fix: resolve codex exec -C repo root eagerly to prevent wrong-project reviews (v0.12.6.0) (#549)

* refactor: remove 6 dead resolver function copies from gen-skill-docs.ts

These functions were moved to scripts/resolvers/{review,design}.ts but the
old copies in gen-skill-docs.ts were never deleted. They are defined but
never called — the RESOLVERS map from resolvers/index.ts is the live
dispatch. The dead copies had already diverged from the live versions.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* fix: resolve codex exec -C repo root eagerly to prevent wrong-project reviews

When codex exec commands run in background bash tasks (e.g., Conductor
workspaces), $(git rev-parse --show-toplevel) evaluates in whatever cwd
the background shell inherits, which may be a different project. Fix by
resolving _REPO_ROOT once at the top of each bash block and referencing
the stored value in -C.

12 occurrences fixed across 4 source files:
- codex/SKILL.md.tmpl (3)
- autoplan/SKILL.md.tmpl (3)
- scripts/resolvers/review.ts (3)
- scripts/resolvers/design.ts (3)

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* test: regression guard for codex exec inline git rev-parse in -C flag

Scans all .tmpl and resolver .ts source files for codex exec commands
that use inline $(git rev-parse --show-toplevel) in the -C flag. This
pattern causes wrong-project reviews in Conductor workspaces. The test
ensures nobody reintroduces the old pattern.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* chore: bump version and changelog (v0.12.6.0)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: address adversarial review findings — codex review cwd, test scope, fail-loud

1. codex review commands now cd to $_REPO_ROOT (review doesn't support -C)
2. Autoplan codex commands converted from prose "Prerequisite" to fenced bash blocks
3. || pwd fallback replaced with hard fail — silent wrong-dir is worse than error
4. Regression test now scans all resolver .ts files + generated SKILL.md files

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

* test: harden regression test — Bun.Glob, SKILL.md scan, codex review check

Fixes three gaps found by adversarial review:
1. fs.readdirSync recursive hits ELOOP on .claude/skills/gstack symlink.
   Switched to Bun.Glob with followSymlinks:false.
2. Generated SKILL.md files now scanned (not just .tmpl sources).
3. New test: codex review commands must not use inline git rev-parse
   (codex review doesn't support -C, so cd "$_REPO_ROOT" is the fix).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Garry Tan
2026-03-26 23:52:05 -06:00
committed by GitHub
parent b343ba2797
commit 18bf4244ac
19 changed files with 198 additions and 732 deletions

View File

@@ -489,6 +489,7 @@ which codex 2>/dev/null && echo "CODEX_AVAILABLE" || echo "CODEX_NOT_AVAILABLE"
1. **Codex design voice** (via Bash):
```bash
TMPERR_DESIGN=$(mktemp /tmp/codex-design-XXXXXXXX)
_REPO_ROOT=$(git rev-parse --show-toplevel) || { echo "ERROR: not in a git repo" >&2; exit 1; }
codex exec "Read the plan file at [plan-file-path]. Evaluate this plan's UI/UX design against these criteria.
HARD REJECTION — flag if ANY apply:
@@ -514,7 +515,7 @@ HARD RULES — first classify as MARKETING/LANDING PAGE vs APP UI vs HYBRID, the
- APP UI: Calm surface hierarchy, dense but readable, utility language, minimal chrome
- UNIVERSAL: CSS variables for colors, no default font stacks, one job per section, cards earn existence
For each finding: what's wrong, what will happen if it ships unresolved, and the specific fix. Be opinionated. No hedging." -C "$(git rev-parse --show-toplevel)" -s read-only -c 'model_reasoning_effort="high"' --enable web_search_cached 2>"$TMPERR_DESIGN"
For each finding: what's wrong, what will happen if it ships unresolved, and the specific fix. Be opinionated. No hedging." -C "$_REPO_ROOT" -s read-only -c 'model_reasoning_effort="high"' --enable web_search_cached 2>"$TMPERR_DESIGN"
```
Use a 5-minute timeout (`timeout: 300000`). After the command completes, read stderr:
```bash