AgentSkillsCN

Code Review

代码审查

SKILL.md
--- frontmatter
name: "Code Review"
id: "code-review"
version: "1.0"
claude_subagent: "code-reviewer"
fallback: "self-check"
min_tad_version: "2.1"
platforms: ["claude", "codex", "gemini"]

Code Review Skill

Purpose

Review code for quality, maintainability, security, and adherence to best practices. Identify issues by severity level (P0-P3).

When to Use

  • •Before committing code changes
  • •During Gate 2 (design review)
  • •During Gate 3 (implementation quality)
  • •During pull request reviews

Checklist

Critical (P0) - Must Pass

  • • No security vulnerabilities (SQL injection, XSS, etc.)
  • • No data loss or corruption risks
  • • No unhandled exceptions that crash the application
  • • No authentication/authorization bypasses
  • • No sensitive data exposure

Important (P1) - Should Pass

  • • No logic errors causing wrong behavior
  • • No race conditions in async code
  • • Features work as specified
  • • Error handling for likely failures
  • • API contracts are correct

Nice-to-have (P2) - Informational

  • • Consistent naming conventions
  • • Descriptive variable/function names
  • • Minimal code duplication
  • • Appropriate documentation
  • • Optimal but functional approach

Suggestions (P3) - Optional

  • • Additional comments where helpful
  • • Alternative approaches to consider
  • • Future refactoring opportunities
  • • Stylistic preferences

Pass Criteria

LevelRequirement
P0Zero issues allowed
P1Zero issues allowed
P2Max 10 issues
P3Unlimited (suggestions)

Evidence Output

Path: .tad/evidence/reviews/{date}-code-review-{task}.md

Execution Contract

  • •Input: file_paths[], context{}, changed_lines[]
  • •Output: {passed: bool, issues: [{severity, file, line, description}], evidence_path: string}
  • •Timeout: 180s
  • •Parallelizable: true

Claude Enhancement

When running on Claude Code, call subagent code-reviewer for deeper analysis. Reference: .claude/skills/code-review/SKILL.md for extended guidance.

Issue Severity Definitions

P0 - Critical

Security vulnerability, data loss, crash. Examples:

  • •SQL injection vulnerability
  • •Unhandled exception that crashes app
  • •Data corruption possible
  • •Authentication bypass
  • •Sensitive data exposure

P1 - High

Logic errors, race conditions, broken features. Examples:

  • •Logic error causing wrong calculation
  • •Race condition in async code
  • •Feature doesn't work as specified
  • •Missing error handling for likely failures
  • •Incorrect API contract

P2 - Medium

Style, naming, minor improvements. Examples:

  • •Inconsistent naming convention
  • •Could use more descriptive variable name
  • •Minor code duplication
  • •Missing documentation
  • •Suboptimal but functional approach

P3 - Low

Suggestions, nice-to-haves. Examples:

  • •Could add more comments
  • •Alternative approach might be cleaner
  • •Consider refactoring in future
  • •Stylistic preference