great_cto

Measured product quality + mechanical pipeline + board server split

• Product-quality harness (P1–P4) — automated 0–100 score for shipped products:
  archetype-aware rubric, fleet scoring, empirical benchmark across 6 generated
  archetypes (89/100), SCORE artifact, UI-heuristic fix.
• QUALITY-DEEPEN #1–#5 — executed-quality harness + deploy gate, per-archetype
  domain contracts, unified quality verdict with trend tracking and gate.
• Pipeline autochain — mechanical agent handoff via shared/pipeline.toml map +
  PostToolUse dispatcher; fleet-wide canonical verdict logging; reviewer family
  unified with HANDOFF fallback in the dispatcher.
• Agent fixes (waves 2–3) — mechanical 72h gate expiry + mid-session reviewer
  nudge hooks, runnable l3-support escalation ladder, coordinator lane-overlap
  check (WPL = Decomposition Matrix), AI-pair contract sync, security-officer
  mode/templates/waiver extraction, shared boilerplate factored into
  agents/_shared/, bd CLI verb drift fixed, db-migration-reviewer wired into
  qa-engineer (was orphaned).
• Board server split — packages/board/server.mjs (3,414 lines, hottest file
  in the repo) split into a 133-line entry + 15 single-responsibility modules under
  packages/board/lib/ (design: docs/arch/ARCH-board-server-split.md). Strictly
  behavior-preserving: full suite 143/143 green, QA report byte-diffed the largest
  moved functions and all 26 routes, security review confirmed CSRF/origin guard
  and body-size cap unchanged. One commit per module — the branch bisects cleanly.
• Docs — README refresh (61 agents, measured quality, escape hatch,
  cross-model review).