feat: extract coding:architecture-dimensions-assistant agent + guide

[bborbe]

Summary

Closes the MAJOR finding from PR #59's bot review: `/coding:architecture-review` Agent B routed to generic `claude` with a ~200-word inline checklist — a "thin wrapper" violation.

This PR extracts the 8-dimension behavioral review into a paired doc + agent:

• `docs/architecture-dimensions-guide.md` — source of truth for the 8 dimensions (data flow, failure & resilience, concurrency & lifecycle, observability, cross-cutting consistency, config/secrets/blast radius, evolvability, architectural drift), severity rubric, output contract, anti-patterns, skip directives
• `agents/architecture-dimensions-assistant.md` — agent that reads the guide and applies it; matches frontmatter conventions of `go-architecture-assistant` / `srp-checker`
• `commands/architecture-review.md` — Agent B prompt collapses from inlined 200-word checklist to one-line "Apply the guide" directive

Why this matters

• Single source of truth. Updating the dimensions checklist now requires one edit (doc) instead of three (command + agent prompt + vault guide)
• Maintainable contract. Agent file documents output format + severity calibration explicitly
• Pattern consistency. Matches Doc↔Agent alignment principle in `CLAUDE.md` — every enforceable guide has a matching agent

Indexed in

• README.md (Workflows & Documentation table + Go agents table)
• llms.txt
• CLAUDE.md Doc↔Agent table

Test plan

• `make precommit` green (30 checks)
• Local self-review loop converged (1 pass; no Must/Should, 2 optional nits)
• Manual smoke test of `/coding:architecture-review` with the new agent (deferred until merge + plugin update — the v0.24.0 smoke test already validated the parent command's plumbing)

[ben-s-pull-request-reviewer]

Now I have the complete picture. Let me run the Step 4b-i candidate computation manually for the changed files:

• CHANGELOG.md → changelog/conventional-prefix-required
• CLAUDE.md → claude-md/agent-context-not-user-docs
• agents/architecture-dimensions-assistant.md → agent-cmd/agent-frontmatter
• commands/architecture-review.md → agent-cmd/command-frontmatter, agent-cmd/command-thin

No Go/Python files, no k8s manifests, no commits — all Go/Python/teamvault/k8s rules have **/*.go, **/main.go, k8s/**, **/*.py, @commits, .env triggers that don't match. The **/*.md trigger on markdown-todo/lowercase-x-for-complete doesn't apply — the diff modifies no checkboxes.

Selector mode — Step 4c-sel CLASSIFY:

rule-id	decision	reason
changelog/conventional-prefix-required	applicable	CHANGELOG.md modified; feat: prefix present
claude-md/agent-context-not-user-docs	applicable	CLAUDE.md modified; entry is agent-operational (correct)
agent-cmd/agent-frontmatter	applicable	agents/architecture-dimensions-assistant.md created
agent-cmd/command-frontmatter	applicable	commands/architecture-review.md modified
agent-cmd/command-thin	applicable	commands/architecture-review.md modified (routing change)
readme/user-facing-not-agent-context	skipped	no violation found; description is user-facing

Selector mode — Step 4d-sel ADJUDICATE:

Reading the relevant ### RULE blocks and diff:

• changelog/conventional-prefix-required: The ## Unreleased bullet starts with feat: — correct conventional prefix. No violation.

• claude-md/agent-context-not-user-docs: The new CLAUDE.md table entry maps architecture-dimensions-guide.md → architecture-dimensions-assistant. This is agent-operational context (maps docs to agents), not user-facing marketing. No violation.

• agent-cmd/agent-frontmatter: New agent has all required fields: name: architecture-dimensions-assistant, description (whole-codebase behavioral review, sibling to go/python-architecture-assistant), model: sonnet, effort: high, tools: Read,Grep,Glob,Bash, color: purple, allowed-tools: Bash(grep:*),Bash(find:*),Bash(awk:*),Bash(git:*). The sibling agent references (go-architecture-assistant, python-architecture-assistant) match existing files. No violation.

• agent-cmd/command-frontmatter: Modified command has all required fields: allowed-tools: Task,Bash(git status:+),Bash(git log:+),Bash(git branch:+),Glob,Read, argument-hint: "[directory]", description. No violation.

• agent-cmd/command-thin: Agent B routing changed from generic claude + 200-word inline checklist to coding:architecture-dimensions-assistant + 1-line prompt. The collapsed routing delegates to a properly documented, guide-paired agent with a richer rubric than the inline checklist. No violation — behavior is preserved and quality is improved.

Step 4d citation validation: No findings emitted → nothing to validate.

Step 3 context check: No .env, no k8s YAMLs, no main.go, no CHANGELOG.md content beyond the bullet → no context-specific conventions needed.

Concerns from plan — all addressed:

• Agent frontmatter ✅ — all 7 fields present and appropriate
• Sibling agent names ✅ — go-architecture-assistant.md and python-architecture-assistant.md both exist
• Guide cross-references ✅ — references go-architecture-assistant/python-architecture-assistant, /coding:architecture-review, docs/architecture-dimensions-guide.md (self-referential, correct)
• Command routing collapse ✅ — functionally equivalent, higher quality (guide-paired, detailed rubric)
• CLAUDE.md table ✅ — architecture-dimensions-guide.md → architecture-dimensions-assistant
• README tables ✅ — guide description and agent description accurate
• llms.txt ✅ — correctly captures guide purpose and paired agent

---

{
  "verdict": "approve",
  "summary": "Clean extraction PR: new `architecture-dimensions-assistant` agent and `docs/architecture-dimensions-guide.md` are well-formed, fully frontmatter-compliant, correctly reference sibling agents, and the `commands/architecture-review.md` routing collapse from generic `claude` + inline checklist to the new guide-paired agent preserves all behavioral coverage. All four judgment rules applicable to the diff (changelog prefix, CLAUDE.md operational context, agent frontmatter, command frontmatter/thin) pass without violation. README and llms.txt entries are accurate.",
  "comments": [],
  "concerns_addressed": [
    "correctness: agent frontmatter complete (name, description, model:sonnet, effort:high, tools, color, allowed-tools) — agents/architecture-dimensions-assistant.md",
    "correctness: sibling agent references 'go-architecture-assistant' and 'python-architecture-assistant' match existing files — agents/architecture-dimensions-assistant.md:3",
    "correctness: guide cross-references consistent (sibling agents, /coding:architecture-review, self) — docs/architecture-dimensions-guide.md:200-205",
    "correctness: command routing collapse preserves behavioral coverage, improves quality — commands/architecture-review.md:59-62",
    "correctness: CLAUDE.md table entry maps architecture-dimensions-guide.md → architecture-dimensions-assistant — CLAUDE.md:79",
    "correctness: README tables updated with accurate descriptions — README.md:173, README.md:223",
    "correctness: llms.txt entry captures guide purpose and paired agent — llms.txt:88",
    "tests: precommit skipped (selector mode) — CI covers lint+test"
  ]
}