[codex] Build Strand cockpit TUI

[TSchonleber]

Summary

• Add the strand cockpit operator entrypoint and a width-safe, gamified Ink dashboard with mission/pulse/shield/reach panels.
• Wire the cockpit to the existing local operator snapshot data: review queue, action stats, guardrails, X health, and follower deltas.
• Add Phase 3/4 metrics plumbing, docs/runbook updates, CLI/status/review surfacing, and focused coverage for the cockpit and actor behavior.
• S5: Skill lifecycle + brainctl integration + Ink renderer bridge (Devin-4)

Why

The existing TUI wrapped badly in normal terminal widths and hid the available controls. This makes the cockpit usable as a pinned terminal: clearer controls, responsive rendering, a ? help menu, and tests that enforce 80-column safety.

S5 Changes (Devin-4)

Skill storage (src/db/schema.sql):

• skill_records table: tracks per-skill usage_count, success_count, token_cost_samples_json (for p50/p95), last_used_at, trust_score, triggers, supersedes, and status (active | retired | draft | queued_draft | queued_retire).
• skill_decisions table: audit trail of accept/reject decisions with 30-day rejection suppression.

Skill lifecycle (src/agent/skills/lifecycle.ts):

• SkillRecordStore: upsert, usage recording, status transitions, per-status listing.
• runNightlyScorer(): queues retire proposals when hit-rate < 0.15, success-rate < 0.5 (n >= 10), or superseded. Respects 30-day rejection suppression.
• acceptProposal() / rejectProposal(): transition skill status and record decisions.

Brainctl integration:

• toBrainctlDecisionEvent(): maps SkillDecision to brainctl-compatible event shape.
• SkillDecisionStore.record() auto-computes suppressed_until on rejection (30 days).

Ink renderer bridge (src/cockpit/ink/bridge.ts):

• createInkBridge(): subscribes to core EventBus, optional type filtering, destroy() cleanup.
• createSkillEventBridge(): filtered bridge for skill.proposal + skill.decision events.
• Classic gamified TUI preserved; this adds the chat-first renderer path alongside it.

Validation

• pnpm typecheck — clean
• pnpm exec vitest run tests/agent/skill-lifecycle.test.ts tests/cockpit/ink-bridge.test.ts — 20 tests pass
• pnpm exec biome check — clean on all touched files

Review & Testing Checklist for Human

• Verify skill_records / skill_decisions schema matches §6 expectations
• Confirm nightly scorer thresholds (0.15 hit-rate, 0.5 success-rate with n>=10) match spec intent
• Check brainctl adapter shape (BrainctlDecisionEvent) — clean adapter path, slots into brainctl skill category when it's added

Notes

• CockpitEvent schema NOT altered — skill.proposal and skill.decision types were already in S0 scaffold
• S4 frontmatter shape consumed, not rewritten — triggers_json mirrors what S4 skill files declare
• Ink bridge depends only on cockpit/core events, maintaining the two-invariant contract

Link to Devin session: https://app.devin.ai/sessions/181c7d5b9b714a65af8850685f1556f5
Requested by: @TSchonleber