feat: AI Executive Briefing + minimal UI pass

[nfaggian]

Summary

Two changes on this branch, both aimed at increasing signal per pixel:

1. AI Executive Briefing — a Generate/Regenerate button on the dashboard produces a 2–3 paragraph leadership summary composed by an LLM from live pipeline, capacity, and delivery data. Works with any OpenAI-compatible provider.
2. Minimal UI pass — every element on screen should convey information that leads to a decision. Removed everything that didn't: non-functional header controls, hardcoded sidebar decoration, and dashboard panels that duplicated info already surfaced by KPIs, charts, or the AI briefing.

Stacked on top of cursor/simplify-codebase-535c. Base branch is set accordingly.

---

Part 1 — AI Executive Briefing

Backend (+237 / -4)

• delta_command.briefing
  • build_context(db) — turns the current Database into a compact prompt (team headline + top 5 overallocated engineers, pipeline totals + late-stage count, delivery status + at-risk projects). Total prompt is ~10 short lines regardless of team size, so token cost stays flat as engineers grow.
  • call_llm(system, user) — async POST to {LLM_BASE_URL}/chat/completions with Authorization: Bearer {LLM_API_KEY}. Uses httpx.AsyncClient so the FastAPI event loop stays free.
  • Custom exceptions (LLMNotConfigured, LLMError) surface as 503 / 502 at the API layer.
• POST /api/briefing — { "briefing": "..." } on success; { "detail": "..." } with a specific reason on failure.
• httpx promoted from dev to runtime dependency.
• 4 new tests covering: context summary shape, missing-key 503, happy path (mocked transport), upstream 500 → 502.

Frontend (+118 / -1)

• ExecutiveBriefingPanel.tsx — Sparkles-marked card with Generate / Regenerate button, spinner, whitespace-preserving prose, and an inline error banner. Placed between KPIs and Action Required so the AI narrative sits next to the raw metrics.
• core/api.ts::generateBriefing() — thin wrapper hitting POST /api/briefing.
• apiFetch now parses {detail} from failed responses so users see actionable guidance (e.g. "LLM_API_KEY is not set. Configure an OpenAI-compatible provider (see backend/README.md).") rather than a generic HTTP status.

Configuration

Set in Cursor Dashboard → Cloud Agents → Secrets or your local env:

# OpenAI (default)
LLM_API_KEY=sk-...

# Groq
LLM_API_KEY=gsk_...
LLM_BASE_URL=https://api.groq.com/openai/v1
LLM_MODEL=llama-3.1-70b-versatile

# Local Ollama (any string works as the key)
LLM_API_KEY=ollama
LLM_BASE_URL=http://localhost:11434/v1
LLM_MODEL=llama3.2

Without a key, the panel renders a clear onboarding hint and the endpoint returns 503 — the app remains fully functional otherwise.

---

Part 2 — Minimal UI pass

Chrome removed (Header.tsx, Sidebar.tsx)

• Header: non-functional Search input, Bell/notification badge, and "NF" avatar all gone. Now: page title + optional data-carrying meta + today's date.
• Sidebar: hardcoded "Q3 Review — Leadership sync" card gone. "Engineering Hub" tagline gone. Brand mark + nav is all that remains.

Dashboard: 10 sections → 6

Removed	Why
KPI card Late-Stage Pipeline	count already carried by Pipeline chart + AI briefing
KPI card In Negotiation	subset of above; visible on the same chart
PipelineHealthPanel (deleted file, -121 lines)	4 stats verbatim duplicate of KPI cards + chart
TeamSnapshotPanel (deleted file)	overallocated list is verbatim in Action Required; other stats are in KPI + Utilization chart
DeliveryOverviewPanel (deleted file, -164 lines)	budget burn is a KPI; status split is in Projects Needing Attention
Footer strip (Last updated · N engineers · N opps)	date is in header; counts are in AI briefing

New structure:

Header              [title]                                    [date]
Row 1 — 4 KPI cards:  Pipeline · Delivery · Budget · Utilization
Row 2 — AI Executive Briefing
Row 3 — Action Required
Row 4 — Pipeline chart          |  Utilization chart
Row 5 — Upcoming Closes         |  Upcoming Milestones
Row 6 — Projects Needing Attention  (rendered only when at-risk > 0)

Other pages

• Opportunities: 4 stat cards → 3 (fold Lost into the Won this cycle card as a small trailing annotation). Delete the redundant won/lost chip strip below the Board/List toggle. Header meta now reads 6 active · 1 won · 1 lost.
• Team: remove the Overview view toggle. The Timeline view already shows current-week utilization and every other week and supports inline editing — Overview was a strictly less capable duplicate. Also deletes the orphaned EngineerCard.tsx.
• Projects: drop the decorative status chip strip (sections below already group by status). 4 stat cards → 3.

Analytics pruning

dashboard-analytics.ts no longer computes 12 fields that only fed the deleted panels. Internal computations that feed the actions[] list (overdue milestones, unassigned engineers, etc.) are kept.

Verification

• ✅ uv run pytest — 13/13 pass
• ✅ npm run build — clean. Bundle size shrank on every page:
  • dashboard 106 kB → 105 kB
  • opportunities 4.57 kB → 4.06 kB
  • projects 4.68 kB → 3.98 kB
  • team 6.41 kB → 4.95 kB
• ✅ Live-checked every page.

Dashboard (full length)

Minimal dashboard

Other pages

Opportunities
Team
Projects

Line-count delta on this branch

Commit	Net
Add LLM briefing (backend + frontend + tests)	+355 / -5
Remove decorative chrome + duplicate dashboard panels	+164 / -641
Total	+519 / -646

The dashboard now presents 4 KPIs, one AI-authored paragraph of narrative, an action list, two charts, and two lookahead lists. Nothing is on screen unless it drives a decision.

_{To show artifacts inline, enable in settings.}