AgentSkillsCN

code-reviewer

提供严谨的代码审查指南，重点关注反模式、架构缺陷、最佳实践与逻辑错误。 在以下场景中使用：审查拉取请求、审计代码质量、重构遗留代码，或确保代码严格遵循编码规范。

SKILL.md
--- frontmatter
name: code-reviewer
description: |
  Rigorous code review guide focusing on anti-patterns, architectural flaws, best practices, and logical errors.
  Use when: reviewing pull requests, auditing code quality, refactoring legacy code, or ensuring adherence to coding standards.

Code Review Expert

This skill transforms the agent into a rigorous code reviewer. The goal is to catch bugs, enforce standards, and improve maintainability before code is merged.

How to execute a Code Review

Follow this hierarchy of importance:

  1. •Correctness: Does it do what it's supposed to do? Are there logic errors?
  2. •Security: Are there vulnerabilities (Injection, XSS, Secrets)?
  3. •Architecture: Does it break boundaries? Is it scalable?
  4. •Readability: Is the intent clear? Are names descriptive?
  5. •Performance: Are there obvious bottlenecks ($O(n^2)$ loops, N+1 queries)?
  6. •Style: Formatting, typo fixes (Least important - use linters for this).

How to identify Anti-Patterns

Look for these red flags:

  • •God Class: A class that knows/does too much. Fix: Extract Delegate/Service.
  • •Shotgun Surgery: A single change requires edits in many classes. Fix: Increase cohesion.
  • •Primitive Obsession: Using int or string for domain concepts like Money or Phone. Fix: Use Value Objects.
  • •Feature Envy: A method accesses data of another object more than its own. Fix: Move method to the data owner.

How to validate Architecture & Design

  • •SOLID: Check Single Responsibility and Dependency Inversion violations.
  • •Coupling: High coupling makes testing hard. Look for new keywords in business logic.
  • •Leaky Abstractions: Does the database schema leak into the UI layer?

Common Warnings & Pitfalls

"Looks Good To Me" (LGTM)

  • •Issue: rubber-stamping code without deep analysis.
  • •Fix: Always find at least one meaningful question or suggestion.

Nitpicking

  • •Issue: Focusing only on formatting/naming.
  • •Fix: Automate style checks. Focus review on logic and architecture.

Logic Errors

  • •Off-by-one: Check loop boundaries (< vs <=).
  • •Null checks: Are edge cases (null, empty list, negative numbers) handled?

Best Practices (Review Etiquette)

RuleExplanation
Critique Code, Not People"This variable name is unclear" vs "You named this badly".
Explain Why"Extract this method to improve testability" vs "Extract this".
Ask Questions"What happens if this list is empty?" (Socratic method).

Deep Dives

References