-      if (!remoteHandlerRef.current) return
-      applyingRemote.current = true
-      const nodes = Array.from(nodesMap.values())
-      const edges = Array.from(edgesMap.values())
-      remoteHandlerRef.current(nodes, edges)
-      queueMicrotask(() => {
-        applyingRemote.current = false
-      })
+    const onStatus = ({ status }: { status: "connected" | "disconnected" | "connecting" }) => {
+      setConnected(status === "connected")
+    }
+    const onSync = (state: boolean) => setSynced(state)
+    const onNodesUpdate = (_event: Y.YMapEvent<CanonicalWorkflowRecord>, transaction: Y.Transaction) => {
+      if (transaction.origin !== LOCAL_ORIGIN) emitRemote()
+    }
+    const onEdgesUpdate = (_event: Y.YMapEvent<CanonicalWorkflowRecord>, transaction: Y.Transaction) => {
+      if (transaction.origin !== LOCAL_ORIGIN) emitRemote()
     }
-    nodesMap.observe(onGraphUpdate)
-    edgesMap.observe(onGraphUpdate)
-
     const onAwareness = () => {
       const states = Array.from(provider.awareness.getStates().entries())
       const nextUsers: RemoteUser[] = []
       const nextCursors: RemoteCursor[] = []
       const localId = provider.awareness.clientID
       for (const [clientId, state] of states) {
-        const s = state as { user?: RemoteUser; cursor?: { x: number; y: number } }
-        if (!s.user) continue
-        if (clientId === localId) continue
-        nextUsers.push({ id: clientId, name: s.user.name, color: s.user.color })
-        if (s.cursor) {
-          nextCursors.push({ id: clientId, name: s.user.name, color: s.user.color, x: s.cursor.x, y: s.cursor.y })
+        const current = state as { user?: RemoteUser; cursor?: { x: number; y: number } }
+        if (!current.user || clientId === localId) continue
+        nextUsers.push({ id: clientId, name: current.user.name, color: current.user.color })
+        if (current.cursor) {
+          nextCursors.push({
+            id: clientId,
+            name: current.user.name,
+            color: current.user.color,
+            x: current.cursor.x,
+            y: current.cursor.y,
+          })
         }
       }
       setUsers(nextUsers)
       setCursors(nextCursors)
     }
+
+    provider.on("status", onStatus)
+    provider.on("sync", onSync)
+    nodesMap.observe(onNodesUpdate)
+    edgesMap.observe(onEdgesUpdate)
     provider.awareness.on("change", onAwareness)
 
     return () => {
-      nodesMap.unobserve(onGraphUpdate)
-      edgesMap.unobserve(onGraphUpdate)
+      nodesMap.unobserve(onNodesUpdate)
+      edgesMap.unobserve(onEdgesUpdate)
       provider.awareness.off("change", onAwareness)
+      provider.off("sync", onSync)
       provider.off("status", onStatus)
       provider.destroy()
       doc.destroy()
       bindingRef.current = null
       setConnected(false)
+      setSynced(false)
       setUsers([])
       setCursors([])
     }
-  }, [enabled, url, room, identity])
-
-  if (!enabled) return NULL_BINDING
-
-  return {
-    connected,
-    users,
-    cursors,
-    publish(nodes, edges) {
-      const b = bindingRef.current
-      if (!b || applyingRemote.current) return
-      b.doc.transact(() => {
-        const nextNodeIds = new Set(nodes.map((n) => n.id))
-        for (const key of Array.from(b.nodesMap.keys())) {
-          if (!nextNodeIds.has(key)) b.nodesMap.delete(key)
-        }
-        for (const node of nodes) b.nodesMap.set(node.id, node)
+  }, [enabled, emitRemote, identity, room, token, url])
+
+  const publish = useCallback((nodes: CanonicalWorkflowRecord[], edges: CanonicalWorkflowRecord[]) => {
+    const binding = bindingRef.current
+    if (!binding || !binding.initialized || applyingRemote.current) return
+    binding.doc.transact(() => {
+      const nextNodeIds = new Set(nodes.map((node) => node.id))
+      for (const key of Array.from(binding.nodesMap.keys())) {
+        if (!nextNodeIds.has(key)) binding.nodesMap.delete(key)
+      }
+      for (const node of nodes) setRecordIfChanged(binding.nodesMap, node)
 
-        const nextEdgeIds = new Set(edges.map((e) => e.id))
-        for (const key of Array.from(b.edgesMap.keys())) {
-          if (!nextEdgeIds.has(key)) b.edgesMap.delete(key)
-        }
-        for (const edge of edges) b.edgesMap.set(edge.id, edge)
-      }, "local")
-    },
-    publishCursor(x, y) {
-      bindingRef.current?.provider.awareness.setLocalStateField("cursor", { x, y })
-    },
-    onRemote(handler) {
-      remoteHandlerRef.current = handler
-      return () => {
-        if (remoteHandlerRef.current === handler) remoteHandlerRef.current = null
+      const nextEdgeIds = new Set(edges.map((edge) => edge.id))
+      for (const key of Array.from(binding.edgesMap.keys())) {
+        if (!nextEdgeIds.has(key)) binding.edgesMap.delete(key)
       }
-    },
-  }
+      for (const edge of edges) setRecordIfChanged(binding.edgesMap, edge)
+    }, LOCAL_ORIGIN)
+  }, [])
+
+  const initialize = useCallback((nodes: CanonicalWorkflowRecord[], edges: CanonicalWorkflowRecord[]) => {
+    const binding = bindingRef.current
+    if (!binding || !binding.provider.synced || binding.initialized) return
+    binding.initialized = true
+    if (binding.nodesMap.size || binding.edgesMap.size) {
+      emitRemote()
+      return
+    }
+    publish(nodes, edges)
+  }, [emitRemote, publish])
+
+  const publishCursor = useCallback((x: number, y: number) => {
+    bindingRef.current?.provider.awareness.setLocalStateField("cursor", { x, y })
+  }, [])
+
+  const onRemote = useCallback((handler: (nodes: CanonicalWorkflowRecord[], edges: CanonicalWorkflowRecord[]) => void) => {
+    remoteHandlerRef.current = handler
+    const binding = bindingRef.current
+    if (binding?.provider.synced && (binding.nodesMap.size || binding.edgesMap.size)) emitRemote()
+    return () => {
+      if (remoteHandlerRef.current === handler) remoteHandlerRef.current = null
+    }
+  }, [emitRemote])
+
+  return useMemo(() => {
+    if (!enabled) return NULL_BINDING
+    return { authorized: Boolean(token), connected, synced, users, cursors, initialize, publish, publishCursor, onRemote }
+  }, [connected, cursors, enabled, initialize, onRemote, publish, publishCursor, synced, token, users])
 }
diff --git a/frontend/lib/inbox/workbench-state.ts b/frontend/lib/inbox/workbench-state.ts
new file mode 100644
index 00000000..f2eb7707
--- /dev/null
+++ b/frontend/lib/inbox/workbench-state.ts
@@ -0,0 +1,53 @@
+export type ApprovalAvailability =
+  | 'loading'
+  | 'ready'
+  | 'no_workspace'
+  | 'workspace_error'
+  | 'inbox_error'
+
+export type ApprovalAvailabilityInput = {
+  workspaceLoading: boolean
+  workspaceError: boolean
+  workspaceCount: number
+  workspaceId: string | null
+  inboxLoading: boolean
+  inboxError: boolean
+}
+
+export function resolveApprovalAvailability({
+  workspaceLoading,
+  workspaceError,
+  workspaceCount,
+  workspaceId,
+  inboxLoading,
+  inboxError,
+}: ApprovalAvailabilityInput): ApprovalAvailability {
+  if (workspaceError) return 'workspace_error'
+  if (workspaceLoading) return 'loading'
+  if (workspaceCount === 0 || !workspaceId) return 'no_workspace'
+  if (inboxError) return 'inbox_error'
+  if (inboxLoading) return 'loading'
+  return 'ready'
+}
+
+const INTERACTIVE_TAGS: Record<string, true> = {
+  A: true,
+  BUTTON: true,
+  INPUT: true,
+  SELECT: true,
+  SUMMARY: true,
+  TEXTAREA: true,
+}
+
+export function shouldIgnoreInboxShortcut({
+  tagName,
+  isContentEditable,
+  withinInteractive,
+}: {
+  tagName?: string
+  isContentEditable?: boolean
+  withinInteractive?: boolean
+}): boolean {
+  if (isContentEditable || withinInteractive) return true
+  return tagName ? INTERACTIVE_TAGS[tagName.toUpperCase()] === true : false
+}
diff --git a/frontend/lib/navigation.ts b/frontend/lib/navigation.ts
index 41b72292..6f3fc67a 100644
--- a/frontend/lib/navigation.ts
+++ b/frontend/lib/navigation.ts
@@ -4,10 +4,10 @@ import {
   Database,
   LayoutDashboard,
   PanelsTopLeft,
-  Settings,
   Settings2,
-  ShieldAlert,
   ShieldCheck,
+  ShieldAlert,
+  Settings,
   Workflow,
   type LucideIcon,
 } from 'lucide-react'
@@ -48,10 +48,10 @@ export const NAV_GROUPS: NavGroup[] = [
       { href: '/studio', label: '项目', icon: PanelsTopLeft, match: ['/studio', '/canvas'] },
       { href: '/plugins', label: '插件中心', icon: Blocks },
       {
-        href: '/schedules',
-        label: '自动化与 Agent',
+        href: '/operations-agents',
+        label: '自动化与智能体',
         icon: Workflow,
-        match: ['/schedules', '/plans', '/agents', '/skills'],
+        match: ['/operations-agents', '/schedules'],
       },
     ],
   },
@@ -94,6 +94,11 @@ export const NAV_GROUPS: NavGroup[] = [
         icon: Settings,
         match: ['/settings'],
       },
+      {
+        href: '/system',
+        label: '系统设置',
+        icon: Settings2,
+      },
     ],
   },
 ]
@@ -107,21 +112,17 @@ export const ROUTE_LABELS: Record<string, string> = {
   '/canvas': '节点工作流（兼容入口）',
   '/plugins': '插件中心',
   '/sources': '数据（兼容入口）',
-  '/schedules': '触发与调度',
-  '/plans': '计划',
+  '/schedules': '自动化与智能体',
   '/tasks': '工作项',
   '/records': '成果与数据',
   '/notifications': '通知',
   '/agents': '智能体',
+  '/operations-agents': '自动化与智能体',
+  '/system': '系统设置',
   '/skills': '技能',
   '/providers': '模型与连接',
-  '/nodes': '执行资源',
   '/workers': 'Worker',
   '/browsers': 'Chrome 池',
   '/control/actions': '控制与审计',
-  '/control/kill-switch': '熔断开关',
-  '/control/advisory-report': '建议报告',
-  '/control/odp-state': 'ODP 状态',
-  '/system': '系统与运维',
-  '/settings': '工作区设置',
+  '/settings': '系统设置',
 }
diff --git a/frontend/lib/rendering/gpu-capabilities.ts b/frontend/lib/rendering/gpu-capabilities.ts
new file mode 100644
index 00000000..d77dbbf8
--- /dev/null
+++ b/frontend/lib/rendering/gpu-capabilities.ts
@@ -0,0 +1,48 @@
+export type GpuRendererBackend = 'pending' | 'webgl2' | 'fallback'
+
+export type GpuCapabilities = {
+  webgl2: boolean
+  webgpuApiPresent: boolean
+}
+
+const UNSUPPORTED_CAPABILITIES: GpuCapabilities = {
+  webgl2: false,
+  webgpuApiPresent: false,
+}
+
+let cachedCapabilities: GpuCapabilities | null = null
+
+/**
+ * Performs an SSR-safe capability probe. This describes what the browser can
+ * create, never the backend currently used by a product surface.
+ */
+export function probeGpuCapabilities(): GpuCapabilities {
+  if (
+    typeof window === 'undefined'
+    || typeof document === 'undefined'
+    || typeof navigator === 'undefined'
+  ) {
+    return UNSUPPORTED_CAPABILITIES
+  }
+  if (cachedCapabilities) return cachedCapabilities
+
+  const webgpuApiPresent = 'gpu' in navigator
+  let webgl2 = false
+  let context: WebGL2RenderingContext | null = null
+  try {
+    const canvas = document.createElement('canvas')
+    context = canvas.getContext('webgl2')
+    webgl2 = Boolean(context && !context.isContextLost())
+  } catch {
+    webgl2 = false
+  } finally {
+    context?.getExtension('WEBGL_lose_context')?.loseContext()
+  }
+
+  cachedCapabilities = { webgl2, webgpuApiPresent }
+  return cachedCapabilities
+}
+
+export function invalidateGpuCapabilitiesAfterContextLoss() {
+  cachedCapabilities = null
+}
diff --git a/frontend/lib/studio/operations-task-model.ts b/frontend/lib/studio/operations-task-model.ts
new file mode 100644
index 00000000..ae841d18
--- /dev/null
+++ b/frontend/lib/studio/operations-task-model.ts
@@ -0,0 +1,40 @@
+export type OperationsTaskStatus =
+  | 'pending'
+  | 'running'
+  | 'done'
+  | 'failed'
+  | 'blocked'
+  | 'partial'
+
+export type OperationsNodeState = {
+  nodeId: string
+  status: string
+  eventCount: number
+}
+
+export type OperationsNodeTask = {
+  id: string
+  label: string
+  note: string
+  status: OperationsTaskStatus
+}
+
+export function toOperationsTaskStatus(status: string): OperationsTaskStatus {
+  if (status === 'completed') return 'done'
+  if (status === 'failed') return 'failed'
+  if (status === 'blocked') return 'blocked'
+  if (status === 'partial' || status === 'partial_success') return 'partial'
+  if (status === 'running') return 'running'
+  return 'pending'
+}
+
+export function buildOperationsNodeTasks(
+  nodeStates: OperationsNodeState[],
+): OperationsNodeTask[] {
+  return nodeStates.map((node) => ({
+    id: node.nodeId,
+    label: node.nodeId,
+    note: `${node.eventCount} events · ${node.status}`,
+    status: toOperationsTaskStatus(node.status),
+  }))
+}
diff --git a/frontend/lib/workflow/persistence.ts b/frontend/lib/workflow/persistence.ts
new file mode 100644
index 00000000..5647c60f
--- /dev/null
+++ b/frontend/lib/workflow/persistence.ts
@@ -0,0 +1,23 @@
+import type { WorkflowProject, WorkflowProjectNode } from './schema'
+
+export function persistableWorkflowProject(project: WorkflowProject): WorkflowProject {
+  const persistableNode = (node: WorkflowProjectNode): WorkflowProjectNode => {
+    const ui = { ...(node.ui ?? {}) }
+    delete ui.runtimeCapability
+    return {
+      ...node,
+      ...(node.ui ? { ui } : {}),
+      ...(node.internals
+        ? {
+            internals: {
+              ...node.internals,
+              nodes: node.internals.nodes.map((item) =>
+                persistableNode(item as WorkflowProjectNode),
+              ),
+            },
+          }
+        : {}),
+    }
+  }
+  return { ...project, nodes: project.nodes.map(persistableNode) }
+}
diff --git a/frontend/scripts/check-inbox-regressions.mjs b/frontend/scripts/check-inbox-regressions.mjs
index 4e646050..fbe6f0c7 100644
--- a/frontend/scripts/check-inbox-regressions.mjs
+++ b/frontend/scripts/check-inbox-regressions.mjs
@@ -4,8 +4,9 @@ import { test } from 'node:test'
 
 const read = (path) => readFile(new URL(`../${path}`, import.meta.url), 'utf8')
 
-test('inbox is built only from currently available operational APIs', async () => {
+test('inbox combines existing operational signals with server-backed human approvals', async () => {
   const page = await read('app/(app)/inbox/page.tsx')
+  const approvalDetail = await read('components/inbox/queue-detail.tsx')
   const hooks = await read('lib/api/hooks.ts')
   const endpoints = await read('lib/api/endpoints.ts')
 
@@ -17,16 +18,24 @@ test('inbox is built only from currently available operational APIs', async () =
   assert.match(hooks, /export function useInfiniteNotificationLogs/)
   assert.match(hooks, /export function useInfiniteControlActions/)
   assert.match(endpoints, /listNotificationLogs = \(params\?: \{ rule_id\?: string; page\?: number; limit\?: number \}\)/)
-  assert.doesNotMatch(page, /useMyWorkspaces|useOperationsInbox|\/workspaces|operations-inbox/)
+  assert.match(page, /useGovernedWorkspaces\(\)/)
+  assert.match(page, /useOperationsInbox\(workspaceId, 'open'\)/)
+  assert.match(approvalDetail, /useDecideOperationsApproval\(\)/)
+  assert.match(approvalDetail, /<AIApproval/)
+  assert.match(approvalDetail, /审批理由（必填）/)
+  assert.match(approvalDetail, /setCardRevision\(\(revision\) => revision \+ 1\)/)
+  assert.match(hooks, /export function useOperationsInbox/)
+  assert.match(hooks, /export function useDecideOperationsApproval/)
 })
 
 test('inbox uses a Linear-style queue while preserving destinations for underlying records', async () => {
   const page = await read('app/(app)/inbox/page.tsx')
+  const detail = await read('components/inbox/queue-detail.tsx')
 
   assert.match(page, /data-testid="inbox-workbench"/)
   assert.match(page, /lg:h-\[calc\(100dvh-3\.5rem\)\]/)
   assert.match(page, /data-testid="inbox-queue-scroll"/)
-  assert.match(page, /data-testid="inbox-detail-scroll"/)
+  assert.match(detail, /data-testid="inbox-detail-scroll"/)
   assert.doesNotMatch(page, /<PageContainer/)
   assert.doesNotMatch(page, /className="overflow-hidden rounded-xl border bg-card shadow-sm"/)
   assert.match(page, /ACTION_CENTER_TABS/)
@@ -43,7 +52,7 @@ test('inbox uses a Linear-style queue while preserving destinations for underlyi
   assert.match(page, /href: `\/tasks\/\$\{task\.id\}`/)
   assert.match(page, /href: '\/notifications'/)
   assert.match(page, /href: '\/control\/actions'/)
-  assert.match(page, /href=\{`\/sources\/\$\{item\.sourceId\}`\}/)
+  assert.match(detail, /href=\{`\/sources\/\$\{item\.sourceId\}`\}/)
 })
 
 test('inbox preserves queue state and progressively loads hundreds-scale signal sets', async () => {
@@ -65,8 +74,8 @@ test('inbox preserves queue state and progressively loads hundreds-scale signal
 test('inbox renders explicit initial, partial, empty, and total failure states', async () => {
   const page = await read('app/(app)/inbox/page.tsx')
 
-  assert.match(page, /const isInitialLoading = queries\.every/)
-  assert.match(page, /const isTotalFailure = queries\.every/)
+  assert.match(page, /const isInitialLoading =\s+queries\.every/)
+  assert.match(page, /const isTotalFailure =\s+queries\.every/)
   assert.match(page, /const partialFailures =/)
   assert.match(page, /暂时无法读取，其余信号仍可处理/)
   assert.match(page, /当前视图已经清空/)
diff --git a/frontend/scripts/check-login-theme-regressions.mjs b/frontend/scripts/check-login-theme-regressions.mjs
index 40eddf7a..793eb2ce 100644
--- a/frontend/scripts/check-login-theme-regressions.mjs
+++ b/frontend/scripts/check-login-theme-regressions.mjs
@@ -14,13 +14,15 @@ test('login keeps the liquid, terminal, and pixel theme switcher', async () => {
   assert.match(login, /<Dither/)
 })
 
-test('login preserves the current auth paths and reduced-motion fallback', async () => {
+test('login uses local credentials and keeps the reduced-motion fallback', async () => {
   const login = await read('app/login/page.tsx')
 
-  assert.match(login, /signInWithOidc/)
-  assert.match(login, /signInWithBootstrap/)
+  assert.match(login, /signInWithPassword/)
   assert.match(login, /enterDevelopmentMode/)
   assert.match(login, /prefers-reduced-motion: reduce/)
+  assert.doesNotMatch(login, /signInWithBootstrap/)
+  assert.doesNotMatch(login, /管理员身份令牌/)
+  assert.doesNotMatch(login, /Fleet API 令牌/)
 })
 
 test('auth defaults return to the project list instead of a contextless workflow', async () => {
diff --git a/frontend/scripts/check-navigation-transition-regressions.mjs b/frontend/scripts/check-navigation-transition-regressions.mjs
index 587e7bae..dc37ab8c 100644
--- a/frontend/scripts/check-navigation-transition-regressions.mjs
+++ b/frontend/scripts/check-navigation-transition-regressions.mjs
@@ -32,7 +32,7 @@ test('persistent application chrome stays outside the routed animation boundary'
   assert.match(shell, /className="[^"]*relative[^"]*z-0[^"]*overflow-x-clip[^"]*bg-background[^"]*"/)
 })
 
-test('sidebar consolidates tasks, notifications, automation, and agents into clear work areas', async () => {
+test('sidebar keeps automation separate from Agent surfaces', async () => {
   const [navigation, sidebar] = await Promise.all([
     read('lib/navigation.ts'),
     read('components/shell/app-sidebar.tsx'),
@@ -43,7 +43,7 @@ test('sidebar consolidates tasks, notifications, automation, and agents into cle
     '任务与通知',
     '项目',
     '插件中心',
-    '自动化与 Agent',
+    '自动化与智能体',
     '执行资源',
     '成果与数据',
     '模型与连接',
@@ -53,9 +53,13 @@ test('sidebar consolidates tasks, notifications, automation, and agents into cle
 
   assert.match(navigation, /href: '\/inbox'/)
   assert.match(navigation, /match: \['\/inbox', '\/tasks', '\/notifications'\]/)
-  assert.match(navigation, /match: \['\/sources', '\/schedules', '\/agents', '\/skills'\]/)
-  assert.match(navigation, /match: \['\/nodes', '\/workers'\]/)
-  assert.match(navigation, /match: \['\/providers', '\/control\/actions'\]/)
+  assert.match(navigation, /match: \['\/operations-agents', '\/schedules'\]/)
+  assert.doesNotMatch(navigation, /label: '自动化与 Agent'/)
+  assert.doesNotMatch(navigation, /match: \[[^\]]*'\/agents'/)
+  assert.doesNotMatch(navigation, /match: \[[^\]]*'\/skills'/)
+  assert.match(navigation, /'\/operations-agents': '自动化与智能体'/)
+  assert.match(navigation, /match: \['\/nodes', '\/workers', '\/browsers'\]/)
+  assert.match(navigation, /match: \['\/providers'\]/)
   for (const group of ['工作台', '构建', '运行与数据', '管理']) {
     assert.match(navigation, new RegExp(`label: '${group}'`))
   }
@@ -69,12 +73,11 @@ test('sidebar consolidates tasks, notifications, automation, and agents into cle
 test('records use a scalable source-to-table explorer with pagination and raw evidence detail', async () => {
   const records = await read('app/(app)/records/page.tsx')
 
-  assert.match(records, /lg:grid-cols-\[15rem_minmax\(0,1fr\)\]/)
+  assert.match(records, /lg:grid-cols-\[minmax\(0,1fr\)_auto\]/)
   assert.doesNotMatch(records, /grid min-h-\[38rem\] overflow-hidden/)
-  assert.match(records, /min-h-\[20rem\]/)
-  assert.match(records, /min-h-\[32rem\]/)
-  assert.match(records, /aria-label="成果数据集"/)
-  assert.match(records, /useSources\(\{ page: 1, limit: 100 \}\)/)
+  assert.match(records, /min-h-80/)
+  assert.match(records, /aria-label="当前数据字段"/)
+  assert.match(records, /useRecords\(\{[\s\S]{0,200}page,[\s\S]{0,100}limit: PAGE_SIZE/)
   assert.match(records, /limit: PAGE_SIZE/)
   assert.match(records, /visibleFields/)
   assert.match(records, /第 \{page\.toLocaleString/)
@@ -95,24 +98,28 @@ test('task and automation sibling routes share their consolidated route tabs', a
     read('app/(app)/skills/page.tsx'),
   ])
 
-  for (const label of ['待处理', '工作项', '通知规则', '数据源', '调度', 'Agent', '技能']) {
+  for (const label of ['待处理', '工作项', '通知规则', '自动化与智能体', 'Agent', '技能']) {
     assert.match(tabs, new RegExp(`label: '${label}'`))
   }
   for (const page of [inbox, tasks, notifications]) {
     assert.match(page, /ACTION_CENTER_TABS/)
   }
-  for (const page of [sources, schedules, agents, skills]) {
+  for (const page of [agents, skills]) {
     assert.match(page, /AUTOMATION_TABS/)
   }
+  assert.match(sources, /redirect\('\/records'\)/)
+  assert.match(schedules, /redirect\('\/operations-agents'\)/)
 })
 
-test('studio uses ordinary creation choices while Agent stays global', async () => {
-  const [studio, templates, shell, header, agentDock] = await Promise.all([
+test('studio keeps Agent conversation global while management has its own entry', async () => {
+  const [studio, templates, shell, header, agentBubble, agentDock, transition] = await Promise.all([
     read('app/(app)/studio/page.tsx'),
     read('app/(app)/studio/templates/page.tsx'),
     read('components/shell/app-shell.tsx'),
     read('components/shell/app-header.tsx'),
+    read('components/shell/global-agent-bubble.tsx'),
     read('components/shell/global-agent-dock.tsx'),
+    read('components/motion/app-route-transition.tsx'),
   ])
 
   assert.match(studio, /\/studio\/templates\?workspace=/)
@@ -126,8 +133,12 @@ test('studio uses ordinary creation choices while Agent stays global', async ()
   assert.match(templates, /搜索模板、节点或用途/)
   assert.match(templates, /可复用的执行链路/)
   assert.doesNotMatch(templates, /改用 Agent 创建/)
+  assert.match(shell, /<GlobalAgentBubble onClick=\{\(\) => \{ setAgentPrompt\(''\); setAgentOpen\(true\) \}\} \/>/)
   assert.match(shell, /<GlobalAgentDock open=\{agentOpen\}/)
-  assert.match(header, /打开全局 Agent/)
+  assert.match(header, /href="\/operations-agents"/)
+  assert.doesNotMatch(header, /onOpenAgent/)
+  assert.match(agentBubble, /fixed bottom-4 right-4/)
+  assert.match(agentBubble, /aria-label="打开全局 Agent"/)
   assert.match(agentDock, /当前上下文/)
   assert.match(agentDock, /new URLSearchParams\(window\.location\.search\)/)
   assert.match(agentDock, /workspace_id: workspaceId/)
@@ -146,6 +157,7 @@ test('studio uses ordinary creation choices while Agent stays global', async ()
   assert.match(agentDock, /仅在后端能解析出唯一授权范围时允许确认写操作/)
   assert.match(agentDock, /\/chat\/confirm/)
   assert.match(agentDock, /queryClient\.invalidateQueries/)
+  assert.match(transition, /'\/operations-agents'/)
 })
 
 test('SSGOI boundary is pathname-keyed, interruptible, and reduced-motion safe', async () => {
diff --git a/frontend/scripts/check-project-workbench-regressions.mjs b/frontend/scripts/check-project-workbench-regressions.mjs
index 68bbb0dd..48a09958 100644
--- a/frontend/scripts/check-project-workbench-regressions.mjs
+++ b/frontend/scripts/check-project-workbench-regressions.mjs
@@ -53,7 +53,7 @@ test('project overview exposes Dify-style API access and logs monitoring with re
   assert.match(operationsPage, /useProjectRuntimeTrace/)
   assert.match(operationsPage, /TRACE_PAGE_SIZE/)
   assert.match(operationsPage, /下一批/)
-  assert.match(operationsPage, /事件时间线/)
+  assert.match(operationsPage, /节点状态与事件/)
   assert.match(navigation, /projects\/\$\{projectId\}\/api/)
   assert.match(navigation, /label: 'API \/ MCP'/)
   assert.match(navigation, /projects\/\$\{projectId\}\/operations/)
diff --git a/frontend/scripts/check-workflow-regressions.mjs b/frontend/scripts/check-workflow-regressions.mjs
index 82af9cde..35c7c705 100644
--- a/frontend/scripts/check-workflow-regressions.mjs
+++ b/frontend/scripts/check-workflow-regressions.mjs
@@ -524,7 +524,7 @@ test('node workflow lives inside a project shell while the legacy canvas route r
     readSource('app/page.tsx'),
   ])
 
-  for (const label of ['概览', '项目', '自动化与 Agent', '任务与通知', '执行资源']) {
+  for (const label of ['概览', '项目', '自动化与智能体', '任务与通知', '执行资源']) {
     assert.match(navigation, new RegExp(`label:\\s*['"]${label}['"]`))
   }
   assert.doesNotMatch(navigation, /href:\s*['"]\/canvas['"][\s\S]{0,80}label:\s*['"]节点工作流['"]/) 
@@ -1404,7 +1404,7 @@ test('the right inspector uses graph contracts instead of manual keys and field
   assert.match(inspector, /onReconnect\(currentEdge, connection\)/)
   assert.match(inspector, /connectNodes\(connection\)/)
   assert.match(inspector, /removeEdgesByIds\(currentEdges\.map\(\(edge\) => edge\.id\)\)/)
-  assert.match(inspector, /<SelectValue placeholder=\{copy\.inputUnbound\}/)
+  assert.match(inspector, /<SelectValue>[\s\S]{0,500}copy\.inputUnbound/)
   assert.doesNotMatch(inspector, /placeholder=["']data\.source["']/)
   assert.doesNotMatch(inspector, /placeholder=["']data\.target["']/)
   assert.doesNotMatch(parameterInterface, /Config \(JSON\)/)
diff --git a/frontend/scripts/smoothui-operations.test.mjs b/frontend/scripts/smoothui-operations.test.mjs
new file mode 100644
index 00000000..e4d517b2
--- /dev/null
+++ b/frontend/scripts/smoothui-operations.test.mjs
@@ -0,0 +1,61 @@
+import assert from 'node:assert/strict'
+import test from 'node:test'
+
+import {
+  buildOperationsNodeTasks,
+  toOperationsTaskStatus,
+} from '../lib/studio/operations-task-model.ts'
+import {
+  resolveApprovalAvailability,
+  shouldIgnoreInboxShortcut,
+} from '../lib/inbox/workbench-state.ts'
+
+test('node task projection derives counts and statuses only from node states', () => {
+  const tasks = buildOperationsNodeTasks([
+    { nodeId: 'fetch-source', status: 'completed', eventCount: 2 },
+    { nodeId: 'approval-gate', status: 'blocked', eventCount: 1 },
+  ])
+
+  assert.equal(tasks.length, 2)
+  assert.equal(tasks.filter((task) => task.status === 'done').length, 1)
+  assert.equal(tasks[0]?.note, '2 events · completed')
+  assert.equal(tasks[1]?.status, 'blocked')
+})
+
+test('node statuses preserve terminal, partial, active, and waiting semantics', () => {
+  assert.equal(toOperationsTaskStatus('completed'), 'done')
+  assert.equal(toOperationsTaskStatus('failed'), 'failed')
+  assert.equal(toOperationsTaskStatus('blocked'), 'blocked')
+  assert.equal(toOperationsTaskStatus('partial'), 'partial')
+  assert.equal(toOperationsTaskStatus('partial_success'), 'partial')
+  assert.equal(toOperationsTaskStatus('running'), 'running')
+  assert.equal(toOperationsTaskStatus('queued'), 'pending')
+  assert.equal(toOperationsTaskStatus('waiting'), 'pending')
+})
+
+test('inbox global shortcuts ignore direct and nested interactive targets', () => {
+  for (const tagName of ['input', 'textarea', 'select', 'button', 'a', 'summary']) {
+    assert.equal(shouldIgnoreInboxShortcut({ tagName }), true, tagName)
+  }
+  assert.equal(shouldIgnoreInboxShortcut({ tagName: 'svg', withinInteractive: true }), true)
+  assert.equal(shouldIgnoreInboxShortcut({ tagName: 'div', isContentEditable: true }), true)
+  assert.equal(shouldIgnoreInboxShortcut({ tagName: 'div' }), false)
+})
+
+test('approval availability exposes loading, prerequisite, and error states', () => {
+  const ready = {
+    workspaceLoading: false,
+    workspaceError: false,
+    workspaceCount: 1,
+    workspaceId: 'workspace-1',
+    inboxLoading: false,
+    inboxError: false,
+  }
+
+  assert.equal(resolveApprovalAvailability({ ...ready, workspaceLoading: true }), 'loading')
+  assert.equal(resolveApprovalAvailability({ ...ready, workspaceError: true }), 'workspace_error')
+  assert.equal(resolveApprovalAvailability({ ...ready, workspaceCount: 0, workspaceId: null }), 'no_workspace')
+  assert.equal(resolveApprovalAvailability({ ...ready, inboxLoading: true }), 'loading')
+  assert.equal(resolveApprovalAvailability({ ...ready, inboxError: true }), 'inbox_error')
+  assert.equal(resolveApprovalAvailability(ready), 'ready')
+})
diff --git a/frontend/tsconfig.json b/frontend/tsconfig.json
index 235db06a..06009111 100644
--- a/frontend/tsconfig.json
+++ b/frontend/tsconfig.json
@@ -34,7 +34,17 @@
     "**/*.ts",
     "**/*.tsx",
     ".next/types/**/*.ts",
-    ".next/dev/types/**/*.ts"
+    ".next/dev/types/**/*.ts",
+    ".next-slider-smoke/types/**/*.ts",
+    ".next-slider-smoke/dev/types/**/*.ts",
+    ".next-eldora-verify/types/**/*.ts",
+    ".next-eldora-verify/dev/types/**/*.ts",
+    ".next-agent-rollout/types/**/*.ts",
+    ".next-agent-rollout/dev/types/**/*.ts",
+    ".next-agent-smoke/types/**/*.ts",
+    ".next-agent-smoke/dev/types/**/*.ts",
+    ".next-agent-webpack/types/**/*.ts",
+    ".next-agent-webpack/dev/types/**/*.ts"
   ],
   "exclude": [
     "node_modules"
diff --git a/openspec/changes/collection-lineage-envelope/proposal.md b/openspec/changes/collection-lineage-envelope/proposal.md
new file mode 100644
index 00000000..997652d1
--- /dev/null
+++ b/openspec/changes/collection-lineage-envelope/proposal.md
@@ -0,0 +1,36 @@
+## Why
+
+The collection audit found two execution planes whose provenance stops at different boundaries: scheduled collection runs do not carry SourceBinding revisions or credential scope into persisted outputs, while durable acquisition artifacts are not connected to normalized records, enrichment, or delivery. As a result, operators cannot prove which source revision, binding, execution, worker/runtime, or artifact produced a record or notification, and old records have no safe compatibility contract.
+
+## What Changes
+
+- Define one immutable collection-lineage envelope for every persisted collection execution and every downstream output or attempt.
+- Propagate source, binding, account/credential, project scope, run/execution, worker/runtime, trace, and artifact references through collection tasks, task runs, acquisition executions, normalized records, enrichment results, and notification/delivery logs.
+- Specify nullable lineage fields and read behavior for records created before this contract.
+- Give notification and delivery attempts their own lineage and idempotency references without inventing quota, capacity, or identifiers.
+- Explicitly bound this P0-1 change: acquisition raw results are lineage-bearing inputs until they are normalized into records; raw results are not themselves asserted to be records, evidence, enriched outputs, or delivered items.
+
+## Capabilities
+
+### New Capabilities
+
+- `collection-lineage-envelope`: An immutable, nullable-compatible provenance contract that links collection source and execution context to records, artifacts, enrichment, trace events, and delivery attempts.
+
+### Modified Capabilities
+
+- Existing scheduled collection and durable acquisition persistence SHALL reference the common lineage contract.
+- Existing record, artifact, enrichment, trace, and notification projections SHALL preserve lineage when they are derived from a collection execution.
+
+## Impact
+
+- Backend collection, acquisition, pipeline, artifact, enrichment, trace, and notification persistence models and their serializers.
+- Worker/runtime adapters and execution boundaries that construct or forward lineage.
+- Migrations and compatibility reads for pre-existing records.
+- Focused contract and integration verification for scheduled and durable acquisition paths.
+
+## Non-Goals
+
+- This change does not define durable schedule occurrence leases, cancellation/lease-loss semantics, stage partial-success state machines, or restart reconciliation (P0-2/P0-3 and later slices).
+- This change does not normalize acquisition raw items or make them enter the record/evidence/AI/delivery path; it defines the boundary and lineage needed by that later slice.
+- This change does not infer provider quota, capacity, token usage, cost, or timing when the provider did not return documented evidence.
+- This change does not synthesize missing source, binding, worker, runtime, trace, artifact, run, or execution IDs.
diff --git a/openspec/changes/collection-lineage-envelope/specs/collection-lineage-envelope/spec.md b/openspec/changes/collection-lineage-envelope/specs/collection-lineage-envelope/spec.md
new file mode 100644
index 00000000..47698f16
--- /dev/null
+++ b/openspec/changes/collection-lineage-envelope/specs/collection-lineage-envelope/spec.md
@@ -0,0 +1,106 @@
+## Purpose
+
+Define the immutable provenance envelope carried by collection executions and their persisted derivatives, while preserving safe reads of records created before the envelope exists.
+
+## ADDED Requirements
+
+### Requirement: Collection lineage is represented by one immutable envelope
+Every collection execution that creates or processes persisted collection output SHALL expose one immutable lineage envelope. The envelope SHALL carry references, when established by the owning system, to the source revision, source binding revision, account and credential revision, project/scope binding, execution, run, worker, runtime, trace, and artifact. A consumer MAY add a derived reference only by preserving the originating envelope; it SHALL NOT mutate the originating values.
+
+#### Scenario: Scheduled collection constructs an attributable envelope
+- **WHEN** a scheduled collection is created from a SourceBinding revision with account, credential, and project scope
+- **THEN** the CollectionTask and TaskRun carry the same source, binding, account/credential, and scope references
+- **AND** the execution and run references are carried into downstream lineage projections
+- **AND** no field is populated from an unrelated source or binding.
+
+#### Scenario: Durable acquisition carries the envelope across execution
+- **WHEN** an AcquisitionExecution dispatches work to an agent, browser, OpenCLI, or another worker runtime
+- **THEN** the worker handoff and resulting artifact references carry the originating envelope
+- **AND** the execution remains attributable to the source and binding revision that authorized it.
+
+### Requirement: Persisted records retain collection lineage
+A normalized or stored collection record derived from a collection execution SHALL retain references to the originating source, binding, account/credential, scope, execution/run, worker/runtime, trace, and artifact envelope. Record identity or deduplication SHALL NOT discard lineage when multiple projections refer to the same stored record.
+
+#### Scenario: Record preserves scheduled-run provenance
+- **WHEN** a scheduled collection item is normalized and stored
+- **THEN** the persisted record references the source revision, binding revision, execution/run, worker/runtime, trace, and raw artifact associated with that item
+- **AND** a reader can distinguish the record's collection lineage from unrelated records.
+
+#### Scenario: Record preserves durable-acquisition provenance
+- **WHEN** a durable acquisition item has crossed the normalization boundary and is stored as a record
+- **THEN** the record references the acquisition execution and its artifact through the same lineage envelope shape
+- **AND** the lineage does not depend on parsing an unstructured log message.
+
+### Requirement: Enrichment and trace projections preserve lineage
+Every persisted enrichment result or enrichment attempt and every persisted collection trace event associated with a record or execution SHALL retain the originating envelope, including the relevant execution/run and worker/runtime references. Trace references SHALL identify the trace context actually used; the system SHALL NOT manufacture trace identity when none was established.
+
+#### Scenario: Enrichment is attributable to its collection
+- **WHEN** an AI or other enrichment processor handles a normalized record
+- **THEN** its result or attempt references the record's originating lineage and artifact
+- **AND** the enrichment projection does not replace collection provenance with only a model/provider reference.
+
+#### Scenario: Worker trace remains tied to runtime
+- **WHEN** a worker emits a collection trace event for queueing, execution, artifact production, or handoff
+- **THEN** the event carries the originating run/execution and worker/runtime references
+- **AND** a trace event from one execution cannot be attached to another execution by position or timestamp alone.
+
+### Requirement: Notification and delivery attempts retain originating lineage
+Each notification or delivery attempt derived from collection output SHALL retain the originating collection envelope and SHALL expose the delivery attempt identity supplied by the existing delivery/idempotency contract. Retries and mixed outcomes SHALL remain linked to the same source, binding, execution/run, worker/runtime, trace, artifact, and record lineage.
+
+#### Scenario: Notification references the collected record
+- **WHEN** a normalized record triggers a notification or external delivery
+- **THEN** the notification log and delivery attempt reference the record and its complete available collection lineage
+- **AND** the delivery projection does not become attributable only to a destination or recipient.
+
+#### Scenario: Retry keeps lineage and delivery identity
+- **WHEN** a delivery attempt is retried after a timeout or provider failure
+- **THEN** the retry retains the originating envelope and record reference
+- **AND** it uses the existing delivery-attempt/idempotency identity rules
+- **AND** it does not create a synthetic collection run or claim success merely because a retry was scheduled.
+
+### Requirement: Legacy records are backward-compatible and nullable
+Lineage fields SHALL be nullable for records, artifacts, enrichment results, trace events, and notification rows created before this contract or created by a path that did not establish a given reference. Compatibility readers SHALL return null for unavailable lineage rather than guessing, rejecting, or relabeling the historical row. New writes SHALL populate every reference that is genuinely available at the boundary that creates the row.
+
+#### Scenario: Historical record remains readable
+- **WHEN** a reader loads a record created before collection lineage was persisted
+- **THEN** the record loads successfully with unavailable lineage fields represented as null
+- **AND** the system does not infer a source revision, execution, worker, runtime, trace, or artifact from timestamps or nearby rows.
+
+#### Scenario: Partially attributable new output remains explicit
+- **WHEN** a new output has a valid run and source reference but no worker runtime or artifact reference was established
+- **THEN** the persisted projection retains the known references and stores the unavailable references as null
+- **AND** it does not block compatibility solely because optional provenance is absent.
+
+### Requirement: The envelope does not invent identifiers or capacity facts
+The lineage contract SHALL use only identifiers and provider usage facts established by the source, execution, worker, runtime, trace, artifact, record, or delivery subsystem. It SHALL NOT invent run IDs, execution IDs, worker/runtime IDs, trace IDs, artifact IDs, delivery IDs, quota, capacity, token usage, cost, or other measurements. Unknown or unavailable values SHALL remain null or explicitly unknown according to the owning subsystem's existing representation.
+
+#### Scenario: Missing identifier stays missing
+- **WHEN** an acquisition response has no provider request ID or artifact ID
+- **THEN** the lineage envelope leaves that reference null/unknown
+- **AND** it does not hash unrelated payload data or assign a locally generated surrogate as the provider identifier.
+
+#### Scenario: No quota is inferred
+- **WHEN** a provider returns collection output without documented quota or capacity usage
+- **THEN** the collection lineage projection records no quota or capacity claim
+- **AND** it does not derive one from item count, elapsed time, retry count, or configured limits.
+
+### Requirement: Acquisition raw results have an explicit pre-normalization boundary
+An acquisition raw result SHALL be treated as a lineage-bearing input/artifact envelope until a separate normalization operation produces a normalized record. Carrying collection lineage on a raw result SHALL NOT by itself make that result a record, evidence item, enriched output, deduplicated item, or notification input. Raw results that cannot be normalized SHALL remain raw with their lineage and failure/unknown status available to the acquisition boundary.
+
+#### Scenario: Raw result is not silently promoted
+- **WHEN** an AcquisitionExecution returns a raw item with a source, execution, worker, runtime, trace, and artifact envelope
+- **THEN** the system may persist that raw item and its lineage as an acquisition result
+- **AND** it SHALL NOT represent the raw item as a normalized record, evidence, enrichment result, or delivered item until normalization explicitly succeeds.
+
+#### Scenario: Normalization creates the downstream link
+- **WHEN** a later normalization operation accepts an acquisition raw item
+- **THEN** it creates or updates a normalized record while preserving the raw item's originating envelope and artifact reference
+- **AND** downstream enrichment and notification lineage begin from that normalized record rather than from an implicit raw-result shortcut.
+
+### Requirement: Lineage propagation does not weaken authorization boundaries
+Lineage propagation SHALL preserve the source, binding, account, credential, and project scope selected by the authorizing execution. A consumer SHALL NOT broaden scope, substitute a credential revision, or use lineage metadata as permission to access an artifact or record.
+
+#### Scenario: Revoked or mismatched binding is not repaired by propagation
+- **WHEN** a worker receives an envelope whose binding or credential revision is no longer authorized
+- **THEN** the worker follows the existing authorization and revocation behavior
+- **AND** it SHALL NOT replace the reference with a current revision merely to produce attributable output.
diff --git a/openspec/changes/collection-lineage-envelope/tasks.md b/openspec/changes/collection-lineage-envelope/tasks.md
new file mode 100644
index 00000000..8242bac4
--- /dev/null
+++ b/openspec/changes/collection-lineage-envelope/tasks.md
@@ -0,0 +1,24 @@
+## 1. Contract
+
+- [x] 1.1 Define the immutable collection-lineage envelope and propagation points.
+- [x] 1.2 Define backward-compatible null behavior for records predating the envelope.
+- [x] 1.3 Define notification/delivery lineage and the acquisition raw-result normalization boundary.
+- [x] 1.4 Explicitly prohibit invented quota, capacity, cost, and identifiers.
+
+## 2. Backend
+
+- [ ] 2.1 Add the canonical lineage envelope and persistence/serialization contract for source, binding, account/credential, project scope, run/execution, worker/runtime, trace, and artifact references.
+- [ ] 2.2 Thread one envelope through scheduled `CollectionTask`/`TaskRun` and durable `AcquisitionExecution` creation and worker handoff without replacing caller-owned identifiers.
+- [ ] 2.3 Add nullable lineage references to normalized records, artifact references, enrichment results, trace events, and notification/delivery attempts; preserve nulls for legacy rows.
+- [ ] 2.4 Ensure notification and delivery retries retain the originating lineage and use only IDs established by the existing idempotency contract.
+- [ ] 2.5 Mark acquisition raw results as pre-normalization inputs and prevent them from being represented as normalized records, evidence, enrichment, or delivery solely by carrying an envelope.
+- [ ] 2.6 Add migrations and compatibility serializers/readers for the selected persistence models.
+
+## 3. Verification
+
+- [ ] 3.1 Verify a scheduled run preserves source/binding/account/credential/project scope, run/execution, worker/runtime, trace, and artifact lineage through record, enrichment, and notification projections.
+- [ ] 3.2 Verify a durable acquisition execution preserves the same lineage through worker/runtime and artifact boundaries, with no invented IDs or quota values.
+- [ ] 3.3 Verify pre-envelope records deserialize and project with null lineage without failing or being relabeled as newly attributable.
+- [ ] 3.4 Verify notification mixed outcomes and retries retain originating lineage and delivery identity.
+- [ ] 3.5 Verify raw acquisition output remains explicitly pre-normalization until a later normalization contract is implemented.
+- [ ] 3.6 Run the repository's focused backend contract/integration checks and strict OpenSpec validation for this change.
diff --git a/openspec/changes/durable-deployment-compass/README.md b/openspec/changes/durable-deployment-compass/README.md
index fbaedfb8..cecc8501 100644
--- a/openspec/changes/durable-deployment-compass/README.md
+++ b/openspec/changes/durable-deployment-compass/README.md
@@ -4,3 +4,14 @@ Make the deployment compass explicit: OpenCLI Admin must be installable,
 persisted, upgraded, observed, backed up, and PTT-accepted on Docker, NAS, and
 Fleet Agent nodes before capabilities are called deployable.
 
+
+## Current Handoff / Closure
+
+- Frontend gate closed; Agent packaging closed.
+- PTT D1-D4 and D6-D7 are evidenced.
+- D0 remains partial because the OpenSpec CLI is unavailable and the Sentrux
+  baseline is missing.
+- D5 is blocked: workstation DNS resolves public RSS hosts to `198.18.0.x`,
+  and the SSRF guard correctly rejects those addresses.
+- Data-node IDE work is intentionally deferred.
+
diff --git a/openspec/changes/durable-deployment-compass/design.md b/openspec/changes/durable-deployment-compass/design.md
index f54d65e1..704012d9 100644
--- a/openspec/changes/durable-deployment-compass/design.md
+++ b/openspec/changes/durable-deployment-compass/design.md
@@ -59,10 +59,11 @@ Each runtime profile must declare:
    - Rationale: It already has compose, image, and installer surfaces. It is the
      shortest path to a real Agent and evidence persistence loop.
 
-3. Shell/systemd Agent remains blocked for runtime support until it distributes
-   runtime adapter packages, not only `agent_server.py`.
-   - Rationale: Current Python install mode can start the server but cannot
-     import the runtime registry package reliably.
+3. Shell/systemd packaging is unblocked at the package-distribution layer; real
+   host PTT remains pending.
+   - Rationale: The installer now downloads and validates the authenticated
+     runtime bundle, but a real host enrollment is still required before runtime
+     support is accepted.
 
 4. Runtime profile support is separate from source support.
    - Rationale: A source may be approved, but a node still needs the runtime
diff --git a/openspec/changes/durable-deployment-compass/tasks.md b/openspec/changes/durable-deployment-compass/tasks.md
index eeb61edc..13213991 100644
--- a/openspec/changes/durable-deployment-compass/tasks.md
+++ b/openspec/changes/durable-deployment-compass/tasks.md
@@ -3,7 +3,7 @@
 - [ ] 1.1 Link `docs/ptt-acceptance.md` to this deployment compass.
 - [ ] 1.2 Split PTT into deployment gates D0-D7 and product workflow gates P0-P6.
 - [ ] 1.3 Add deployment support states: proposed, dry-run, ptt-ready, supported, blocked, deprecated.
-- [ ] 1.4 Document the current blocker: shell/systemd Agent installs only `agent_server.py` and not runtime packages.
+- [ ] 1.4 Document the shell/systemd package-distribution status and remaining real-host PTT gap.
 
 ## 2. Deployment Profiles
 
@@ -23,7 +23,7 @@
 
 - [x] 4.1 Package `backend/agent_runtimes` and `backend/miniflow` into the Docker Agent image.
 - [x] 4.2 Add a unit test that prevents Docker Agent runtime packaging regression.
-- [ ] 4.3 Fix shell/systemd Python installer to distribute runtime adapter packages with `agent_server.py`.
+- [x] 4.3 Fix shell/systemd Python installer to distribute runtime adapter packages with `agent_server.py`. (Installer downloads and validates the authenticated runtime bundle before extraction; focused packaging/API tests pass.)
 - [ ] 4.4 Add runtime version reporting and health probes for MiniFlow, OpenTabs, OpenCLI, and future profiles.
 - [ ] 4.5 Add runtime allowlist and workflow directory allowlist for NAS/MiniFlow execution.
 - [ ] 4.6 Add Agent log tail, failure count, current task, deployed image/tag, and deploy type to inventory/status.
diff --git a/openspec/changes/gaojixing-live-business-chain/proposal.md b/openspec/changes/gaojixing-live-business-chain/proposal.md
new file mode 100644
index 00000000..836cf7f0
--- /dev/null
+++ b/openspec/changes/gaojixing-live-business-chain/proposal.md
@@ -0,0 +1,53 @@
+## Why
+
+Gaojixing is currently represented by adjacent research/channel and workflow surfaces, but the repository does not yet define one attributable business chain from a live question to an accepted business outcome. The existing Doubao channel can invoke `opencli doubao ask`, extract answer URLs, and best-effort capture a conversation URL; its focused tests use mocked adapter output. The collection audit also identifies a broader gap: raw acquisition, normalization, evidence, enrichment, and delivery are not one durable lineage contract, and delivery transport success is not the business outcome.
+
+Without a strict contract, a fixture or mock can look like a live result, a changed question can be mistaken for the configured question, a citation can be detached from the answer that produced it, and an HTTP 202/send result can be reported as business success without destination acknowledgement.
+
+## What Changes
+
+- Define Gaojixing as a real business workflow chain: publish a live-capable Doubao/chat capability, resolve an authenticated session, execute an immutable question package, capture answer/citation/conversation evidence, normalize and record it, and report transport and business outcomes separately.
+- Snapshot the exact question package at execution time and compute a stable digest over its canonical representation; downstream artifacts and events reference that snapshot rather than mutable source configuration.
+- Define evidence objects for the raw answer, extracted citations, and conversation reference, with explicit capture status and no claim that a citation was verified merely because a URL was extracted.
+- Carry one lineage identity through collect, normalize, record, evidence, and delivery projections so an operator can reconstruct the chain.
+- Require capability readiness and live-session/network gates before a run may claim live execution. Fixture and mock paths remain explicitly labeled and cannot satisfy live acceptance.
+- Separate delivery transport state (attempted/accepted/failed/unknown) from business outcome (confirmed/unconfirmed/failed/unknown), requiring a destination ACK or equivalent documented confirmation for `confirmed`.
+- Fail closed for missing capability publication, unauthenticated or unverified session, unavailable network, missing answer, malformed evidence, lineage mismatch, or absent destination confirmation; never downgrade to fixture/mock success.
+
+## Capabilities
+
+### New Capabilities
+
+- `gaojixing-live-business-chain`: A live-capable, evidence-bearing Gaojixing question workflow with immutable package identity, end-to-end lineage, explicit fixture separation, and fail-closed business acceptance.
+
+### Modified Capabilities
+
+- Existing `doubao_research` collection SHALL remain usable, but its answer/citation/conversation fields SHALL be mapped into the Gaojixing evidence and lineage contract when used by this chain.
+- Existing normalize/record/evidence and delivery projections SHALL preserve Gaojixing package, run, and artifact lineage and expose transport versus business outcome separately.
+- Existing capability catalogs/readiness projections SHALL distinguish published, configured, authenticated, network-allowed, executable, and live-accepted states; a catalog entry alone SHALL NOT imply live readiness.
+
+## Impact
+
+- Backend capability publication/readiness, Doubao/OpenCLI adapter invocation, question-package persistence, canonical digesting, evidence and lineage models, normalization/record sinks, run events, and delivery/outcome projections.
+- Authentication/session binding, network policy, and destination acknowledgement adapters.
+- Focused contract and integration verification, including explicit fixture/mock tests that cannot pass live acceptance gates.
+- No frontend or backend implementation is included in this read-only proposal.
+
+## External Prerequisites for Live Acceptance
+
+A live acceptance run is possible only when all of the following are true:
+
+1. The Gaojixing/Doubao capability is published by the runtime capability catalog with an executable adapter and an explicit live (not fixture/mock) mode.
+2. An authenticated Doubao browser/OpenCLI session is available to the worker, with the required site/session binding and a session health check that passes immediately before execution.
+3. The worker has permitted network access to Doubao and any destination/citation endpoint required by the chosen question package; no SSRF or domain-policy block is active.
+4. The configured destination is reachable and returns a documented acknowledgement tied to the delivery idempotency key (or an equivalent durable business confirmation).
+5. The acceptance operator has a concrete immutable question package and permission to run it; no missing or guessed credentials, capability, evidence, or destination result may be substituted.
+
+These prerequisites are not proven by the repository's mocked unit tests or by a local fixture run. A run lacking any prerequisite MUST remain blocked/unknown and MUST NOT be reported as live business success.
+
+## Non-Goals
+
+- This change does not claim that the current repository already has a Gaojixing capability publication, authenticated live session, public network access, or destination ACK.
+- This change does not define a new AI provider, bypass captcha or login controls, or infer capacity, quota, token usage, citation validity, or business success.
+- This change does not replace the general collection-lineage envelope; it defines the Gaojixing contract and its required use of that lineage.
+- This change does not make fixture/mock execution a fallback for live execution, nor does it turn a transport response into business confirmation.
diff --git a/openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md b/openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md
new file mode 100644
index 00000000..d8741f0c
--- /dev/null
+++ b/openspec/changes/gaojixing-live-business-chain/specs/gaojixing-live-business-chain/spec.md
@@ -0,0 +1,131 @@
+## Purpose
+
+Define Gaojixing as an attributable live business workflow rather than an unqualified channel demo: an immutable question package is executed through a published and ready Doubao/chat capability, its answer and evidence enter the shared data chain, and delivery is accepted only when the business destination confirms the outcome.
+
+## ADDED Requirements
+
+### Requirement: Live capability readiness is explicit and fail-closed
+A Gaojixing run claiming live execution SHALL require a published capability with an executable adapter, an explicitly live mode, a valid authenticated persistent-session binding, a passing session health check (using provider `whoami` when the primary identity lookup is unavailable), and permitted network access. Capability publication, configuration, authentication, session health, network permission, and executable readiness SHALL be represented as distinct states. CAPTCHA detection SHALL fail closed. A catalog entry or configured channel alone SHALL NOT imply live readiness.
+
+
+#### Scenario: Capability is published but not executable
+- **WHEN** the catalog contains Gaojixing/Doubao metadata but no executable adapter or live mode
+- **THEN** the run is blocked with a precise capability-readiness reason
+- **AND** it does not invoke a fixture/mock adapter or report live success.
+
+#### Scenario: Authenticated session or network is unavailable
+- **WHEN** the adapter is published but the required session health check or network policy check fails
+- **THEN** the run remains blocked or retryable according to the typed failure
+- **AND** no answer, evidence, record, or business-success projection is fabricated from the missing prerequisite.
+
+### Requirement: Execution uses an immutable question-package snapshot and digest
+Before dispatch, Gaojixing SHALL persist the exact question package used for the run, including the effective question and relevant execution options, as an immutable snapshot. It SHALL compute a deterministic digest from a canonical representation of that snapshot. Prompt construction, evidence, lineage, replay, and audit projections SHALL reference the same snapshot and digest; mutable source configuration SHALL NOT replace it after dispatch.
+
+#### Scenario: Runtime parameters override configured question
+- **WHEN** a run supplies a runtime question that differs from the source's configured question
+- **THEN** the snapshot contains the effective runtime question and its canonical digest
+- **AND** the adapter, answer evidence, and downstream record all reference that snapshot rather than the stale source value.
+
+#### Scenario: Source changes after dispatch
+- **WHEN** the source question is edited after a run snapshot has been persisted
+- **THEN** the in-flight or replayed run continues to reference its original snapshot and digest
+- **AND** evidence from the edited source cannot be attached to that run.
+
+### Requirement: Answer, citation, and conversation evidence are independently attributable
+A live response SHALL capture the raw assistant answer as an evidence artifact and SHALL represent citation extraction and conversation capture as separate evidence projections linked to the same package digest, run/execution, and answer artifact. Each projection SHALL expose capture status and source metadata. Extracting a URL from answer text SHALL NOT assert that the citation is externally verified, and a missing conversation reference SHALL remain unknown rather than being guessed.
+
+#### Scenario: Answer contains citations and a conversation URL
+- **WHEN** a live Doubao response returns assistant text, one or more URLs, and a conversation reference
+- **THEN** the raw answer, ordered de-duplicated citation list, and conversation evidence are persisted with the same run and package digest
+- **AND** citation capture identifies answer-text extraction separately from any future citation verification.
+
+#### Scenario: Conversation status lookup fails
+- **WHEN** the answer is captured but the conversation status lookup fails or returns no conversation id
+- **THEN** answer evidence remains attributable and citation capture may remain available
+- **AND** conversation evidence is recorded as unavailable/unknown without inventing a URL or failing closed solely because this optional reference was not established.
+
+### Requirement: Accepted live answers cross the normalize-to-record-to-evidence lineage
+A successfully captured live answer SHALL cross the shared normalize and dedupe boundary before it is represented as a normalized business record. The normalized record, raw answer artifact, citation and conversation evidence, enrichment/trace projections, and any delivery attempt SHALL retain the package snapshot/digest and originating run/execution lineage, including established source/binding and worker/runtime references. A raw response alone SHALL NOT be treated as a record or delivered business item.
+
+#### Scenario: Live answer becomes a stored record
+- **WHEN** a captured answer is accepted by normalization
+- **THEN** the resulting record links to the exact question-package digest, raw answer artifact, citation/conversation evidence, and execution lineage
+- **AND** deduplication does not discard the ability to trace the record to that run and artifact.
+
+#### Scenario: Normalization rejects the answer
+- **WHEN** the raw answer cannot satisfy the normalizer or evidence contract
+- **THEN** it remains a raw failed/unaccepted artifact with its lineage and typed reason
+- **AND** no normalized record, enrichment success, or delivery business outcome is created.
+
+### Requirement: Delivery transport and business outcome are separate
+Every Gaojixing delivery SHALL persist an idempotent delivery-attempt identity and separate transport status from business outcome. Transport status SHALL describe what the client/provider reported (for example attempted, accepted, failed, or unknown). Business outcome SHALL be confirmed only by a documented destination ACK or equivalent durable confirmation tied to that delivery attempt; transport acceptance, HTTP 202, enqueue success, or lack of an error SHALL not by itself mean business success.
+
+#### Scenario: Destination accepts transport but does not acknowledge business effect
+- **WHEN** the delivery provider accepts a request but no destination ACK or equivalent confirmation is received
+- **THEN** transport is recorded as accepted (or the provider's precise status)
+- **AND** business outcome remains unconfirmed or unknown and the run is not reported as completed business success.
+
+#### Scenario: Destination ACK arrives
+- **WHEN** a destination returns a documented acknowledgement tied to the delivery idempotency key and record/package lineage
+- **THEN** the delivery attempt records the ACK evidence and business outcome confirmed
+- **AND** retries remain linked to the same delivery identity rather than creating a second business outcome.
+
+### Requirement: Fixture and mock execution cannot satisfy live acceptance
+Fixture and mock adapters, responses, sessions, citations, conversation URLs, records, and delivery outcomes SHALL be explicitly labeled with their non-live mode and provenance. They SHALL be usable for deterministic contract verification but SHALL NOT satisfy a live capability/session/network gate or produce a live business-accepted outcome. A live path SHALL NOT silently fall back to fixture/mock behavior.
+
+#### Scenario: Fixture is used for deterministic verification
+- **WHEN** a test or preview runs Gaojixing with fixture/mock input
+- **THEN** every resulting artifact and projection is marked fixture/mock with its source provenance
+- **AND** the run is excluded from live-acceptance reporting and cannot be counted as a live business outcome.
+
+#### Scenario: Live dependency disappears
+- **WHEN** a live run loses capability readiness, session, network, or destination access
+- **THEN** it fails or remains explicitly unknown/blocked with a typed reason
+- **AND** it does not substitute fixture/mock output to reach a successful terminal state.
+
+### Requirement: Gaojixing fails closed on missing or contradictory evidence
+The chain SHALL fail closed when a required live prerequisite, answer, package digest, lineage reference, or delivery confirmation is missing or contradictory. It SHALL preserve partial evidence and typed failure/unknown state for diagnosis, but SHALL NOT infer citations, conversation identity, lineage, provider capacity, transport success, or business outcome from timestamps, nearby records, payload hashes unrelated to the package, or configured defaults.
+
+#### Scenario: Evidence belongs to another package or run
+- **WHEN** an answer, citation, conversation, record, or delivery event has a package digest or run lineage that does not match the current execution
+- **THEN** the projection is rejected or quarantined with a lineage-mismatch reason
+- **AND** the current run cannot claim live business acceptance.
+
+#### Scenario: Required answer is empty or malformed
+- **WHEN** the live adapter returns no assistant text or an unparseable response
+- **THEN** collection terminates with an actionable typed failure
+- **AND** no empty answer is normalized, recorded, delivered, or reported as a successful business result.
+
+### Requirement: Live acceptance exposes external prerequisites and honest terminal state
+The Gaojixing acceptance surface SHALL state whether capability publication, authenticated session health, network access, and destination ACK prerequisites were actually observed for the run. It SHALL distinguish completed-live, blocked, failed, partial, unconfirmed, unknown, and fixture/mock outcomes. Repository tests or local fixtures SHALL NOT be presented as proof that external live prerequisites were met.
+
+#### Scenario: All live prerequisites and destination confirmation are present
+- **WHEN** the published live capability, authenticated healthy session, permitted network, captured evidence, normalized lineage, and matching destination ACK are all observed
+- **THEN** the run may be reported as completed live business success
+- **AND** the acceptance record retains the prerequisite observations and evidence references.
+
+#### Scenario: An external prerequisite is absent
+- **WHEN** any required live prerequisite is unavailable or not observed
+- **THEN** the acceptance result identifies the exact missing prerequisite and remains blocked, failed, partial, unconfirmed, or unknown as appropriate
+- **AND** it MUST NOT be upgraded to completed live business success by a mock, fixture, transport response, or inferred state.
+
+### Requirement: Workflow v2 and project-scoped projections are explicit
+Accepted Gaojixing execution SHALL follow the Workflow v2 stages `source → normalize → accept → sink`. The service SHALL expose project-scoped APIs for run detail, run events, evidence, and projections, and every response SHALL preserve the run, project, source, package-digest, and execution lineage.
+
+#### Scenario: Project requests run evidence
+- **WHEN** an authorized project-scoped client requests run detail, events, evidence, or projections
+- **THEN** the response is limited to that project and retains complete source-to-sink lineage
+- **AND** cross-project records are not disclosed.
+
+### Requirement: Canonical dedupe and downstream replay are deterministic and idempotent
+The chain SHALL compute dedupe identity from a deterministic canonical representation and SHALL distinguish stored sink references from duplicate-skipped sink references. Downstream replay SHALL consume persisted source evidence and its lineage without invoking the provider again; repeated replay of the same canonical input SHALL be idempotent.
+
+#### Scenario: Replay an accepted source evidence artifact
+- **WHEN** a downstream replay is requested for persisted source evidence
+- **THEN** normalization, acceptance, and sink processing reuse the canonical package and evidence lineage without another provider call
+- **AND** the result records whether the sink reference was stored or skipped as a duplicate.
+
+#### Scenario: Equivalent canonical inputs are replayed twice
+- **WHEN** equivalent inputs produce the same canonical dedupe identity
+- **THEN** the first sink reference is stored and the later reference is marked duplicate-skipped
+- **AND** both outcomes retain complete links to the source evidence, run, project, package digest, and replay execution.
diff --git a/openspec/changes/gaojixing-live-business-chain/tasks.md b/openspec/changes/gaojixing-live-business-chain/tasks.md
new file mode 100644
index 00000000..62e42a7f
--- /dev/null
+++ b/openspec/changes/gaojixing-live-business-chain/tasks.md
@@ -0,0 +1,39 @@
+## 1. Contract
+
+- [x] 1.1 Define the Gaojixing live business chain and its immutable question-package snapshot/digest.
+- [x] 1.2 Define answer, citation, and conversation evidence capture states and provenance.
+- [x] 1.3 Define normalize → record → evidence lineage and separate delivery transport from business outcome.
+- [x] 1.4 Define fixture/mock separation, live prerequisites, and fail-closed blockers.
+
+## 2. Capability and execution
+
+- [x] 2.1 Publish Gaojixing/Doubao as a capability only when its executable adapter and declared live mode are available; expose publication, configuration, authentication, network, and executable readiness separately.
+- [x] 2.2 Resolve and health-check the authenticated Doubao/OpenCLI session immediately before execution; use provider `whoami` as the identity fallback, bind the session identity and scope to the run without persisting secrets, and fail closed on CAPTCHA.
+- [x] 2.3 Persist the canonical question package snapshot and deterministic digest before dispatch, and use that snapshot for the prompt, evidence, lineage, replay, and audit views.
+- [x] 2.4 Capture the raw answer, extracted citations, conversation reference, adapter/request metadata, and capture status as separate evidence objects; preserve unknown or unavailable fields as unknown/null.
+- [x] 2.5 Reject empty/malformed answers, missing required evidence, capability/session/network blockers, or question-package digest mismatches without creating a live-success result.
+
+## 3. Normalize, record, and delivery
+
+- [x] 3.1 Route the accepted live answer through the Workflow v2 `source → normalize → accept → sink` pipeline and shared canonical dedupe boundary, retaining package digest, run/execution, worker/runtime, source/binding, and raw-artifact references.
+- [x] 3.2 Persist a normalized record and its evidence links so every record can be traced back to the exact question, answer artifact, citation capture, conversation reference, and execution lineage; expose project-scoped run detail, events, evidence, and projection APIs.
+- [x] 3.3 Add delivery attempt state that records transport outcome independently from business outcome, with deterministic canonical dedupe identity and retry lineage.
+- [x] 3.4 Mark business outcome `confirmed` only from a documented destination ACK/equivalent confirmation tied to the delivery attempt; otherwise expose `unconfirmed` or `unknown` even when transport is accepted.
+- [x] 3.5 Ensure delivery failure, timeout, mixed outcomes, and missing ACK are visible as partial/blocked states and never silently promoted to completed business success.
+- [x] 3.6 Support idempotent downstream replay from persisted source evidence without another provider call; retain sink references for stored and duplicate-skipped outcomes with complete lineage.
+
+## 4. Fixture and mock boundaries
+
+- [ ] 4.1 Label every fixture/mock run and artifact with its mode and provenance; prevent fixture/mock evidence from satisfying live acceptance.
+- [ ] 4.2 Remove or reject implicit fixture/mock fallback when a live capability, session, network, answer, or destination ACK is unavailable.
+- [ ] 4.3 Provide explicit fixture-only contract coverage for deterministic package digests, evidence shape, lineage propagation, and transport/business outcome separation without calling live services.
+
+## 5. Verification and live acceptance
+
+- [ ] 5.1 Verify capability publication/readiness reports precise blockers for missing adapter, authentication, session health, or network permission.
+- [ ] 5.2 Verify a live run stores one immutable question snapshot/digest and carries it through answer, citation, conversation, normalized record, evidence, and delivery projections.
+- [ ] 5.3 Verify a changed source question after dispatch cannot alter the run's snapshot or digest and cannot mix evidence from another run.
+- [ ] 5.4 Verify fixture/mock output is visibly non-live and cannot produce live business acceptance.
+- [ ] 5.5 Verify transport accepted without destination ACK remains unconfirmed/unknown, while a matching destination ACK produces confirmed business outcome.
+- [ ] 5.6 Verify timeout, captcha, unauthenticated session, network denial, empty answer, malformed citation/conversation capture, lineage mismatch, and destination failure all fail closed with actionable typed state.
+- [ ] 5.7 Run focused contract/integration checks and a live acceptance only after all external prerequisites in proposal.md are supplied; no live claim may be based on mocks or fixtures.
diff --git a/openspec/changes/local-codex-agent-runtime/proposal.md b/openspec/changes/local-codex-agent-runtime/proposal.md
new file mode 100644
index 00000000..89510691
--- /dev/null
+++ b/openspec/changes/local-codex-agent-runtime/proposal.md
@@ -0,0 +1,52 @@
+## Why
+
+The local workstation has a usable Codex CLI, but the Operations Agent UI currently treats `codex` as an automation label rather than an executable runtime. Selecting Codex can persist an Automation record, yet no Fleet runtime capability, local runner, event adapter, or terminal result contract proves that the task actually ran.
+
+The same gap affects the three Agent Starters introduced for the control plane:
+
+- 运行简报 Agent
+- 系统回顾 Agent
+- 异常跟进 Agent
+
+The product must also be honest about provider capacity. Local Claude Code and Codex CLIs do not provide a stable, portable five-hour quota API that the control plane can infer. The UI must report `unknown` when usage cannot be read rather than inventing remaining capacity or cutting a task short based on elapsed time.
+
+## What Changes
+
+- Add a governed `runtime.codex` local Agent runner and Fleet capability contract.
+- Execute Codex on a registered user workstation through the existing Agent transport instead of spawning a user binary from the web/API process.
+- Translate Codex process output into the existing Agent event set: started, text, tool_call, tool_result, state, done, and error.
+- Add health/readiness evidence for the Codex binary and its working directory before dispatch.
+- Make the three Agent Starters idempotent, runnable, and traceable through the existing Automation and Operations Agent APIs.
+- Reuse the existing SmoothUI Agent Avatar and Switchboard Card components for starter discovery and runtime status; do not add a parallel visual component system.
+- Raise the governed default execution timeout to the deep-run profile while keeping a hard upper bound and explicit approval mode.
+- Add a provider-capacity projection with three states: measured, unavailable, and not applicable. Never infer provider quota from wall-clock duration.
+
+## Capabilities
+
+### New Capabilities
+
+- `codex-local-runtime`: Governed Codex CLI execution on a registered local Agent node with streamed events and terminal evidence.
+- `agent-starter-installation`: Idempotent installation and traceable execution of the three first-party Agent Starters.
+- `provider-capacity-visibility`: Honest provider usage state with optional provider-specific usage adapters and an explicit unknown state.
+
+### Modified Capabilities
+
+- `operations-agent-runtime`: Add `runtime.codex` as an optional runtime binding; preserve existing MiniFlow and Pi contracts.
+- `agent-execution-depth`: Use the deep-run timeout profile without treating it as a quota bypass.
+- `operations-agent-ui`: Use existing SmoothUI primitives for Agent starter cards and runtime state.
+
+## Impact
+
+- Backend Agent runtime registry, local Agent runner, Fleet capability projection, and event normalization.
+- Operations Agent runtime binding validation and dispatch timeout defaults.
+- Automation starter installation and duplicate detection.
+- Provider usage/capacity projection APIs and UI.
+- SmoothUI-backed Operations Agent page and provider page.
+- Focused runtime, API, and frontend interaction tests.
+
+## Non-Goals
+
+- Do not claim to read Claude Code, Codex, or any provider's five-hour quota when no supported usage endpoint exists.
+- Do not bypass local Agent registration, workspace permissions, approval gates, or project path confinement.
+- Do not spawn arbitrary user commands from the API server merely because a browser machine has a CLI installed.
+- Do not introduce a second run/event persistence model outside Operations Agent Run and the existing Fleet transport.
diff --git a/openspec/changes/local-codex-agent-runtime/specs/local-codex-agent-runtime/spec.md b/openspec/changes/local-codex-agent-runtime/specs/local-codex-agent-runtime/spec.md
new file mode 100644
index 00000000..700d213a
--- /dev/null
+++ b/openspec/changes/local-codex-agent-runtime/specs/local-codex-agent-runtime/spec.md
@@ -0,0 +1,107 @@
+# Local Codex Agent Runtime
+
+## ADDED Requirements
+
+### Requirement: Codex runtime readiness is explicit
+
+The system MUST expose a readiness result for a registered local Agent before dispatching a Codex run.
+
+The readiness result MUST include:
+
+- whether the `codex` binary is present;
+- the detected Codex version when available;
+- the permitted project root and working directory;
+- the runtime capability identifier `runtime.codex`;
+- a typed failure reason when the binary, path, or Agent connection is unavailable.
+
+The readiness result MUST NOT include API keys, session tokens, or raw provider credentials.
+
+#### Scenario: Local Codex is installed
+
+- **WHEN** a registered Agent reports a valid Codex binary and permitted project path
+- **THEN** the Fleet capability inventory MUST mark `runtime.codex` as ready
+- **AND** the Operations Agent UI MUST allow a run to be submitted
+
+#### Scenario: Local Codex is missing
+
+- **WHEN** the Agent cannot resolve the configured Codex binary
+- **THEN** readiness MUST be `blocked`
+- **AND** the UI MUST show the missing-binary reason
+- **AND** the API MUST NOT enqueue a run that cannot start
+
+### Requirement: Codex execution uses the existing Agent transport
+
+The control plane MUST dispatch Codex through an authenticated registered Agent connection.
+
+The API server MUST NOT invoke a user workstation's Codex binary directly from a browser request or by assuming that a local CLI exists on the API host.
+
+#### Scenario: Codex run dispatch
+
+- **WHEN** a permitted Operations Agent run targets `runtime.codex`
+- **THEN** the registered Agent MUST receive the task through the existing transport
+- **AND** the control plane MUST persist the run as `running`
+- **AND** streamed runtime events MUST be normalized into the existing event envelope
+
+### Requirement: Codex terminal evidence is durable
+
+A Codex run MUST finish with exactly one terminal `done` or `error` event.
+
+The run MUST persist:
+
+- normalized text/tool/state events as available;
+- the terminal result or typed error;
+- process exit status when available;
+- timeout or cancellation reason when applicable;
+- the runtime identity and Codex version used for the run.
+
+#### Scenario: Codex exits successfully
+
+- **WHEN** the Codex process exits with a successful terminal response
+- **THEN** the Operations Agent Run MUST become `completed`
+- **AND** its output MUST be available to the activity view
+
+#### Scenario: Codex exceeds the deep-run timeout
+
+- **WHEN** the Codex process exceeds the configured deep-run timeout
+- **THEN** the Agent MUST cancel the process
+- **AND** the run MUST become `failed` with a typed timeout reason
+- **AND** the system MUST NOT claim that the provider quota was exhausted
+
+### Requirement: Provider capacity is honest
+
+The control plane MUST distinguish between measured usage and unavailable usage.
+
+A provider capacity projection MUST use one of:
+
+- `measured`: a documented provider usage endpoint returned valid data;
+- `unavailable`: the provider or local CLI does not expose a supported usage endpoint;
+- `not_applicable`: the runtime does not have provider quota semantics.
+
+#### Scenario: Local Claude Code or Codex has no quota API
+
+- **WHEN** the local CLI cannot report its five-hour usage window
+- **THEN** the UI MUST show `unavailable`
+- **AND** it MUST NOT render a guessed remaining percentage
+- **AND** task execution MUST continue according to runtime completion, approval, and timeout policy
+
+### Requirement: Starter installation is idempotent
+
+The platform MUST provide stable first-party starter keys for:
+
+- `daily-run-brief`
+- `weekly-system-review`
+- `anomaly-follow-up`
+
+Installing starters MUST skip an existing starter with the same workspace and stable key.
+
+#### Scenario: First installation
+
+- **WHEN** an administrator installs the starter pack in a Workspace
+- **THEN** all missing starters MUST be created through the existing Automation API
+- **AND** each starter MUST retain its executor, schedule, approval mode, and lineage
+
+#### Scenario: Repeated installation
+
+- **WHEN** the same starter pack is installed again
+- **THEN** existing starters MUST NOT be duplicated
+- **AND** the response MUST report created and skipped counts
diff --git a/openspec/changes/local-codex-agent-runtime/tasks.md b/openspec/changes/local-codex-agent-runtime/tasks.md
new file mode 100644
index 00000000..1b53b6e6
--- /dev/null
+++ b/openspec/changes/local-codex-agent-runtime/tasks.md
@@ -0,0 +1,41 @@
+## 1. Contract
+
+- [x] 1.1 Define the `runtime.codex` capability, transport boundary, supported event mapping, and terminal result contract.
+- [x] 1.2 Define the deep-run execution profile, timeout ceiling, approval behavior, project-path confinement, and cancellation semantics.
+- [x] 1.3 Define provider-capacity states (`measured`, `unavailable`, `not_applicable`) and prohibit inferred quota values.
+- [x] 1.4 Define the three Agent Starter templates, stable keys, schedules, executor bindings, and idempotent installation rules.
+
+## 2. Local Codex Runtime
+
+- [x] 2.1 Add a local Agent Codex runner that validates `codex --version`, working directory, branch, and configured binary before dispatch.
+- [x] 2.2 Register `runtime.codex` in the Agent capability registry and expose readiness diagnostics without leaking credentials.
+- [x] 2.3 Stream Codex output into the existing Agent event envelope and persist state/output/error through Operations Agent Run.
+- [x] 2.4 Enforce workspace path confinement, approval mode, timeout, cancellation, and non-zero exit handling.
+- [x] 2.5 Add focused adapter tests for healthy, missing-binary, invalid-path, timeout, cancellation, tool-event, and terminal-result cases.
+
+## 3. Agent Starters
+
+- [x] 3.1 Add idempotent create/install behavior for 运行简报 Agent, 系统回顾 Agent, and 异常跟进 Agent.
+- [x] 3.2 Bind each starter to its intended executor/runtime and expose the selected binding in the UI.
+- [ ] 3.3 Persist starter-created automation/run lineage and show recent status in the Operations Agent activity view.
+- [ ] 3.4 Add API tests for first install, repeated install, partial install failure, and starter run dispatch.
+
+## 4. Capacity Visibility
+
+- [ ] 4.1 Add provider usage adapter interfaces without assuming a shared quota schema.
+- [ ] 4.2 Implement measured usage only for providers with a documented usage endpoint.
+- [x] 4.3 Return explicit unknown state for local Claude Code/Codex when the CLI cannot report quota.
+- [ ] 4.4 Add UI states and tests proving unknown capacity never renders as zero, 100%, or a guessed remaining amount.
+
+## 5. UI Integration
+
+- [x] 5.1 Keep SmoothUI Switchboard Card and Agent Avatar as the starter and runtime visual primitives.
+- [ ] 5.2 Add deep-run state, runtime readiness, capacity state, and approval state to the existing Agent surfaces.
+- [ ] 5.3 Add responsive browser smoke coverage for starter installation, runtime status, and unknown capacity.
+
+## 6. Verification
+
+- [x] 6.1 Run focused backend runtime and Operations Agent tests.
+- [ ] 6.2 Run frontend lint, typecheck, build, and targeted interaction checks.
+- [x] 6.3 Run `openspec validate local-codex-agent-runtime --strict`.
+- [x] 6.4 Run Code Intel/Sentrux session gates and record provider-capacity limitations as explicit evidence.
diff --git a/openspec/changes/real-node-io-webhook-runtime/tasks.md b/openspec/changes/real-node-io-webhook-runtime/tasks.md
index b46ce8a7..46efcffb 100644
--- a/openspec/changes/real-node-io-webhook-runtime/tasks.md
+++ b/openspec/changes/real-node-io-webhook-runtime/tasks.md
@@ -26,7 +26,7 @@
 
 - [x] 4.1 Add backend integration tests for compile, resource resolution, blocked states, webhook ingress, run events, and EvidenceBatch projection.
 - [x] 4.2 Add frontend tests or fixture assertions for catalog projection, inspector binding, mini/full node views, runtime patches, result, and trace.
-- [ ] 4.3 Run `npm run typecheck:frontend` and `npm run lint:frontend` (targeted workflow assertions pass, but the canonical worktree still has unresolved cross-branch TypeScript imports; repository-wide lint is also blocked by generated `frontend/dist` debt).
+- [x] 4.3 Run `npm run typecheck:frontend` and `npm run lint:frontend` (both canonical root commands pass; `frontend/dist` is absent and excluded by `frontend/eslint.config.mjs`; prior cross-branch import blocker is not reproducible in the current worktree).
 - [x] 4.4 Run targeted pytest suites for workflow compile, OpenCLI HDA trace, run events, webhook, and EvidenceBatch APIs.
 - [x] 4.5 Run `openspec validate real-node-io-webhook-runtime --strict`.
 - [x] 4.6 Run Code Intel Pipeline normal mode after implementation and record Sentrux gate/check status (orchestration and doctor pass; Sentrux Pro/rules are available, but the baseline is missing and the pipeline stage hangs without output, so the owned process was stopped after bounded waiting).
diff --git a/openspec/changes/workflow-editor-design-pipeline-v090/brief.md b/openspec/changes/workflow-editor-design-pipeline-v090/brief.md
new file mode 100644
index 00000000..b1e044c5
--- /dev/null
+++ b/openspec/changes/workflow-editor-design-pipeline-v090/brief.md
@@ -0,0 +1,24 @@
+# Brief: workflow-editor-design-pipeline-v090
+
+## Problem
+
+Design the OpenCLI workflow editor for operators building demand-driven, Yjs-collaborative, evidence-grounded automations; preserve the Dark Ops Console, make node authoring immediate, surface collaboration, lifecycle, lineage, recovery, and grounded-state truth without generic AI styling.
+
+## Framework
+
+nextjs
+
+## Reference Sites
+
+- None provided.
+
+## Template Evidence
+
+- None provided.
+
+## Acceptance
+
+- Resolve material product ambiguity through grill-with-docs.
+- Assess scope before synthesis.
+- Generate a project-specific DESIGN.md with explicit source decisions.
+- Continue through implementation and normal design-pipeline QA.
diff --git a/openspec/changes/workflow-editor-design-pipeline-v090/component-capabilities.json b/openspec/changes/workflow-editor-design-pipeline-v090/component-capabilities.json
new file mode 100644
index 00000000..b546150b
--- /dev/null
+++ b/openspec/changes/workflow-editor-design-pipeline-v090/component-capabilities.json
@@ -0,0 +1,7 @@
+{
+  "capabilityResults": {},
+  "directQuery": "工作流编辑器：节点选择器、类型化端口、Yjs 协同状态、保存确认、生命周期、需求标签、grounded Evidence、离线恢复、键盘操作",
+  "directQueryResults": 0,
+  "searchedCapabilities": [],
+  "zeroResultInconclusive": false
+}
diff --git a/openspec/changes/workflow-editor-design-pipeline-v090/component-inventory.json b/openspec/changes/workflow-editor-design-pipeline-v090/component-inventory.json
new file mode 100644
index 00000000..e1d819c6
--- /dev/null
+++ b/openspec/changes/workflow-editor-design-pipeline-v090/component-inventory.json
@@ -0,0 +1,24 @@
+{
+  "blockers": [],
+  "brief": "工作流编辑器支持节点选择、类型化端口连接、Yjs 协同状态、持久化确认、验证发布、需求标签、grounded Evidence、离线恢复、键盘操作和减弱动效",
+  "capabilities": [
+    "aria.control-semantics",
+    "focus.management",
+    "form.validation",
+    "keyboard.navigation"
+  ],
+  "components": [
+    "custom-component"
+  ],
+  "registryHash": "71fe3a1b807de5359554c0b46fe8dff4848c98d1d2ec123c5647b996f3d15017",
+  "requiredChecks": [
+    "aria-semantics",
+    "focus-order",
+    "focus-restoration",
+    "invalid-state",
+    "keyboard-navigation",
+    "validation-message"
+  ],
+  "schema": "design-pipeline.component-capability.v1",
+  "status": "ready"
+}
diff --git a/openspec/changes/workflow-editor-design-pipeline-v090/decisions/grill.md b/openspec/changes/workflow-editor-design-pipeline-v090/decisions/grill.md
new file mode 100644
index 00000000..f595ca93
--- /dev/null
+++ b/openspec/changes/workflow-editor-design-pipeline-v090/decisions/grill.md
@@ -0,0 +1,61 @@
+# Workflow Editor Design Decisions
+
+## Product boundary
+
+The target is the existing OpenCLI Studio workflow editor, not a new editor shell. It serves business operators, workflow engineers, and reviewers who build demand-driven automations and inspect real runtime evidence.
+
+## Authority
+
+- The original demand is the authority for generated labels, search terms, and node proposals.
+- React Flow remains the canvas and interaction engine.
+- Yjs/y-websocket remains the incremental collaboration engine.
+- Studio Workflow Draft and immutable Workflow Version remain the durable product authority.
+- Evidence, Lineage, grounded status, Workspace ownership, and audit remain first-party OpenCLI responsibilities.
+
+## Operating pressure
+
+Operators work in dense, long-running desktop sessions. They need immediate node authoring, truthful capability state, explicit recovery, and stable project context. Network or collaboration failure must not lose edits.
+
+## Primary workflows
+
+1. Enter or refine a business demand.
+2. Review demand-derived labels and proposed nodes.
+3. Add, connect, configure, or remove nodes with immediate local feedback.
+4. Observe collaboration state and durable save acknowledgement separately.
+5. Validate the current revision, publish an immutable version, run it, and inspect Trace, Evidence, Lineage, and grounded gates.
+6. Recover from disconnected collaboration, capability gaps, validation errors, and failed runs without hidden fallback behavior.
+
+## Visual direction
+
+Refine the existing Dark Ops Console rather than redesign it. Preserve compact density, zinc surfaces, one primary blue, role-based signal colors, 4px spacing rhythm, restrained radii, Noto Sans SC for UI text, and IBM Plex Mono for identifiers, ports, metrics, and trace data.
+
+Reject decorative gradients, glass-heavy cards, oversized SaaS marketing layouts, anonymous AI sparkle styling, and continuous motion for static state.
+
+## Collaboration behavior
+
+- A local node mutation appears immediately.
+- LIVE, SYNCING, OFFLINE, and TOKEN REQUIRED are textual states, not color-only indicators.
+- Durable snapshot acknowledgement is visible independently from local rendering.
+- Offline mode keeps the current full-Draft PUT fallback.
+- Validate and publish force a current durable snapshot.
+
+## Grounding behavior
+
+- Model-discovered entities stay unverified until a concrete readable URL supports the exact product identity.
+- Grounded and ungrounded states show source count, missing evidence, and recovery action.
+- Project cards summarize demand labels and grounding counts; node configuration retains the original demand and generation provenance.
+
+## Accessibility and motion
+
+Target WCAG 2.2 AA. Preserve keyboard node selection, palette search, visible focus, readable status text, reduced-motion substitutions, and no motion-dependent state communication. Motion explains spatial entry, selection, connection, and asynchronous acknowledgement; it never decorates idle state.
+
+## Performance budget
+
+- Local node appearance: under 100 ms implementation target; current browser acceptance remains below 350 ms including palette interaction.
+- Durable collaboration acknowledgement: under 1.5 s.
+- Offline fallback acknowledgement: under 1.5 s.
+- No full graph PUT for each connected edit.
+
+## Open questions
+
+No material product decision remains for the foundation. Exact visual changes still require direction comparison and bounded desktop/mobile browser evidence before implementation.
diff --git a/openspec/changes/workflow-editor-design-pipeline-v090/design-synthesis.json b/openspec/changes/workflow-editor-design-pipeline-v090/design-synthesis.json
new file mode 100644
index 00000000..a6f4f685
--- /dev/null
+++ b/openspec/changes/workflow-editor-design-pipeline-v090/design-synthesis.json
@@ -0,0 +1,59 @@
+{
+  "schema": "design-pipeline.design-synthesis.v1",
+  "changeId": "workflow-editor-design-pipeline-v090",
+  "status": "ready-to-implement",
+  "stage": "implementation",
+  "initializedAt": "2026-08-26T15:40:42.444Z",
+  "updatedAt": "2026-08-26T15:47:41.761Z",
+  "artifactRoot": "openspec/changes/workflow-editor-design-pipeline-v090",
+  "projectRoot": "../../..",
+  "problem": "Design the OpenCLI workflow editor for operators building demand-driven, Yjs-collaborative, evidence-grounded automations; preserve the Dark Ops Console, make node authoring immediate, surface collaboration, lifecycle, lineage, recovery, and grounded-state truth without generic AI styling.",
+  "inputs": {
+    "mode": "requirements-only",
+    "framework": "nextjs",
+    "references": [],
+    "templates": []
+  },
+  "interaction": {
+    "mode": "human-in-loop",
+    "nextCommand": "Implement from DESIGN.md, change design.md, and tasks.md; then run normal design-pipeline QA.",
+    "message": "All synthesis gates passed. Continue without another interaction pause."
+  },
+  "scope": {
+    "budget": "program",
+    "threshold": 60,
+    "status": "fit",
+    "score": 37,
+    "surprise": false,
+    "counts": {
+      "surfaces": 3,
+      "workflows": 4,
+      "integrations": 3,
+      "unknowns": 2,
+      "decisions": 8
+    }
+  },
+  "wayfinder": {
+    "required": false,
+    "status": "not-required",
+    "mapUrl": null
+  },
+  "output": {
+    "path": "DESIGN.md",
+    "status": "validated",
+    "sha256": "150086d05ca2940d2e741501de2f3d86c99454415d846abf832c0a752d8ffbd6",
+    "validatedAt": "2026-08-26T15:47:27.918Z"
+  },
+  "evidence": [
+    {
+      "type": "grill",
+      "path": "openspec/changes/workflow-editor-design-pipeline-v090/decisions/grill.md",
+      "recordedAt": "2026-08-26T15:41:08.154Z"
+    },
+    {
+      "type": "design-validation",
+      "path": "DESIGN.md",
+      "recordedAt": "2026-08-26T15:47:27.918Z"
+    }
+  ]
+}
diff --git a/openspec/changes/workflow-editor-design-pipeline-v090/design.md b/openspec/changes/workflow-editor-design-pipeline-v090/design.md
new file mode 100644
index 00000000..0ecb82b6
--- /dev/null
+++ b/openspec/changes/workflow-editor-design-pipeline-v090/design.md
@@ -0,0 +1,65 @@
+# Change Design
+
+## Requirements Map
+
+| Requirement | Design consequence |
+| --- | --- |
+| Immediate node authoring | Render local mutation first; never wait for persistence before showing the node. |
+| Yjs collaboration | Show LIVE, SYNCING, OFFLINE, and TOKEN REQUIRED in compact text; keep presence secondary. |
+| Durable authority | Show local rendering and acknowledged revision as separate facts. |
+| Demand-driven labels | Keep original demand, generated labels, and generation provenance visible in project and node context. |
+| Grounded evidence | Show grounded count, readable source URLs, missing evidence, and recovery action; never promote model discovery alone. |
+| Failure recovery | Preserve disconnected full-Draft PUT, visible retry, revision conflict, and validation blockers. |
+| Dense workflow engineering | Preserve compact cards, typed ports, one Dock, keyboard paths, and low-noise canvas surfaces. |
+
+## Evidence Adoption Matrix
+
+| Source | Observed property | Decision | Reason | Target |
+| --- | --- | --- | --- | --- |
+| Existing `DESIGN.md` and Dark Ops Console | Compact operational density and truth-first states | Adopt | Established product identity | Project foundation and Studio shell |
+| Existing React Flow implementation | Canvas, selection, typed handles, viewport | Adopt | Mature installed execution layer | Workflow canvas |
+| Existing Yjs/y-websocket implementation | Incremental graph collaboration | Adopt | Mature installed execution layer | Draft authoring |
+| Design Pipeline v0.9.0 | Foundation, motion, component-first, evidence gates | Adopt | Prevents design-by-guessing | This change lifecycle |
+| Public Yjs demo server | Unscoped external room | Reject | Violates project ownership and privacy | None |
+| Generic AI dashboard styling | Gradients, glass, oversized cards, sparkle treatments | Reject | Conflicts with operational trust and density | None |
+| Evidence-first permanent split view | Persistent large review pane | Reject as default | Reduces authoring space | Keep in Evidence workspace |
+
+## Product Design Output
+
+Project foundation: `DESIGN.md` (`sha256:150086d05ca2940d2e741501de2f3d86c99454415d846abf832c0a752d8ffbd6`).
+
+Selected direction: Calm Live Operations.
+
+Implementation targets:
+
+- Collaboration badge reports protocol state and scoped room without dominating the canvas.
+- Lifecycle strip reports durable revision independently from LIVE state.
+- Node picker and local canvas update remain immediate.
+- Workflow Dock owns outline, configuration, last run, and Trace.
+- Project surfaces show demand labels and grounded counts before entering the graph.
+- Evidence and data workbenches receive deeper audit content rather than duplicating it in every node.
+
+## Component Strategy
+
+Reuse project-owned Button, Badge, Tooltip, Tabs, Select, Dialog, Sheet, WorkflowNode, CommandPalette, Workflow Dock, lifecycle strip, status badges, and data-state components. React Flow, Zustand, Yjs, and y-websocket remain runtime providers. No new UI framework or collaboration engine is authorized by this change.
+
+## Accessibility
+
+- WCAG 2.2 AA target.
+- Status always includes text or an icon.
+- Keyboard add/select/delete/undo, palette search, Dock navigation, and visible focus remain required.
+- Reduced motion uses static state and immediate positioning.
+- Long Chinese, mixed-script, and identifier strings must expose their complete accessible names.
+
+## Performance
+
+- Local node appearance remains below 350ms in browser acceptance, with a sub-100ms implementation target once palette interaction is excluded.
+- Connected persistence is asynchronous and does not issue per-edit full graph PUT.
+- Durable acknowledgement target is below 1.5s.
+- Offline fallback remains below 1.5s.
+
+## Spec Reconciliation
+
+- Graybox evidence: not required for this bounded refinement; the existing rendered workflow editor is the implementation authority.
+- Reference: browser-verified `/studio/workflow` surface and the project foundations.
+- Reconciliation: preserve canvas topology, project navigation, canonical node contracts, and status semantics. Change only the collaboration/persistence behavior and the clarity of its state representation.
diff --git a/openspec/changes/workflow-editor-design-pipeline-v090/directions.md b/openspec/changes/workflow-editor-design-pipeline-v090/directions.md
new file mode 100644
index 00000000..4c83d5e8
--- /dev/null
+++ b/openspec/changes/workflow-editor-design-pipeline-v090/directions.md
@@ -0,0 +1,34 @@
+# Directions
+
+## Direction A — Calm Live Operations
+
+Selected.
+
+- Keep the existing Dark Ops Console and compact canvas.
+- Make local authoring, collaboration, durable save, validation, publication, execution, and evidence separate visible states.
+- Use one right-side Workflow Dock rather than more floating surfaces.
+- Let demand provenance and grounded evidence appear where operators make decisions.
+- Use bounded semantic motion only for direct manipulation, spatial navigation, connection creation, and acknowledgement.
+
+Fit: highest. It preserves current product identity and repairs the exact trust and latency problems reported by operators.
+
+Risk: dense status language can become noisy. Mitigation: progressive disclosure and one primary status per region.
+
+## Direction B — Evidence-First Review Desk
+
+- Give Evidence and Lineage equal visual weight to the canvas.
+- Use a persistent split view for source URLs, grounded claims, and acceptance gates.
+- Optimize for reviewers rather than authors.
+
+Fit: strong for audit sessions, weak for everyday authoring. Rejected as the default direction because it reduces canvas space and makes simple edits feel like review work. Retain as the project Evidence workspace, not the editor shell.
+
+## Direction C — Ambient Collaborative Canvas
+
+- Emphasize presence, live cursors, animated edge flow, and multiplayer identity.
+- Treat collaboration as the primary visual theme.
+
+Fit: low. Rejected because it overstates collaboration, adds motion noise, and risks turning operational truth into decorative activity. Presence remains a compact status layer inside Direction A.
+
+## Decision
+
+Adopt Direction A. Use Direction B only when the operator explicitly enters Evidence or Trace. Reject Direction C's ambient motion and collaboration-first visual identity.
diff --git a/openspec/changes/workflow-editor-design-pipeline-v090/events.jsonl b/openspec/changes/workflow-editor-design-pipeline-v090/events.jsonl
new file mode 100644
index 00000000..6981ac64
--- /dev/null
+++ b/openspec/changes/workflow-editor-design-pipeline-v090/events.jsonl
@@ -0,0 +1,5 @@
+{"evidence":["legacy-events:5","legacy-events-sha256:8b58ec40b73e5c04546d9be0f9f4411db0b7d3a6e601728cd4f7539b4a0f4084"],"files":["state.json","events.jsonl"],"id":"dpe-3c402c28ddb08786a8d4","nextActions":["Assess the declared surfaces, workflows, integrations, unknowns, and decisions.","Implement from DESIGN.md, change design.md, and tasks.md; then run normal design-pipeline QA."],"phase":"implementation","schema":"design-pipeline.event.v2","seq":1,"summary":"Migrated legacy event history to the v2 event contract","ts":"2026-08-26T15:52:13.104Z","type":"state-repair"}
+{"evidence":["openspec/changes/workflow-editor-design-pipeline-v090/qa.md"],"files":["DESIGN.md","MOTION.md","openspec/changes/workflow-editor-design-pipeline-v090/design.md","openspec/changes/workflow-editor-design-pipeline-v090/motion.md"],"id":"dpe-f3657e818ef99ece58e3","nextActions":["Review runtime, accessibility, performance, and provenance evidence."],"phase":"gate-review","schema":"design-pipeline.event.v2","seq":2,"summary":"Implemented and documented the workflow editor collaboration and design foundations; entering evidence review.","ts":"2026-08-26T15:53:10.063Z","type":"state-advance"}
+{"evidence":["DESIGN.md","MOTION.md","openspec/changes/workflow-editor-design-pipeline-v090/qa.md"],"files":["openspec/changes/workflow-editor-design-pipeline-v090/qa.md"],"id":"dpe-6f78517a6becab449b50","nextActions":["Confirm release readiness and archive the design change."],"phase":"verification","schema":"design-pipeline.event.v2","seq":3,"summary":"Gate review passed: foundations, browser behavior, collaboration ownership, fallback, build, tests, and cleanup evidence are complete.","ts":"2026-08-26T15:53:27.414Z","type":"state-advance"}
+{"evidence":["openspec/changes/workflow-editor-design-pipeline-v090/qa.md"],"files":[],"id":"dpe-bc0d962dab1f1e67b751","nextActions":["Archive the completed change while retaining DESIGN.md and MOTION.md as project foundations."],"phase":"release-readiness","schema":"design-pipeline.event.v2","seq":4,"summary":"Verification is complete and the workflow editor design contract is ready to become the reusable project authority.","ts":"2026-08-26T15:53:45.523Z","type":"state-advance"}
+{"evidence":["openspec/changes/workflow-editor-design-pipeline-v090/toolchain-plan.json"],"files":["DESIGN.md","MOTION.md","openspec/changes/workflow-editor-design-pipeline-v090/qa.md"],"id":"dpe-e274bc6ee080764f5e26","nextActions":["Use the validated foundations for future workflow editor design changes."],"phase":"archive","schema":"design-pipeline.event.v2","seq":5,"summary":"Archived the completed Design Pipeline v0.9.0 workflow editor change; project DESIGN.md and MOTION.md remain active foundations.","ts":"2026-08-26T15:54:15.485Z","type":"state-advance"}
diff --git a/openspec/changes/workflow-editor-design-pipeline-v090/handoff.md b/openspec/changes/workflow-editor-design-pipeline-v090/handoff.md
new file mode 100644
index 00000000..e05dc418
--- /dev/null
+++ b/openspec/changes/workflow-editor-design-pipeline-v090/handoff.md
@@ -0,0 +1,18 @@
+# Handoff
+
+<!-- DESIGN-PIPELINE:DESIGN-SYNTHESIS:START -->
+## Requirements-Driven DESIGN.md Synthesis
+
+- Manifest: `openspec/changes/workflow-editor-design-pipeline-v090/design-synthesis.json`
+- Status: ready-to-implement
+- Stage: implementation
+- Product design output: `DESIGN.md`
+- Project motion foundation: `MOTION.md`
+- Change design: `openspec/changes/workflow-editor-design-pipeline-v090/design.md`
+- Input mode: requirements-only
+- Scope: 37/60 (fit)
+- Next: Implement from DESIGN.md, change design.md, and tasks.md; then run normal design-pipeline QA.
+
+Template and reference inputs are attributed evidence only. Product requirements and recorded
+decisions determine the synthesized design.
+<!-- DESIGN-PIPELINE:DESIGN-SYNTHESIS:END -->
diff --git a/openspec/changes/workflow-editor-design-pipeline-v090/motion.md b/openspec/changes/workflow-editor-design-pipeline-v090/motion.md
new file mode 100644
index 00000000..5bad6141
--- /dev/null
+++ b/openspec/changes/workflow-editor-design-pipeline-v090/motion.md
@@ -0,0 +1,49 @@
+# Motion
+
+## Foundation
+
+- Project foundation: `MOTION.md`
+- Foundation SHA-256: `31c6d70594b5c0c7d396544ff84a2e31e30795c657a24592869cf5f26094331e`
+- Posture: minimal
+- Selected primitive: `reveal.trim-line`
+
+## Scene: Node Authoring
+
+- Trigger: operator selects a node from the picker or confirms an Agent proposal.
+- Local layer: node appears immediately at the resolved canvas position.
+- Selection layer: 70ms press response, then 120ms selected-surface response.
+- Persistence layer: status changes to saving without moving the node; durable acknowledgement changes text and icon once revision advances.
+- Interruption: undo or delete removes/retargets the local state immediately and cancels obsolete acknowledgement.
+
+## Scene: Edge Creation
+
+- Trigger: a valid typed connection is committed.
+- Primitive: `reveal.trim-line` for one bounded causal reveal.
+- Duration: up to 200ms.
+- Invalid connection: no reveal; show the typed contract error and preserve focus.
+- Reduced motion: render the complete static edge immediately.
+
+## Scene: Collaboration State
+
+- LIVE, SYNCING, OFFLINE, and TOKEN REQUIRED change through 120ms surface/icon response only.
+- Remote graph updates never replay entry animation for the whole graph.
+- Remote selection or cursor movement does not produce trails.
+- Reconnection does not flash every node or edge.
+
+## Scene: Workflow Dock
+
+- Trigger: open, close, switch mode, or select a node.
+- Duration: 160ms for mode response; 200ms for bounded panel entry/exit.
+- Interruption: retarget to the latest mode and keep canvas focus recovery deterministic.
+- Reduced motion: no translation; content appears immediately or with near-zero opacity duration.
+
+## Runtime Bindings
+
+- CSS: press, response, status, compact disclosure.
+- WAAPI: optional interruptible panel/spatial choreography.
+- React Flow: canvas viewport and edge geometry.
+- No GSAP, Anime.js, Canvas renderer, PixiJS, WebGL, or WebGPU adoption.
+
+## Evidence
+
+Browser acceptance must record local node appearance, durable revision acknowledgement, LIVE/OFFLINE state, disconnected fallback, keyboard cleanup, and reduced-motion behavior before the change can claim visual completion.
diff --git a/openspec/changes/workflow-editor-design-pipeline-v090/proposal.md b/openspec/changes/workflow-editor-design-pipeline-v090/proposal.md
new file mode 100644
index 00000000..2b87afce
--- /dev/null
+++ b/openspec/changes/workflow-editor-design-pipeline-v090/proposal.md
@@ -0,0 +1,13 @@
+# Proposal: workflow-editor-design-pipeline-v090
+
+## Why
+
+The project needs a requirements-driven reusable DESIGN.md before implementation.
+
+## Problem
+
+Design the OpenCLI workflow editor for operators building demand-driven, Yjs-collaborative, evidence-grounded automations; preserve the Dark Ops Console, make node authoring immediate, surface collaboration, lifecycle, lineage, recovery, and grounded-state truth without generic AI styling.
+
+## Safety
+
+Reference sites and templates are attributed evidence, not product authority or copy targets.
diff --git a/openspec/changes/workflow-editor-design-pipeline-v090/qa.md b/openspec/changes/workflow-editor-design-pipeline-v090/qa.md
new file mode 100644
index 00000000..3572b467
--- /dev/null
+++ b/openspec/changes/workflow-editor-design-pipeline-v090/qa.md
@@ -0,0 +1,62 @@
+# QA
+
+## Release integrity
+
+- Release: `design-pipeline v0.9.0`
+- Asset: `design-pipeline-skill.tgz`
+- Expected SHA-256: `1c6ed651590dc255b96ae8a0a2bdfd87bf3d9649d72f47c06a76041cdb68e957`
+- Actual SHA-256: `1c6ed651590dc255b96ae8a0a2bdfd87bf3d9649d72f47c06a76041cdb68e957`
+- License: MIT
+- Pipeline doctor: ready; no missing requirements.
+
+## Foundations
+
+- `DESIGN.md`: ready
+- Design SHA-256: `150086d05ca2940d2e741501de2f3d86c99454415d846abf832c0a752d8ffbd6`
+- `MOTION.md`: ready
+- Motion SHA-256: `31c6d70594b5c0c7d396544ff84a2e31e30795c657a24592869cf5f26094331e`
+- Motion posture: minimal
+- Primitive: `reveal.trim-line`
+
+## Scope
+
+- Budget: program
+- Score: 37/60
+- Surfaces: 3
+- Workflows: 4
+- Integrations: 3
+- Unknowns: 2
+- Design decisions: 8
+- Wayfinder: not required.
+
+## Runtime evidence
+
+Actual surface: `/studio/workflow` for project `gaojixing-competitor-demand`.
+
+- Collaboration state rendered as `LIVE` with the canonical Workspace/Project/Workflow room.
+- Unauthorized WebSocket upgrade returned HTTP 401.
+- Local node appearance measured between approximately 263ms and 321ms including picker interaction.
+- Durable Yjs revision acknowledgement measured approximately 1.44s.
+- Disconnected local appearance measured approximately 153ms.
+- Disconnected full-Draft PUT acknowledgement measured approximately 996ms.
+- Persisted collaboration writes recorded `updated_by_user_id=collaboration-service`.
+- Disconnected fallback recorded `updated_by_user_id=local-development-user`.
+- Verification nodes were removed; final graph returned to six nodes.
+
+## Engineering gates
+
+- Collaboration API and demand routing tests: 3 passed.
+- Frontend production build: passed.
+- TypeScript: passed.
+- Static page generation: 42/42.
+- Collaboration Node syntax: passed.
+- Docker Compose configuration: passed.
+- API, frontend, Agent, and collaboration containers: healthy.
+
+## Accessibility and reduced motion
+
+The foundation requires textual collaboration and grounding states, visible focus, keyboard authoring paths, and static reduced-motion substitutes. No new decorative runtime or motion dependency was adopted. Browser acceptance verified the real desktop authoring route; a separate mobile visual refinement was not part of this bounded implementation.
+
+## Claim
+
+The design foundation and implemented collaboration behavior are verified for the current desktop workflow editor. This change does not claim a new visual redesign, pixel reconstruction, or completed mobile redesign.
diff --git a/openspec/changes/workflow-editor-design-pipeline-v090/state.json b/openspec/changes/workflow-editor-design-pipeline-v090/state.json
new file mode 100644
index 00000000..ce08a8f3
--- /dev/null
+++ b/openspec/changes/workflow-editor-design-pipeline-v090/state.json
@@ -0,0 +1,104 @@
+{
+  "blockers": [],
+  "changeId": "workflow-editor-design-pipeline-v090",
+  "evidence": [
+    "openspec/changes/workflow-editor-design-pipeline-v090/qa.md",
+    "DESIGN.md",
+    "MOTION.md",
+    "openspec/changes/workflow-editor-design-pipeline-v090/toolchain-plan.json"
+  ],
+  "extensions": {
+    "legacy": {
+      "artifactRoot": "openspec/changes/workflow-editor-design-pipeline-v090",
+      "capabilities": {
+        "fallbacks": [],
+        "missing": []
+      },
+      "decisions": [
+        "Treat reference sites and DESIGN.md templates as attributed evidence, not product authority"
+      ],
+      "designSynthesis": {
+        "manifest": "openspec/changes/workflow-editor-design-pipeline-v090/design-synthesis.json",
+        "output": "DESIGN.md",
+        "stage": "implementation",
+        "status": "ready-to-implement"
+      },
+      "gbrain": {
+        "detected": false,
+        "paths": [],
+        "syncPlanned": false
+      },
+      "motion": {
+        "implementationLibrary": "unknown",
+        "motionSpec": "openspec/changes/workflow-editor-design-pipeline-v090/motion.md",
+        "reducedMotion": "pending",
+        "required": false
+      },
+      "openSpec": {
+        "changeId": "workflow-editor-design-pipeline-v090",
+        "detected": true,
+        "paths": [
+          "openspec/changes/workflow-editor-design-pipeline-v090"
+        ]
+      },
+      "projectRoot": ".",
+      "qa": {
+        "evidenceRoot": "openspec/changes/workflow-editor-design-pipeline-v090/qa.md",
+        "scores": {},
+        "status": "not-run"
+      },
+      "surfaces": [
+        "DESIGN.md"
+      ]
+    },
+    "legacyEvents": {
+      "count": 5,
+      "sha256": "8b58ec40b73e5c04546d9be0f9f4411db0b7d3a6e601728cd4f7539b4a0f4084"
+    },
+    "legacyStatus": "in-progress",
+    "unknown": {
+      "motionFoundation": {
+        "path": "MOTION.md",
+        "posture": "minimal",
+        "primitiveRegistry": "design-pipeline.motion-primitives.v1",
+        "selectedPrimitives": [
+          "reveal.trim-line"
+        ],
+        "sha256": "31c6d70594b5c0c7d396544ff84a2e31e30795c657a24592869cf5f26094331e",
+        "status": "ready"
+      }
+    }
+  },
+  "foundations": {
+    "design": {
+      "path": "DESIGN.md",
+      "sha256": "150086d05ca2940d2e741501de2f3d86c99454415d846abf832c0a752d8ffbd6",
+      "status": "ready",
+      "validator": "designer-pipeline foundation check --kind design"
+    },
+    "motion": {
+      "path": "MOTION.md",
+      "sha256": "31c6d70594b5c0c7d396544ff84a2e31e30795c657a24592869cf5f26094331e",
+      "status": "ready",
+      "validator": "designer-pipeline foundation check --kind motion"
+    }
+  },
+  "lastEventSeq": 5,
+  "migration": {
+    "sourceNextField": "nextActions",
+    "sourcePhaseField": "phase",
+    "sourceSchema": "design-pipeline.state.v1",
+    "sourceSha256": "e2d93306d0d4113b496a067613a494c4cf97fdff9bd0889ec11816444ec7e3d5",
+    "sourceStatus": "in-progress"
+  },
+  "nextActions": [
+    "Use the validated foundations for future workflow editor design changes."
+  ],
+  "phase": "archive",
+  "registryVersion": "design-pipeline.phases.v2",
+  "revision": 5,
+  "sceneRuntime": null,
+  "schema": "design-pipeline.state.v2",
+  "status": "complete",
+  "updatedAt": "2026-08-26T15:54:15.485Z"
+}
diff --git a/openspec/changes/workflow-editor-design-pipeline-v090/tasks.md b/openspec/changes/workflow-editor-design-pipeline-v090/tasks.md
new file mode 100644
index 00000000..c6407218
--- /dev/null
+++ b/openspec/changes/workflow-editor-design-pipeline-v090/tasks.md
@@ -0,0 +1,16 @@
+# Tasks
+
+- [x] Verify Design Pipeline v0.9.0 release checksum.
+- [x] Run pipeline doctor against the target repository.
+- [x] Record the workflow editor product boundary and design decisions.
+- [x] Assess scope against the program budget.
+- [x] Select Calm Live Operations over review-first and ambient-collaboration directions.
+- [x] Repair and validate the reusable project `DESIGN.md`.
+- [x] Create and validate the reusable project `MOTION.md`.
+- [x] Bind the change to existing React Flow, Zustand, Yjs, y-websocket, Studio Draft, Version, Evidence, and Lineage boundaries.
+- [x] Preserve disconnected full-Draft PUT fallback and validation/publication durability.
+- [x] Verify the real workflow editor shows scoped LIVE collaboration.
+- [x] Measure local node appearance and durable revision acknowledgement.
+- [x] Verify authenticated room rejection and snapshot ownership.
+- [x] Remove browser verification nodes and restore the six-node canonical graph.
+- [x] Record QA evidence and foundation hashes.
diff --git a/openspec/changes/workflow-editor-design-pipeline-v090/toolchain-plan.json b/openspec/changes/workflow-editor-design-pipeline-v090/toolchain-plan.json
new file mode 100644
index 00000000..a8bcaaa2
--- /dev/null
+++ b/openspec/changes/workflow-editor-design-pipeline-v090/toolchain-plan.json
@@ -0,0 +1,150 @@
+{
+  "blockers": [],
+  "brief": "Refine the existing Next.js OpenCLI workflow editor with immediate React Flow node authoring, scoped Yjs collaboration, durable Studio revisions, demand provenance, grounded Evidence, accessible recovery states, and the existing Dark Ops Console design system.",
+  "evidence": [
+    "registry:2026-08-14",
+    "mengto:3f4c22d10055d3fdddb17248d59d0c1b731cb8d3",
+    "tool:design-pipeline/core@bundled",
+    "tool:design-pipeline/feedback-loop@bundled",
+    "tool:MengTo/skills@3f4c22d10055d3fdddb17248d59d0c1b731cb8d3",
+    "tool:wevm/frog@702c4eadefe175a208290a99c36e7f4ac1390e53"
+  ],
+  "framework": "react",
+  "graphics": null,
+  "invocations": [
+    {
+      "command": null,
+      "kind": "agent-route",
+      "owner": "agent",
+      "target": "design-pipeline/core",
+      "toolId": "design-pipeline/core"
+    },
+    {
+      "command": null,
+      "kind": "agent-route",
+      "owner": "agent",
+      "target": "design-pipeline/feedback-loop",
+      "toolId": "design-pipeline/feedback-loop"
+    },
+    {
+      "command": null,
+      "kind": "agent-route",
+      "owner": "agent",
+      "target": "MengTo/skills",
+      "toolId": "MengTo/skills"
+    },
+    {
+      "command": null,
+      "kind": "governed-review",
+      "owner": "agent",
+      "status": "review",
+      "target": "wevm/frog",
+      "toolId": "wevm/frog"
+    }
+  ],
+  "primaryRouteId": "design-pipeline/core",
+  "probes": [
+    {
+      "command": null,
+      "kind": "catalog",
+      "status": "available",
+      "toolId": "design-pipeline/core"
+    },
+    {
+      "command": null,
+      "kind": "catalog",
+      "status": "available",
+      "toolId": "design-pipeline/feedback-loop"
+    },
+    {
+      "command": null,
+      "kind": "catalog",
+      "status": "available",
+      "toolId": "MengTo/skills"
+    },
+    {
+      "command": null,
+      "kind": "governed-review",
+      "status": "review",
+      "toolId": "wevm/frog"
+    }
+  ],
+  "recommendedSkills": [],
+  "reviews": [
+    "wevm/frog is review"
+  ],
+  "routingContext": null,
+  "schema": "design-pipeline.toolchain-plan.v1",
+  "sourceHashes": {
+    "adapterRegistry": "00cfdb3b60c4388c248103164e340a39dc16719ff85367cd4586ece4649c729d",
+    "frontendRegistry": "8ef8f8c46e103619e081ad0b9f983d41d4317e41e384b4e72e9362f3b98d4f75",
+    "graphicsCatalog": "9baabd9656a05cf7ebf3a95399e11e348db50422ca068f7b00abee0caf1bc090"
+  },
+  "status": "ready",
+  "styling": {
+    "id": "none",
+    "label": "None"
+  },
+  "tools": [
+    {
+      "id": "design-pipeline/core",
+      "mode": "project-owned",
+      "source": "frontend-stack",
+      "status": "ready"
+    },
+    {
+      "id": "design-pipeline/feedback-loop",
+      "mode": "project-owned",
+      "source": "frontend-stack",
+      "status": "ready"
+    },
+    {
+      "id": "MengTo/skills",
+      "mode": "bundled-index",
+      "source": "frontend-stack",
+      "status": "ready"
+    },
+    {
+      "id": "wevm/frog",
+      "mode": "governed-candidate",
+      "source": "frontend-stack",
+      "status": "review"
+    }
+  ],
+  "uiLibrary": {
+    "id": "none",
+    "label": "None",
+    "license": "project",
+    "source": "project-owned"
+  },
+  "verification": [
+    {
+      "evidenceTypes": [
+        "design-workflow"
+      ],
+      "receiptSchema": "design-pipeline.toolchain-receipt.v1",
+      "toolId": "design-pipeline/core"
+    },
+    {
+      "evidenceTypes": [
+        "friction-logging"
+      ],
+      "receiptSchema": "design-pipeline.toolchain-receipt.v1",
+      "toolId": "design-pipeline/feedback-loop"
+    },
+    {
+      "evidenceTypes": [
+        "design-workflow"
+      ],
+      "receiptSchema": "design-pipeline.toolchain-receipt.v1",
+      "toolId": "MengTo/skills"
+    },
+    {
+      "evidenceTypes": [
+        "friction-logging"
+      ],
+      "receiptSchema": "design-pipeline.toolchain-receipt.v1",
+      "toolId": "wevm/frog"
+    }
+  ]
+}
diff --git a/openspec/changes/workflow-editor-design-pipeline-v090/toolchain-request.json b/openspec/changes/workflow-editor-design-pipeline-v090/toolchain-request.json
new file mode 100644
index 00000000..3dfbc400
--- /dev/null
+++ b/openspec/changes/workflow-editor-design-pipeline-v090/toolchain-request.json
@@ -0,0 +1,30 @@
+{
+  "schema": "design-pipeline.toolchain-request.v1",
+  "framework": "react",
+  "brief": "Refine the existing Next.js OpenCLI workflow editor with immediate React Flow node authoring, scoped Yjs collaboration, durable Studio revisions, demand provenance, grounded Evidence, accessible recovery states, and the existing Dark Ops Console design system.",
+  "existing": {
+    "framework": "Next.js 16",
+    "canvas": "@xyflow/react 12.11.2",
+    "state": "zustand 5.0.14",
+    "collaboration": "yjs 13.6.31 + y-websocket 3.0.0",
+    "styles": "project CSS tokens and existing UI primitives",
+    "motion": "CSS, WAAPI, Motion One where already present"
+  },
+  "requested": {
+    "mode": "refine",
+    "componentOrigin": "project-owned",
+    "runtimeAdoption": "reuse-installed-only"
+  },
+  "capabilities": [
+    "node-canvas",
+    "typed-connections",
+    "incremental-collaboration",
+    "durable-save-acknowledgement",
+    "workflow-lifecycle",
+    "demand-provenance",
+    "grounded-evidence",
+    "offline-recovery",
+    "keyboard-authoring",
+    "reduced-motion"
+  ]
+}
diff --git a/scripts/install-agent.sh b/scripts/install-agent.sh
index 5fd80138..849156c2 100755
--- a/scripts/install-agent.sh
+++ b/scripts/install-agent.sh
@@ -296,6 +296,33 @@ install_python() {
   fi
   touch "$AGENT_DIR/backend/__init__.py"
 
+  # ── Download runtime adapters and MiniFlow for native Agents ──────────────
+  RUNTIME_BUNDLE="$AGENT_DIR/agent-runtime.tar.gz"
+  RUNTIME_BUNDLE_URL="${CENTRAL_API_URL%/}/api/v1/nodes/install/agent-runtime.tar.gz"
+  command -v tar >/dev/null 2>&1 || die "tar is required to install Agent runtime packages"
+  info "Downloading Agent runtime packages from center..."
+  if command -v curl >/dev/null 2>&1; then
+    curl -fsSL "${AUTH_HEADER[@]}" "$RUNTIME_BUNDLE_URL" -o "$RUNTIME_BUNDLE"
+  elif command -v wget >/dev/null 2>&1; then
+    if [[ -n "$AGENT_API_TOKEN" ]]; then
+      wget --header="Authorization: Bearer $AGENT_API_TOKEN" \
+        -qO "$RUNTIME_BUNDLE" "$RUNTIME_BUNDLE_URL"
+    else
+      wget -qO "$RUNTIME_BUNDLE" "$RUNTIME_BUNDLE_URL"
+    fi
+  else
+    die "Neither curl nor wget found — cannot download Agent runtime packages"
+  fi
+  tar -tzf "$RUNTIME_BUNDLE" | while IFS= read -r entry; do
+    [[ "$entry" == backend/agent_runtimes/*.py \
+      || "$entry" == backend/miniflow/*.py \
+      || "$entry" == backend/security/*.py ]] || \
+      die "Unexpected path in Agent runtime package: $entry"
+  done
+  tar --no-same-owner --no-same-permissions -xzf "$RUNTIME_BUNDLE" -C "$AGENT_DIR"
+  rm -f "$RUNTIME_BUNDLE"
+  info "Installed Agent runtime packages under $AGENT_DIR/backend"
+
   # ── Install Python dependencies ───────────────────────────────────────────
   info "Installing Python dependencies..."
   VENV_DIR="$AGENT_DIR/venv"
diff --git a/scripts/install.sh b/scripts/install.sh
index a79c9a36..5a308589 100644
--- a/scripts/install.sh
+++ b/scripts/install.sh
@@ -61,14 +61,12 @@ replace_env() {
 }
 
 api_token="$(random_hex 32)"
-bootstrap_token="$(random_hex 32)"
 credential_encryption_key="$(random_fernet)"
 if [ -z "$credential_encryption_key" ]; then
   echo "Failed to generate CREDENTIAL_ENCRYPTION_KEY" >&2
   exit 1
 fi
 replace_env API_AUTH_TOKEN "$api_token"
-replace_env BOOTSTRAP_ADMIN_TOKEN "$bootstrap_token"
 replace_env SECRET_KEY "$(random_hex 32)"
 replace_env CREDENTIAL_ENCRYPTION_KEY "$credential_encryption_key"
 chmod 600 "$INSTALL_DIR/.env"
@@ -91,6 +89,5 @@ done
 
 printf '\nOpenCLI Admin %s is ready.\n' "$VERSION"
 printf 'URL: http://localhost:%s\n' "${FRONTEND_PORT:-3010}"
-printf 'BOOTSTRAP_ADMIN_TOKEN: %s\n' "$bootstrap_token"
-printf 'API_AUTH_TOKEN: %s\n' "$api_token"
-printf 'Use BOOTSTRAP_ADMIN_TOKEN in the first login field and API_AUTH_TOKEN in the optional fleet field. Both are stored in %s/.env\n' "$INSTALL_DIR"
+printf 'Local login: admin / admin (change it later in Account Settings)\n'
+printf 'API_AUTH_TOKEN is generated for Fleet/Agent/API transport and stored in %s/.env\n' "$INSTALL_DIR"
diff --git a/tests/integration/test_agents_api.py b/tests/integration/test_agents_api.py
index 50a2d110..cbafd76c 100644
--- a/tests/integration/test_agents_api.py
+++ b/tests/integration/test_agents_api.py
@@ -32,6 +32,56 @@ async def test_create_agent(client, agent_data):
     assert "id" in data
 
 
+@pytest.mark.asyncio
+async def test_create_paw_agent_round_trips_safe_processor_config(client, agent_data):
+    response = await client.post(
+        "/api/v1/agents",
+        json={
+            **agent_data,
+            "name": "Offline PAW Agent",
+            "processor_type": "paw",
+            "processor_config": {"max_tokens": 128},
+        },
+    )
+
+    assert response.status_code == 201
+    data = response.json()["data"]
+    assert data["processor_type"] == "paw"
+    assert data["processor_config"] == {"max_tokens": 128}
+
+
+@pytest.mark.asyncio
+async def test_paw_agent_api_rejects_blank_or_unapproved_config(client, agent_data):
+    blank = await client.post(
+        "/api/v1/agents",
+        json={**agent_data, "processor_type": "paw", "prompt_template": "   "},
+    )
+    unsupported = await client.post(
+        "/api/v1/agents",
+        json={
+            **agent_data,
+            "processor_type": "paw",
+            "processor_config": {"endpoint": "http://example.test"},
+        },
+    )
+
+    assert blank.status_code == 422
+    assert unsupported.status_code == 422
+
+
+@pytest.mark.asyncio
+async def test_paw_agent_update_validates_merged_state(client, agent_data):
+    created = await client.post(
+        "/api/v1/agents",
+        json={**agent_data, "processor_type": "paw", "processor_config": {"max_tokens": 8}},
+    )
+    agent_id = created.json()["data"]["id"]
+
+    response = await client.patch(f"/api/v1/agents/{agent_id}", json={"prompt_template": " "})
+
+    assert response.status_code == 422
+
+
 @pytest.mark.asyncio
 async def test_list_agents_after_create(client, agent_data):
     await client.post("/api/v1/agents", json=agent_data)
diff --git a/tests/integration/test_auth_api.py b/tests/integration/test_auth_api.py
index 226fc3e6..d6a62f73 100644
--- a/tests/integration/test_auth_api.py
+++ b/tests/integration/test_auth_api.py
@@ -6,6 +6,8 @@
 change is visible immediately and undone automatically after each test.
 """
 
+import secrets
+
 import pytest
 
 from backend.config import get_settings
@@ -15,6 +17,7 @@
     get_request_identity,
     identity_dependency,
 )
+from backend.security.local_auth import DEFAULT_LOCAL_ADMIN_PASSWORD_HASH
 
 TOKEN = "fleet-test-token"
 
@@ -138,3 +141,67 @@ async def test_health_exempt_and_leaks_nothing(client, auth_enabled):
     response = await client.get("/health")
     assert response.status_code == 200
     assert response.json() == {"status": "ok"}
+
+
+@pytest.mark.asyncio
+async def test_local_admin_login_uses_simple_credentials(client, auth_enabled):
+    response = await client.post(
+        "/api/v1/auth/login",
+        json={"username": "admin", "password": "admin"},
+    )
+    assert response.status_code == 200
+    payload = response.json()["data"]
+    assert payload["token_type"] == "bearer"
+    assert payload["using_default_password"] is True
+
+    identity = await client.get(
+        "/api/v1/auth/me",
+        headers={"Authorization": f"Bearer {payload['access_token']}"},
+    )
+    assert identity.status_code == 200
+    assert identity.json()["data"]["auth_method"] == "local"
+
+
+@pytest.mark.asyncio
+async def test_local_admin_login_rejects_wrong_password(client, auth_disabled):
+    wrong_password = f"wrong-{secrets.token_hex(8)}"
+    response = await client.post(
+        "/api/v1/auth/login",
+        json={"username": "admin", "password": wrong_password},
+    )
+    assert response.status_code == 401
+
+
+
+@pytest.mark.asyncio
+async def test_local_admin_can_change_password(client, auth_disabled, monkeypatch, tmp_path):
+    new_password = f"local-{secrets.token_hex(8)}"
+    monkeypatch.setenv("ENV_FILE_PATH", str(tmp_path / ".env"))
+    monkeypatch.setenv("LOCAL_ADMIN_PASSWORD_HASH", DEFAULT_LOCAL_ADMIN_PASSWORD_HASH)
+    get_settings.cache_clear()
+    try:
+        login = await client.post(
+            "/api/v1/auth/login",
+            json={"username": "admin", "password": "admin"},
+        )
+        token = login.json()["data"]["access_token"]
+        changed = await client.post(
+            "/api/v1/auth/password",
+            headers={"Authorization": f"Bearer {token}"},
+            json={"current_password": "admin", "new_password": new_password},
+        )
+        assert changed.status_code == 200
+        assert (
+            await client.post(
+                "/api/v1/auth/login",
+                json={"username": "admin", "password": "admin"},
+            )
+        ).status_code == 401
+        assert (
+            await client.post(
+                "/api/v1/auth/login",
+                json={"username": "admin", "password": new_password},
+            )
+        ).status_code == 200
+    finally:
+        get_settings.cache_clear()
diff --git a/tests/integration/test_auth_ws.py b/tests/integration/test_auth_ws.py
index 942ac43f..be52f3cf 100644
--- a/tests/integration/test_auth_ws.py
+++ b/tests/integration/test_auth_ws.py
@@ -148,6 +148,38 @@ def test_nodes_ws_register_with_runtimes_accepted(test_client, auth_enabled):
         assert reply["type"] == "registered"
 
 
+
+def test_nodes_ws_register_with_runtime_capabilities_accepted(test_client, auth_enabled):
+    msg = {
+        **REGISTER_MSG_NODES,
+        "runtimes": ["pi", "prime-agent"],
+        "runtime_capabilities": {
+            "pi": ["streaming", "tool_events"],
+            "prime-agent": ["streaming", "subagents", "tool_events"],
+        },
+    }
+    with test_client.websocket_connect(
+        "/api/v1/nodes/ws", headers={"Authorization": f"Bearer {TOKEN}"}
+    ) as ws:
+        ws.send_json(msg)
+        assert ws.receive_json()["type"] == "registered"
+
+
+def test_nodes_ws_rejects_capability_manifest_for_unadvertised_runtime(
+    test_client, auth_enabled
+):
+    msg = {
+        **REGISTER_MSG_NODES,
+        "runtimes": ["pi"],
+        "runtime_capabilities": {"prime-agent": ["streaming"]},
+    }
+    with pytest.raises(WebSocketDisconnect):
+        with test_client.websocket_connect(
+            "/api/v1/nodes/ws", headers={"Authorization": f"Bearer {TOKEN}"}
+        ) as ws:
+            ws.send_json(msg)
+            ws.receive_json()
+
 def test_nodes_ws_register_invalid_runtimes_type_rejected(test_client, auth_enabled):
     msg = {**REGISTER_MSG_NODES, "runtimes": "not-a-list"}
     with pytest.raises(WebSocketDisconnect):
diff --git a/tests/integration/test_local_workspace_api.py b/tests/integration/test_local_workspace_api.py
new file mode 100644
index 00000000..cef51c81
--- /dev/null
+++ b/tests/integration/test_local_workspace_api.py
@@ -0,0 +1,23 @@
+import pytest
+
+
+@pytest.mark.asyncio
+async def test_local_admin_gets_default_workspace_membership(client):
+    login = await client.post(
+        "/api/v1/auth/login",
+        json={"username": "admin", "password": "admin"},
+    )
+    token = login.json()["data"]["access_token"]
+    headers = {"Authorization": f"Bearer {token}"}
+
+    workspaces = await client.get("/api/v1/governance/workspaces", headers=headers)
+    assert workspaces.status_code == 200
+    workspace = workspaces.json()["data"][0]
+    assert workspace["slug"] == "opencli-default"
+
+    automations = await client.get(
+        f"/api/v1/workspaces/{workspace['id']}/automations",
+        headers=headers,
+    )
+    assert automations.status_code == 200
+    assert automations.json()["data"] == []
diff --git a/tests/integration/test_studio_collaboration_api.py b/tests/integration/test_studio_collaboration_api.py
new file mode 100644
index 00000000..f3ddb341
--- /dev/null
+++ b/tests/integration/test_studio_collaboration_api.py
@@ -0,0 +1,92 @@
+import pytest
+
+from tests.fixtures.workflow_conformance import workflow_conformance_project
+
+
+async def _bootstrap_workflow(client):
+    workspace_id = (await client.get("/api/v1/workspaces")).json()["data"][0]["id"]
+    response = await client.post(
+        f"/api/v1/workspaces/{workspace_id}/projects/bootstrap",
+        json={
+            "project": {"name": "Collaboration project", "slug": "collaboration"},
+            "workflow": {"name": "Collaborative workflow", "graph": workflow_conformance_project()},
+        },
+    )
+    assert response.status_code == 201, response.text
+    data = response.json()["data"]
+    return workspace_id, data["project"]["id"], data["primary_workflow"]["id"]
+
+
+def _room(workspace_id: str, project_id: str, workflow_id: str) -> str:
+    return f"workspace:{workspace_id}:project:{project_id}:workflow:{workflow_id}"
+
+
+@pytest.mark.asyncio
+async def test_collaboration_authorize_returns_only_the_owning_workflow_room(client):
+    workspace_id, project_id, workflow_id = await _bootstrap_workflow(client)
+    room = _room(workspace_id, project_id, workflow_id)
+
+    authorized = await client.post(
+        "/api/v1/internal/collaboration/authorize", json={"room": room}
+    )
+    assert authorized.status_code == 200, authorized.text
+    assert authorized.json()["data"] == {"room": room}
+
+    malformed = await client.post(
+        "/api/v1/internal/collaboration/authorize",
+        json={"room": f"workspace:{workspace_id}:workflow:{workflow_id}"},
+    )
+    assert malformed.status_code == 400
+
+    other_project = await client.post(
+        "/api/v1/internal/collaboration/authorize",
+        json={"room": _room(workspace_id, "not-the-owner", workflow_id)},
+    )
+    assert other_project.status_code == 404
+
+
+@pytest.mark.asyncio
+async def test_collaboration_snapshot_replaces_graph_entries_and_preserves_draft_metadata(client):
+    workspace_id, project_id, workflow_id = await _bootstrap_workflow(client)
+    room = _room(workspace_id, project_id, workflow_id)
+    draft_url = (
+        f"/api/v1/workspaces/{workspace_id}/projects/{project_id}"
+        f"/workflows/{workflow_id}/draft"
+    )
+    original = (await client.get(draft_url)).json()["data"]["graph"]
+    nodes = [
+        {
+            **node,
+            "ui": {
+                **(node.get("ui") or {}),
+                "position": {"x": 240 + index, "y": 120 + index},
+            },
+        }
+        for index, node in enumerate(original["nodes"])
+    ]
+    edges = [
+        {**edge, "label": f"collaborative-{index}"}
+        for index, edge in enumerate(original["edges"])
+    ]
+
+    snapshot = await client.post(
+        "/api/v1/internal/collaboration/snapshot",
+        json={
+            "room": room,
+            "data": {
+                "nodes": {"type": "Map", "content": {node["id"]: node for node in nodes}},
+                "edges": {"type": "Map", "content": {edge["id"]: edge for edge in edges}},
+            },
+        },
+    )
+    assert snapshot.status_code == 200, snapshot.text
+    assert snapshot.json()["data"] == {"revision": 2}
+
+    persisted = (await client.get(draft_url)).json()["data"]
+    assert persisted["revision"] == 2
+    assert persisted["updated_by_user_id"] == "collaboration-service"
+    assert persisted["graph"]["name"] == original["name"]
+    assert persisted["graph"]["settings"] == original["settings"]
+    assert persisted["graph"]["adapters"] == original["adapters"]
+    assert persisted["graph"]["nodes"] == nodes
+    assert persisted["graph"]["edges"] == edges
diff --git a/tests/integration/test_studio_lifecycle_api.py b/tests/integration/test_studio_lifecycle_api.py
index cae45749..2ccfa871 100644
--- a/tests/integration/test_studio_lifecycle_api.py
+++ b/tests/integration/test_studio_lifecycle_api.py
@@ -33,8 +33,7 @@ async def _create_studio_workflow(client, *, graph: dict | None = None) -> dict:
     project = result["project"]
     workflow = result["primary_workflow"]
     base_url = (
-        f"/api/v1/workspaces/{workspace_id}/projects/{project['id']}"
-        f"/workflows/{workflow['id']}"
+        f"/api/v1/workspaces/{workspace_id}/projects/{project['id']}/workflows/{workflow['id']}"
     )
     return {"workflow": workflow, "base_url": base_url}
 
@@ -76,8 +75,8 @@ async def test_studio_legacy_graph_extensions_and_nested_nulls_round_trip(client
     first_draft = first.json()["data"]
     first_graph = first_draft["graph"]
     assert first_graph["legacyExtension"] == graph["legacyExtension"]
-    assert first_graph["nodes"][0]["legacyNodeExtension"] == (
-        graph["nodes"][0]["legacyNodeExtension"]
+    assert (
+        first_graph["nodes"][0]["legacyNodeExtension"] == (graph["nodes"][0]["legacyNodeExtension"])
     )
     assert "sourceAnchor" not in first_graph["nodes"][0]
 
@@ -91,8 +90,9 @@ async def test_studio_legacy_graph_extensions_and_nested_nulls_round_trip(client
     assert updated.status_code == 200, updated.text
     updated_graph = updated.json()["data"]["graph"]
     assert updated_graph["legacyExtension"] == first_graph["legacyExtension"]
-    assert updated_graph["nodes"][0]["legacyNodeExtension"] == (
-        first_graph["nodes"][0]["legacyNodeExtension"]
+    assert (
+        updated_graph["nodes"][0]["legacyNodeExtension"]
+        == (first_graph["nodes"][0]["legacyNodeExtension"])
     )
 
     reloaded = await client.get(draft_url)
@@ -160,8 +160,7 @@ async def test_studio_validation_rejects_an_isolated_source_node(client):
         {
             "code": "isolated_source_node",
             "message": (
-                'Workflow source node "isolated-http-source" is not connected '
-                "to a downstream node"
+                'Workflow source node "isolated-http-source" is not connected to a downstream node'
             ),
             "node_id": "isolated-http-source",
             "edge_id": None,
@@ -226,7 +225,7 @@ async def test_studio_api_run_is_version_bound_idempotent_and_visible_in_logs(
     request = {
         "inputs": {"topic": "OpenCLI ecosystem"},
         "response_mode": "async",
-        "user": "server-worker",
+        "initiated_by": "server-worker",
     }
     headers = {
         "Idempotency-Key": "nightly-project-job",
@@ -251,7 +250,7 @@ async def test_studio_api_run_is_version_bound_idempotent_and_visible_in_logs(
     ).model_dump(mode="json")
     assert row.request["input"]["payload"] == request["inputs"]
     assert row.request["input"]["source"] == "external"
-    assert row.request["input"]["sourceId"] == request["user"]
+    assert row.request["input"]["sourceId"] == request["initiated_by"]
     assert row.request["trigger"]["requestId"] == headers["X-Request-ID"]
 
     project_url = created["base_url"].split("/workflows/", 1)[0]
@@ -274,7 +273,7 @@ async def test_studio_api_run_is_version_bound_idempotent_and_visible_in_logs(
     trace_data = trace.json()["data"]
     assert trace_data["workflow_version"] == published["version"]
     assert trace_data["inputs"] == request["inputs"]
-    assert trace_data["user"] == request["user"]
+    assert trace_data["user"] == request["initiated_by"]
     assert trace_data["trace"]["projection"]["runId"] == projection["runId"]
     paged_trace = await client.get(
         f"{created['base_url']}/runs/{projection['runId']}/trace",
@@ -287,10 +286,36 @@ async def test_studio_api_run_is_version_bound_idempotent_and_visible_in_logs(
         "afterSequence": 0,
         "limit": 1,
     }
-    assert (
-        paged_trace.json()["data"]["trace"]["nextAfterSequence"]
-        == paged_events[0]["sequence"]
-    )
+    assert paged_trace.json()["data"]["trace"]["nextAfterSequence"] == paged_events[0]["sequence"]
+
+    scoped_run_url = f"{created['base_url']}/runs/{projection['runId']}"
+    scoped_projection = await client.get(scoped_run_url)
+    scoped_events = await client.get(f"{scoped_run_url}/events")
+    scoped_batches = await client.get(f"{scoped_run_url}/evidence-batches")
+    scoped_evidence = await client.get(f"{scoped_run_url}/projection")
+    assert scoped_projection.status_code == 200, scoped_projection.text
+    assert scoped_projection.json()["data"]["runId"] == projection["runId"]
+    assert scoped_events.status_code == 200, scoped_events.text
+    assert scoped_events.json()["data"] == trace_data["trace"]["events"]
+    assert scoped_batches.status_code == 200, scoped_batches.text
+    assert scoped_evidence.status_code == 200, scoped_evidence.text
+
+    workspace_url = project_url.rsplit("/projects/", 1)[0]
+    for wrong_path in (
+        scoped_run_url.replace(workspace_url, f"{workspace_url}-other", 1),
+        scoped_run_url.replace(
+            project_url,
+            f"{workspace_url}/projects/not-the-owning-project/workflows/{created['workflow']['id']}",
+            1,
+        ),
+        scoped_run_url.replace(
+            created["base_url"],
+            f"{project_url}/workflows/not-the-owning-workflow",
+            1,
+        ),
+    ):
+        denied = await client.get(wrong_path)
+        assert denied.status_code == 404, (wrong_path, denied.text)
 
     generic_run_url = f"/api/v1/workflows/runs/{projection['runId']}"
     for suffix in (
@@ -482,9 +507,7 @@ async def test_image_canvas_document_is_fixed_to_snapshot_during_validation_and_
     )
     assert updated.status_code == 200, updated.text
 
-    validation = await client.post(
-        f"{created['base_url']}/draft/validation-runs", json={}
-    )
+    validation = await client.post(f"{created['base_url']}/draft/validation-runs", json={})
     assert validation.status_code == 201, validation.text
     assert validation.json()["data"]["valid"] is True
     published = await client.post(
diff --git a/tests/integration/test_system_config_api.py b/tests/integration/test_system_config_api.py
new file mode 100644
index 00000000..36cc2e38
--- /dev/null
+++ b/tests/integration/test_system_config_api.py
@@ -0,0 +1,51 @@
+import pytest
+
+from backend.config import get_settings
+
+
+@pytest.mark.asyncio
+async def test_system_config_exposes_runtime_sections(client):
+    response = await client.get("/api/v1/system/config")
+    assert response.status_code == 200
+    data = response.json()["data"]
+    assert data["collection_mode"] == "local"
+    assert "agent_pool_endpoints" in data
+    assert "credential_encryption_configured" in data
+    assert "effective_cdp_endpoints" in data
+    assert data["effective_cdp_endpoints"] == (
+        data["agent_pool_endpoints"]
+        if data["agent_pool_endpoints"]
+        else [data["opencli_cdp_endpoint"]]
+    )
+    assert "control_kill_switch" in data
+
+
+@pytest.mark.asyncio
+async def test_system_config_updates_safe_runtime_fields(client, monkeypatch, tmp_path):
+    monkeypatch.setenv("ENV_FILE_PATH", str(tmp_path / ".env"))
+    monkeypatch.setenv("AGENT_POOL_ENDPOINTS", "")
+    get_settings.cache_clear()
+    try:
+        response = await client.patch(
+            "/api/v1/system/config",
+            json={
+                "collection_mode": "agent",
+                "local_max_concurrent_pipelines": 12,
+                "default_timezone": "Asia/Shanghai",
+                "control_kill_switch": True,
+                "agent_pool_endpoints": "http://agent-1:19222, http://agent-2:19222",
+            },
+        )
+        assert response.status_code == 200
+        data = response.json()["data"]
+        assert data["collection_mode"] == "agent"
+        assert data["local_max_concurrent_pipelines"] == 12
+        assert data["default_timezone"] == "Asia/Shanghai"
+        assert data["control_kill_switch"] is True
+        assert data["agent_pool_endpoints"] == [
+            "http://agent-1:19222",
+            "http://agent-2:19222",
+        ]
+        assert data["effective_cdp_endpoints"] == data["agent_pool_endpoints"]
+    finally:
+        get_settings.cache_clear()
diff --git a/tests/integration/test_workflow_capabilities_api.py b/tests/integration/test_workflow_capabilities_api.py
index bbac4194..b8c3564a 100644
--- a/tests/integration/test_workflow_capabilities_api.py
+++ b/tests/integration/test_workflow_capabilities_api.py
@@ -126,12 +126,7 @@ async def _native_lifecycle_capability(client, monkeypatch, tools):
     monkeypatch.setattr(
         "backend.workflow.capability_projection.list_workflow_tool_capabilities",
         lambda: registry.model_copy(
-            update={
-                "tools": [
-                    tool for tool in registry.tools if tool.id not in native_ids
-                ]
-                + tools
-            }
+            update={"tools": [tool for tool in registry.tools if tool.id not in native_ids] + tools}
         ),
     )
     response = await client.get("/api/v1/workflows/capabilities")
@@ -184,9 +179,7 @@ async def test_compile_reports_webhook_notify_contract_without_live_delivery(cli
 
 
 @pytest.mark.asyncio
-async def test_native_lifecycle_readiness_requires_exact_complete_action_set(
-    client, monkeypatch
-):
+async def test_native_lifecycle_readiness_requires_exact_complete_action_set(client, monkeypatch):
     tools = _native_lifecycle_tools()
 
     capability = await _native_lifecycle_capability(client, monkeypatch, tools)
@@ -238,11 +231,7 @@ async def test_native_lifecycle_readiness_fails_closed_for_invalid_action_set(
     client, monkeypatch, case, expected_detail
 ):
     tools = _native_lifecycle_tools()
-    target = next(
-        tool
-        for tool in tools
-        if tool.executor.params.get("action") == "report.answers"
-    )
+    target = next(tool for tool in tools if tool.executor.params.get("action") == "report.answers")
     if case == "missing":
         tools.remove(target)
     elif case == "duplicate":
@@ -328,6 +317,7 @@ async def test_workflow_capabilities_project_real_backend_surfaces(client, monke
         "intelligence.source.rsshub",
         "intelligence.source.rss-bridge",
         "intelligence.source.http",
+        "intelligence.source.doubao-research",
     ):
         assert catalog[source_id]["status"] == "runnable"
         assert catalog[source_id]["backendAvailable"] is True
@@ -336,6 +326,11 @@ async def test_workflow_capabilities_project_real_backend_surfaces(client, monke
             {"name": "out", "type": "items[]"}
         ]
         assert "network.fetch" in catalog[source_id]["manifest"]["permissions"]
+    assert catalog["intelligence.source.doubao-research"]["channelType"] == "doubao_research"
+    assert (
+        "canFetchNetwork"
+        in catalog["intelligence.source.doubao-research"]["manifest"]["permissions"]
+    )
     assert catalog["intelligence.source.opencli-slot"]["status"] == "runnable"
     assert catalog["intelligence.source.opencli-slot"]["backendAvailable"] is True
     assert catalog["intelligence.source.opencli-slot"]["runtimeBinding"]
@@ -502,15 +497,12 @@ async def test_workflow_capabilities_project_real_backend_surfaces(client, monke
     assert triggers["trigger.webhook"]["status"] == "runnable"
     assert triggers["trigger.webhook"]["missing"] == []
     assert triggers["trigger.webhook"]["runtimeBinding"] == "workflow.trigger.webhook_input"
-    assert triggers["trigger.webhook"]["manifest"]["contract"]["status"] == (
-        "dispatch_only"
+    assert triggers["trigger.webhook"]["manifest"]["contract"]["status"] == ("dispatch_only")
+    assert triggers["trigger.webhook"]["manifest"]["contract"]["configGate"] == {"required": []}
+    assert (
+        "workflow-webhook-ingress-api"
+        in triggers["trigger.webhook"]["manifest"]["contract"]["fixtureCoverage"]["cases"]
     )
-    assert triggers["trigger.webhook"]["manifest"]["contract"]["configGate"] == {
-        "required": []
-    }
-    assert "workflow-webhook-ingress-api" in triggers["trigger.webhook"]["manifest"][
-        "contract"
-    ]["fixtureCoverage"]["cases"]
 
     resources = {item["id"]: item for item in data["resources"]}
     fleet_resource = resources["resource.workflow-fleet-runtime"]
@@ -689,9 +681,7 @@ def fake_run(*args, **kwargs):
         return subprocess.CompletedProcess(
             args=args[0],
             returncode=0,
-            stdout=json.dumps(
-                [{"site": "bbc", "name": "news", "access": "read", "args": []}]
-            ),
+            stdout=json.dumps([{"site": "bbc", "name": "news", "access": "read", "args": []}]),
             stderr="",
         )
 
diff --git a/tests/integration/test_workflow_opencli_hda_trace_api.py b/tests/integration/test_workflow_opencli_hda_trace_api.py
index 5db05916..bf615c5a 100644
--- a/tests/integration/test_workflow_opencli_hda_trace_api.py
+++ b/tests/integration/test_workflow_opencli_hda_trace_api.py
@@ -423,14 +423,11 @@ async def collect(_self, config, parameters):
             {},
         )
     ]
-    events = (await client.get(f"/api/v1/workflows/runs/{data['runId']}/events")).json()[
-        "data"
-    ]
+    events = (await client.get(f"/api/v1/workflows/runs/{data['runId']}/events")).json()["data"]
     source_partial = next(
         event
         for event in events
-        if event["nodeId"] == "multi-source-opencli::source-bbc"
-        and event["eventType"] == "partial"
+        if event["nodeId"] == "multi-source-opencli::source-bbc" and event["eventType"] == "partial"
     )
     assert source_partial["details"]["itemCount"] == 1
     assert source_partial["details"]["agentDispatch"]["protocol"] == "local"
@@ -443,9 +440,25 @@ async def collect(_self, config, parameters):
     record = (await db_session.execute(select(CollectedRecord))).scalar_one()
     assert record.workflow_id == project["id"]
     assert record.workflow_run_id == data["runId"]
-    assert record.raw_data["_workflowLineage"][0]["nodeId"] == (
-        "multi-source-opencli::source-bbc"
+    assert record.raw_data["_workflowLineage"][0]["nodeId"] == ("multi-source-opencli::source-bbc")
+    assert sink_partial["details"]["storedRefs"][0]["outcome"] == "stored"
+
+    duplicate_response = await client.post("/api/v1/workflows/runs", json={"project": project})
+    assert duplicate_response.status_code == 202, duplicate_response.text
+    duplicate_events = (
+        await client.get(
+            f"/api/v1/workflows/runs/{duplicate_response.json()['data']['runId']}/events"
+        )
+    ).json()["data"]
+    duplicate_sink_partial = next(
+        event
+        for event in duplicate_events
+        if event["nodeId"] == "record-sink" and event["eventType"] == "partial"
     )
+    assert duplicate_sink_partial["details"]["storedRecordCount"] == 0
+    assert duplicate_sink_partial["details"]["skippedRecordCount"] == 1
+    assert duplicate_sink_partial["details"]["storedRefs"][0]["outcome"] == "skipped"
+    assert duplicate_sink_partial["details"]["storedRefs"][0]["recordId"] == record.id
 
 
 def _native_first_loop_project() -> dict:
@@ -1058,8 +1071,7 @@ async def test_opencli_hda_trace_accepts_ai_source_slots_without_static_internal
     assert data["dispatches"][0]["sourceBindingRevisionId"] == "binding-video-r3"
     assert data["dispatches"][0]["sourceBindingRevisionNumber"] == 3
     assert (
-        data["dispatches"][0]["iii"]["payload"]["source_binding_revision_id"]
-        == "binding-video-r3"
+        data["dispatches"][0]["iii"]["payload"]["source_binding_revision_id"] == "binding-video-r3"
     )
     assert data["dispatches"][1]["site"] == "xiaohongshu"
 
@@ -1070,9 +1082,7 @@ async def test_opencli_hda_collects_per_source_failures_without_blocking_package
     monkeypatch,
 ):
     project = _multi_source_opencli_hda_project()
-    project["nodes"][0]["params"] = {
-        "execution": {"failureMode": "collect-per-source"}
-    }
+    project["nodes"][0]["params"] = {"execution": {"failureMode": "collect-per-source"}}
 
     async def fake_dispatch(dispatch, fleet_match, *, node):
         if dispatch.sourceGroup == "video":
@@ -1113,12 +1123,11 @@ async def fake_dispatch(dispatch, fleet_match, *, node):
     states = {state["nodeId"]: state for state in data["nodeStates"]}
     assert states["multi-source-opencli"]["status"] == "completed"
     assert states["multi-source-opencli::source-xiaohongshu"]["status"] == "failed"
-    events = (
-        await client.get("/api/v1/workflows/runs/run-collect-per-source/events")
-    ).json()["data"]
+    events = (await client.get("/api/v1/workflows/runs/run-collect-per-source/events")).json()[
+        "data"
+    ]
     assert not any(
-        event["nodeId"] == "multi-source-opencli"
-        and event["eventType"] == "blocked"
+        event["nodeId"] == "multi-source-opencli" and event["eventType"] == "blocked"
         for event in events
     )
 
@@ -1129,9 +1138,7 @@ async def test_opencli_hda_collect_per_source_blocks_package_when_all_sources_fa
     monkeypatch,
 ):
     project = _multi_source_opencli_hda_project()
-    project["nodes"][0]["params"] = {
-        "execution": {"failureMode": "collect-per-source"}
-    }
+    project["nodes"][0]["params"] = {"execution": {"failureMode": "collect-per-source"}}
 
     async def fake_dispatch(dispatch, fleet_match, *, node):
         return [], {
@@ -1163,13 +1170,10 @@ async def fake_dispatch(dispatch, fleet_match, *, node):
     assert states["multi-source-opencli::source-bilibili"]["status"] == "failed"
     assert states["multi-source-opencli::source-xiaohongshu"]["status"] == "failed"
     events = (
-        await client.get(
-            "/api/v1/workflows/runs/run-collect-per-source-all-failed/events"
-        )
+        await client.get("/api/v1/workflows/runs/run-collect-per-source-all-failed/events")
     ).json()["data"]
     assert not any(
-        event["nodeId"] == "multi-source-opencli"
-        and event["eventType"] == "completed"
+        event["nodeId"] == "multi-source-opencli" and event["eventType"] == "completed"
         for event in events
     )
 
@@ -1670,14 +1674,11 @@ async def fake_dispatch(dispatch, fleet_match):
     )
 
     assert response.status_code == 202
-    events = (
-        await client.get(f"/api/v1/workflows/runs/{run_id}/events")
-    ).json()["data"]
+    events = (await client.get(f"/api/v1/workflows/runs/{run_id}/events")).json()["data"]
     blocked = next(
         event
         for event in events
-        if event["nodeId"] == "tool-twitter-post"
-        and event["eventType"] == "blocked"
+        if event["nodeId"] == "tool-twitter-post" and event["eventType"] == "blocked"
     )
     assert blocked["blockReason"]["code"] == expected_code
     assert dispatched == []
@@ -1762,13 +1763,9 @@ async def fake_dispatch(dispatch, fleet_match):
     assert fleet_matches[0].selected.endpoint == "http://agent-x:19823"
 
     events = (
-        await client.get(
-            "/api/v1/workflows/runs/run-opencli-write-authorized/events"
-        )
+        await client.get("/api/v1/workflows/runs/run-opencli-write-authorized/events")
     ).json()["data"]
-    action_events = [
-        event for event in events if event["nodeId"] == "tool-twitter-post"
-    ]
+    action_events = [event for event in events if event["nodeId"] == "tool-twitter-post"]
     assert [event["eventType"] for event in action_events] == [
         "queued",
         "started",
@@ -1777,9 +1774,7 @@ async def fake_dispatch(dispatch, fleet_match):
         "tool_call_completed",
         "completed",
     ]
-    tool_call = next(
-        event for event in action_events if event["eventType"] == "tool_call_started"
-    )
+    tool_call = next(event for event in action_events if event["eventType"] == "tool_call_started")
     assert tool_call["details"]["resourceRequirement"]["mutationMode"] == "write"
     resource_resolution = tool_call["details"]["resourceResolution"]
     assert resource_resolution["status"] == "resolved"
@@ -2418,9 +2413,7 @@ async def __aexit__(self, exc_type, exc, tb):
         async def request(self, method, url, **kwargs):
             assert method == "GET"
             expected_query = (
-                {"format": "json", "q": "OpenCLI"}
-                if url.endswith("/bilibili")
-                else None
+                {"format": "json", "q": "OpenCLI"} if url.endswith("/bilibili") else None
             )
             assert kwargs["params"] == expected_query
             return FakeResponse(live_payloads[url])
diff --git a/tests/integration/test_workflow_patch_api.py b/tests/integration/test_workflow_patch_api.py
index 679d6f35..c76d4dc2 100644
--- a/tests/integration/test_workflow_patch_api.py
+++ b/tests/integration/test_workflow_patch_api.py
@@ -586,6 +586,64 @@ async def test_demand_draft_assembles_xiaohongshu_need_into_native_nodes(client)
     )
 
 
+@pytest.mark.asyncio
+async def test_demand_draft_routes_competitor_research_to_doubao(client):
+    project = _valid_workflow_project()
+    text = (
+        "围绕高吉星燕窝酸 DHA 藻油，持续发现同类竞品；"
+        "标签和搜索词必须由这条需求生成。先按每批 5 条轮转，"
+        "核验品牌、产品、配方、价格、渠道、监管、投诉、召回和检测风险；"
+        "只有含可读取具体 URL 的实体才能标记 grounded。"
+    )
+
+    response = await client.post(
+        "/api/v1/workflows/demand-draft",
+        json={"project": project, "text": text, "locale": "zh-CN"},
+    )
+
+    assert response.status_code == 200
+    data = response.json()["data"]
+    assert data["valid"] is True
+    assert data["missing_capabilities"] == []
+
+    adapters = {adapter["id"]: adapter for adapter in data["project"]["adapters"]}
+    adapter = adapters["source-doubao-research-capture"]
+    assert adapter["provider"] == "doubao_research"
+    assert adapter["config"]["channelType"] == "doubao_research"
+    assert adapter["config"]["settle_seconds"] == 240
+
+    nodes = {node["id"]: node for node in data["project"]["nodes"]}
+    source = nodes["source-doubao-research"]
+    assert source["ui"]["catalogId"] == "intelligence.source.doubao-research"
+    assert source["ui"]["label"] == "高吉星燕窝酸 DHA 藻油 · 需求研究"
+    assert source["params"]["question"] == text
+    assert source["params"]["rotationBatchSize"] == 5
+    assert source["params"]["demand"]["labelSource"] == "demand"
+    assert source["params"]["demand"]["labels"] == [
+        "高吉星燕窝酸 DHA 藻油",
+        "竞品",
+        "品牌",
+        "产品",
+        "配方",
+        "价格",
+        "渠道",
+        "监管",
+        "投诉",
+        "召回",
+        "检测",
+        "风险",
+    ]
+    assert "source-linux-do" not in nodes
+    assert "source-36kr" not in nodes
+
+    runtime_nodes = {
+        node["id"]: node for node in data["compile"]["plan"]["runtime"]["nodes"]
+    }
+    runtime_source = runtime_nodes["source-doubao-research"]
+    assert runtime_source["runtime"]["binding"]["binding_id"] == "workflow.source.fetch"
+    assert runtime_source["runtime"]["binding"]["input"]["channelType"] == "doubao_research"
+
+
 @pytest.mark.asyncio
 async def test_demand_draft_assembles_multi_source_need_through_merge(client):
     project = _valid_workflow_project()
diff --git a/tests/unit/agent_runtimes/test_base.py b/tests/unit/agent_runtimes/test_base.py
index 88d9f884..7ed09352 100644
--- a/tests/unit/agent_runtimes/test_base.py
+++ b/tests/unit/agent_runtimes/test_base.py
@@ -11,8 +11,11 @@
     RuntimeAdapter,
     RuntimeCapabilities,
     RuntimeInvocationError,
+    event_artifact,
+    event_audit,
     event_done,
     event_error,
+    event_evidence,
     event_started,
     event_state,
     event_text,
@@ -21,6 +24,7 @@
 )
 from backend.agent_runtimes.registry import (
     _REGISTRY,
+    available_runtime_capabilities,
     available_runtimes,
     get_runtime,
     list_runtime_types,
@@ -100,6 +104,12 @@ def test_event_error_default_error_type_none():
     assert ev["error_type"] is None
 
 
+def test_evidence_event_shapes_are_in_closed_protocol():
+    assert event_artifact("t1", {"kind": "report"})["type"] == "artifact"
+    assert event_evidence("t1", {"kind": "citation"})["type"] == "evidence"
+    assert event_audit("t1", {"action": "read"})["type"] == "audit"
+
+
 @pytest.mark.parametrize(
     "builder,args",
     [
@@ -108,6 +118,9 @@ def test_event_error_default_error_type_none():
         (event_tool_call, ("t1", "n")),
         (event_tool_result, ("t1", "n")),
         (event_state, ("t1", {})),
+        (event_artifact, ("t1", {})),
+        (event_evidence, ("t1", {})),
+        (event_audit, ("t1", {})),
         (event_done, ("t1",)),
         (event_error, ("t1", "m")),
     ],
@@ -260,6 +273,8 @@ def test_pi_adapter_registered_by_default_discovery():
     assert get_runtime("pi").runtime_type == "pi"
     assert "miniflow" in list_runtime_types()
     assert get_runtime("miniflow").runtime_type == "miniflow"
+    assert "prime-agent" in list_runtime_types()
+    assert get_runtime("prime-agent").runtime_type == "prime-agent"
 
 
 def test_available_runtimes_filters_by_is_available(monkeypatch):
@@ -282,6 +297,9 @@ def is_available(cls) -> bool:
     available = available_runtimes()
     assert "available_runtime" in available
     assert "unavailable_runtime" not in available
+    capabilities = available_runtime_capabilities()
+    assert capabilities["available_runtime"] == ["streaming"]
+    assert "unavailable_runtime" not in capabilities
 
 
 def test_available_runtimes_excludes_adapters_without_is_available():
diff --git a/tests/unit/agent_runtimes/test_codex_adapter.py b/tests/unit/agent_runtimes/test_codex_adapter.py
new file mode 100644
index 00000000..09971360
--- /dev/null
+++ b/tests/unit/agent_runtimes/test_codex_adapter.py
@@ -0,0 +1,188 @@
+"""Focused contract tests for the registered-Agent Codex runtime adapter."""
+
+import asyncio
+import sys
+from typing import Any
+
+import pytest
+
+from backend.agent_runtimes.base import AgentTask
+from backend.agent_runtimes.codex_adapter import CodexRuntimeAdapter
+from backend.agent_runtimes.registry import get_runtime, list_runtime_types
+
+_FAKE_CODEX = r'''
+import json
+import sys
+import time
+
+if "--version" in sys.argv:
+    print("codex-cli 1.2.3", flush=True)
+    raise SystemExit(0)
+
+def emit(value):
+    print(json.dumps(value), flush=True)
+
+emit({"type": "thread.started", "thread_id": "thread-1"})
+emit({"type": "item.started", "item": {"id": "call-1", "type": "command_execution", "command": "ls"}})
+emit({"type": "item.completed", "item": {"id": "call-1", "type": "command_execution", "command": "ls", "aggregated_output": "README", "exit_code": 0}})
+emit({"type": "item.completed", "item": {"type": "agent_message", "text": "hello"}})
+emit({"type": "turn.completed", "usage": {"input_tokens": 1}})
+'''
+
+_FAKE_CODEX_HANG = r'''
+import sys
+import time
+if "--version" in sys.argv:
+    print("codex-cli 1.2.3", flush=True)
+    raise SystemExit(0)
+time.sleep(30)
+'''
+
+
+def _write_fake(tmp_path, source: str) -> str:
+    path = tmp_path / "fake_codex.py"
+    path.write_text(source, encoding="utf-8")
+    return str(path)
+
+
+def _config(script: str, **overrides: Any) -> dict[str, Any]:
+    result: dict[str, Any] = {
+        "binary": sys.executable,
+        "args": [script],
+    }
+    result.update(overrides)
+    return result
+
+
+async def test_missing_binary_is_terminal_error():
+    events = [
+        event
+        async for event in CodexRuntimeAdapter().invoke(
+            AgentTask(task_id="missing", workflow="exec", config={"binary": "not-a-codex-binary"})
+        )
+    ]
+    assert events == [
+        {
+            "type": "error",
+            "task_id": "missing",
+            "message": "codex binary not found: 'not-a-codex-binary'",
+            "error_type": "FileNotFoundError",
+        }
+    ]
+
+
+async def test_readiness_reports_version_and_confined_paths(tmp_path):
+    script = _write_fake(tmp_path, _FAKE_CODEX)
+    readiness = await CodexRuntimeAdapter().readiness(
+        _config(script, project_root=str(tmp_path), cwd=str(tmp_path))
+    )
+    assert readiness.status == "ready"
+    assert readiness.runtime == "codex"
+    assert readiness.binary_present is True
+    assert readiness.version == "codex-cli 1.2.3"
+    assert readiness.permitted_project_root == str(tmp_path.resolve())
+    assert readiness.working_directory == str(tmp_path.resolve())
+    assert not hasattr(readiness, "api_key")
+
+
+def test_validate_config_accepts_registered_agent_config(tmp_path):
+    errors = CodexRuntimeAdapter().validate_config(
+        {
+            "binary": "codex",
+            "project_root": str(tmp_path),
+            "cwd": str(tmp_path),
+            "timeout_seconds": 120,
+            "permission_mode": "approval_required",
+        }
+    )
+    assert errors == []
+
+
+async def test_invalid_working_directory_is_blocked(tmp_path):
+    outside = tmp_path / "outside"
+    readiness = await CodexRuntimeAdapter().readiness(
+        {
+            "binary": sys.executable,
+            "project_root": str(tmp_path),
+            "cwd": str(outside),
+        }
+    )
+    assert readiness.status == "blocked"
+    assert readiness.reason_code == "invalid_path"
+
+
+async def test_event_normalization_and_terminal_result(tmp_path):
+    script = _write_fake(tmp_path, _FAKE_CODEX)
+    events = [
+        event
+        async for event in CodexRuntimeAdapter().invoke(
+            AgentTask(
+                task_id="happy",
+                workflow="exec",
+                input={"message": "say hello"},
+                config=_config(script, cwd=str(tmp_path), project_root=str(tmp_path)),
+            )
+        )
+    ]
+    assert [event["type"] for event in events] == [
+        "started",
+        "state",
+        "state",
+        "tool_call",
+        "tool_result",
+        "text",
+        "state",
+        "done",
+    ]
+    assert events[-1]["result"]["text"] == "hello"
+    assert events[-1]["result"]["codex_version"] == "codex-cli 1.2.3"
+    assert sum(event["type"] in {"done", "error"} for event in events) == 1
+
+
+async def test_timeout_stops_process_and_yields_typed_error(tmp_path):
+    script = _write_fake(tmp_path, _FAKE_CODEX_HANG)
+    events = [
+        event
+        async for event in CodexRuntimeAdapter().invoke(
+            AgentTask(
+                task_id="timeout",
+                workflow="exec",
+                config=_config(
+                    script,
+                    cwd=str(tmp_path),
+                    project_root=str(tmp_path),
+                    timeout_seconds=0.1,
+                ),
+            )
+        )
+    ]
+    assert events[-1]["type"] == "error"
+    assert events[-1]["error_type"] == "TimeoutError"
+    assert sum(event["type"] in {"done", "error"} for event in events) == 1
+
+
+async def test_cancellation_stops_process(tmp_path):
+    script = _write_fake(tmp_path, _FAKE_CODEX_HANG)
+    adapter = CodexRuntimeAdapter()
+    task = AgentTask(
+        task_id="cancel",
+        workflow="exec",
+        config=_config(script, cwd=str(tmp_path), project_root=str(tmp_path), timeout_seconds=30),
+    )
+    started = asyncio.Event()
+
+    async def consume():
+        async for event in adapter.invoke(task):
+            if event["type"] == "started":
+                started.set()
+
+    consumer = asyncio.create_task(consume())
+    await started.wait()
+    consumer.cancel()
+    with pytest.raises(asyncio.CancelledError):
+        await asyncio.wait_for(consumer, timeout=1)
+
+
+def test_codex_registered():
+    assert "codex" in list_runtime_types()
+    assert get_runtime("codex").runtime_type == "codex"
diff --git a/tests/unit/agent_runtimes/test_miniflow_adapter.py b/tests/unit/agent_runtimes/test_miniflow_adapter.py
index 172798e8..b56c6a0a 100644
--- a/tests/unit/agent_runtimes/test_miniflow_adapter.py
+++ b/tests/unit/agent_runtimes/test_miniflow_adapter.py
@@ -62,6 +62,25 @@ def test_miniflow_runtime_registered_and_available_by_default():
     assert "miniflow" in available_runtimes()
 
 
+
+async def test_builtin_readiness_workflow_is_deployment_independent(monkeypatch):
+    monkeypatch.delenv("MINIFLOW_WORKFLOW_ROOT", raising=False)
+    adapter = MiniFlowRuntimeAdapter()
+    task = AgentTask(
+        task_id="t-builtin",
+        workflow="builtin.read_only_readiness",
+        config={"_clock": _clock(), "_sleep": _no_sleep},
+    )
+
+    events = [event async for event in adapter.invoke(task)]
+
+    assert events[-1]["type"] == "done"
+    assert events[-1]["result"]["workflow"] == "operations-read-only-readiness"
+    assert [step["name"] for step in events[-1]["result"]["steps"]] == [
+        "inspect-runtime-root",
+        "verify-runtime-bundle",
+    ]
+
 def test_validate_config_rejects_bad_values():
     adapter = MiniFlowRuntimeAdapter()
 
diff --git a/tests/unit/agent_runtimes/test_prime_agent_adapter.py b/tests/unit/agent_runtimes/test_prime_agent_adapter.py
new file mode 100644
index 00000000..e1de9e58
--- /dev/null
+++ b/tests/unit/agent_runtimes/test_prime_agent_adapter.py
@@ -0,0 +1,51 @@
+from backend.agent_runtimes.base import AgentTask
+from backend.agent_runtimes.prime_agent_adapter import PrimeAgentRuntimeAdapter
+from backend.agent_runtimes.registry import get_runtime
+
+
+def test_prime_agent_is_registered_as_distinct_runtime():
+    adapter = get_runtime("prime-agent")
+
+    assert isinstance(adapter, PrimeAgentRuntimeAdapter)
+    assert adapter.binary_name == "prime-agent"
+    assert "subagents" in adapter.capabilities.names()
+    assert "model_selection" in adapter.capabilities.names()
+
+
+def test_prime_agent_rpc_argv_keeps_provider_and_model_separate():
+    adapter = PrimeAgentRuntimeAdapter()
+
+    argv = adapter._compose_argv(
+        {},
+        provider="openrouter",
+        model="anthropic/claude-sonnet",
+    )
+
+    assert argv == [
+        "prime-agent",
+        "--provider",
+        "openrouter",
+        "--model",
+        "anthropic/claude-sonnet",
+        "--mode",
+        "rpc",
+    ]
+
+
+def test_prime_agent_uses_runtime_neutral_prompt_contract():
+    adapter = PrimeAgentRuntimeAdapter()
+    task = AgentTask(
+        task_id="sales-run",
+        workflow="sales-agent",
+        instructions="Act as a sales researcher",
+        input={"prompt": "Research product demand"},
+        provider="openrouter",
+        model="anthropic/claude-sonnet",
+        required_capabilities=("streaming", "subagents"),
+    )
+
+    assert adapter._compose_request(task) == {
+        "type": "prompt",
+        "id": "sales-run",
+        "message": "Act as a sales researcher\n\nResearch product demand",
+    }
diff --git a/tests/unit/api/test_automation_starters.py b/tests/unit/api/test_automation_starters.py
new file mode 100644
index 00000000..cb164402
--- /dev/null
+++ b/tests/unit/api/test_automation_starters.py
@@ -0,0 +1,114 @@
+import pytest
+from fastapi import FastAPI
+from httpx import ASGITransport, AsyncClient
+from sqlalchemy import func, select
+
+from backend.api.v1.automations import router
+from backend.database import get_db
+from backend.models.automation import Automation
+from backend.models.identity import User, Workspace, WorkspaceMembership, WorkspaceRole
+from backend.security.identity import RequestIdentity, get_request_identity
+from backend.services.automation_starter_service import install_starters
+
+
+async def _authorized_client(db_session, *, role=WorkspaceRole.ADMIN, subject="starter-admin"):
+    user = User(subject=subject)
+    workspace = Workspace(name=f"Starter {subject}", slug=f"starter-{subject}")
+    db_session.add_all((user, workspace))
+    await db_session.flush()
+    db_session.add(
+        WorkspaceMembership(workspace_id=workspace.id, user_id=user.id, role=role)
+    )
+    await db_session.commit()
+
+    app = FastAPI()
+    app.include_router(router)
+
+    async def override_db():
+        yield db_session
+
+    async def override_identity():
+        return RequestIdentity(subject=subject)
+
+    app.dependency_overrides[get_db] = override_db
+    app.dependency_overrides[get_request_identity] = override_identity
+    return app, workspace, user
+
+
+@pytest.mark.asyncio
+async def test_install_starters_is_idempotent_and_stores_metadata(db_session):
+    app, workspace, user = await _authorized_client(db_session)
+    async with AsyncClient(transport=ASGITransport(app=app), base_url="http://test") as client:
+        first = await client.post(f"/workspaces/{workspace.id}/automations/starters/install")
+        second = await client.post(f"/workspaces/{workspace.id}/automations/starters/install")
+
+    assert first.status_code == second.status_code == 200
+    assert first.json()["data"]["created_count"] == 3
+    assert first.json()["data"]["skipped_count"] == 0
+    assert second.json()["data"]["created_count"] == 0
+    assert second.json()["data"]["skipped_count"] == 3
+    rows = (
+        await db_session.scalars(
+            select(Automation).where(Automation.workspace_id == workspace.id)
+        )
+    ).all()
+    assert {row.starter_key for row in rows} == {
+        "daily-run-brief",
+        "weekly-system-review",
+        "anomaly-follow-up",
+    }
+    assert all(row.executor == "codex" for row in rows)
+    assert all(row.approval_mode == "suggest_changes" for row in rows)
+    assert all(row.project["starter_key"] == row.starter_key for row in rows)
+    assert all(row.project["lineage"] == "first-party-agent-starter" for row in rows)
+    assert all(row.created_by_user_id == user.id for row in rows)
+    assert all(row.enabled is False for row in rows)
+    assert {row.schedule for row in rows} == {
+        "daily@09:00",
+        "weekly@09:00",
+        "weekdays@09:00",
+    }
+
+
+@pytest.mark.asyncio
+async def test_starter_install_requires_manage_agent_permission(db_session):
+    app, workspace, _ = await _authorized_client(
+        db_session, role=WorkspaceRole.VIEWER, subject="starter-viewer"
+    )
+    async with AsyncClient(transport=ASGITransport(app=app), base_url="http://test") as client:
+        response = await client.post(f"/workspaces/{workspace.id}/automations/starters/install")
+
+    assert response.status_code == 403
+    assert await db_session.scalar(
+        select(func.count()).select_from(Automation).where(Automation.workspace_id == workspace.id)
+    ) == 0
+
+
+@pytest.mark.asyncio
+async def test_failed_starter_pack_rolls_back_all_rows(db_session, monkeypatch):
+    user = User(subject="starter-failure")
+    workspace = Workspace(name="Starter failure", slug="starter-failure")
+    db_session.add_all((user, workspace))
+    await db_session.flush()
+
+    original_flush = db_session.flush
+    calls = 0
+
+    async def fail_on_second_flush(*args, **kwargs):
+        nonlocal calls
+        calls += 1
+        if calls == 2:
+            raise RuntimeError("simulated starter failure")
+        return await original_flush(*args, **kwargs)
+
+    monkeypatch.setattr(db_session, "flush", fail_on_second_flush)
+    with pytest.raises(RuntimeError, match="simulated starter failure"):
+        await install_starters(
+            db_session,
+            workspace_id=workspace.id,
+            created_by_user_id=user.id,
+        )
+
+    assert await db_session.scalar(
+        select(func.count()).select_from(Automation).where(Automation.workspace_id == workspace.id)
+    ) == 0
diff --git a/tests/unit/api/test_automations.py b/tests/unit/api/test_automations.py
index 10ebce69..d83c8afa 100644
--- a/tests/unit/api/test_automations.py
+++ b/tests/unit/api/test_automations.py
@@ -7,7 +7,7 @@
 from backend.security.identity import RequestIdentity, get_request_identity
 
 
-async def test_admin_can_create_list_and_pause_automation(db_session):
+async def test_admin_can_create_list_and_update_disabled_automation_draft(db_session):
     user = User(subject="automation-admin")
     workspace = Workspace(name="Automation", slug="automation")
     db_session.add_all((user, workspace))
@@ -33,11 +33,14 @@ async def override_identity():
             "executor": "codex",
             "schedule": "daily@09:00",
             "timezone": "Asia/Shanghai",
+            "enabled": False,
         })
         automation_id = created.json()["data"]["id"]
         listed = await client.get(f"/workspaces/{workspace.id}/automations")
-        paused = await client.patch(f"/workspaces/{workspace.id}/automations/{automation_id}", json={"enabled": False})
+        updated = await client.patch(f"/workspaces/{workspace.id}/automations/{automation_id}", json={"name": "Updated review"})
 
     assert created.status_code == 201
     assert listed.json()["data"][0]["name"] == "Daily review"
-    assert paused.json()["data"]["enabled"] is False
+    assert updated.json()["data"]["name"] == "Updated review"
+    assert updated.json()["data"]["enabled"] is False
+    assert updated.json()["data"]["revision"] == 2
diff --git a/tests/unit/api/test_internal_agent_runs.py b/tests/unit/api/test_internal_agent_runs.py
new file mode 100644
index 00000000..308ab62c
--- /dev/null
+++ b/tests/unit/api/test_internal_agent_runs.py
@@ -0,0 +1,163 @@
+from fastapi import HTTPException
+
+from backend import ws_agent_manager
+from backend.api.v1 import internal_agent_runs
+from backend.models import (
+    EdgeNode,
+    OperationsAgentIdentity,
+    OperationsAgentRun,
+    PublishedOperationsAgentVersion,
+    Team,
+    User,
+    Workspace,
+)
+
+
+async def _seed_run(db_session, *, status: str) -> OperationsAgentRun:
+    user = User(subject=f"internal-dispatch-{status}")
+    workspace = Workspace(name=f"Internal {status}", slug=f"internal-{status}")
+    db_session.add_all((user, workspace))
+    await db_session.flush()
+    team = Team(workspace_id=workspace.id, name="Operations", slug="operations")
+    db_session.add(team)
+    await db_session.flush()
+    agent = OperationsAgentIdentity(
+        workspace_id=workspace.id,
+        owning_team_id=team.id,
+        name=f"Internal {status}",
+        current_profile_version=1,
+        current_published_version=1,
+    )
+    db_session.add(agent)
+    await db_session.flush()
+    db_session.add(
+        PublishedOperationsAgentVersion(
+            operations_agent_id=agent.id,
+            version=1,
+            draft_revision=1,
+            instructions="Read only",
+            model_configuration={
+                "agent_contract": {
+                    "schema_version": "agent.contract.v2",
+                    "role": "scheduled_reviewer",
+                    "input_schema": {"type": "object"},
+                    "output_schema": {"type": "object"},
+                    "state_schema": {"type": "object"},
+                    "required_capabilities": ["streaming"],
+                    "tool_policy": {},
+                    "budget": {},
+                    "quality_gates": [],
+                    "evidence_requirements": [],
+                },
+                "runtime_binding": {
+                    "schema_version": "agent.runtime-binding.v2",
+                    "workflow": "builtin.read_only_readiness",
+                    "preferred_agent_urls": ["http://internal-agent:19823"],
+                    "preferred_runtimes": ["miniflow"],
+                    "model_binding": None,
+                    "config": {"timeout_seconds": 60},
+                    "dispatch_timeout_seconds": 60,
+                },
+            },
+            tool_configuration={},
+            published_by_user_id=user.id,
+            reason="Internal dispatch test",
+        )
+    )
+    db_session.add(
+        EdgeNode(
+            url="http://internal-agent:19823",
+            label="Internal runtime",
+            protocol="ws",
+            mode="cdp",
+            node_type="docker",
+            status="online",
+            runtimes=["miniflow"],
+            runtime_capabilities={"miniflow": ["streaming"]},
+        )
+    )
+    run = OperationsAgentRun(
+        workspace_id=workspace.id,
+        operations_agent_id=agent.id,
+        published_version=1,
+        profile_version=1,
+        trigger_type="scheduled",
+        target_resource_type="automation",
+        target_resource_id="automation-test",
+        input_payload={},
+        state_payload={},
+        status=status,
+        started_by_user_id=user.id,
+    )
+    db_session.add(run)
+    await db_session.commit()
+    return run
+
+
+def _use_session(db_session, monkeypatch) -> None:
+    class SessionContext:
+        async def __aenter__(self):
+            return db_session
+
+        async def __aexit__(self, *_exc):
+            return False
+
+    monkeypatch.setattr(internal_agent_runs, "AsyncSessionLocal", SessionContext)
+
+
+async def test_terminal_run_returns_without_connected_agent(db_session, monkeypatch):
+    run = await _seed_run(db_session, status="completed")
+    _use_session(db_session, monkeypatch)
+    monkeypatch.setattr(ws_agent_manager, "is_connected", lambda _url: False)
+    scheduled = []
+    monkeypatch.setattr(internal_agent_runs, "schedule_operations_agent_run", scheduled.append)
+
+    response = await internal_agent_runs.dispatch_scheduled_agent_run(run.id)
+
+    assert response.data.status == "completed"
+    assert scheduled == []
+
+
+async def test_queued_run_stays_queued_when_agent_disconnected(db_session, monkeypatch):
+    run = await _seed_run(db_session, status="queued")
+    _use_session(db_session, monkeypatch)
+    monkeypatch.setattr(ws_agent_manager, "is_connected", lambda _url: False)
+    scheduled = []
+    monkeypatch.setattr(internal_agent_runs, "schedule_operations_agent_run", scheduled.append)
+
+    try:
+        await internal_agent_runs.dispatch_scheduled_agent_run(run.id)
+    except HTTPException as exc:
+        assert exc.status_code == 503
+    else:
+        raise AssertionError("disconnected queued run must return 503")
+    assert scheduled == []
+    await db_session.refresh(run)
+    assert run.status == "queued"
+
+
+async def test_running_duplicate_returns_without_dispatch(db_session, monkeypatch):
+    run = await _seed_run(db_session, status="running")
+    _use_session(db_session, monkeypatch)
+    monkeypatch.setattr(ws_agent_manager, "is_connected", lambda _url: False)
+    scheduled = []
+    monkeypatch.setattr(internal_agent_runs, "schedule_operations_agent_run", scheduled.append)
+
+    response = await internal_agent_runs.dispatch_scheduled_agent_run(run.id)
+
+    assert response.data.status == "running"
+    assert scheduled == []
+
+
+async def test_connected_queued_run_is_scheduled_without_waiting(db_session, monkeypatch):
+    run = await _seed_run(db_session, status="queued")
+    _use_session(db_session, monkeypatch)
+    monkeypatch.setattr(ws_agent_manager, "is_connected", lambda _url: True)
+    scheduled = []
+    monkeypatch.setattr(internal_agent_runs, "schedule_operations_agent_run", scheduled.append)
+
+    response = await internal_agent_runs.dispatch_scheduled_agent_run(run.id)
+
+    assert response.data.status == "queued"
+    assert scheduled == [run.id]
+    assert response.data.execution_binding["runtime"] == "miniflow"
diff --git a/tests/unit/api/test_nodes_install_script.py b/tests/unit/api/test_nodes_install_script.py
index 70e8fce5..2cb479bf 100644
--- a/tests/unit/api/test_nodes_install_script.py
+++ b/tests/unit/api/test_nodes_install_script.py
@@ -1,5 +1,11 @@
 """Install-script rendering tests for edge node bootstrap."""
 
+import io
+import os
+import subprocess
+import sys
+import tarfile
+
 import pytest
 
 from backend.api.v1.nodes import _install_script_template
@@ -140,3 +146,70 @@ async def test_managed_opencli_patch_is_served_to_native_agents(client):
     assert "OPENCLI_ADMIN_REMOTE_DAEMON_ROUTE_V1" in response.text
     assert "'dist', 'src', 'daemon.js'" in response.text
     assert "'dist', 'src', 'browser', 'daemon-transport.js'" in response.text
+
+
+def test_native_installer_downloads_and_extracts_runtime_bundle():
+    body = (
+        __import__("pathlib").Path(__file__).parents[3]
+        / "scripts"
+        / "install-agent.sh"
+    ).read_text(encoding="utf-8")
+
+    assert "/api/v1/nodes/install/agent-runtime.tar.gz" in body
+    assert 'tar --no-same-owner --no-same-permissions -xzf "$RUNTIME_BUNDLE"' in body
+    assert 'backend/agent_runtimes/*.py' in body
+    assert 'backend/miniflow/*.py' in body
+    assert 'backend/security/*.py' in body
+
+
+@pytest.mark.asyncio
+async def test_runtime_bundle_contains_native_agent_packages(client):
+    response = await client.get("/api/v1/nodes/install/agent-runtime.tar.gz")
+
+    assert response.status_code == 200
+    assert response.headers["content-type"].startswith("application/gzip")
+    with tarfile.open(fileobj=io.BytesIO(response.content), mode="r:gz") as archive:
+        names = set(archive.getnames())
+
+    assert "backend/agent_runtimes/registry.py" in names
+    assert "backend/agent_runtimes/opentabs_adapter.py" in names
+    assert "backend/agent_runtimes/prime_agent_adapter.py" in names
+    assert "backend/miniflow/runner.py" in names
+    assert "backend/security/url_guard.py" in names
+    assert all(
+        name.startswith(("backend/agent_runtimes/", "backend/miniflow/", "backend/security/"))
+        for name in names
+    )
+
+
+@pytest.mark.asyncio
+async def test_runtime_bundle_imports_in_native_agent_layout(client, tmp_path):
+    response = await client.get("/api/v1/nodes/install/agent-runtime.tar.gz")
+    assert response.status_code == 200
+
+    backend_root = tmp_path / "backend"
+    backend_root.mkdir()
+    (backend_root / "__init__.py").touch()
+    with tarfile.open(fileobj=io.BytesIO(response.content), mode="r:gz") as archive:
+        archive.extractall(tmp_path, filter="data")
+
+    env = os.environ.copy()
+    env["PYTHONPATH"] = str(tmp_path)
+    result = subprocess.run(
+        [
+            sys.executable,
+            "-c",
+            (
+                "from backend.agent_runtimes.registry import list_runtime_types; "
+                "assert {'bbx', 'miniflow', 'opentabs', 'pi', 'prime-agent'} "
+                "<= set(list_runtime_types())"
+            ),
+        ],
+        cwd=tmp_path,
+        env=env,
+        capture_output=True,
+        text=True,
+        check=False,
+    )
+
+    assert result.returncode == 0, result.stderr
diff --git a/tests/unit/api/test_nodes_upsert.py b/tests/unit/api/test_nodes_upsert.py
index bd4e13fa..9b5b3263 100644
--- a/tests/unit/api/test_nodes_upsert.py
+++ b/tests/unit/api/test_nodes_upsert.py
@@ -17,14 +17,24 @@
 @pytest.mark.asyncio
 async def test_upsert_node_persists_runtimes_on_create(db_session):
     node = await _upsert_node(
-        db_session, "http://agent:1", runtimes=["pi", "stub"]
+        db_session,
+        "http://agent:1",
+        runtimes=["pi", "prime-agent"],
+        runtime_capabilities={
+            "pi": ["streaming"],
+            "prime-agent": ["streaming", "subagents"],
+        },
     )
     await db_session.commit()
 
     result = await db_session.execute(select(EdgeNode).where(EdgeNode.url == "http://agent:1"))
     fetched = result.scalar_one()
-    assert fetched.runtimes == ["pi", "stub"]
-    assert node.runtimes == ["pi", "stub"]
+    assert fetched.runtimes == ["pi", "prime-agent"]
+    assert fetched.runtime_capabilities == {
+        "pi": ["streaming"],
+        "prime-agent": ["streaming", "subagents"],
+    }
+    assert node.runtime_capabilities == fetched.runtime_capabilities
 
 
 @pytest.mark.asyncio
@@ -36,6 +46,7 @@ async def test_upsert_node_without_runtimes_leaves_column_null(db_session):
     fetched = result.scalar_one()
     assert fetched.runtimes is None
     assert node.runtimes is None
+    assert node.runtime_capabilities is None
 
 
 @pytest.mark.asyncio
@@ -54,10 +65,21 @@ async def test_upsert_node_reregister_without_runtimes_preserves_existing(db_ses
     """A re-register handshake that omits `runtimes` (e.g. an older agent
     build, or the HTTP registration path which never sends it) must not wipe
     out a previously-advertised runtimes list."""
-    await _upsert_node(db_session, "http://agent:4", runtimes=["pi"])
+    await _upsert_node(
+        db_session,
+        "http://agent:4",
+        runtimes=["pi"],
+        runtime_capabilities={"pi": ["streaming"]},
+    )
     await db_session.commit()
 
-    updated = await _upsert_node(db_session, "http://agent:4", runtimes=None)
+    updated = await _upsert_node(
+        db_session,
+        "http://agent:4",
+        runtimes=None,
+        runtime_capabilities=None,
+    )
     await db_session.commit()
 
     assert updated.runtimes == ["pi"]
+    assert updated.runtime_capabilities == {"pi": ["streaming"]}
diff --git a/tests/unit/api/test_operations_agents.py b/tests/unit/api/test_operations_agents.py
index 3b117055..9bc40fbf 100644
--- a/tests/unit/api/test_operations_agents.py
+++ b/tests/unit/api/test_operations_agents.py
@@ -1,7 +1,6 @@
 from fastapi import FastAPI
 from httpx import ASGITransport, AsyncClient
 from sqlalchemy import select
-
 from backend import ws_agent_manager
 from backend.api.v1.operations_agents import router
 from backend.database import get_db
@@ -59,7 +58,8 @@ async def _add_member(db_session, workspace: Workspace, role: WorkspaceRole, suf
 
 def _agent_contract() -> dict:
     return {
-        "schema_version": "agent.contract.v1",
+        "schema_version": "agent.contract.v2",
+        "role": "operations_reviewer",
         "input_schema": {
             "type": "object",
             "properties": {"target_id": {"type": "string"}},
@@ -74,6 +74,11 @@ def _agent_contract() -> dict:
             "type": "object",
             "properties": {"last_target_id": {"type": "string"}},
         },
+        "required_capabilities": ["streaming"],
+        "tool_policy": {},
+        "budget": {},
+        "quality_gates": [],
+        "evidence_requirements": [],
     }
 
 
@@ -88,15 +93,17 @@ async def _runtime_binding(db_session, monkeypatch) -> dict:
             node_type="shell",
             status="online",
             runtimes=["pi"],
+            runtime_capabilities={"pi": ["streaming", "tool_events"]},
         )
     )
     await db_session.commit()
     monkeypatch.setattr(ws_agent_manager, "is_connected", lambda url: url == agent_url)
     return {
-        "schema_version": "agent.runtime-binding.v1",
-        "agent_url": agent_url,
-        "runtime": "pi",
+        "schema_version": "agent.runtime-binding.v2",
         "workflow": "operations-agent",
+        "preferred_agent_urls": [agent_url],
+        "preferred_runtimes": ["pi"],
+        "model_binding": None,
         "config": {},
     }
 
@@ -118,6 +125,49 @@ async def test_maintainer_creates_agent_with_observe_only_profile(db_session):
     assert data["disabled"] is False
 
 
+
+async def test_maintainer_creates_agent_in_only_workspace_team_when_owner_is_omitted(
+    db_session,
+):
+    _, workspace, team = await _seed_member(db_session, WorkspaceRole.MAINTAINER)
+    client = await _client(db_session, "maintainer-primary")
+
+    async with client:
+        response = await client.post(
+            f"/workspaces/{workspace.id}/operations-agents",
+            json={"name": "Runtime readiness advisor"},
+        )
+
+    assert response.status_code == 201
+    assert response.json()["data"]["owning_team_id"] == team.id
+
+
+async def test_multi_team_workspace_lists_choices_and_requires_explicit_owner(db_session):
+    _, workspace, first_team = await _seed_member(
+        db_session,
+        WorkspaceRole.MAINTAINER,
+    )
+    second_team = Team(workspace_id=workspace.id, name="Research", slug="research")
+    db_session.add(second_team)
+    await db_session.commit()
+    client = await _client(db_session, "maintainer-primary")
+
+    async with client:
+        teams = await client.get(
+            f"/workspaces/{workspace.id}/operations-agents/teams"
+        )
+        omitted = await client.post(
+            f"/workspaces/{workspace.id}/operations-agents",
+            json={"name": "Needs explicit Team"},
+        )
+
+    assert teams.status_code == 200
+    assert {row["id"] for row in teams.json()["data"]} == {
+        first_team.id,
+        second_team.id,
+    }
+    assert omitted.status_code == 422
+
 async def test_maintainer_can_assign_suggest_but_not_automatic(db_session):
     _, workspace, team = await _seed_member(db_session, WorkspaceRole.MAINTAINER)
     client = await _client(db_session, "maintainer-primary")
@@ -250,7 +300,10 @@ async def test_disabling_agent_revokes_profile_and_prevents_new_assignments(
             json={
                 "revision": 1,
                 "instructions": "Observe one target",
-                "model_configuration": {"runtime_binding": runtime_binding},
+                "model_configuration": {
+                    "agent_contract": _agent_contract(),
+                    "runtime_binding": runtime_binding,
+                },
             },
         )
         await client.post(
@@ -259,7 +312,11 @@ async def test_disabling_agent_revokes_profile_and_prevents_new_assignments(
         )
         started = await client.post(
             f"/workspaces/{workspace.id}/operations-agents/{agent_id}/runs",
-            json={"target_resource_type": "plan", "target_resource_id": "daily-news"},
+            json={
+                "target_resource_type": "plan",
+                "target_resource_id": "daily-news",
+                "input_payload": {"target_id": "daily-news"},
+            },
         )
         disabled = await client.patch(
             f"/workspaces/{workspace.id}/operations-agents/{agent_id}",
@@ -420,7 +477,7 @@ async def test_agent_draft_reads_and_rejects_stale_revision(db_session):
     assert current.json()["data"]["instructions"] == "First reviewed draft"
 
 
-async def test_agent_contract_v1_is_validated_and_published_in_versioned_configuration(db_session):
+async def test_agent_contract_v2_is_validated_and_published_in_versioned_configuration(db_session):
     _, workspace, team = await _seed_member(db_session, WorkspaceRole.MAINTAINER)
     client = await _client(db_session, "maintainer-primary")
     contract = _agent_contract()
@@ -466,7 +523,7 @@ async def test_agent_contract_v1_is_validated_and_published_in_versioned_configu
     assert selected_version.json()["data"] == published.json()["data"]
 
 
-async def test_agent_contract_v1_rejects_invalid_json_schema_before_draft_write(db_session):
+async def test_agent_contract_v2_rejects_invalid_json_schema_before_draft_write(db_session):
     _, workspace, team = await _seed_member(db_session, WorkspaceRole.MAINTAINER)
     client = await _client(db_session, "maintainer-primary")
     contract = _agent_contract()
@@ -504,11 +561,11 @@ async def test_agent_contract_v1_rejects_invalid_json_schema_before_draft_write(
                 "model_configuration": {"agent_contract": contract},
             },
         )
-        unsupported_runtime = await client.put(
+        legacy_runtime_binding = await client.put(
             f"/workspaces/{workspace.id}/operations-agents/{agent_id}/draft",
             json={
                 "revision": 1,
-                "instructions": "V1 instructions must be consumed",
+                "instructions": "Legacy bindings must be migrated before use",
                 "model_configuration": {
                     "runtime_binding": {
                         "schema_version": "agent.runtime-binding.v1",
@@ -527,10 +584,11 @@ async def test_agent_contract_v1_rejects_invalid_json_schema_before_draft_write(
                 "instructions": "Workspace config must not choose the edge process",
                 "model_configuration": {
                     "runtime_binding": {
-                        "schema_version": "agent.runtime-binding.v1",
-                        "agent_url": "http://agent.test:19823",
-                        "runtime": "pi",
+                        "schema_version": "agent.runtime-binding.v2",
                         "workflow": "status",
+                        "preferred_agent_urls": ["http://agent.test:19823"],
+                        "preferred_runtimes": ["pi"],
+                        "model_binding": None,
                         "config": {
                             "binary": "python",
                             "args": ["-c", "print('unsafe')"],
@@ -543,7 +601,7 @@ async def test_agent_contract_v1_rejects_invalid_json_schema_before_draft_write(
     assert response.status_code == 422
     assert remote_reference.status_code == 422
     assert remote_dynamic_reference.status_code == 422
-    assert unsupported_runtime.status_code == 422
+    assert legacy_runtime_binding.status_code == 422
     assert unsafe_launch_config.status_code == 422
     draft = await db_session.scalar(
         select(OperationsAgentDraft).where(OperationsAgentDraft.operations_agent_id == agent_id)
@@ -552,7 +610,7 @@ async def test_agent_contract_v1_rejects_invalid_json_schema_before_draft_write(
     assert draft.instructions == ""
 
 
-async def test_publish_revalidates_persisted_agent_contract_v1(db_session):
+async def test_publish_revalidates_persisted_agent_contract_v2(db_session):
     _, workspace, team = await _seed_member(db_session, WorkspaceRole.MAINTAINER)
     client = await _client(db_session, "maintainer-primary")
 
@@ -570,7 +628,7 @@ async def test_publish_revalidates_persisted_agent_contract_v1(db_session):
     draft.model_configuration = {
         "agent_contract": {
             **_agent_contract(),
-            "schema_version": "agent.contract.v2",
+            "schema_version": "agent.contract.v3",
         }
     }
     await db_session.commit()
@@ -643,6 +701,8 @@ async def test_agent_run_validates_and_persists_contract_input_and_state(db_sess
     assert started.status_code == 201
     assert started.json()["data"]["input_payload"] == {"target_id": "daily-news"}
     assert started.json()["data"]["state_payload"] == {"last_target_id": "yesterday"}
+    assert started.json()["data"]["execution_binding"]["runtime"] == "pi"
+    assert started.json()["data"]["evidence_payload"] is None
     stored = await db_session.get(OperationsAgentRun, started.json()["data"]["id"])
     assert stored.input_payload == {"target_id": "daily-news"}
     assert stored.state_payload == {"last_target_id": "yesterday"}
@@ -660,7 +720,11 @@ async def test_agent_run_requires_published_runtime_binding(db_session):
         agent_id = created.json()["data"]["id"]
         await client.put(
             f"/workspaces/{workspace.id}/operations-agents/{agent_id}/draft",
-            json={"revision": 1, "instructions": "Inspect one target"},
+            json={
+                "revision": 1,
+                "instructions": "Inspect one target",
+                "model_configuration": {"agent_contract": _agent_contract()},
+            },
         )
         await client.post(
             f"/workspaces/{workspace.id}/operations-agents/{agent_id}/versions",
@@ -676,7 +740,7 @@ async def test_agent_run_requires_published_runtime_binding(db_session):
 
     assert response.status_code == 409
     assert response.json()["detail"] == (
-        "Published Agent Version requires an AgentRuntimeBindingV1"
+        "Published Agent Version requires an AgentRuntimeBindingV2"
     )
 
 
@@ -698,7 +762,10 @@ async def test_automatic_agent_behavior_release_requires_admin_and_governed_gate
             json={
                 "revision": 1,
                 "instructions": "Adjust qualified schedules",
-                "model_configuration": {"runtime_binding": runtime_binding},
+                "model_configuration": {
+                    "agent_contract": _agent_contract(),
+                    "runtime_binding": runtime_binding,
+                },
             },
         )
         await admin_client.post(
@@ -726,7 +793,11 @@ async def test_automatic_agent_behavior_release_requires_admin_and_governed_gate
         )
         blocked_run = await admin_publish_client.post(
             f"/workspaces/{workspace.id}/operations-agents/{agent_id}/runs",
-            json={"target_resource_type": "plan", "target_resource_id": "daily-news"},
+            json={
+                "target_resource_type": "plan",
+                "target_resource_id": "daily-news",
+                "input_payload": {"target_id": "daily-news"},
+            },
         )
 
     assert denied.status_code == 403
@@ -755,7 +826,10 @@ async def test_operator_run_binds_published_and_profile_versions_then_pauses(
             json={
                 "revision": 1,
                 "instructions": "Inspect one target",
-                "model_configuration": {"runtime_binding": runtime_binding},
+                "model_configuration": {
+                    "agent_contract": _agent_contract(),
+                    "runtime_binding": runtime_binding,
+                },
             },
         )
         await admin_client.post(
@@ -767,7 +841,11 @@ async def test_operator_run_binds_published_and_profile_versions_then_pauses(
     async with operator_client:
         started = await operator_client.post(
             f"/workspaces/{workspace.id}/operations-agents/{agent_id}/runs",
-            json={"target_resource_type": "plan", "target_resource_id": "daily-news"},
+            json={
+                "target_resource_type": "plan",
+                "target_resource_id": "daily-news",
+                "input_payload": {"target_id": "daily-news"},
+            },
         )
         run_id = started.json()["data"]["id"]
     admin_profile_client = await _client(db_session, "admin-primary")
@@ -781,7 +859,10 @@ async def test_operator_run_binds_published_and_profile_versions_then_pauses(
             json={
                 "revision": 2,
                 "instructions": "Inspect two targets",
-                "model_configuration": {"runtime_binding": runtime_binding},
+                "model_configuration": {
+                    "agent_contract": _agent_contract(),
+                    "runtime_binding": runtime_binding,
+                },
             },
         )
         await admin_profile_client.post(
@@ -845,3 +926,5 @@ async def test_agent_without_published_version_cannot_run(db_session):
         )
 
     assert response.status_code == 409
+
+
diff --git a/tests/unit/channels/test_doubao_research_channel.py b/tests/unit/channels/test_doubao_research_channel.py
index d2475c92..ef6820a4 100644
--- a/tests/unit/channels/test_doubao_research_channel.py
+++ b/tests/unit/channels/test_doubao_research_channel.py
@@ -22,17 +22,12 @@ def test_conversation_url_extracts_chat_id():
         '[{"Status": "Connected", "Url": '
         '"https://www.doubao.com/chat/38436240748612354", "Title": "x"}]'
     )
-    assert (
-        _conversation_url(status)
-        == "https://www.doubao.com/chat/38436240748612354"
-    )
+    assert _conversation_url(status) == "https://www.doubao.com/chat/38436240748612354"
 
 
 def test_conversation_url_ignores_root_chat():
     # A freshly opened /chat page has no conversation id yet — must not be picked up.
-    status = (
-        '[{"Status": "Connected", "Url": "https://www.doubao.com/chat", "Title": "x"}]'
-    )
+    status = '[{"Status": "Connected", "Url": "https://www.doubao.com/chat", "Title": "x"}]'
     assert _conversation_url(status) == ""
 
 
@@ -53,6 +48,39 @@ async def fake_run(command):
     assert result.success
     assert result.items[0]["title"] == "麻将机"
     assert result.items[0]["citations"] == [{"url": "https://example.com/source"}]
+
+
+@pytest.mark.asyncio
+async def test_collect_reads_settled_research_answer(monkeypatch):
+    calls = []
+    sleeps = []
+
+    async def fake_run(command):
+        calls.append(command[2])
+        if command[2] == "ask":
+            return 0, '[{"Role":"assistant","Text":"正在理解任务要求"}]', ""
+        return 0, '[{"Role":"assistant","Text":"最终报告 https://example.com/source"}]', ""
+
+    async def fake_sleep(seconds):
+        sleeps.append(seconds)
+
+    monkeypatch.setattr("backend.channels.doubao_research_channel._run_doubao_command", fake_run)
+    monkeypatch.setattr("backend.channels.doubao_research_channel.asyncio.sleep", fake_sleep)
+
+    result = await DoubaoResearchChannel().collect(
+        {
+            "question": "测试",
+            "settle_seconds": 35,
+            "capture_conversation_url": False,
+        },
+        {},
+    )
+
+    assert result.success
+    assert calls == ["ask", "read"]
+    assert sleeps == [35]
+    assert result.items[0]["content"] == "最终报告 https://example.com/source"
+    assert result.items[0]["citations"] == [{"url": "https://example.com/source"}]
     assert result.metadata["citation_count"] == 1
 
 
@@ -84,10 +112,14 @@ async def fake_run(command):
         if command[2] == "ask":
             return 0, '[{"Role":"assistant","Text":"回答"}]', ""
         if command[2] == "status":
-            return 0, (
-                '[{"Status": "Connected", "Url": '
-                '"https://www.doubao.com/chat/12345", "Title": "t"}]'
-            ), ""
+            return (
+                0,
+                (
+                    '[{"Status": "Connected", "Url": '
+                    '"https://www.doubao.com/chat/12345", "Title": "t"}]'
+                ),
+                "",
+            )
         return 0, "", ""
 
     monkeypatch.setattr("backend.channels.doubao_research_channel._run_doubao_command", fake_run)
@@ -117,10 +149,14 @@ async def fake_run(command):
 @pytest.mark.asyncio
 async def test_collect_classifies_captcha_block(monkeypatch):
     async def fake_run(command):
-        return 1, "", (
-            "ok: false\nerror:\n  code: COMMAND_EXEC\n"
-            "  message: Doubao blocked the request with a verification challenge\n"
-            "  help: 'Detected challenge signal: iframe[src*=\"captcha\"]'"
+        return (
+            1,
+            "",
+            (
+                "ok: false\nerror:\n  code: COMMAND_EXEC\n"
+                "  message: Doubao blocked the request with a verification challenge\n"
+                "  help: 'Detected challenge signal: iframe[src*=\"captcha\"]'"
+            ),
         )
 
     monkeypatch.setattr("backend.channels.doubao_research_channel._run_doubao_command", fake_run)
@@ -130,6 +166,22 @@ async def fake_run(command):
     assert result.error_type == "captcha_challenge"
 
 
+@pytest.mark.asyncio
+async def test_collect_classifies_adapter_timeout(monkeypatch):
+    async def fake_run(command):
+        raise TimeoutError("adapter timed out")
+
+    monkeypatch.setattr(
+        "backend.channels.doubao_research_channel._run_doubao_command",
+        fake_run,
+    )
+    result = await DoubaoResearchChannel().collect({"question": "测试"}, {})
+
+    assert not result.success
+    assert result.error_type == "TimeoutError"
+    assert result.error == "Doubao request timed out"
+
+
 @pytest.mark.asyncio
 async def test_collect_does_not_classify_generic_error(monkeypatch):
     async def fake_run(command):
@@ -148,3 +200,97 @@ def test_source_schema_accepts_doubao_research_channel():
     )
     assert source.channel_type == "doubao_research"
 
+
+@pytest.mark.asyncio
+async def test_health_check_accepts_unknown_login_with_authenticated_provider_probe(monkeypatch):
+    calls = []
+
+    async def fake_run(command):
+        calls.append(command)
+        if command[2] == "status":
+            return (
+                0,
+                (
+                    '[{"Status":"Connected","Login":"Unknown","Title":"豆包工作 - '
+                    '字节跳动旗下 AI 智能助手","Url":"https://www.doubao.com/chat/?from_login=1"}]'
+                ),
+                "",
+            )
+        return 0, '[{"Name":"authenticated account","Email":"user@example.com"}]', ""
+
+    monkeypatch.setattr("backend.channels.doubao_research_channel._run_doubao_command", fake_run)
+
+    assert await DoubaoResearchChannel().health_check({"site_session": "persistent"})
+    assert [command[2] for command in calls] == ["status", "whoami"]
+    assert all(command[-2:] == ["--site-session", "persistent"] for command in calls)
+
+
+@pytest.mark.asyncio
+async def test_health_check_rejects_explicitly_logged_out_status(monkeypatch):
+    async def fake_run(command):
+        return 0, '[{"Status":"Connected","Login":"false"}]', ""
+
+    monkeypatch.setattr("backend.channels.doubao_research_channel._run_doubao_command", fake_run)
+
+    assert not await DoubaoResearchChannel().health_check()
+
+
+@pytest.mark.asyncio
+async def test_health_check_rejects_captcha_status(monkeypatch):
+    async def fake_run(command):
+        return 1, "", "Doubao blocked the request with a verification challenge"
+
+    monkeypatch.setattr("backend.channels.doubao_research_channel._run_doubao_command", fake_run)
+
+    assert not await DoubaoResearchChannel().health_check()
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    "status",
+    [
+        '[{"Status":"Connected","Login":"Unknown","Title":"Other provider","Url":"https://example.com/chat"}]',
+        '[{"Status":"Connected","Login":"Unknown","Title":"豆包工作","Url":"https://www.doubao.com/chat/"},'
+        '{"Status":"Connected","Login":"Unknown","Title":"豆包工作","Url":"https://www.doubao.com/chat/"}]',
+    ],
+)
+async def test_health_check_rejects_wrong_or_ambiguous_workspace(monkeypatch, status):
+    async def fake_run(command):
+        return 0, status, ""
+
+    monkeypatch.setattr("backend.channels.doubao_research_channel._run_doubao_command", fake_run)
+
+    assert not await DoubaoResearchChannel().health_check()
+
+
+@pytest.mark.asyncio
+async def test_health_check_rejects_provider_probe_error(monkeypatch):
+    async def fake_run(command):
+        if command[2] == "status":
+            return (
+                0,
+                (
+                    '[{"Status":"Connected","Login":"Unknown","Title":"豆包工作 - '
+                    '字节跳动旗下 AI 智能助手","Url":"https://www.doubao.com/chat/"}]'
+                ),
+                "",
+            )
+        return 1, "", "whoami failed"
+
+    monkeypatch.setattr("backend.channels.doubao_research_channel._run_doubao_command", fake_run)
+
+    assert not await DoubaoResearchChannel().health_check()
+
+
+@pytest.mark.asyncio
+async def test_health_check_preserves_explicit_logged_in_path(monkeypatch):
+    calls = []
+
+    async def fake_run(command):
+        calls.append(command)
+        return 0, '[{"Status":"Connected","Login":"authenticated"}]', ""
+
+    monkeypatch.setattr("backend.channels.doubao_research_channel._run_doubao_command", fake_run)
+
+    assert await DoubaoResearchChannel().health_check()
+    assert [command[2] for command in calls] == ["status"]
diff --git a/tests/unit/pipeline/test_ai_processor.py b/tests/unit/pipeline/test_ai_processor.py
index 29f2ca16..7bc1c3d9 100644
--- a/tests/unit/pipeline/test_ai_processor.py
+++ b/tests/unit/pipeline/test_ai_processor.py
@@ -59,6 +59,51 @@ async def test_process_with_ai_enriches_records():
     assert result == 2
 
 
+@pytest.mark.asyncio
+async def test_process_with_ai_keeps_failed_records_unenriched():
+    records = [MagicMock(), MagicMock()]
+    for record in records:
+        record.ai_enrichment = None
+        record.status = "normalized"
+
+    mock_processor = AsyncMock()
+    mock_processor.process = AsyncMock(
+        return_value=ProcessingResult(
+            success=False,
+            enrichments=[{}, {"summary": "second"}],
+            failed_indices={0},
+        )
+    )
+
+    with patch("backend.pipeline.ai_processor.get_processor", return_value=mock_processor):
+        result = await process_with_ai(records, {"processor_type": "paw"}, resolve_provider=False)
+
+    assert records[0].ai_enrichment is None
+    assert records[0].status == "normalized"
+    assert records[1].ai_enrichment == {"summary": "second"}
+    assert records[1].status == "ai_processed"
+    assert result == 1
+
+
+
+@pytest.mark.asyncio
+async def test_process_with_ai_rejects_malformed_processor_result(caplog):
+    records = [MagicMock(), MagicMock()]
+    for record in records:
+        record.ai_enrichment = None
+        record.status = "normalized"
+    mock_processor = AsyncMock()
+    mock_processor.process = AsyncMock(
+        return_value=ProcessingResult(success=True, enrichments=[{"summary": "only-one"}])
+    )
+
+    with patch("backend.pipeline.ai_processor.get_processor", return_value=mock_processor):
+        result = await process_with_ai(records, {"processor_type": "paw"}, resolve_provider=False)
+
+    assert result == 0
+    assert all(record.ai_enrichment is None and record.status == "normalized" for record in records)
+    assert _logged(caplog, "processor.contract_invalid")
+
 # ─── model-provider runtime PR-F (decision #9): DataSource.ai_config <-> ModelProvider ─────
 # soft dual-track convergence at the ai_config -> processor-config seam.
 
diff --git a/tests/unit/pipeline/test_collection_lineage.py b/tests/unit/pipeline/test_collection_lineage.py
new file mode 100644
index 00000000..d69459c8
--- /dev/null
+++ b/tests/unit/pipeline/test_collection_lineage.py
@@ -0,0 +1,151 @@
+"""Focused P0-1 collection lineage propagation tests."""
+
+from unittest.mock import AsyncMock, patch
+
+import pytest
+from sqlalchemy import select
+
+from backend.models.notification import NotificationLog, NotificationRule
+from backend.models.record import CollectedRecord
+from backend.models.source import DataSource
+from backend.models.task import CollectionTask
+from backend.notifiers.base import NotificationPayload
+from backend.pipeline.notifier_dispatch import dispatch_notifications
+from backend.pipeline.sinks import CollectionLineage, LegacyDbSink, RunContext
+
+
+@pytest.mark.asyncio
+async def test_run_context_lineage_persists_on_legacy_record(db_session):
+    source = DataSource(
+        name="Lineage Source",
+        channel_type="unknown",
+        channel_config={},
+    )
+    db_session.add(source)
+    await db_session.flush()
+    task = CollectionTask(source_id=source.id, parameters={})
+    db_session.add(task)
+    await db_session.flush()
+
+    ctx = RunContext(
+        task_id=task.id,
+        source_id=source.id,
+        provider="unknown",
+        run_id="task-run-1",
+        trace_id="trace-1",
+        source_revision_id="source-revision-1",
+        source_binding_revision_id="binding-revision-1",
+        worker_id="worker-1",
+        runtime_id="runtime-1",
+        trace_ref="trace-artifact-1",
+        artifact_refs=[{"kind": "raw", "ref": "artifact-1"}],
+    )
+    session_cm = AsyncMock()
+    session_cm.__aenter__ = AsyncMock(return_value=db_session)
+    session_cm.__aexit__ = AsyncMock(return_value=False)
+
+    with patch("backend.database.AsyncSessionLocal", return_value=session_cm):
+        result = await LegacyDbSink().write_batch(
+            ctx,
+            [{"title": "Attributed", "url": "https://example.test/item"}],
+        )
+
+    assert result.accepted == 1
+    record = (await db_session.execute(select(CollectedRecord))).scalar_one()
+    assert record.lineage == {
+        "task_id": task.id,
+        "source_id": source.id,
+        "provider": "unknown",
+        "ingest_mode": "snapshot",
+        "collection_run_id": "task-run-1",
+        "acquisition_execution_id": None,
+        "source_revision_id": "source-revision-1",
+        "source_binding_revision_id": "binding-revision-1",
+        "account_revision_id": None,
+        "credential_revision_id": None,
+        "project_id": None,
+        "scope_ref": None,
+        "worker_id": "worker-1",
+        "runtime_id": "runtime-1",
+        "trace_id": "trace-1",
+        "trace_ref": "trace-artifact-1",
+        "artifact_refs": [{"kind": "raw", "ref": "artifact-1"}],
+    }
+
+
+@pytest.mark.asyncio
+async def test_notification_payload_and_log_keep_record_lineage(db_session):
+    source = DataSource(name="Notify Source", channel_type="rss", channel_config={})
+    db_session.add(source)
+    await db_session.flush()
+    task = CollectionTask(source_id=source.id, parameters={})
+    db_session.add(task)
+    await db_session.flush()
+    record = CollectedRecord(
+        task_id=task.id,
+        source_id=source.id,
+        raw_data={"title": "raw"},
+        normalized_data={"title": "normalized"},
+        content_hash="lineage-notification-hash",
+        status="normalized",
+        lineage=CollectionLineage(
+            task_id=task.id,
+            source_id=source.id,
+            collection_run_id="task-run-2",
+            acquisition_execution_id="acquisition-2",
+            trace_id="trace-2",
+            artifact_refs=("artifact-2",),
+        ).to_dict(),
+    )
+    db_session.add(record)
+    rule = NotificationRule(
+        name="Lineage Rule",
+        trigger_event="on_new_record",
+        notifier_type="webhook",
+        notifier_config={},
+        enabled=True,
+    )
+    db_session.add(rule)
+    await db_session.flush()
+
+    notifier = AsyncMock()
+    notifier.send = AsyncMock(return_value=True)
+    session_cm = AsyncMock()
+    session_cm.__aenter__ = AsyncMock(return_value=db_session)
+    session_cm.__aexit__ = AsyncMock(return_value=False)
+    with (
+        patch("backend.pipeline.notifier_dispatch.get_notifier", return_value=notifier),
+        patch("backend.database.AsyncSessionLocal", return_value=session_cm),
+    ):
+        outcome = await dispatch_notifications(db_session, source.id, [record])
+
+    assert outcome == {"sent": 1, "failed": 0}
+    payload = notifier.send.await_args.args[1]
+    assert isinstance(payload, NotificationPayload)
+    assert payload.lineage == record.lineage
+    log = (await db_session.execute(select(NotificationLog))).scalar_one()
+    assert log.lineage == record.lineage
+
+
+@pytest.mark.asyncio
+async def test_pre_lineage_record_remains_readable(db_session):
+    source = DataSource(name="Legacy Source", channel_type="rss", channel_config={})
+    db_session.add(source)
+    await db_session.flush()
+    task = CollectionTask(source_id=source.id, parameters={})
+    db_session.add(task)
+    await db_session.flush()
+    record = CollectedRecord(
+        task_id=task.id,
+        source_id=source.id,
+        raw_data={},
+        normalized_data={},
+        content_hash="legacy-null-lineage-hash",
+        status="normalized",
+    )
+    db_session.add(record)
+    await db_session.flush()
+
+    loaded = await db_session.get(CollectedRecord, record.id)
+    assert loaded is not None
+    assert loaded.lineage is None
diff --git a/tests/unit/security/test_local_auth.py b/tests/unit/security/test_local_auth.py
new file mode 100644
index 00000000..f7ec4532
--- /dev/null
+++ b/tests/unit/security/test_local_auth.py
@@ -0,0 +1,17 @@
+from backend.security.local_auth import (
+    DEFAULT_LOCAL_ADMIN_PASSWORD_HASH,
+    hash_password,
+    verify_password,
+)
+
+
+def test_default_password_hash_accepts_admin_only():
+    assert verify_password("admin", DEFAULT_LOCAL_ADMIN_PASSWORD_HASH)
+    assert not verify_password("wrong", DEFAULT_LOCAL_ADMIN_PASSWORD_HASH)
+
+
+def test_password_hash_round_trip():
+    encoded = hash_password("new-local-password")
+    assert encoded.startswith("scrypt$")
+    assert verify_password("new-local-password", encoded)
+    assert not verify_password("admin", encoded)
diff --git a/tests/unit/test_agent_image_runtime_packaging.py b/tests/unit/test_agent_image_runtime_packaging.py
index eb1f9078..a08b1b59 100644
--- a/tests/unit/test_agent_image_runtime_packaging.py
+++ b/tests/unit/test_agent_image_runtime_packaging.py
@@ -10,6 +10,8 @@ def test_agent_image_packages_runtime_adapter_modules():
     assert "COPY backend/agent_runtimes/ ./backend/agent_runtimes/" in dockerfile
     assert "COPY backend/miniflow/ ./backend/miniflow/" in dockerfile
 
+    assert "COPY backend/security/ ./backend/security/" in dockerfile
+
 
 def test_public_images_package_opencli_without_a_private_checkout():
     main_image = (ROOT / "Dockerfile").read_text(encoding="utf-8")
@@ -22,6 +24,8 @@ def test_public_images_package_opencli_without_a_private_checkout():
         assert "2233admin/OhMyOpenCLI" not in dockerfile
         assert "git clone ${OHMYOPENCLI_REPO}" not in dockerfile
 
+    assert "command -v npm" in agent_image
+
 
 def test_native_adapter_pack_install_requires_an_explicit_repository():
     windows = (ROOT / "scripts" / "install-managed-opencli.ps1").read_text(
diff --git a/tests/unit/test_agent_runtime_selection.py b/tests/unit/test_agent_runtime_selection.py
new file mode 100644
index 00000000..e983c396
--- /dev/null
+++ b/tests/unit/test_agent_runtime_selection.py
@@ -0,0 +1,139 @@
+import pytest
+
+from backend.models.edge_node import EdgeNode
+from backend.schemas.operations_agent import AgentContractV2, AgentRuntimeBindingV2
+from backend.services import agent_runtime_selection
+
+
+def _contract(*capabilities: str) -> AgentContractV2:
+    return AgentContractV2.model_validate(
+        {
+            "schema_version": "agent.contract.v2",
+            "role": "sales_researcher",
+            "input_schema": {"type": "object"},
+            "output_schema": {"type": "object"},
+            "state_schema": {"type": "object"},
+            "required_capabilities": list(capabilities),
+            "tool_policy": {},
+            "budget": {},
+            "quality_gates": [],
+            "evidence_requirements": [],
+        }
+    )
+
+
+def _binding(**overrides) -> AgentRuntimeBindingV2:
+    value = {
+        "schema_version": "agent.runtime-binding.v2",
+        "workflow": "sales-agent",
+        "preferred_agent_urls": [],
+        "preferred_runtimes": ["codex"],
+        "model_binding": None,
+        "config": {},
+    }
+    value.update(overrides)
+    return AgentRuntimeBindingV2.model_validate(value)
+
+
+async def test_selector_uses_capabilities_before_runtime_preference(db_session, monkeypatch):
+    node = EdgeNode(
+        url="http://capability-agent:19823",
+        label="Capability node",
+        protocol="ws",
+        mode="cdp",
+        node_type="docker",
+        status="online",
+        runtimes=["codex", "prime-agent"],
+        runtime_capabilities={
+            "codex": ["streaming", "workspace_read"],
+            "prime-agent": ["streaming", "subagents", "workspace_read"],
+        },
+    )
+    db_session.add(node)
+    await db_session.commit()
+    monkeypatch.setattr(
+        agent_runtime_selection.ws_agent_manager,
+        "is_connected",
+        lambda url: url == node.url,
+    )
+
+    selected = await agent_runtime_selection.select_agent_runtime(
+        db_session,
+        contract=_contract("streaming", "subagents"),
+        binding=_binding(),
+    )
+
+    assert selected["runtime"] == "prime-agent"
+    assert selected["capabilities"] == ["streaming", "subagents", "workspace_read"]
+
+
+async def test_model_binding_requires_model_selection_capability(db_session, monkeypatch):
+    node = EdgeNode(
+        url="http://model-agent:19823",
+        label="Model node",
+        protocol="ws",
+        mode="cdp",
+        node_type="docker",
+        status="online",
+        runtimes=["codex", "prime-agent"],
+        runtime_capabilities={
+            "codex": ["streaming"],
+            "prime-agent": ["model_selection", "streaming"],
+        },
+    )
+    db_session.add(node)
+    await db_session.commit()
+    monkeypatch.setattr(
+        agent_runtime_selection.ws_agent_manager,
+        "is_connected",
+        lambda _url: True,
+    )
+    binding = _binding(
+        model_binding={
+            "schema_version": "agent.model-binding.v1",
+            "provider": "openrouter",
+            "model": "anthropic/claude-sonnet",
+            "auth_profile": "sales",
+        }
+    )
+
+    selected = await agent_runtime_selection.select_agent_runtime(
+        db_session,
+        contract=_contract("streaming"),
+        binding=binding,
+    )
+
+    assert selected["runtime"] == "prime-agent"
+    assert selected["provider"] == "openrouter"
+    assert selected["model"] == "anthropic/claude-sonnet"
+    assert "auth_profile" in selected
+
+
+async def test_selector_fails_closed_without_capability_match(db_session, monkeypatch):
+    node = EdgeNode(
+        url="http://read-only-agent:19823",
+        label="Read-only node",
+        protocol="ws",
+        mode="cdp",
+        node_type="docker",
+        status="online",
+        runtimes=["codex"],
+        runtime_capabilities={"codex": ["streaming", "workspace_read"]},
+    )
+    db_session.add(node)
+    await db_session.commit()
+    monkeypatch.setattr(
+        agent_runtime_selection.ws_agent_manager,
+        "is_connected",
+        lambda _url: True,
+    )
+
+    with pytest.raises(
+        agent_runtime_selection.RuntimeSelectionError,
+        match="workspace_write",
+    ):
+        await agent_runtime_selection.select_agent_runtime(
+            db_session,
+            contract=_contract("workspace_write"),
+            binding=_binding(),
+        )
diff --git a/tests/unit/test_agent_server.py b/tests/unit/test_agent_server.py
index 21a36962..28d681e2 100644
--- a/tests/unit/test_agent_server.py
+++ b/tests/unit/test_agent_server.py
@@ -68,6 +68,24 @@ def test_collect_auth_fails_closed_and_accepts_exact_bearer(monkeypatch):
     agent_server._require_collect_auth("Bearer secret")
 
 
+def test_center_proxy_honors_no_proxy_for_center_host(monkeypatch):
+    monkeypatch.setattr(agent_server, "_CENTRAL_API_URL", "http://api:8000")
+    monkeypatch.setattr(agent_server, "_HTTP_PROXY", "http://proxy.test:7897")
+    monkeypatch.setattr(agent_server, "_HTTPS_PROXY", "http://proxy.test:7897")
+    monkeypatch.setattr(agent_server, "proxy_bypass", lambda host: host == "api")
+
+    assert agent_server._center_proxy() is None
+
+
+def test_center_proxy_keeps_proxy_for_external_center(monkeypatch):
+    monkeypatch.setattr(agent_server, "_CENTRAL_API_URL", "https://center.example")
+    monkeypatch.setattr(agent_server, "_HTTP_PROXY", "http://proxy.test:7897")
+    monkeypatch.setattr(agent_server, "_HTTPS_PROXY", "http://secure-proxy.test:7897")
+    monkeypatch.setattr(agent_server, "proxy_bypass", lambda _host: False)
+
+    assert agent_server._center_proxy() == "http://secure-proxy.test:7897"
+
+
 # ── _register_with_center attaches Authorization header ────────────────────
 
 
@@ -102,6 +120,11 @@ async def test_register_with_center_attaches_authorization_header(monkeypatch):
     monkeypatch.setattr(agent_server, "_AGENT_API_TOKEN", "secret-token")
     monkeypatch.setattr(agent_server, "_CENTRAL_API_URL", "http://center.example")
     monkeypatch.setattr(agent_server, "available_runtimes", lambda: ["opentabs"])
+    monkeypatch.setattr(
+        agent_server,
+        "available_runtime_capabilities",
+        lambda: {"opentabs": ["browser", "tool_events"]},
+    )
 
     import httpx
     monkeypatch.setattr(httpx, "AsyncClient", _FakeAsyncClient)
@@ -112,6 +135,9 @@ async def test_register_with_center_attaches_authorization_header(monkeypatch):
         "Authorization": "Bearer secret-token"
     }
     assert _FakeAsyncClient.last_post_kwargs["json"]["runtimes"] == ["opentabs"]
+    assert _FakeAsyncClient.last_post_kwargs["json"]["runtime_capabilities"] == {
+        "opentabs": ["browser", "tool_events"]
+    }
     assert _FakeAsyncClient.last_post_args == ("http://center.example/api/v1/nodes/register",)
 
 
@@ -120,6 +146,7 @@ async def test_register_with_center_sends_empty_headers_without_token(monkeypatc
     monkeypatch.setattr(agent_server, "_AGENT_API_TOKEN", "")
     monkeypatch.setattr(agent_server, "_CENTRAL_API_URL", "http://center.example")
     monkeypatch.setattr(agent_server, "available_runtimes", lambda: [])
+    monkeypatch.setattr(agent_server, "available_runtime_capabilities", lambda: {})
 
     import httpx
     monkeypatch.setattr(httpx, "AsyncClient", _FakeAsyncClient)
diff --git a/tests/unit/test_automation_schedule.py b/tests/unit/test_automation_schedule.py
new file mode 100644
index 00000000..b1d90513
--- /dev/null
+++ b/tests/unit/test_automation_schedule.py
@@ -0,0 +1,69 @@
+from datetime import UTC, datetime
+
+import pytest
+
+from backend.automation_schedule import automation_fire_times, parse_automation_schedule
+
+
+def test_supported_schedule_strings_parse_to_expected_kinds():
+    assert parse_automation_schedule("hourly").kind == "hourly"
+    assert parse_automation_schedule("daily@09:30").hour == 9
+    assert parse_automation_schedule("weekdays@18:05").kind == "weekdays"
+    assert parse_automation_schedule("weekly@07:00").kind == "weekly"
+
+
+@pytest.mark.parametrize(
+    "value",
+    ["daily", "daily@24:00", "weekly@monday@09:00", "on_anomaly", "hourly@10:00"],
+)
+def test_unsupported_schedule_strings_fail_closed(value):
+    with pytest.raises(ValueError):
+        parse_automation_schedule(value)
+
+
+def test_daily_schedule_uses_automation_timezone():
+    fires = automation_fire_times(
+        "daily@09:00",
+        "Asia/Shanghai",
+        datetime(2026, 8, 26, 0, 59, tzinfo=UTC),
+        datetime(2026, 8, 26, 1, 0, tzinfo=UTC),
+    )
+    assert fires == [datetime(2026, 8, 26, 1, 0, tzinfo=UTC)]
+
+
+def test_weekdays_and_weekly_use_local_calendar():
+    monday_window = (
+        datetime(2026, 8, 24, 8, 59, tzinfo=UTC),
+        datetime(2026, 8, 24, 9, 0, tzinfo=UTC),
+    )
+    saturday_window = (
+        datetime(2026, 8, 22, 8, 59, tzinfo=UTC),
+        datetime(2026, 8, 22, 9, 0, tzinfo=UTC),
+    )
+    assert automation_fire_times("weekdays@09:00", "UTC", *monday_window)
+    assert automation_fire_times("weekly@09:00", "UTC", *monday_window)
+    assert not automation_fire_times("weekdays@09:00", "UTC", *saturday_window)
+    assert not automation_fire_times("weekly@09:00", "UTC", *saturday_window)
+
+
+def test_dst_nonexistent_wall_time_is_skipped():
+    fires = automation_fire_times(
+        "daily@02:30",
+        "America/New_York",
+        datetime(2024, 3, 10, 6, 0, tzinfo=UTC),
+        datetime(2024, 3, 10, 8, 0, tzinfo=UTC),
+    )
+    assert fires == []
+
+
+def test_dst_ambiguous_wall_time_has_two_distinct_occurrences():
+    fires = automation_fire_times(
+        "daily@01:30",
+        "America/New_York",
+        datetime(2024, 11, 3, 4, 0, tzinfo=UTC),
+        datetime(2024, 11, 3, 7, 0, tzinfo=UTC),
+    )
+    assert fires == [
+        datetime(2024, 11, 3, 5, 30, tzinfo=UTC),
+        datetime(2024, 11, 3, 6, 30, tzinfo=UTC),
+    ]
diff --git a/tests/unit/test_automation_schedule_service.py b/tests/unit/test_automation_schedule_service.py
new file mode 100644
index 00000000..9ed6879b
--- /dev/null
+++ b/tests/unit/test_automation_schedule_service.py
@@ -0,0 +1,295 @@
+from datetime import UTC, datetime
+
+import pytest
+import pytest_asyncio
+from sqlalchemy import func, select
+from sqlalchemy.ext.asyncio import async_sessionmaker
+from backend import ws_agent_manager
+from backend.models.automation import Automation
+from backend.models.edge_node import EdgeNode
+from backend.models.identity import Team, User, Workspace
+from backend.models.operations_agent import (
+    AgentPermissionProfile,
+    AgentProfileMode,
+    OperationsAgentIdentity,
+    OperationsAgentRun,
+    PublishedOperationsAgentVersion,
+)
+from backend.services import scheduled_run_recovery as recovery
+from backend.services import automation_schedule_service as service
+
+
+async def _seed_bound_automation(
+    db_session,
+    *,
+    profile_mode: AgentProfileMode = AgentProfileMode.OBSERVE_ONLY,
+    runtime_capabilities: dict[str, list[str]] | None = None,
+    input_schema: dict | None = None,
+):
+    user = User(subject=f"scheduler-{profile_mode.value}")
+    workspace = Workspace(name="Scheduled", slug=f"scheduled-{profile_mode.value}")
+    db_session.add_all((user, workspace))
+    await db_session.flush()
+    team = Team(workspace_id=workspace.id, name="Operations", slug="operations")
+    db_session.add(team)
+    await db_session.flush()
+    agent = OperationsAgentIdentity(
+        workspace_id=workspace.id,
+        owning_team_id=team.id,
+        name="Read-only scheduler agent",
+        current_profile_version=1,
+        current_published_version=1,
+    )
+    db_session.add(agent)
+    await db_session.flush()
+    profile = AgentPermissionProfile(
+        operations_agent_id=agent.id,
+        version=1,
+        mode=profile_mode,
+        assigned_by_user_id=user.id,
+        reason="test profile",
+    )
+    version = PublishedOperationsAgentVersion(
+        operations_agent_id=agent.id,
+        version=1,
+        draft_revision=1,
+        instructions="Read only",
+        model_configuration={
+            "agent_contract": {
+                "schema_version": "agent.contract.v2",
+                "role": "operations_reviewer",
+                "input_schema": input_schema or {"type": "object"},
+                "output_schema": {"type": "object"},
+                "state_schema": {"type": "object"},
+                "required_capabilities": ["streaming"],
+                "tool_policy": {},
+                "budget": {},
+                "quality_gates": [],
+                "evidence_requirements": [],
+            },
+            "runtime_binding": {
+                "schema_version": "agent.runtime-binding.v2",
+                "workflow": "builtin.read_only_readiness",
+                "preferred_agent_urls": ["http://scheduler-agent:19823"],
+                "preferred_runtimes": ["miniflow"],
+                "model_binding": None,
+                "config": {"timeout_seconds": 60},
+                "dispatch_timeout_seconds": 60,
+            },
+        },
+        tool_configuration={},
+        published_by_user_id=user.id,
+        reason="test version",
+    )
+    node = EdgeNode(
+        url="http://scheduler-agent:19823",
+        label="Scheduler runtime",
+        protocol="ws",
+        mode="cdp",
+        node_type="docker",
+        status="online",
+        runtimes=list((runtime_capabilities or {"miniflow": ["streaming"]}).keys()),
+        runtime_capabilities=runtime_capabilities or {"miniflow": ["streaming"]},
+    )
+    automation = Automation(
+        workspace_id=workspace.id,
+        revision=1,
+        operations_agent_id=agent.id,
+        operations_agent_version=1,
+        name="Daily readiness",
+        prompt="Check readiness without mutations",
+        precheck=None,
+        executor="operations-agent",
+        schedule="daily@12:00",
+        session_mode="fresh",
+        approval_mode=profile_mode.value,
+        project={},
+        enabled=True,
+        created_by_user_id=user.id,
+    )
+    db_session.add_all((profile, version, node, automation))
+    await db_session.commit()
+    return automation, agent, profile
+
+
+@pytest_asyncio.fixture
+async def scheduler_factory(db_engine, monkeypatch):
+    factory = async_sessionmaker(db_engine, expire_on_commit=False)
+    monkeypatch.setattr(service, "AsyncSessionLocal", factory)
+    monkeypatch.setattr(recovery, "AsyncSessionLocal", factory)
+    monkeypatch.setattr(service, "schedule_operations_agent_run", lambda _run_id: None)
+    return factory
+
+
+async def test_duplicate_scans_claim_one_durable_occurrence(
+    db_session,
+    scheduler_factory,
+):
+    automation, agent, profile = await _seed_bound_automation(db_session)
+    scheduled_for = datetime(2026, 8, 24, 12, 0, tzinfo=UTC)
+
+    first, first_created = await service.claim_scheduled_automation_run(
+        automation.id,
+        scheduled_for,
+    )
+    second, second_created = await service.claim_scheduled_automation_run(
+        automation.id,
+        scheduled_for,
+    )
+
+    assert first is not None and second is not None
+    assert first_created is True
+    assert second_created is False
+    assert first.id == second.id
+    assert first.trigger_type == "scheduled"
+    assert first.trigger_reference == f"automation:{automation.id}:20260824T1200Z"
+    assert first.automation_revision == 1
+    assert first.automation_snapshot["name"] == "Daily readiness"
+    assert first.scheduled_for == scheduled_for
+    assert first.schedule_timezone == "UTC"
+    assert first.operations_agent_id == agent.id
+    assert await recovery.list_queued_scheduled_run_ids() == [first.id]
+    assert first.execution_binding["runtime"] == "miniflow"
+    await recovery.recover_operations_agent_runs_on_startup()
+    persisted = await db_session.get(OperationsAgentRun, first.id)
+    assert persisted is not None and persisted.status == "queued"
+    assert first.published_version == 1
+    assert first.profile_version == profile.version
+    count = await db_session.scalar(
+        select(func.count()).select_from(OperationsAgentRun).where(
+            OperationsAgentRun.automation_id == automation.id
+        )
+    )
+    assert count == 1
+
+
+async def test_pause_recheck_prevents_claim(
+    db_session,
+    scheduler_factory,
+):
+    automation, _, _ = await _seed_bound_automation(db_session)
+    automation.enabled = False
+    await db_session.commit()
+
+    run, created = await service.claim_scheduled_automation_run(
+        automation.id,
+        datetime(2026, 8, 24, 12, 0, tzinfo=UTC),
+    )
+
+    assert run is None
+    assert created is False
+
+
+
+
+async def test_manual_run_rejects_offline_bound_runtime(
+    db_session,
+    monkeypatch,
+):
+    automation, _, _ = await _seed_bound_automation(db_session)
+    monkeypatch.setattr(ws_agent_manager, "is_connected", lambda _url: False)
+
+    with pytest.raises(service.AutomationBindingError, match="no connected"):
+        await service.create_bound_automation_run(
+            db_session,
+            automation,
+            trigger_type="manual",
+            started_by_user_id=automation.created_by_user_id,
+        )
+
+    count = await db_session.scalar(
+        select(func.count()).select_from(OperationsAgentRun).where(
+            OperationsAgentRun.automation_id == automation.id
+        )
+    )
+    assert count == 0
+
+
+async def test_scheduled_offline_runtime_remains_queued_for_recovery(
+    db_session,
+    scheduler_factory,
+    monkeypatch,
+):
+    automation, _, _ = await _seed_bound_automation(db_session)
+    monkeypatch.setattr(ws_agent_manager, "is_connected", lambda _url: False)
+
+    run, created = await service.claim_scheduled_automation_run(
+        automation.id,
+        datetime(2026, 8, 24, 12, 0, tzinfo=UTC),
+    )
+
+    assert run is not None and created is True
+    assert run.status == "queued"
+    assert await recovery.list_queued_scheduled_run_ids() == [run.id]
+
+
+async def test_structural_binding_failure_persists_one_failed_occurrence(
+    db_session,
+    scheduler_factory,
+):
+    automation, agent, profile = await _seed_bound_automation(
+        db_session,
+        runtime_capabilities={"pi": []},
+    )
+    scheduled_for = datetime(2026, 8, 24, 12, 0, tzinfo=UTC)
+
+    first, first_created = await service.claim_scheduled_automation_run(
+        automation.id,
+        scheduled_for,
+    )
+    second, second_created = await service.claim_scheduled_automation_run(
+        automation.id,
+        scheduled_for,
+    )
+
+    assert first is not None and second is not None
+    assert first_created is True
+    assert second_created is False
+    assert first.id == second.id
+    assert first.status == "failed"
+    assert "required capabilities" in first.error_message
+    assert first.trigger_reference == f"automation:{automation.id}:20260824T1200Z"
+    assert first.scheduled_for == scheduled_for
+    assert first.automation_revision == automation.revision
+    assert first.automation_snapshot["id"] == automation.id
+    assert first.operations_agent_id == agent.id
+    assert first.published_version == automation.operations_agent_version
+    assert first.profile_version == profile.version
+    count = await db_session.scalar(
+        select(func.count()).select_from(OperationsAgentRun).where(
+            OperationsAgentRun.automation_id == automation.id,
+            OperationsAgentRun.scheduled_for == scheduled_for,
+        )
+    )
+    assert count == 1
+async def test_low_risk_automatic_profile_is_incompatible(db_session):
+    automation, _, _ = await _seed_bound_automation(
+        db_session,
+        profile_mode=AgentProfileMode.LOW_RISK_AUTOMATIC,
+    )
+
+    with pytest.raises(service.AutomationBindingError, match="Low-Risk Automatic"):
+        await service.validate_automation_binding(db_session, automation)
+
+
+async def test_runtime_must_satisfy_agent_capabilities(db_session):
+    automation, _, _ = await _seed_bound_automation(
+        db_session,
+        runtime_capabilities={"pi": []},
+    )
+
+    with pytest.raises(service.AutomationBindingError, match="required capabilities"):
+        await service.validate_automation_binding(db_session, automation)
+
+
+async def test_automation_payload_must_satisfy_pinned_agent_contract(db_session):
+    automation, _, _ = await _seed_bound_automation(
+        db_session,
+        input_schema={
+            "type": "object",
+            "required": ["unsupported_required_field"],
+        },
+    )
+
+    with pytest.raises(service.AutomationBindingError, match="incompatible"):
+        await service.validate_automation_binding(db_session, automation)
diff --git a/tests/unit/test_gaojixing_runtime.py b/tests/unit/test_gaojixing_runtime.py
new file mode 100644
index 00000000..25952ae5
--- /dev/null
+++ b/tests/unit/test_gaojixing_runtime.py
@@ -0,0 +1,1081 @@
+from types import SimpleNamespace
+from unittest.mock import AsyncMock
+
+import pytest
+
+from backend.channels.base import ChannelResult
+from backend.workflow import gaojixing_runtime as runtime
+from backend.workflow.gaojixing_runtime import (
+    GAOJIXING_CAPABILITY_ID,
+    GaojixingReadinessError,
+    build_question_package,
+    capture_live_doubao,
+    map_capture_item,
+)
+from backend.workflow.opencli_hda_tracer import (
+    _execute_gaojixing_fixture_source,
+    _execute_gaojixing_source,
+    _store_record_sink_outputs,
+    replay_downstream_from_persisted_gaojixing_source,
+)
+
+
+def test_question_package_uses_runtime_question_and_stable_digest():
+    first = build_question_package(
+        node_params={"question": "configured", "sourceGroup": "research"},
+        adapter_config={"site_session": "persistent", "settle_seconds": 35},
+        runtime_payload={"question": "runtime"},
+    )
+    second = build_question_package(
+        node_params={"question": "other", "sourceGroup": "research"},
+        adapter_config={"site_session": "persistent", "settle_seconds": 35},
+        runtime_payload={"question": "runtime"},
+    )
+
+    assert first.question == "runtime"
+    assert first.digest == second.digest
+    assert len(first.digest) == 64
+    assert first.options["settle_seconds"] == 35
+
+
+def test_question_package_remains_immutable_after_source_changes():
+    node_params = {"question": "original", "sourceGroup": "before"}
+    package = build_question_package(
+        node_params=node_params,
+        adapter_config={"settle_seconds": 35},
+        runtime_payload={},
+    )
+
+    node_params["question"] = "edited"
+    node_params["sourceGroup"] = "after"
+
+    assert package.question == "original"
+    assert package.options["sourceGroup"] == "before"
+    with pytest.raises(TypeError):
+        package.options["sourceGroup"] = "mutated"
+    assert package.digest == build_question_package(
+        node_params={"question": "original", "sourceGroup": "before"},
+        adapter_config={"settle_seconds": 35},
+        runtime_payload={},
+    ).digest
+
+
+def test_question_package_accepts_published_run_query_input():
+    package = build_question_package(
+        node_params={},
+        adapter_config={},
+        runtime_payload={"query": "高吉星燕窝酸 DHA 藻油"},
+    )
+
+    assert package.question == "高吉星燕窝酸 DHA 藻油"
+
+
+def test_question_package_requires_effective_question():
+    with pytest.raises(GaojixingReadinessError) as error:
+        build_question_package(node_params={}, adapter_config={}, runtime_payload={})
+
+    assert error.value.code == "gaojixing_question_required"
+
+
+@pytest.mark.asyncio
+async def test_capture_fails_closed_when_capability_missing(monkeypatch):
+    package = build_question_package(
+        node_params={"question": "q"}, adapter_config={}, runtime_payload={}
+    )
+
+    async def should_not_probe(self, _config):
+        raise AssertionError("session probe must not run for a missing capability")
+
+    monkeypatch.setattr(runtime.DoubaoResearchChannel, "health_check", should_not_probe)
+    with pytest.raises(GaojixingReadinessError) as error:
+        await capture_live_doubao(
+            package=package,
+            node_params={"capabilityId": "missing.capture"},
+            adapter_config={},
+            network_allowed=True,
+        )
+
+    assert error.value.code == "gaojixing_capability_missing"
+
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    ("adapter_config", "network_allowed", "expected_code"),
+    [
+        ({"capabilityAvailable": False}, True, "gaojixing_capability_missing"),
+        ({"adapterAvailable": False}, True, "gaojixing_adapter_missing"),
+        ({"authenticated": False}, True, "gaojixing_authentication_required"),
+        ({"sessionAvailable": False}, True, "gaojixing_session_unavailable"),
+        ({}, False, "gaojixing_network_denied"),
+    ],
+)
+async def test_capture_reports_each_configured_readiness_blocker(
+    monkeypatch,
+    adapter_config,
+    network_allowed,
+    expected_code,
+):
+    package = build_question_package(
+        node_params={"question": "q"}, adapter_config={}, runtime_payload={}
+    )
+
+    async def should_not_probe(self, _config):
+        raise AssertionError("session probe must not run after a readiness blocker")
+
+    monkeypatch.setattr(runtime.DoubaoResearchChannel, "health_check", should_not_probe)
+    with pytest.raises(GaojixingReadinessError) as error:
+        await capture_live_doubao(
+            package=package,
+            node_params={},
+            adapter_config=adapter_config,
+            network_allowed=network_allowed,
+        )
+
+    assert error.value.code == expected_code
+
+
+@pytest.mark.asyncio
+async def test_capture_reports_captcha_from_session_probe(monkeypatch):
+    package = build_question_package(
+        node_params={"question": "q"}, adapter_config={}, runtime_payload={}
+    )
+
+    async def unhealthy(self, _config):
+        return False
+
+    async def captcha(self, _config):
+        return "captcha_challenge"
+
+    monkeypatch.setattr(runtime.DoubaoResearchChannel, "health_check", unhealthy)
+    monkeypatch.setattr(runtime.DoubaoResearchChannel, "readiness_code", captcha)
+
+    with pytest.raises(GaojixingReadinessError) as error:
+        await capture_live_doubao(
+            package=package,
+            node_params={},
+            adapter_config={},
+            network_allowed=True,
+        )
+
+    assert error.value.code == "gaojixing_captcha_challenge"
+    assert error.value.details["site"] == "doubao"
+
+@pytest.mark.asyncio
+async def test_capture_uses_live_channel_after_health_probe(monkeypatch):
+    package = build_question_package(
+        node_params={"question": "q"}, adapter_config={}, runtime_payload={}
+    )
+    calls = []
+
+    async def healthy(self, _config):
+        calls.append("health")
+        return True
+
+    async def collect(self, config, parameters):
+        calls.append((config["question"], parameters["question"]))
+        return ChannelResult.ok([{"content": "answer", "citations": [], "conversation_url": ""}])
+
+    monkeypatch.setattr(runtime.DoubaoResearchChannel, "health_check", healthy)
+    monkeypatch.setattr(runtime.DoubaoResearchChannel, "collect", collect)
+    result = await capture_live_doubao(
+        package=package,
+        node_params={},
+        adapter_config={"capabilityId": GAOJIXING_CAPABILITY_ID},
+        network_allowed=True,
+    )
+
+    assert result.success
+    assert calls == ["health", ("q", "q")]
+
+
+def test_capture_mapping_keeps_package_and_independent_evidence():
+    package = build_question_package(
+        node_params={"question": "q"}, adapter_config={}, runtime_payload={}
+    )
+    mapped = map_capture_item(
+        {
+            "content": "answer https://example.test/source",
+            "citations": [{"url": "https://example.test/source"}],
+            "citation_capture": "answer_url_extraction",
+            "conversation_url": "https://www.doubao.com/chat/123",
+        },
+        package=package,
+        workflow_id="wf",
+        run_id="run",
+        node_id="source",
+        artifact_id="artifact",
+    )
+
+    evidence = mapped["gaojixing"]["evidence"]
+    assert mapped["packageDigest"] == package.digest
+    assert evidence["answer"]["artifactId"] == "artifact"
+    assert evidence["citations"]["verified"] is False
+    assert evidence["conversation"]["status"] == "captured"
+    assert mapped["dedupe"] == {
+        "type": "source-identity",
+        "field": "conversation_url",
+        "value": "https://www.doubao.com/chat/123",
+        "status": "unique",
+    }
+
+
+def test_capture_mapping_does_not_invent_malformed_optional_evidence():
+    package = build_question_package(
+        node_params={"question": "q"}, adapter_config={}, runtime_payload={}
+    )
+    mapped = map_capture_item(
+        {
+            "content": "answer",
+            "citations": "not-a-citation-list",
+            "conversation_url": {"url": "https://www.doubao.com/chat/guessed"},
+        },
+        package=package,
+        workflow_id="wf",
+        run_id="run",
+        node_id="source",
+        artifact_id="artifact",
+    )
+
+    evidence = mapped["gaojixing"]["evidence"]
+    assert evidence["citations"] == {
+        "status": "empty",
+        "capture": "answer_url_extraction",
+        "verified": False,
+        "items": [],
+    }
+    assert evidence["conversation"] == {"status": "unknown", "url": None}
+
+
+class _Emitter:
+    def __init__(self):
+        self.events = []
+
+    def emit(self, node, event_type, **kwargs):
+        self.events.append((node.id, event_type, kwargs))
+
+
+@pytest.mark.asyncio
+async def test_hda_live_source_branch_maps_capture_output(monkeypatch):
+    node = SimpleNamespace(
+        id="gaojixing-source",
+        kind="source",
+        adapter=None,
+        depends_on=[],
+        params={"question": "configured", "sourceGroup": "gaojixing"},
+        runtime={
+            "binding": {
+                "binding_id": "workflow.source.fetch",
+                "input": {
+                    "channelType": "doubao_research",
+                    "liveMode": "live",
+                    "adapterConfig": {"capabilityId": GAOJIXING_CAPABILITY_ID},
+                },
+            }
+        },
+    )
+    body = SimpleNamespace(
+        input=SimpleNamespace(payload={"question": "runtime"}),
+        project=SimpleNamespace(
+            id="workflow",
+            agentPermissions=SimpleNamespace(canFetchNetwork=True),
+        ),
+    )
+    emitter = _Emitter()
+    outputs = {}
+
+    async def fake_capture(**kwargs):
+        assert kwargs["package"].question == "runtime"
+        return ChannelResult.ok(
+            [
+                {
+                    "content": "answer",
+                    "citations": [{"url": "https://example.test"}],
+                    "conversation_url": "https://www.doubao.com/chat/123",
+                }
+            ]
+        )
+
+    monkeypatch.setattr("backend.workflow.opencli_hda_tracer.capture_live_doubao", fake_capture)
+    await _execute_gaojixing_source(
+        node,
+        body=body,
+        run_id="run",
+        workflow_id="workflow",
+        trace_id="trace",
+        outputs_by_node=outputs,
+        emitter=emitter,
+        session=None,
+    )
+
+    item = outputs["gaojixing-source"][0]
+    raw = item["raw"]
+    assert raw["gaojixing"]["mode"] == "live"
+    assert raw["gaojixing"]["evidence"]["packageDigest"] == raw["packageDigest"]
+    assert item["lineage"][0]["artifact"] == "gaojixing.capture"
+    assert [event[1] for event in emitter.events] == ["partial", "completed"]
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    ("capture_result", "expected_code"),
+    [
+        (ChannelResult.fail("Doubao request timed out", error_type="TimeoutError"), "TimeoutError"),
+        (
+            ChannelResult.fail("CAPTCHA challenge", error_type="captcha_challenge"),
+            "captcha_challenge",
+        ),
+        (ChannelResult.ok([{"content": " "}]), "gaojixing_answer_missing"),
+    ],
+)
+async def test_hda_live_source_fails_closed_on_capture_errors(
+    monkeypatch,
+    capture_result,
+    expected_code,
+):
+    node = SimpleNamespace(
+        id="gaojixing-source",
+        kind="source",
+        adapter=None,
+        depends_on=[],
+        params={"question": "q", "sourceGroup": "gaojixing"},
+        runtime={
+            "binding": {
+                "input": {
+                    "channelType": "doubao_research",
+                    "liveMode": "live",
+                    "adapterConfig": {"capabilityId": GAOJIXING_CAPABILITY_ID},
+                }
+            }
+        },
+    )
+    body = SimpleNamespace(
+        input=SimpleNamespace(payload={}),
+        project=SimpleNamespace(
+            id="workflow",
+            agentPermissions=SimpleNamespace(canFetchNetwork=True),
+        ),
+    )
+    emitter = _Emitter()
+    outputs = {}
+
+    async def fake_capture(**_kwargs):
+        return capture_result
+
+    monkeypatch.setattr("backend.workflow.opencli_hda_tracer.capture_live_doubao", fake_capture)
+    await _execute_gaojixing_source(
+        node,
+        body=body,
+        run_id="run",
+        workflow_id="workflow",
+        trace_id="trace",
+        outputs_by_node=outputs,
+        emitter=emitter,
+        session=None,
+    )
+
+    assert outputs["gaojixing-source"] == []
+    assert emitter.events[-1][1] == "failed"
+    assert emitter.events[-1][2]["block_reason"].code == expected_code
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    ("mode", "provenance", "business_outcome"),
+    [
+        ("live", "opencli:doubao", "unconfirmed"),
+        ("fixture", "fixture:contract-vector", "fixture"),
+    ],
+)
+async def test_replay_preserves_provenance_for_webhook_delivery(
+    monkeypatch,
+    mode,
+    provenance,
+    business_outcome,
+):
+    source_node = SimpleNamespace(
+        id="source",
+        kind="source",
+        params={},
+        adapter=None,
+        runtime={"binding": {"input": {"channelType": "doubao_research", "liveMode": mode}}},
+    )
+    source_input = SimpleNamespace(
+        sourceId=None,
+        model_copy=lambda *, update: SimpleNamespace(**{**source_input.__dict__, **update}),
+    )
+    source_request = SimpleNamespace(
+        project=object(),
+        input=source_input,
+        model_copy=lambda *, update, deep: SimpleNamespace(**{**source_request.__dict__, **update}),
+    )
+    source_run = SimpleNamespace(
+        projection=SimpleNamespace(status="completed", workflowId="workflow"),
+        studio_workflow_version_id="version",
+        workflow_version_id=None,
+        request=source_request,
+        events=[
+            SimpleNamespace(
+                nodeId="source",
+                eventType="partial",
+                details={
+                    "channelType": "doubao_research",
+                    "capabilityId": "chat-ai.capture",
+                    "package": {"digest": "digest"},
+                    "artifactId": "artifact",
+                    "evidence": {
+                        "mode": mode,
+                        "provenance": provenance,
+                        "packageDigest": "digest",
+                        "runId": "source-run",
+                        "workflowId": "workflow",
+                        "nodeId": "source",
+                        "answer": {"artifactId": "artifact", "text": "persisted answer"},
+                        "citations": {"items": []},
+                        "conversation": {"url": "https://example.test/conversation"},
+                    },
+                },
+            ),
+            SimpleNamespace(nodeId="source", eventType="completed"),
+        ],
+    )
+    record = SimpleNamespace(
+        raw_data={
+            "gaojixing": {"artifactId": "artifact", "package": {"digest": "digest"}},
+            "_workflowLineage": [
+                {
+                    "nodeId": "source",
+                    "artifact": "gaojixing.capture",
+                    "artifactId": "artifact",
+                    "packageDigest": "digest",
+                }
+            ],
+        }
+    )
+    session = SimpleNamespace(
+        execute=AsyncMock(
+            return_value=SimpleNamespace(scalars=lambda: SimpleNamespace(all=lambda: [record]))
+        )
+    )
+    monkeypatch.setattr(
+        "backend.workflow.opencli_hda_tracer._load_workflow_run",
+        AsyncMock(side_effect=[source_run, None]),
+    )
+    monkeypatch.setattr(
+        "backend.workflow.opencli_hda_tracer.compile_workflow_project",
+        lambda _project: SimpleNamespace(
+            valid=True, plan=SimpleNamespace(runtime=SimpleNamespace(nodes=[source_node]))
+        ),
+    )
+    start = AsyncMock(return_value=SimpleNamespace(runId="replay-run"))
+    monkeypatch.setattr("backend.workflow.opencli_hda_tracer.start_workflow_run", start)
+
+    replay = await replay_downstream_from_persisted_gaojixing_source(
+        "source-run",
+        expected_workflow_id="workflow",
+        expected_studio_workflow_version_id="version",
+        session=session,
+    )
+
+    assert replay.runId == "replay-run"
+    request = start.await_args.args[0]
+    assert request.input.sourceId == "source-run"
+    replay_raw = request.sourceOutputs["source"][0]["raw"]
+    assert replay_raw["gaojixing"]["artifactId"] == "artifact"
+    assert replay_raw["gaojixing"]["provenance"] == provenance
+    assert request.sourceOutputs["source"][0]["lineage"][0]["mode"] == "persisted-replay"
+    assert start.await_args.kwargs["replay_source_node_ids"] == {"source"}
+
+    from backend.notifiers.base import NotificationSendResult
+    from backend.workflow.webhook_delivery import execute_workflow_webhook_delivery
+
+    class _Notifier:
+        async def send(self, _config, _payload):
+            return NotificationSendResult(success=True, response_data=None)
+
+    monkeypatch.setattr(
+        "backend.workflow.webhook_delivery.get_notifier",
+        lambda _kind: _Notifier(),
+    )
+    delivery = await execute_workflow_webhook_delivery(
+        {},
+        request.sourceOutputs["source"],
+        workflow_id="workflow",
+        run_id="replay-run",
+        node_id="notify",
+    )
+    assert delivery["mode"] == mode
+    assert delivery["provenance"] == provenance
+    assert delivery["businessOutcome"] == business_outcome
+    assert delivery["liveAccepted"] is False
+
+
+@pytest.mark.asyncio
+async def test_record_sink_persists_gaojixing_refs_and_collection_lineage(monkeypatch):
+    package = build_question_package(
+        node_params={"question": "q"}, adapter_config={}, runtime_payload={}
+    )
+    raw = map_capture_item(
+        {
+            "content": "answer",
+            "citations": [{"url": "https://example.test/source"}],
+            "conversation_url": "https://www.doubao.com/chat/123",
+        },
+        package=package,
+        workflow_id="workflow",
+        run_id="run",
+        node_id="source",
+        artifact_id="artifact",
+    )
+    source_node = SimpleNamespace(
+        id="source",
+        kind="source",
+        adapter=None,
+        depends_on=[],
+        params={"sourceGroup": "gaojixing"},
+        runtime={
+            "binding": {
+                "input": {"channelType": "doubao_research"},
+                "binding_id": "workflow.source.fetch",
+            }
+        },
+    )
+    sink_node = SimpleNamespace(
+        id="sink",
+        kind="sink",
+        adapter=None,
+        depends_on=["source"],
+        params={},
+        runtime={"binding": {"binding_id": "workflow.record-sink.records", "input": {}}},
+    )
+
+    class _Session:
+        async def execute(self, _statement):
+            return SimpleNamespace(
+                scalars=lambda: [SimpleNamespace(id="record-id", content_hash=captured["hash"])]
+            )
+
+        async def flush(self):
+            return None
+
+    captured = {}
+
+    async def fake_materialize(*args, **kwargs):
+        return "source-id", "task-id"
+
+    async def fake_store(*args, **kwargs):
+        captured["hash"] = args[3][0][2]
+        captured["lineage"] = kwargs["lineage"]
+        return [SimpleNamespace(id="record-id")], 0
+
+    monkeypatch.setattr(
+        "backend.workflow.opencli_hda_tracer._materialize_source_task",
+        fake_materialize,
+    )
+    monkeypatch.setattr("backend.workflow.opencli_hda_tracer.store_records", fake_store)
+    stored, skipped = await _store_record_sink_outputs(
+        sink_node,
+        [{"raw": raw, "lineage": [{"artifact": "gaojixing.capture", "nodeId": "source"}]}],
+        run_id="run",
+        workflow_id="workflow",
+        target="records",
+        session=_Session(),
+        runtime_nodes_by_id={"source": source_node},
+        materialized_source_tasks={},
+    )
+
+    assert skipped == 0
+    assert stored[0]["recordId"] == "record-id"
+    assert stored[0]["raw"]["packageDigest"] == package.digest
+    assert stored[0]["normalizedData"]["packageDigest"] == package.digest
+    artifact_refs = captured["lineage"]["artifact_refs"]
+    assert artifact_refs[1]["artifactId"] == "artifact"
+    assert artifact_refs[1]["packageDigest"] == package.digest
+    assert captured["lineage"]["collection_run_id"] == "run"
+
+
+@pytest.mark.asyncio
+async def test_gaojixing_delivery_distinguishes_transport_from_business_ack(monkeypatch):
+    from backend.notifiers.base import NotificationSendResult
+    from backend.workflow.webhook_delivery import execute_workflow_webhook_delivery
+
+    package = build_question_package(
+        node_params={"question": "q"}, adapter_config={}, runtime_payload={}
+    )
+    raw = map_capture_item(
+        {"content": "answer", "citations": [], "conversation_url": ""},
+        package=package,
+        workflow_id="workflow",
+        run_id="run",
+        node_id="source",
+        artifact_id="artifact",
+    )
+    captured = {}
+
+    class _Notifier:
+        async def send(self, config, payload):
+            captured["payload"] = payload
+            return NotificationSendResult(success=True, response_data=None)
+
+    monkeypatch.setattr(
+        "backend.workflow.webhook_delivery.get_notifier",
+        lambda _kind: _Notifier(),
+    )
+    result = await execute_workflow_webhook_delivery(
+        {"target": "business", "config": {"url": "https://example.test/hook"}},
+        [{"raw": raw, "lineage": [{"nodeId": "source"}]}],
+        workflow_id="workflow",
+        run_id="run",
+        node_id="notify",
+    )
+
+    assert result["transportStatus"] == "accepted"
+    assert result["businessOutcome"] == "unconfirmed"
+    assert result["ackEvidence"] is None
+    assert result["packageDigest"] == package.digest
+    assert captured["payload"].delivery_id == result["deliveryAttemptId"]
+
+
+@pytest.mark.asyncio
+async def test_matching_destination_ack_confirms_one_idempotent_business_outcome(monkeypatch):
+    from backend.notifiers.base import NotificationSendResult
+    from backend.workflow.webhook_delivery import execute_workflow_webhook_delivery
+
+    package = build_question_package(
+        node_params={"question": "q"}, adapter_config={}, runtime_payload={}
+    )
+    raw = map_capture_item(
+        {"content": "answer"},
+        package=package,
+        workflow_id="workflow",
+        run_id="run",
+        node_id="source",
+        artifact_id="artifact",
+    )
+
+    class _Notifier:
+        async def send(self, _config, payload):
+            return NotificationSendResult(
+                success=True,
+                response_data={
+                    "businessAck": True,
+                    "deliveryAttemptId": payload.delivery_id,
+                },
+            )
+
+    monkeypatch.setattr(
+        "backend.workflow.webhook_delivery.get_notifier",
+        lambda _kind: _Notifier(),
+    )
+    first = await execute_workflow_webhook_delivery(
+        {},
+        [{"raw": raw, "lineage": [{"nodeId": "source"}]}],
+        workflow_id="workflow",
+        run_id="run",
+        node_id="notify",
+    )
+    retry = await execute_workflow_webhook_delivery(
+        {},
+        [{"raw": raw, "lineage": [{"nodeId": "source"}]}],
+        workflow_id="workflow",
+        run_id="run",
+        node_id="notify",
+    )
+
+    assert first["deliveryAttemptId"] == retry["deliveryAttemptId"]
+    assert first["transportStatus"] == "accepted"
+    assert first["businessOutcome"] == "confirmed"
+    assert first["liveAccepted"] is True
+    assert first["ackEvidence"]["matchesDeliveryAttempt"] is True
+
+
+@pytest.mark.asyncio
+async def test_unmatched_destination_ack_remains_unconfirmed(monkeypatch):
+    from backend.notifiers.base import NotificationSendResult
+    from backend.workflow.webhook_delivery import execute_workflow_webhook_delivery
+
+    package = build_question_package(
+        node_params={"question": "q"}, adapter_config={}, runtime_payload={}
+    )
+    raw = map_capture_item(
+        {"content": "answer"},
+        package=package,
+        workflow_id="workflow",
+        run_id="run",
+        node_id="source",
+        artifact_id="artifact",
+    )
+
+    class _Notifier:
+        async def send(self, _config, _payload):
+            return NotificationSendResult(
+                success=True,
+                response_data={"businessAck": True, "deliveryAttemptId": "another-attempt"},
+            )
+
+    monkeypatch.setattr(
+        "backend.workflow.webhook_delivery.get_notifier",
+        lambda _kind: _Notifier(),
+    )
+    result = await execute_workflow_webhook_delivery(
+        {},
+        [{"raw": raw, "lineage": [{"nodeId": "source"}]}],
+        workflow_id="workflow",
+        run_id="run",
+        node_id="notify",
+    )
+
+    assert result["transportStatus"] == "accepted"
+    assert result["businessOutcome"] == "unconfirmed"
+    assert result["liveAccepted"] is False
+    assert result["ackEvidence"]["matchesDeliveryAttempt"] is False
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    ("failure_kind", "expected_code"),
+    [
+        ("notifier", "webhook_delivery_failed"),
+        ("network", "webhook_delivery_network_error"),
+    ],
+)
+async def test_destination_failures_preserve_typed_unknown_business_state(
+    monkeypatch,
+    failure_kind,
+    expected_code,
+):
+    from backend.notifiers.base import NotificationSendResult
+    from backend.workflow.webhook_delivery import (
+        WorkflowWebhookDeliveryError,
+        execute_workflow_webhook_delivery,
+    )
+
+    package = build_question_package(
+        node_params={"question": "q"}, adapter_config={}, runtime_payload={}
+    )
+    raw = map_capture_item(
+        {"content": "answer"},
+        package=package,
+        workflow_id="workflow",
+        run_id="run",
+        node_id="source",
+        artifact_id="artifact",
+    )
+
+    class _Notifier:
+        async def send(self, _config, _payload):
+            if failure_kind == "network":
+                raise OSError("destination timed out")
+            return NotificationSendResult(success=False, response_data=None)
+
+    monkeypatch.setattr(
+        "backend.workflow.webhook_delivery.get_notifier",
+        lambda _kind: _Notifier(),
+    )
+    with pytest.raises(WorkflowWebhookDeliveryError) as error:
+        await execute_workflow_webhook_delivery(
+            {},
+            [{"raw": raw, "lineage": [{"nodeId": "source"}]}],
+            workflow_id="workflow",
+            run_id="run",
+            node_id="notify",
+        )
+
+    assert error.value.code == expected_code
+    assert error.value.details["transportStatus"] == "failed"
+    assert error.value.details["businessOutcome"] == "unknown"
+    assert error.value.details["deliveryAttemptId"]
+
+
+@pytest.mark.asyncio
+async def test_fixture_source_labels_digest_evidence_and_lineage_without_live_dispatch(
+    monkeypatch,
+):
+    node = SimpleNamespace(
+        id="fixture-source",
+        kind="source",
+        adapter=None,
+        depends_on=[],
+        params={
+            "question": "configured",
+            "sourceGroup": "gaojixing",
+            "fixtureItems": [
+                {
+                    "content": "fixture answer",
+                    "citations": [{"url": "https://fixture.test/citation"}],
+                    "conversation_url": "https://fixture.test/conversation",
+                }
+            ],
+        },
+        runtime={
+            "binding": {
+                "binding_id": "workflow.source.fetch",
+                "input": {
+                    "channelType": "doubao_research",
+                    "liveMode": "fixture",
+                    "adapterConfig": {"fixtureProvenance": "fixture:contract-vector"},
+                },
+            }
+        },
+    )
+    body = SimpleNamespace(
+        input=SimpleNamespace(payload={"question": "runtime"}),
+        project=SimpleNamespace(
+            id="workflow",
+            agentPermissions=SimpleNamespace(canFetchNetwork=False),
+        ),
+        sourceOutputs={},
+    )
+    emitter = _Emitter()
+    outputs = {}
+
+    async def should_not_dispatch_live(**_kwargs):
+        raise AssertionError("fixture execution must not dispatch the live channel")
+
+    monkeypatch.setattr(
+        "backend.workflow.opencli_hda_tracer.capture_live_doubao", should_not_dispatch_live
+    )
+    await _execute_gaojixing_fixture_source(
+        node,
+        body=body,
+        run_id="run",
+        workflow_id="workflow",
+        outputs_by_node=outputs,
+        emitter=emitter,
+    )
+
+    raw = outputs["fixture-source"][0]["raw"]
+    evidence = raw["gaojixing"]["evidence"]
+    assert raw["gaojixing"]["mode"] == "fixture"
+    assert raw["gaojixing"]["provenance"] == "fixture:contract-vector"
+    assert evidence["mode"] == "fixture"
+    assert evidence["provenance"] == "fixture:contract-vector"
+    assert raw["packageDigest"] == build_question_package(
+        node_params=node.params,
+        adapter_config={"fixtureProvenance": "fixture:contract-vector"},
+        runtime_payload={"question": "runtime"},
+    ).digest
+    assert outputs["fixture-source"][0]["lineage"][0]["mode"] == "fixture"
+    assert (
+        outputs["fixture-source"][0]["lineage"][0]["provenance"]
+        == "fixture:contract-vector"
+    )
+    assert emitter.events[-2][2]["details"]["liveAccepted"] is False
+
+
+@pytest.mark.asyncio
+async def test_live_source_never_falls_back_to_fixture_items_when_session_is_unavailable(
+    monkeypatch,
+):
+    node = SimpleNamespace(
+        id="live-source",
+        kind="source",
+        adapter=None,
+        depends_on=[],
+        params={"question": "q", "fixtureItems": [{"content": "must not be used"}]},
+        runtime={
+            "binding": {
+                "input": {
+                    "channelType": "doubao_research",
+                    "liveMode": "live",
+                    "adapterConfig": {"capabilityId": GAOJIXING_CAPABILITY_ID},
+                }
+            }
+        },
+    )
+    body = SimpleNamespace(
+        input=SimpleNamespace(payload={}),
+        project=SimpleNamespace(
+            id="workflow",
+            agentPermissions=SimpleNamespace(canFetchNetwork=True),
+        ),
+    )
+    emitter = _Emitter()
+    outputs = {}
+
+    async def unavailable(**_kwargs):
+        raise GaojixingReadinessError("gaojixing_session_unavailable", "session unavailable")
+
+    monkeypatch.setattr("backend.workflow.opencli_hda_tracer.capture_live_doubao", unavailable)
+    await _execute_gaojixing_source(
+        node,
+        body=body,
+        run_id="run",
+        workflow_id="workflow",
+        trace_id="trace",
+        outputs_by_node=outputs,
+        emitter=emitter,
+        session=None,
+    )
+
+    assert outputs["live-source"] == []
+    assert [event[1] for event in emitter.events] == ["blocked"]
+    details = emitter.events[0][2]
+    assert details["details"] == {"nodeId": "live-source"}
+    assert details["block_reason"].code == "gaojixing_session_unavailable"
+
+
+@pytest.mark.asyncio
+async def test_mock_delivery_ack_remains_mock_business_outcome(monkeypatch):
+    from backend.notifiers.base import NotificationSendResult
+    from backend.workflow.webhook_delivery import execute_workflow_webhook_delivery
+
+    package = build_question_package(
+        node_params={"question": "q"}, adapter_config={}, runtime_payload={}
+    )
+    raw = map_capture_item(
+        {"content": "mock answer", "citations": [], "conversation_url": ""},
+        package=package,
+        workflow_id="workflow",
+        run_id="run",
+        node_id="source",
+        artifact_id="artifact",
+        mode="mock",
+        provenance="mock:contract-vector",
+    )
+    captured = {}
+
+    class _Notifier:
+        async def send(self, _config, payload):
+            captured["payload"] = payload
+            return NotificationSendResult(success=True, response_data={"businessAck": True})
+
+    monkeypatch.setattr(
+        "backend.workflow.webhook_delivery.get_notifier",
+        lambda _kind: _Notifier(),
+    )
+    result = await execute_workflow_webhook_delivery(
+        {"target": "business", "config": {"url": "https://example.test/hook"}},
+        [{"raw": raw, "lineage": [{"nodeId": "source", "mode": "mock"}]}],
+        workflow_id="workflow",
+        run_id="run",
+        node_id="notify",
+    )
+
+    assert result["transportStatus"] == "accepted"
+    assert result["businessOutcome"] == "mock"
+    assert result["ackEvidence"]["status"] == "confirmed"
+    assert result["liveAccepted"] is False
+    assert result["mode"] == "mock"
+    assert result["provenance"] == "mock:contract-vector"
+    assert captured["payload"].data["mode"] == "mock"
+
+
+@pytest.mark.asyncio
+async def test_delivery_rejects_mixed_live_and_mock_contexts():
+    from backend.workflow.webhook_delivery import (
+        WorkflowWebhookDeliveryError,
+        execute_workflow_webhook_delivery,
+    )
+
+    package = build_question_package(
+        node_params={"question": "q"}, adapter_config={}, runtime_payload={}
+    )
+    live = map_capture_item(
+        {"content": "live answer"},
+        package=package,
+        workflow_id="workflow",
+        run_id="run",
+        node_id="source",
+        artifact_id="live-artifact",
+    )
+    mock = map_capture_item(
+        {"content": "mock answer"},
+        package=package,
+        workflow_id="workflow",
+        run_id="run",
+        node_id="source",
+        artifact_id="mock-artifact",
+        mode="mock",
+        provenance="mock:contract-vector",
+    )
+
+    with pytest.raises(WorkflowWebhookDeliveryError) as error:
+        await execute_workflow_webhook_delivery(
+            {},
+            [{"raw": live}, {"raw": mock}],
+            workflow_id="workflow",
+            run_id="run",
+            node_id="notify",
+        )
+
+    assert error.value.code == "gaojixing_delivery_context_mismatch"
+    assert error.value.details["inconsistentFields"] == ["mode", "provenance"]
+
+
+@pytest.mark.asyncio
+async def test_delivery_rejects_contradictory_evidence_envelope():
+    from backend.workflow.webhook_delivery import (
+        WorkflowWebhookDeliveryError,
+        execute_workflow_webhook_delivery,
+    )
+
+    package = build_question_package(
+        node_params={"question": "q"}, adapter_config={}, runtime_payload={}
+    )
+    raw = map_capture_item(
+        {"content": "answer"},
+        package=package,
+        workflow_id="workflow",
+        run_id="run",
+        node_id="source",
+        artifact_id="artifact",
+    )
+    raw["gaojixing"]["evidence"]["mode"] = "mock"
+
+    with pytest.raises(WorkflowWebhookDeliveryError) as error:
+        await execute_workflow_webhook_delivery(
+            {},
+            [{"raw": raw}],
+            workflow_id="workflow",
+            run_id="run",
+            node_id="notify",
+        )
+
+    assert error.value.code == "gaojixing_delivery_evidence_mismatch"
+
+
+@pytest.mark.asyncio
+async def test_delivery_rejects_cross_run_and_package_lineage():
+    from backend.workflow.webhook_delivery import (
+        WorkflowWebhookDeliveryError,
+        execute_workflow_webhook_delivery,
+    )
+
+    package = build_question_package(
+        node_params={"question": "q"}, adapter_config={}, runtime_payload={}
+    )
+    raw = map_capture_item(
+        {"content": "answer"},
+        package=package,
+        workflow_id="workflow",
+        run_id="run",
+        node_id="source",
+        artifact_id="artifact",
+    )
+
+    with pytest.raises(WorkflowWebhookDeliveryError) as error:
+        await execute_workflow_webhook_delivery(
+            {},
+            [
+                {
+                    "raw": raw,
+                    "lineage": [
+                        {
+                            "nodeId": "source",
+                            "packageDigest": "another-package",
+                            "runId": "another-run",
+                        }
+                    ],
+                }
+            ],
+            workflow_id="workflow",
+            run_id="run",
+            node_id="notify",
+        )
+
+    assert error.value.code == "gaojixing_delivery_lineage_mismatch"
+    assert error.value.details["mismatchedFields"] == [
+        "sourceLineage.packageDigest",
+        "sourceLineage.runId",
+    ]
diff --git a/tests/unit/test_operations_agent_runtime_service.py b/tests/unit/test_operations_agent_runtime_service.py
index f6e711aa..3fe399c6 100644
--- a/tests/unit/test_operations_agent_runtime_service.py
+++ b/tests/unit/test_operations_agent_runtime_service.py
@@ -1,6 +1,6 @@
 import asyncio
 
-from sqlalchemy import update
+from sqlalchemy import select, update
 from sqlalchemy.ext.asyncio import AsyncSession, async_sessionmaker
 
 from backend.models.identity import Team, User, Workspace
@@ -16,7 +16,8 @@
 def _model_configuration() -> dict:
     return {
         "agent_contract": {
-            "schema_version": "agent.contract.v1",
+            "schema_version": "agent.contract.v2",
+            "role": "operations_reviewer",
             "input_schema": {
                 "type": "object",
                 "properties": {"target_id": {"type": "string"}},
@@ -31,17 +32,48 @@ def _model_configuration() -> dict:
                 "type": "object",
                 "properties": {"last_target_id": {"type": "string"}},
             },
+            "required_capabilities": ["streaming", "tool_events"],
+            "tool_policy": {"allow": ["read"]},
+            "budget": {"max_turns": 8},
+            "quality_gates": [],
+            "evidence_requirements": [],
         },
         "runtime_binding": {
-            "schema_version": "agent.runtime-binding.v1",
-            "agent_url": "http://agent-runtime.test:19823",
-            "runtime": "pi",
+            "schema_version": "agent.runtime-binding.v2",
             "workflow": "operations-agent",
+            "preferred_agent_urls": ["http://agent-runtime.test:19823"],
+            "preferred_runtimes": ["pi"],
+            "model_binding": {
+                "schema_version": "agent.model-binding.v1",
+                "provider": "openrouter",
+                "model": "anthropic/claude-sonnet",
+                "auth_profile": "operations",
+            },
             "config": {},
         },
     }
 
 
+def _stub_runtime_selection(monkeypatch) -> None:
+    async def select_agent_runtime(*args, **kwargs):
+        return {
+            "schema_version": "agent.runtime-selection.v1",
+            "agent_url": "http://agent-runtime.test:19823",
+            "runtime": "pi",
+            "workflow": "operations-agent",
+            "capabilities": ["model_selection", "streaming", "tool_events"],
+            "provider": "openrouter",
+            "model": "anthropic/claude-sonnet",
+            "auth_profile": "operations",
+        }
+
+    monkeypatch.setattr(
+        operations_agent_runtime_service,
+        "select_agent_runtime",
+        select_agent_runtime,
+    )
+
+
 async def _seed_run(db_session) -> OperationsAgentRun:
     user = User(subject="runtime-owner")
     workspace = Workspace(name="Runtime workspace", slug="runtime-workspace")
@@ -107,18 +139,25 @@ async def test_dispatch_uses_existing_runtime_protocol_and_validates_result(
     run = await _seed_run(db_session)
     session_factory = async_sessionmaker(db_engine, class_=AsyncSession, expire_on_commit=False)
     monkeypatch.setattr(operations_agent_runtime_service, "AsyncSessionLocal", session_factory)
+    _stub_runtime_selection(monkeypatch)
     captured_task = {}
 
     async def send_agent_task(agent_url, task, on_event, timeout):
         captured_task.update(task)
         assert agent_url == "http://agent-runtime.test:19823"
-        assert timeout == 600.0
+        assert timeout == 1800.0
         await on_event(
             {
                 "type": "state",
                 "state": {"last_target_id": "daily-news"},
             }
         )
+        await on_event(
+            {
+                "type": "audit",
+                "audit": {"action": "read", "api_key": "must-not-persist"},
+            }
+        )
         return {
             "type": "done",
             "result": {"summary": "Target inspected"},
@@ -137,10 +176,60 @@ async def send_agent_task(agent_url, task, on_event, timeout):
     assert captured_task["instructions"] == "Inspect the requested target"
     assert captured_task["input"] == {"target_id": "daily-news"}
     assert captured_task["config"]["permission_mode"] == "observe_only"
+    assert captured_task["config"]["timeout_seconds"] == 1800
+    assert captured_task["provider"] == "openrouter"
+    assert captured_task["model"] == "anthropic/claude-sonnet"
+    assert captured_task["required_capabilities"] == ["streaming", "tool_events"]
+    assert captured_task["budget"] == {"max_turns": 8}
     assert run.status == "completed"
     assert run.state_payload == {"last_target_id": "daily-news"}
     assert run.output_payload == {"summary": "Target inspected"}
     assert run.error_message is None
+    assert run.execution_binding["runtime"] == "pi"
+    assert run.evidence_payload["schema_version"] == "agent.run-evidence.v1"
+    assert run.evidence_payload["audit"][1]["api_key"] == "[REDACTED]"
+
+
+async def test_duplicate_dispatch_is_harmless_after_queued_claim(
+    db_engine,
+    db_session,
+    monkeypatch,
+):
+    run = await _seed_run(db_session)
+    session_factory = async_sessionmaker(
+        db_engine,
+        class_=AsyncSession,
+        expire_on_commit=False,
+    )
+    monkeypatch.setattr(operations_agent_runtime_service, "AsyncSessionLocal", session_factory)
+    _stub_runtime_selection(monkeypatch)
+    started = asyncio.Event()
+    release = asyncio.Event()
+    calls = 0
+
+    async def send_agent_task(agent_url, task, on_event, timeout):
+        nonlocal calls
+        calls += 1
+        started.set()
+        await release.wait()
+        return {"type": "done", "result": {"summary": "Target inspected"}}
+
+    monkeypatch.setattr(
+        operations_agent_runtime_service,
+        "send_agent_task",
+        send_agent_task,
+    )
+    first = asyncio.create_task(
+        operations_agent_runtime_service.dispatch_operations_agent_run(run.id)
+    )
+    await started.wait()
+    await operations_agent_runtime_service.dispatch_operations_agent_run(run.id)
+    release.set()
+    await first
+
+    await db_session.refresh(run)
+    assert calls == 1
+    assert run.status == "completed"
 
 
 async def test_dispatch_fails_closed_when_runtime_output_breaks_contract(
@@ -149,6 +238,7 @@ async def test_dispatch_fails_closed_when_runtime_output_breaks_contract(
     run = await _seed_run(db_session)
     session_factory = async_sessionmaker(db_engine, class_=AsyncSession, expire_on_commit=False)
     monkeypatch.setattr(operations_agent_runtime_service, "AsyncSessionLocal", session_factory)
+    _stub_runtime_selection(monkeypatch)
 
     async def send_agent_task(agent_url, task, on_event, timeout):
         return {"type": "done", "result": {"unexpected": True}}
@@ -167,6 +257,49 @@ async def send_agent_task(agent_url, task, on_event, timeout):
     assert "output_schema" in run.error_message
 
 
+async def test_dispatch_fails_closed_when_required_evidence_is_missing(
+    db_engine, db_session, monkeypatch
+):
+    run = await _seed_run(db_session)
+    version = await db_session.scalar(
+        select(PublishedOperationsAgentVersion).where(
+            PublishedOperationsAgentVersion.operations_agent_id
+            == run.operations_agent_id
+        )
+    )
+    configuration = _model_configuration()
+    configuration["agent_contract"]["evidence_requirements"] = ["citation"]
+    version.model_configuration = configuration
+    await db_session.commit()
+    session_factory = async_sessionmaker(
+        db_engine,
+        class_=AsyncSession,
+        expire_on_commit=False,
+    )
+    monkeypatch.setattr(
+        operations_agent_runtime_service,
+        "AsyncSessionLocal",
+        session_factory,
+    )
+    _stub_runtime_selection(monkeypatch)
+
+    async def send_agent_task(agent_url, task, on_event, timeout):
+        return {"type": "done", "result": {"summary": "No evidence"}}
+
+    monkeypatch.setattr(
+        operations_agent_runtime_service,
+        "send_agent_task",
+        send_agent_task,
+    )
+
+    await operations_agent_runtime_service.dispatch_operations_agent_run(run.id)
+
+    await db_session.refresh(run)
+    assert run.status == "failed"
+    assert "citation" in run.error_message
+    assert run.evidence_payload["lineage"][0]["type"] == "agent_runtime"
+
+
 async def test_dispatch_rejects_automatic_profile_without_governed_gateway(
     db_engine, db_session, monkeypatch
 ):
@@ -198,6 +331,7 @@ async def test_paused_run_is_not_overwritten_by_terminal_result(db_engine, db_se
     run = await _seed_run(db_session)
     session_factory = async_sessionmaker(db_engine, class_=AsyncSession, expire_on_commit=False)
     monkeypatch.setattr(operations_agent_runtime_service, "AsyncSessionLocal", session_factory)
+    _stub_runtime_selection(monkeypatch)
 
     async def send_agent_task(agent_url, task, on_event, timeout):
         async with session_factory() as session:
diff --git a/tests/unit/test_operations_agent_schema.py b/tests/unit/test_operations_agent_schema.py
new file mode 100644
index 00000000..d6f599e1
--- /dev/null
+++ b/tests/unit/test_operations_agent_schema.py
@@ -0,0 +1,63 @@
+import pytest
+from pydantic import ValidationError
+
+from backend.schemas.operations_agent import (
+    DEFAULT_DEEP_RUN_TIMEOUT_SECONDS,
+    MAX_DEEP_RUN_TIMEOUT_SECONDS,
+    AgentModelBindingV1,
+    AgentRuntimeBindingV2,
+)
+
+
+def _binding(**overrides):
+    values = {
+        "schema_version": "agent.runtime-binding.v2",
+        "preferred_agent_urls": ["https://agent.example.test"],
+        "preferred_runtimes": ["pi"],
+        "workflow": "operations-agent",
+    }
+    values.update(overrides)
+    return values
+
+
+def test_deep_run_timeout_defaults_are_bounded():
+    binding = AgentRuntimeBindingV2.model_validate(_binding())
+
+    assert binding.dispatch_timeout_seconds == DEFAULT_DEEP_RUN_TIMEOUT_SECONDS == 1800
+    assert MAX_DEEP_RUN_TIMEOUT_SECONDS == 3600
+
+
+def test_runtime_preferences_accept_prime_agent_without_binding_business_role():
+    binding = AgentRuntimeBindingV2.model_validate(
+        _binding(preferred_runtimes=["prime-agent", "codex"])
+    )
+    assert binding.preferred_runtimes == ["prime-agent", "codex"]
+
+
+def test_dispatch_timeout_enforces_hard_bounds():
+    with pytest.raises(ValidationError):
+        AgentRuntimeBindingV2.model_validate(
+            _binding(dispatch_timeout_seconds=MAX_DEEP_RUN_TIMEOUT_SECONDS + 1)
+        )
+    with pytest.raises(ValidationError):
+        AgentRuntimeBindingV2.model_validate(_binding(dispatch_timeout_seconds=0))
+
+
+def test_inner_timeout_override_enforces_same_hard_bounds():
+    with pytest.raises(ValidationError):
+        AgentRuntimeBindingV2.model_validate(
+            _binding(config={"timeout_seconds": MAX_DEEP_RUN_TIMEOUT_SECONDS + 1})
+        )
+    with pytest.raises(ValidationError):
+        AgentRuntimeBindingV2.model_validate(_binding(config={"timeout_seconds": 0}))
+
+def test_model_binding_rejects_control_plane_credentials():
+    with pytest.raises(ValidationError):
+        AgentModelBindingV1.model_validate(
+            {
+                "schema_version": "agent.model-binding.v1",
+                "provider": "openrouter",
+                "model": "anthropic/claude-sonnet",
+                "api_key": "must-stay-on-edge",
+            }
+        )
diff --git a/tests/unit/test_paw_processor.py b/tests/unit/test_paw_processor.py
new file mode 100644
index 00000000..96850507
--- /dev/null
+++ b/tests/unit/test_paw_processor.py
@@ -0,0 +1,304 @@
+import json
+from types import SimpleNamespace
+from unittest.mock import MagicMock, patch
+
+import pytest
+
+from backend.processors.paw_processor import PawProcessor
+from backend.processors.registry import get_processor
+
+
+class FakeResponse:
+    def __init__(self, payload, *, fail=False):
+        self.payload = payload
+        self.fail = fail
+        self.closed = False
+
+    def raise_for_status(self):
+        if self.fail:
+            raise RuntimeError("sidecar unavailable")
+
+    async def aiter_bytes(self):
+        yield json.dumps(self.payload, allow_nan=True).encode("utf-8")
+
+    async def aclose(self):
+        self.closed = True
+
+
+class FakeClient:
+    def __init__(self, responses):
+        self.responses = iter(responses)
+        self.requests = []
+
+    async def __aenter__(self):
+        return self
+
+    async def __aexit__(self, *_args):
+        return None
+
+    def build_request(self, method, url, json):
+        return method, url, json
+
+    async def send(self, request, *, stream):
+        method, url, payload = request
+        assert method == "POST"
+        assert stream is True
+        self.requests.append((url, payload))
+        return next(self.responses)
+
+
+def _record(title="Market note"):
+    record = MagicMock()
+    record.normalized_data = {"title": title, "content": "Short normalized content"}
+    record.raw_data = {"cookies": "must-not-send", "html": "<html>full document</html>"}
+    return record
+
+
+def _settings():
+    return SimpleNamespace(
+        paw_runtime_url="http://paw-runtime:8097",
+        paw_runtime_timeout_seconds=4,
+        paw_program_id="0123456789abcdef",
+        paw_max_tokens=512,
+    )
+
+
+def _response(enrichment=None, **overrides):
+    return {
+        "contractVersion": overrides.get("contractVersion", "opencli.paw.runtime.v1"),
+        "programId": overrides.get("programId", "0123456789abcdef"),
+        "enrichment": {"summary": "ok"} if enrichment is None else enrichment,
+    }
+
+
+def test_paw_processor_is_registered():
+    assert get_processor("paw").processor_type == "paw"
+
+@pytest.mark.asyncio
+async def test_paw_rejects_missing_program_identity_before_http():
+    settings = _settings()
+    settings.paw_program_id = ""
+    with patch("backend.processors.paw_processor.get_settings", return_value=settings), patch(
+        "backend.processors.paw_processor.guarded_async_client"
+    ) as guarded:
+        result = await PawProcessor().process([_record()], "{{title}}", {})
+
+    assert result.success is False
+    assert result.error == "runtime.program_not_configured"
+    guarded.assert_not_awaited()
+
+
+
+@pytest.mark.asyncio
+async def test_paw_posts_only_rendered_normalized_short_text_and_validates_response():
+    client = FakeClient([FakeResponse(_response())])
+    with patch("backend.processors.paw_processor.get_settings", return_value=_settings()), patch(
+        "backend.processors.paw_processor.guarded_async_client",
+        return_value=(client, "http://paw-runtime:8097"),
+    ) as guarded:
+        result = await PawProcessor().process(
+            [_record()], "Classify {{title}}", {"max_tokens": 32}
+        )
+
+    assert result.success is True
+    assert result.enrichments == [{"summary": "ok"}]
+    assert client.requests == [
+        (
+            "http://paw-runtime:8097/v1/enrich",
+            {
+                "programId": "0123456789abcdef",
+                "input": "Classify Market note",
+                "maxTokens": 32,
+            },
+        )
+    ]
+    assert guarded.await_args.kwargs["allow_private"] is True
+
+
+
+@pytest.mark.asyncio
+async def test_paw_accepts_enrichment_matching_optional_output_schema():
+    client = FakeClient([FakeResponse(_response({"summary": "ok"}))])
+    with patch("backend.processors.paw_processor.get_settings", return_value=_settings()), patch(
+        "backend.processors.paw_processor.guarded_async_client",
+        return_value=(client, "http://paw-runtime:8097"),
+    ):
+        result = await PawProcessor().process(
+            [_record()],
+            "{{title}}",
+            {
+                "output_schema": {
+                    "type": "object",
+                    "required": ["summary"],
+                    "properties": {"summary": {"type": "string"}},
+                }
+            },
+        )
+
+    assert result.success is True
+    assert result.enrichments == [{"summary": "ok"}]
+
+
+@pytest.mark.asyncio
+async def test_paw_rejects_enrichment_not_matching_optional_output_schema():
+    client = FakeClient([FakeResponse(_response({"summary": 3}))])
+    with patch("backend.processors.paw_processor.get_settings", return_value=_settings()), patch(
+        "backend.processors.paw_processor.guarded_async_client",
+        return_value=(client, "http://paw-runtime:8097"),
+    ):
+        result = await PawProcessor().process(
+            [_record()],
+            "{{title}}",
+            {"output_schema": {"type": "object", "properties": {"summary": {"type": "string"}}}},
+        )
+
+    assert result.success is False
+    assert result.failed_indices == {0}
+    assert result.enrichments == [{}]
+
+
+@pytest.mark.asyncio
+async def test_paw_rejects_nonlocal_endpoint_and_sensitive_or_blank_prompts():
+    settings = _settings()
+    settings.paw_runtime_url = "https://example.test:8097"
+    with patch("backend.processors.paw_processor.get_settings", return_value=settings), patch(
+        "backend.processors.paw_processor.guarded_async_client"
+    ) as guarded:
+        result = await PawProcessor().process([_record()], "{{title}}", {})
+    assert result.error == "runtime.endpoint_rejected: paw.local_only"
+    guarded.assert_not_awaited()
+
+    with patch("backend.processors.paw_processor.get_settings", return_value=_settings()), patch(
+        "backend.processors.paw_processor.guarded_async_client"
+    ) as guarded:
+        sensitive = await PawProcessor().process([_record()], "{{auth_token}}", {})
+        blank = await PawProcessor().process([_record()], "   ", {})
+    assert sensitive.error == "request.sensitive_placeholder"
+    assert blank.failed_indices == {0}
+    guarded.assert_not_awaited()
+
+
+@pytest.mark.asyncio
+async def test_paw_rejects_invalid_output_schema_before_contacting_sidecar():
+    with patch("backend.processors.paw_processor.get_settings", return_value=_settings()), patch(
+        "backend.processors.paw_processor.guarded_async_client"
+    ) as guarded:
+        result = await PawProcessor().process(
+            [_record()], "{{title}}", {"output_schema": {"type": "not-a-json-schema-type"}}
+        )
+
+    assert result.success is False
+    assert result.error == "request.output_schema_invalid"
+    guarded.assert_not_awaited()
+
+@pytest.mark.asyncio
+async def test_paw_rejects_identity_or_contract_mismatch_without_enrichment():
+    client = FakeClient(
+        [
+            FakeResponse(_response(programId="other.program")),
+            FakeResponse(_response(contractVersion="opencli.other.v1")),
+        ]
+    )
+    with patch("backend.processors.paw_processor.get_settings", return_value=_settings()), patch(
+        "backend.processors.paw_processor.guarded_async_client",
+        return_value=(client, "http://paw-runtime:8097"),
+    ):
+        result = await PawProcessor().process([_record(), _record()], "{{title}}", {})
+
+    assert result.success is False
+    assert result.failed_indices == {0, 1}
+    assert result.enrichments == [{}, {}]
+
+
+@pytest.mark.asyncio
+async def test_paw_rejects_oversized_prompt_before_sidecar_call():
+    client = FakeClient([])
+    with patch("backend.processors.paw_processor.get_settings", return_value=_settings()), patch(
+        "backend.processors.paw_processor.guarded_async_client",
+        return_value=(client, "http://paw-runtime:8097"),
+    ):
+        result = await PawProcessor().process([_record("x" * 8_193)], "{{title}}", {})
+
+    assert result.success is False
+    assert result.failed_indices == {0}
+    assert client.requests == []
+
+
+@pytest.mark.asyncio
+async def test_paw_rejects_empty_rendered_input_without_contacting_sidecar():
+    with patch("backend.processors.paw_processor.get_settings", return_value=_settings()), patch(
+        "backend.processors.paw_processor.guarded_async_client"
+    ) as guarded:
+        result = await PawProcessor().process([_record()], "", {})
+
+    assert result.success is False
+    assert result.failed_indices == {0}
+    guarded.assert_not_awaited()
+
+
+@pytest.mark.asyncio
+async def test_paw_rejects_invalid_token_config_without_contacting_sidecar():
+    with patch("backend.processors.paw_processor.get_settings", return_value=_settings()):
+        result = await PawProcessor().process([_record()], "{{title}}", {"max_tokens": 513})
+
+    assert result.success is False
+    assert result.error == "request.max_tokens_invalid"
+
+
+@pytest.mark.asyncio
+async def test_paw_rejects_guard_failure_without_contacting_sidecar():
+    from backend.security.url_guard import SSRFValidationError
+
+    with patch("backend.processors.paw_processor.get_settings", return_value=_settings()), patch(
+        "backend.processors.paw_processor.guarded_async_client",
+        side_effect=SSRFValidationError("blocked"),
+    ):
+        result = await PawProcessor().process([_record()], "{{title}}", {})
+
+    assert result.success is False
+    assert result.error.startswith("runtime.endpoint_rejected:")
+
+
+@pytest.mark.asyncio
+async def test_paw_rejects_non_json_or_oversized_enrichment():
+    client = FakeClient(
+        [
+            FakeResponse(_response({"not_json": {1}})),
+            FakeResponse(_response({"large": "x" * 65_537})),
+        ]
+    )
+    with patch("backend.processors.paw_processor.get_settings", return_value=_settings()), patch(
+        "backend.processors.paw_processor.guarded_async_client",
+        return_value=(client, "http://paw-runtime:8097"),
+    ):
+        result = await PawProcessor().process([_record(), _record()], "{{title}}", {})
+
+    assert result.success is False
+    assert result.failed_indices == {0, 1}
+
+
+@pytest.mark.asyncio
+async def test_paw_rejects_nonfinite_json_enrichment():
+    client = FakeClient([FakeResponse(_response({"score": float("nan")}))])
+    with patch("backend.processors.paw_processor.get_settings", return_value=_settings()), patch(
+        "backend.processors.paw_processor.guarded_async_client",
+        return_value=(client, "http://paw-runtime:8097"),
+    ):
+        result = await PawProcessor().process([_record()], "{{title}}", {})
+
+    assert result.success is False
+    assert result.failed_indices == {0}
+
+
+@pytest.mark.asyncio
+async def test_paw_continues_after_sidecar_failure():
+    client = FakeClient([FakeResponse({}, fail=True), FakeResponse(_response({"tag": "news"}))])
+    with patch("backend.processors.paw_processor.get_settings", return_value=_settings()), patch(
+        "backend.processors.paw_processor.guarded_async_client",
+        return_value=(client, "http://paw-runtime:8097"),
+    ):
+        result = await PawProcessor().process([_record(), _record()], "{{title}}", {})
+
+    assert result.success is False
+    assert result.failed_indices == {0}
+    assert result.enrichments == [{}, {"tag": "news"}]
diff --git a/tests/unit/test_provider_capacity.py b/tests/unit/test_provider_capacity.py
new file mode 100644
index 00000000..26b42fbc
--- /dev/null
+++ b/tests/unit/test_provider_capacity.py
@@ -0,0 +1,79 @@
+from datetime import datetime, timezone
+from types import SimpleNamespace
+
+import pytest
+from pydantic import ValidationError
+
+from backend.schemas.provider import ModelProviderRead
+from backend.schemas.provider_capacity import (
+    ProviderCapacityRead,
+    ProviderCapacityState,
+)
+
+
+def _provider(**overrides):
+    values = {
+        "id": "provider-1",
+        "name": "Local Claude",
+        "provider_type": "local",
+        "base_url": None,
+        "api_key": None,
+        "default_model": "local-model",
+        "notes": None,
+        "enabled": True,
+        "created_at": datetime.now(timezone.utc),
+        "updated_at": datetime.now(timezone.utc),
+    }
+    values.update(overrides)
+    return SimpleNamespace(**values)
+
+
+def test_unavailable_capacity_serializes_without_guessed_quota():
+    projection = ProviderCapacityRead.unavailable()
+
+    payload = projection.model_dump(mode="json")
+
+    assert payload["state"] == "unavailable"
+    assert payload["usage"] is None
+    assert "remaining_percent" not in payload
+    assert "remaining" not in payload
+
+
+def test_model_provider_projection_defaults_to_unavailable():
+    read = ModelProviderRead.from_model(_provider())
+
+    assert read.capacity.state is ProviderCapacityState.UNAVAILABLE
+    assert read.capacity.usage is None
+    assert read.capacity.model_dump(mode="json")["state"] == "unavailable"
+
+
+def test_measured_capacity_requires_explicit_adapter_data():
+    projection = ProviderCapacityRead.measured(
+        {"window": "five_hour", "used_units": 12},
+        source="documented-provider-usage-v1",
+    )
+
+    assert projection.state is ProviderCapacityState.MEASURED
+    assert projection.usage == {"window": "five_hour", "used_units": 12}
+    assert projection.source == "documented-provider-usage-v1"
+
+    with pytest.raises(ValidationError, match="requires adapter usage data"):
+        ProviderCapacityRead(state=ProviderCapacityState.MEASURED)
+
+
+def test_not_applicable_capacity_has_no_usage_payload():
+    projection = ProviderCapacityRead.not_applicable()
+
+    assert projection.model_dump(mode="json") == {
+        "state": "not_applicable",
+        "usage": None,
+        "measured_at": None,
+        "source": None,
+        "reason": "Runtime has no provider quota semantics",
+    }
+
+    with pytest.raises(ValidationError, match="must not include measured usage data"):
+        ProviderCapacityRead(
+            state=ProviderCapacityState.UNAVAILABLE,
+            usage={"remaining": 0},
+        )
diff --git a/tests/unit/test_scheduler.py b/tests/unit/test_scheduler.py
index f1e7dd05..f54ae5b6 100644
--- a/tests/unit/test_scheduler.py
+++ b/tests/unit/test_scheduler.py
@@ -10,13 +10,17 @@
 
 
 @pytest.fixture(autouse=True)
-def _reset_warned_bad_cron():
+def _reset_warned_bad_cron(monkeypatch):
     """AUDIT C2's warn-once cache is a process-lifetime module-level set by
     design — reset it around every test so test execution order can't leak
     warn state between cases (see triage doc P3-6 on shared process state)."""
     import backend.scheduler as sched_module
 
     sched_module._warned_bad_cron.clear()
+    monkeypatch.setattr(
+        "backend.services.automation_schedule_service.dispatch_due_automations",
+        AsyncMock(return_value=[]),
+    )
     yield
     sched_module._warned_bad_cron.clear()
 
@@ -305,6 +309,36 @@ async def mock_sleep(seconds):
     )
 
 
+@pytest.mark.asyncio
+async def test_scheduler_loop_recovers_queued_scheduled_agent_runs():
+    from backend.scheduler import _scheduler_loop
+
+    t0 = datetime(2024, 6, 1, 12, 0, 0, tzinfo=timezone.utc)
+    t1 = datetime(2024, 6, 1, 12, 1, 0, tzinfo=timezone.utc)
+    iteration = 0
+
+    async def mock_sleep(_seconds):
+        nonlocal iteration
+        iteration += 1
+        if iteration >= 3:
+            raise asyncio.CancelledError()
+
+    recover = AsyncMock(return_value=["scheduled-run-1"])
+    with (
+        patch("asyncio.sleep", side_effect=mock_sleep),
+        patch("backend.scheduler._now", side_effect=[t0, t1]),
+        patch("backend.scheduler._get_enabled_schedules", new=AsyncMock(return_value=[])),
+        patch("backend.executor.get_executor", return_value=AsyncMock()),
+        patch(
+            "backend.services.scheduled_run_recovery.recover_queued_scheduled_runs_local",
+            recover,
+        ),
+    ):
+        await _scheduler_loop()
+
+    recover.assert_awaited_once()
+
+
 @pytest.mark.asyncio
 async def test_scheduler_loop_skips_when_no_fire_in_window():
     """_scheduler_loop does not dispatch schedules with zero fires in the window."""
diff --git a/tests/unit/test_workflow_osint_source_catalog.py b/tests/unit/test_workflow_osint_source_catalog.py
index 3031abb9..3d6d63d4 100644
--- a/tests/unit/test_workflow_osint_source_catalog.py
+++ b/tests/unit/test_workflow_osint_source_catalog.py
@@ -18,6 +18,12 @@
         "source-rsshub-feed",
         "rss",
     ),
+    (
+        "intelligence.source.doubao-research",
+        {"question": "高吉星燕窝酸 DHA 藻油"},
+        "source-doubao-research-capture",
+        "doubao_research",
+    ),
 )
 
 
diff --git a/tests/unit/test_workflow_record_hygiene.py b/tests/unit/test_workflow_record_hygiene.py
index 39ebb21a..80642e48 100644
--- a/tests/unit/test_workflow_record_hygiene.py
+++ b/tests/unit/test_workflow_record_hygiene.py
@@ -10,6 +10,41 @@
     execute_record_hygiene,
 )
 
+
+def _gaojixing_item(
+    *,
+    source_identity: str | None = "https://example.test/conversations/one",
+    package_digest: str = "package-digest",
+    artifact_id: str = "artifact-1",
+    source_id: str = "doubao",
+    run_id: str = "run-1",
+    node_id: str = "source-1",
+) -> dict:
+    raw: dict = {
+        "title": "Captured answer",
+        "url": source_identity or "",
+        "content": "Answer",
+        "source_id": source_id,
+        "published_at": "2026-08-26T00:00:00Z",
+        "packageDigest": package_digest,
+        "questionPackage": {"digest": package_digest},
+        "answerArtifactId": artifact_id,
+        "gaojixing": {
+            "package": {"digest": package_digest},
+            "artifactId": artifact_id,
+            "evidence": {"runId": run_id, "nodeId": node_id},
+        },
+    }
+    if source_identity:
+        raw["dedupe"] = {
+            "type": "source-identity",
+            "field": "conversation_url",
+            "value": source_identity,
+            "status": "unique",
+        }
+    return {"raw": raw}
+
+
 CONTEXT = {"runId": "run-1", "nodeId": "hygiene-1"}
 RAW_ITEMS = [
     {
@@ -71,6 +106,70 @@ def test_normalize_profile_reuses_pipeline_semantics_and_preserves_refs():
     assert result.metrics["normalizedCount"] == 1
 
 
+def test_normalize_canonical_source_identity_accepts_and_preserves_lineage():
+    candidate = execute_record_hygiene(
+        "normalize",
+        [_gaojixing_item()],
+        {},
+        {"runId": "run-1", "nodeId": "normalize-1"},
+    ).records[0]
+    accepted = execute_record_hygiene(
+        "accept",
+        [candidate],
+        {"mode": "automatic_with_review"},
+        {"runId": "run-1", "nodeId": "accept-1"},
+    )
+
+    assert accepted.rejected == []
+    assert accepted.records[0]["status"] == "accepted"
+    assert candidate["dedupe"]["type"] == "canonical-dedupe"
+    assert candidate["dedupe"]["basis"]["packageDigest"] == "package-digest"
+    assert candidate["dedupe"]["artifact"] == {
+        "artifactId": "artifact-1",
+        "packageDigest": "package-digest",
+    }
+    assert candidate["raw"]["gaojixing"]["evidence"] == {"runId": "run-1", "nodeId": "source-1"}
+    assert candidate["lineage"][-1] == {
+        "step": "normalize",
+        "index": 0,
+        "runId": "run-1",
+        "nodeId": "normalize-1",
+    }
+
+
+def test_canonical_source_identity_is_retry_stable_and_business_scoped():
+    first = execute_record_hygiene(
+        "normalize", [_gaojixing_item(artifact_id="artifact-1", run_id="run-1")], {}, CONTEXT
+    ).records[0]
+    retry = execute_record_hygiene(
+        "normalize", [_gaojixing_item(artifact_id="artifact-2", run_id="run-2")], {}, CONTEXT
+    ).records[0]
+    other = execute_record_hygiene(
+        "normalize",
+        [_gaojixing_item(source_identity="https://example.test/conversations/two")],
+        {},
+        CONTEXT,
+    ).records[0]
+
+    assert first["dedupe"]["identity"] == retry["dedupe"]["identity"]
+    assert first["dedupe"]["fingerprint"] == retry["dedupe"]["fingerprint"]
+    assert first["dedupe"]["artifact"]["artifactId"] != retry["dedupe"]["artifact"]["artifactId"]
+    assert first["dedupe"]["identity"] != other["dedupe"]["identity"]
+
+
+def test_missing_canonical_source_identity_remains_review_required():
+    candidate = execute_record_hygiene(
+        "normalize", [_gaojixing_item(source_identity=None)], {}, CONTEXT
+    )
+    result = execute_record_hygiene(
+        "accept", candidate.records, {"mode": "automatic_with_review"}, CONTEXT
+    )
+
+    assert result.records == []
+    assert result.rejected[0]["disposition"] == "review"
+    assert result.rejected[0]["rejection"]["reasons"] == ["dedupe_required"]
+
+
 def test_dedupe_profile_parses_compound_key_and_uses_24_hour_window():
     candidates = execute_record_hygiene("normalize", RAW_ITEMS, {}, CONTEXT).records
     result = execute_record_hygiene(
diff --git a/tests/unit/worker/test_beat_schedule.py b/tests/unit/worker/test_beat_schedule.py
index d2654270..0c3942da 100644
--- a/tests/unit/worker/test_beat_schedule.py
+++ b/tests/unit/worker/test_beat_schedule.py
@@ -6,6 +6,8 @@
 from celery.schedules import crontab
 
 from backend.worker.beat_schedule import parse_cron_expression
+from backend.worker.celery_app import celery_app
+from backend.worker import tasks as worker_tasks
 
 
 def test_parse_cron_expression_valid():
@@ -16,3 +18,31 @@ def test_parse_cron_expression_valid():
 def test_parse_cron_expression_wrong_field_count():
     with pytest.raises(ValueError, match="need 5 fields"):
         parse_cron_expression("* * * *")
+
+
+def test_automation_scheduler_tick_is_wired_into_celery_beat():
+    entry = celery_app.conf.beat_schedule["scheduled-automation-tick"]
+    assert entry["task"] == "run_automation_scheduler_tick"
+    assert entry["schedule"] == 60.0
+
+
+def test_automation_tick_enqueues_every_queued_scheduled_run(monkeypatch):
+    monkeypatch.setattr(
+        worker_tasks,
+        "post_control_plane",
+        lambda _path, _payload: {
+            "success": True,
+            "data": {"queued_run_ids": ["run-1", "run-2"]},
+        },
+    )
+    enqueued: list[str] = []
+    monkeypatch.setattr(
+        worker_tasks.dispatch_scheduled_operations_agent_run,
+        "delay",
+        enqueued.append,
+    )
+
+    result = worker_tasks.run_automation_scheduler_tick.run()
+
+    assert enqueued == ["run-1", "run-2"]
+    assert result["dispatch_enqueued_run_ids"] == ["run-1", "run-2"]