Code Review Skill

Universal skill for automated code review using AI agents to identify bugs, compliance issues, and quality concerns.

Compatible Platforms

This skill works with: Claude, Cursor, Codex, GitHub Copilot, and other LLM-based assistants.

Overview

This skill enables AI-powered code review that:

• •Checks code compliance with project guidelines
• •Identifies potential bugs and logic errors
• •Provides confidence-scored feedback to reduce false positives
• •Links directly to problematic code sections
• •Integrates with pull request workflows

How to Use

Quick Setup

Copy the content below and paste it into:

• •Claude: Project Knowledge or use as command
• •Cursor: .cursorrules file or as instructions
• •Codex: .codex/skills/code-review/SKILL.md
• •Copilot: .github/copilot-instructions.md

Skill Content

Purpose

Perform thorough code review focusing on:

1. •Guidelines Compliance: Check adherence to project-specific guidelines (CLAUDE.md, CONTRIBUTING.md, style guides)
2. •Bug Detection: Identify obvious bugs, logic errors, and potential runtime issues
3. •Quality Assurance: Ensure code quality meets project standards
4. •Context Analysis: Consider git history and related changes

Review Process

When performing code review, follow these steps:

1. Initial Assessment

Determine if review is needed:

• •Skip closed or draft pull requests
• •Skip trivial changes (typos, formatting only)
• •Skip automated PRs (dependency updates, generated code)
• •Check if already reviewed

2. Gather Context

Collect relevant information:

• •Read project guidelines (CLAUDE.md, CONTRIBUTING.md, etc.)
• •Review PR description and intent
• •Understand modified files and their purpose
• •Check related git history if needed

3. Multi-Perspective Analysis

Review from multiple angles:

Guideline Compliance:

• •Check for violations of documented project rules
• •Verify naming conventions and code style
• •Ensure architectural patterns are followed
• •Validate that guidelines apply to modified files

Bug Detection:

• •Look for syntax errors and type mismatches
• •Identify logic errors in conditionals and loops
• •Check for null/undefined handling issues
• •Spot resource leaks (memory, connections, files)
• •Verify error handling and edge cases

Quality Assessment:

• •Evaluate code clarity and maintainability
• •Check for security vulnerabilities
• •Assess performance implications
• •Verify proper use of APIs and libraries

4. Confidence Scoring

For each issue found, assign a confidence score (0-100):

• •0-25: Low confidence, likely false positive
• •26-50: Moderate confidence, might be real
• •51-75: High confidence, probably real
• •76-100: Very high confidence, definitely real

Only report issues with confidence ≥ 80

5. Issue Reporting

For each valid issue, provide:

Description:

Brief explanation of the problem

Evidence:

• •Quote the problematic code
• •Reference specific guideline if applicable
• •Explain why it's an issue

Location:

• •File path and line numbers
• •Link to code (use format: file.ext:L10-L15)

Suggested Fix:

• •Provide concrete solution when possible
• •Keep suggestions small and actionable
• •Only suggest if you're confident in the fix

High Signal Focus

DO flag these issues:

• •Syntax errors, type errors, missing imports
• •Clear logic errors that will produce wrong results
• •Unambiguous guideline violations with explicit rules
• •Security vulnerabilities (SQL injection, XSS, etc.)
• •Resource leaks and memory issues
• •Missing error handling for critical operations

DO NOT flag these:

• •Code style preferences (unless in guidelines)
• •Subjective quality suggestions
• •Nitpicks that don't affect functionality
• •Issues linters will automatically catch
• •Pre-existing problems not in the changes
• •Potential issues that depend on unknown state

False Positive Prevention

Avoid flagging:

• •Pre-existing issues not introduced in this change
• •Code that looks wrong but is actually correct
• •Pedantic nitpicks a senior engineer would ignore
• •General quality concerns unless required by guidelines
• •Issues with explicit suppression comments

Output Format

When Issues Found

## Code Review Results

Found [N] issues:

### Issue 1: [Brief Title]
**Confidence:** [85]/100
**Category:** [Bug/Guideline/Security/Performance]

**Problem:**
[Clear explanation of the issue]

**Location:**
file.ext:L10-L15

**Evidence:**

[relevant code snippet]

**Guideline Reference:** (if applicable)
[Quote from CLAUDE.md or other guideline]

**Suggested Fix:**
[Concrete solution]

---

[Repeat for each issue]

When No Issues Found

## Code Review Results

No issues found. Checked for:
- Bugs and logic errors
- Project guideline compliance
- Security vulnerabilities
- Code quality concerns

Platform-Specific Notes

For Claude

• •Can launch parallel agents for comprehensive review
• •Use gh CLI for GitHub integration
• •Support for inline comments on PRs

For Cursor

• •Review on-demand as you write code
• •Integrate with IDE linting and testing
• •Focus on current file or selection

For Codex

• •Structured skill with YAML frontmatter
• •Triggered automatically on relevant tasks
• •Can invoke scripts for deeper analysis

For Copilot

• •Continuous review as code is written
• •Focus on immediate context
• •Suggestions inline with editing

Best Practices

1. •Read Guidelines First: Always check project-specific rules before reviewing
2. •Be Specific: Reference exact lines and files
3. •Provide Context: Explain why something is an issue
4. •Suggest Solutions: Offer concrete fixes when possible
5. •Prioritize Severity: Focus on critical issues first
6. •Respect Intent: Consider the PR's stated goals
7. •Avoid Nitpicking: Only flag issues that matter
8. •Verify Claims: Double-check before flagging

Integration Examples

With Git Workflow

git checkout feature-branch
[request code review from AI]
[make fixes based on feedback]
git commit -m "Address review feedback"

In Pull Requests

Review checklist:
- [ ] Follows project guidelines
- [ ] No obvious bugs
- [ ] Proper error handling
- [ ] Security concerns addressed
- [ ] Performance considerations

Configuration

Adjust these parameters as needed:

• •Confidence Threshold: Default 80 (range: 0-100)
• •Focus Areas: Bugs, guidelines, security, performance
• •Scope: Full PR, specific files, or code selection
• •Depth: Quick scan vs. comprehensive review

Requirements

• •Access to code repository
• •Project guidelines (CLAUDE.md, CONTRIBUTING.md, etc.)
• •Understanding of project's language and frameworks
• •For GitHub integration: gh CLI installed

Credits

Based on work by Boris Cherny (Anthropic) Adapted for universal AI agent compatibility