v1.29.0.0 feat: worktree-aware gbrain code sources via path-hash IDs and CWD pin

[garrytan]

Summary

Conductor sibling worktrees of the same repo no longer collide on a shared
gstack-code-<slug> source ID. /sync-gbrain derives a path-hashed source ID
per worktree, runs gbrain sources attach to write .gbrain-source in the
worktree root (kubectl-style CWD context), and removes the legacy unsuffixed
source on first new-format sync.

Bug fixes (surfaced by /codex adversarial review during /ship):

• Silent attach failure now reported as stage failure
• Startup preamble checks per-worktree pin, not global state
• Code stage no longer skipped on remote-MCP installs (early-exit was in SKILL template)
• Source registration routes through lib/gbrain-sources.ts only (deleted the orchestrator's near-duplicate)

Requires gbrain v0.30.0+. Spike report: ~/.gstack/projects/garrytan-gstack/2026-05-08-gbrain-split-engine-spike.md.

Test Coverage

[+] bin/gstack-gbrain-sync.ts
  ├── deriveCodeSourceId() with pathhash
  │   ├── [★★★ TESTED] worktree-distinct ids
  │   ├── [★★★ TESTED] same-path stable across runs
  │   ├── [★★  TESTED] long remote, no-origin, pathological basename (existing)
  ├── runCodeImport async + lib ensureSourceRegistered
  │   └── [★★  TESTED] dry-run orchestration unchanged
  ├── sources attach step
  │   └── [★★  TESTED] dry-run preview includes attach
  └── legacy source cleanup
      └── [★★  TESTED] dry-run preview shows new flow

[+] sync-gbrain/SKILL.md.tmpl + preamble
  └── [★★  TESTED] gen-skill-docs freshness (0 STALE)

COVERAGE: 100% | Tests: 405 → 408 (+3 new, +1 updated)

Pre-Landing Review

No critical issues found in the diff (no SQL/data safety, no LLM trust boundary, no auth surfaces touched). All 3 codex adversarial findings auto-fixed before commit:

• HIGH: silent attach failure → stage now returns ok:false with reason
• HIGH: preamble checks .gbrain-source in cwd, not global state
• MEDIUM: legacy gstack-code-<slug> source removed on first new-format sync

Plan Completion

All applicable items DONE:

• ✅ Phase 0 spike (gbrain v0.30 verified, --db absent, sources attach is the routing primitive)
• ✅ Worktree-aware source ID via path hash
• ✅ Delete ensureSourceRegisteredSync duplicate (Codex Feedback #12)
• ✅ Add gbrain sources attach step
• ✅ Path 4 early-exit removed from SKILL template (Codex refactor: reorganize codebase into modular structure #2)
• ✅ CLAUDE.md guidance block updated for per-worktree story
• ✅ .gbrain-source in .gitignore
• ✅ Tests for worktree-aware behavior

⏭️ Codex #14 detect-layer extension, #15 schema migration, tests #24-28: N/A — original plan was architected for Option A (per-worktree HTTP serve + project-scoped MCP namespace) which the spike invalidated. Option B (CLI via Bash) doesn't have the surfaces those tests covered.

TODOS

No TODO items completed in this PR.

Test plan

• bun test test/gstack-gbrain-sync.test.ts test/gbrain-sources.test.ts test/gen-skill-docs.test.ts → 408 pass / 0 fail
• Type check + bundle: bun build --target=bun bin/gstack-gbrain-sync.ts succeeds
• gen:skill-docs --host all --dry-run reports 0 STALE
• Manual smoke: gbrain sources attach writes .gbrain-source in worktree, git check-ignore matches the new entry
• Live /sync-gbrain --full in this worktree — full reindex of 53K LOC takes ~25-35 min; user runs when convenient

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith with what you need.}

• Let Codesmith autofix CI failures and bot reviews

[github-actions]

E2E Evals: ✅ PASS

58/58 tests passed | $7.58 total cost | 12 parallel runners

Suite	Result	Status	Cost
e2e-browse	4/4	✅	$0.17
e2e-deploy	6/6	✅	$1.22
e2e-design	4/4	✅	$0.81
e2e-plan	8/8	✅	$1.83
e2e-qa-workflow	3/3	✅	$1
e2e-review	6/6	✅	$1.68
e2e-workflow	3/3	✅	$0.39
llm-judge	24/24	✅	$0.48

---

12x ubicloud-standard-2 (Docker: pre-baked toolchain + deps) | wall clock ≈ slowest suite