---

name: pre-review-gate description: Run the local "ready for review" gate before asking for human review or opening a PR. Use to (1) run the repo’s lint/typecheck/tests (prefer the same commands CI runs), then (2) run codex review as a final automated review pass, and optionally (3) run CodeRabbit CLI (coderabbit --prompt-only) to catch additional issues and reduce GitHub Actions feedback loops.

Pre-Review Gate

Goal

Run the same quality gates you expect in CI, then run an automated code review pass locally before considering changes “ready for human review”.

1) Determine the project’s canonical commands

Prefer exactly what CI runs.

1. If the repo has GitHub Actions workflows (.github/workflows/*.yml), extract the lint/typecheck/test commands from run: steps and use those locally.
2. Otherwise, look for a project runner (in this order):
   • justfile (just lint, just typecheck, just test)
   • Makefile (make lint, make test)
   • package.json scripts (npm|pnpm|yarn run lint|typecheck|test)
   • Language-native defaults (only if there is no repo runner):
     • Python: ruff check ., pyright, pytest
     • TypeScript: eslint ., tsc -p . --noEmit, vitest
     • Go: golangci-lint run, go test ./...
     • Rust: cargo fmt --check, cargo clippy, cargo test

If the repo uses a monorepo task runner (Nx/Turbo/Bazel/etc.), use it rather than ad-hoc per-package commands.

2) Run quality gates (stop on first failure)

Run in this order:

1. Lint / formatting checks
2. Type checking
3. Tests

If anything fails, fix it and re-run from the failing step forward.

3) Final gate: run Codex local review

Use codex review as the last gate after lint/typecheck/tests pass.

Recommended modes:

• Working tree review (best during iteration): codex review --uncommitted
• Branch review vs base (best before opening a PR): codex review --base main

To increase review thoroughness/recall, use the bundled wrapper script (it feeds a high‑recall prompt and asks Codex to gather staged/unstaged/untracked diffs):

• ~/.codex/skills/pre-review-gate/scripts/codex_review_thorough.sh --uncommitted
• ~/.codex/skills/pre-review-gate/scripts/codex_review_thorough.sh --base main

If you prefer the built-in codex review --uncommitted / --base modes, you can still make output more expansive via config overrides:

• codex review --uncommitted -c 'model_verbosity="high"' -c 'model_reasoning_effort="xhigh"'

If you need stricter guidance, pass a short custom prompt such as:

• “Focus on correctness, missing tests, and footguns. List Blockers first, then Suggested improvements, then Nitpicks. Include file paths and concrete fixes.”

4) Optional: run CodeRabbit CLI (second opinion)

If CodeRabbit CLI is installed/configured, run it after codex review to catch additional issues:

• Uncommitted changes: coderabbit --prompt-only --type uncommitted
• Against a base branch: coderabbit --prompt-only --base main

Notes:

• CodeRabbit CLI can take ~7–30+ minutes; run it in the background if needed and come back when it finishes.
• Avoid running it more than ~3 times per change set to keep iterations efficient.

5) Produce a “Gate Report”

Conclude with a short report:

• Commands run + pass/fail (and any failures fixed)
• Codex review: blockers + actions taken
• CodeRabbit (if run): blockers + actions taken
• Verdict: Ready for human review / Not ready (with next steps)