AgentSkillsCN

big-review

运行 ~/code/dotfiles/bin/review.sh，执行多模型代码审查，评估发现的问题，修复有效的缺陷（在条件允许的情况下附上测试用例），为无效的发现添加清晰的注释，随后运行必要的检查。此操作宜在创建 Pull Request 之前执行，或在用户提出深度审查需求时使用。

SKILL.md
--- frontmatter
name: big-review
description:
  Run ~/code/dotfiles/bin/review.sh to perform a multi-model review, evaluate findings, fix valid
  issues (with tests when possible), add clarifying comments for invalid findings, then run required
  checks. Use before PR creation or when the user asks for a deep review.

Big Review

Overview

Run the local review.sh script to get multi-model review findings, then triage and address valid issues before final validation and PR creation.

Workflow

1) Preflight

  • •Confirm the script exists and is executable: ~/code/dotfiles/bin/review.sh.
  • •Ensure gh, gemini, claude, codex, jq, and script are available (the script checks this).
  • •Do not use --post unless explicitly asked.
  • •Expect a long runtime (15+ minutes); set any command timeout to >= 30 minutes.

2) Run the review

  • •Run: ~/code/dotfiles/bin/review.sh.
  • •Capture the consolidated output (P0-P3 items or "No issues found.").

3) Triage findings

For each item:

  • •Valid + worth fixing
    • •If tests exist for the area, add or extend coverage for the issue.
    • •Run the test and confirm it fails.
    • •Apply the fix and re-run the test to confirm it passes.
  • •Invalid or not worth fixing
    • •Add a brief code comment explaining why it is safe or intentional.

4) Re-run review if needed

  • •Re-run review.sh only if you made substantive changes that affect earlier findings.

5) Final validation

  • •Run repo-required checks (for this repo: corepack yarn check).
  • •Note any warnings but avoid unrelated changes unless required.

Notes

  • •The review script already forbids modifications while running; only edit after it finishes.
  • •If review.sh fails (missing tools or empty output), fall back to manual review and proceed.