v1.26.1.0 fix: gbrain-sync orchestrator resolves sibling via import.meta.dir

[garrytan]

Summary

Source fix (1 line): bin/gstack-gbrain-sync.ts:222 — runBrainSyncPush now resolves its sibling gstack-brain-sync binary via join(import.meta.dir, "gstack-brain-sync") instead of a hardcoded ~/.claude/skills/gstack/bin/... path. This matches the pattern already in runMemoryIngest at line 193.

The orchestrator's curated-git-push stage now fires from any host install — Claude Code, Codex CLI, dev workspace — instead of silently no-op'ing with "skipped (gstack-brain-sync not installed)" outside the Claude Code install root. Closes Codex outside-voice finding M9.

Maintenance bundled in:

• Test ratchet: plan-review preamble byte budget bumped 33 KB → 34 KB. v1.25.1.0 (recommendation pattern) + v1.26.0.0 (gbrain sync block) added intentional preamble bytes; the ratchet just hadn't been bumped to track. Comment in the test already authorizes this kind of intentional-growth bump.
• Golden fixtures: claude-ship-SKILL.md and factory-ship-SKILL.md regenerated against the live /ship template. v1.25.1.0 added the canonical Recommendation: <action> because ... line to the adversarial subagent prompts; the goldens hadn't been re-baked.

Commits

0fd81ffd chore: bump version and changelog (v1.26.1.0)
5dcb93a5 chore: bump plan-review preamble ratchet + regenerate ship goldens
9c638da2 fix: gbrain-sync orchestrator resolves brain-sync sibling via import.meta.dir

Test Coverage

New regression test in test/gstack-gbrain-sync.test.ts runs the orchestrator with a synthetic HOME (no ~/.claude/skills/gstack tree), then asserts the lying-skip path is no longer taken. Pre-fix: "skipped (gstack-brain-sync not installed)". Post-fix: real spawn output (or honest spawn-error code), never the misleading skip message.

The existing runMemoryIngest precedent has no path-resolution test; this PR establishes the pattern. Out-of-scope to retroactively cover runMemoryIngest here.

Tests: 456 → 457 (+1 new). 0 fails.

End-to-end smoke from this Conductor workspace:

bun run bin/gstack-gbrain-sync.ts --incremental --no-code --no-memory --quiet
# .gbrain-sync-state.json: {"name":"brain-sync","ran":true,"ok":true,"summary":"curated artifacts pushed"}

Pre-Landing Review

Diff scope: 24 source-impacting lines + 5 lines of fixture regen. No SQL, no LLM trust boundary, no conditional side effects, no auth surface. Claude adversarial subagent skipped for proportionality — the Codex M9 finding IS the outside voice on record for the source change, and the maintenance work (ratchet + goldens) is mechanical sync to existing template state.

Plan Review

/plan-eng-review ran on this branch and returned CLEAR with 1 test gap resolved (the regression test above). See plan file for full review report.

Test plan

• bun test test/gstack-gbrain-sync.test.ts — 9 pass (8 existing + 1 new regression)
• bun test test/gen-skill-docs.test.ts — 374 pass, including the Option A budget test that was red on origin/main
• bun test test/host-config.test.ts — 73 pass, including all three ship-skill golden-file regressions
• End-to-end smoke from a non-~/.claude/skills/gstack/ workspace returns summary: "curated artifacts pushed" instead of the misleading skip message

🤖 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

0/0 tests passed | $0 total cost | 12 parallel runners

Suite	Result	Status	Cost

---

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