review
针对分阶段更改或指定文件的常规代码审查。在审查任何语言的拉取请求、提交或代码更改时使用。
📖 中文导读:针对分阶段更改或指定文件的常规代码审查。在审查任何语言的拉取请求、提交或代码更改时使用。 以下正文为该技能的原始 SKILL.md 内容(英文),上方安装命令可直接使用。
name: review description: General code review for staged changes or specified files. Use when reviewing pull requests, commits, or code changes across any language. allowed-tools: ["shell", "read_file", "apply_patch"] metadata: short-description: Review code changes comprehensively
Code Review
Read these references:
~/.config/agent/domain/review/process.md- Review workflow~/.config/agent/domain/review/priorities.md- Review focus areas
Instructions
Review staged changes (or specified files/diff) in this priority order:
Note: When receiving review feedback, use the review-reception skill instead of this one.
Review Focus (Priority Order)
Security issues
- Injection vulnerabilities (SQL, command, XSS)
- Secrets exposure (hardcoded keys, tokens)
- Authentication/authorization bypasses
- Insecure cryptography, weak randomness
Runtime safety
- Nil/null pointer dereferences
- Race conditions, data races
- Resource leaks (file handles, connections, memory)
- Buffer overflows, out-of-bounds access
Code correctness
- Logic errors, incorrect algorithms
- Edge cases not handled
- Error handling: all errors checked and wrapped
- State management issues
Language idiomaticity
- Go: Uber Go Style Guide, effective Go patterns
- Rust: ownership patterns, lifetime management
- Python: PEP 8, pythonic idioms
- Follow language-specific best practices
Performance
- Unnecessary allocations (especially in hot paths)
- Inefficient algorithms (O(n²) where O(n) possible)
- Repeated expensive operations
- Missing caching opportunities
Test coverage
- Missing tests for new functionality
- Untested edge cases
- Insufficient error case testing
Review Process
Per AGENT.md Code Review:
- Be thorough and pedantic - catch every issue
- Present interactively - discuss each issue separately
- Order by severity - most critical first
- Offer fixes - for each issue, offer to make the change
- Prove when possible - run tests/lint to demonstrate issues
Arguments
Target: ${ARGUMENTS:-staged changes}