codex_pr-review-toolkit

在决定为 PR 选择何种评审技能时，需综合考虑评论、测试、错误排查、类型检查、代码质量优化以及代码简化等多个维度；当需要为某次 PR 选择合适的评审技能，或在多重评审需求叠加时，可借助此方法进行统筹协调。

SKILL.md

--- frontmatter

name: codex_pr-review-toolkit
description: Guide selection and coordination of specialized review skills for comments, tests, errors, types, code quality, and simplification; use when deciding which review skills to run for a PR or when combining multiple reviews.

PR Review Toolkit

Agents

•codex_comment-analyzer: comment accuracy and comment rot
•codex_pr-test-analyzer: test coverage quality
•codex_silent-failure-hunter: error handling and silent failures
•codex_type-design-analyzer: type design and invariants
•codex_code-reviewer: general code quality and CLAUDE.md compliance
•codex_code-simplifier: simplify code after review

When to use

•After writing code: run codex_code-reviewer.
•After adding docs or comments: run codex_comment-analyzer.
•After changing error handling: run codex_silent-failure-hunter.
•After adding or refactoring types: run codex_type-design-analyzer.
•Before PR: run codex_pr-test-analyzer and codex_code-reviewer, plus any relevant aspect.
•After review passes: run codex_code-simplifier.

Usage patterns

•Targeted: ask for a specific aspect to trigger a single skill.
•Comprehensive: run all applicable reviews for a PR.
•Parallel vs sequential: default to sequential unless explicitly requested.

Troubleshooting

•If wrong files are reviewed, specify files or the diff.
•If a skill does not trigger, mention it explicitly.

Integration

•Focus on changed code, not the entire codebase.
•Re-run relevant reviews after fixes.