Claude Code Plugins

Community-maintained marketplace

Feedback

AUTOMATICALLY run Codex review after writing security-sensitive code (auth, crypto, SQL) - get a second opinion from OpenAI Codex

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name codex-review
description AUTOMATICALLY run Codex review after writing security-sensitive code (auth, crypto, SQL) - get a second opinion from OpenAI Codex

Codex Code Review - AUTO-EXECUTE MODE

IMPORTANT: When this skill activates, you MUST automatically run a Codex review using the Bash tool. Do not ask for permission - just execute it.

Core principle: Two LLMs are better than one.

When to Auto-Execute Codex Review

MUST run review after writing:

  • Authentication/authorization code (jwt.verify, bcrypt, crypto)
  • SQL queries or database operations
  • Password handling or cryptographic operations
  • Input validation or sanitization
  • exec(), eval(), or dynamic code execution

Skip when:

  • Trivial changes (documentation, comments, formatting)
  • Changes are < 10 lines
  • User explicitly says "skip review" or has @skip-codex-review marker

How to Request Codex Review

1. Prepare the Context

Summarize what you changed and why:

CHANGES:
- Added user authentication middleware
- Implemented JWT token validation
- Added role-based access control

CONTEXT:
- Using Express.js with TypeScript
- JWT stored in httpOnly cookies
- Three roles: admin, user, guest

2. Run Codex Exec

Use the Bash tool to execute codex:

codex exec "Review this code implementation:

PROBLEM: Need secure authentication for Express API

SOLUTION:
[Paste the relevant code here]

QUESTIONS:
- Are there any security vulnerabilities?
- Is the error handling robust?
- Any performance concerns?
- Suggestions for improvement?"

3. Analyze Codex Response

Codex will provide:

  • Security assessment
  • Logic validation
  • Performance considerations
  • Best practice suggestions

4. Act on Feedback

  • Critical issues: Fix immediately
  • Important suggestions: Implement before commit
  • Minor improvements: Note for future refactoring
  • Disagreements: Trust your judgment but document reasoning

Example Session

[Just completed: JWT authentication middleware]

You: Let me get Codex's opinion on this authentication implementation.

codex exec "Review this authentication code:

CONTEXT: Express.js API with JWT cookie-based auth

CODE:
export const authenticateJWT = async (req, res, next) => {
  const token = req.cookies.jwt;
  if (!token) return res.status(401).json({ error: 'Unauthorized' });

  try {
    const decoded = jwt.verify(token, process.env.JWT_SECRET);
    req.user = await User.findById(decoded.userId);
    next();
  } catch (err) {
    return res.status(401).json({ error: 'Invalid token' });
  }
};

QUESTIONS:
- Security vulnerabilities?
- Error handling sufficient?
- Performance implications?
"

[Codex responds]:
SECURITY:
- ✅ Good: httpOnly cookies, JWT verification
- ⚠️ Issue: No JWT_SECRET validation on startup
- ⚠️ Issue: User query could fail, should handle null

PERFORMANCE:
- ⚠️ DB query on every request - consider caching decoded token

SUGGESTIONS:
- Add JWT_SECRET existence check at startup
- Handle null user case
- Consider Redis for token caching
- Add rate limiting

You: [Fix critical issues: JWT_SECRET check, null user handling]
You: [Note for later: token caching, rate limiting]

Integration with Existing Workflows

With code-reviewer subagent:

  • Use code-reviewer for architectural review
  • Use codex-review for security/implementation details
  • Complementary perspectives

With TDD:

  • Write tests first
  • Implement code
  • Run codex-review before commit
  • Fix issues, ensure tests still pass

With git workflow:

  • Make changes
  • Run codex-review
  • Fix issues
  • Commit with confidence

Skip Conditions

Automatic skip:

  • File contains @skip-codex-review comment
  • Environment variable: SKIP_CODEX_REVIEW=1
  • Session tracking: Won't nag repeatedly

Manual skip:

  • Simply proceed without running codex
  • Skill is suggestive, not blocking

Tips for Effective Codex Reviews

1. Be specific in prompts:

# Good
codex exec "Review for SQL injection vulnerabilities in this query builder"

# Less effective
codex exec "Review this code"

2. Provide context:

  • What problem are you solving?
  • What constraints exist?
  • What specific concerns do you have?

3. Focus reviews:

  • Security review
  • Performance review
  • Logic review
  • Best practices review

4. Batch related changes:

  • Review 3-5 related files together
  • Show how they interact
  • Get holistic feedback

Common Patterns

Security Review

codex exec "Security review for authentication code:
[code]
Focus on: injection attacks, token handling, timing attacks"

Performance Review

codex exec "Performance review for data processing:
[code]
Focus on: algorithmic complexity, database queries, memory usage"

Logic Review

codex exec "Logic review for workflow engine:
[code]
Focus on: edge cases, state transitions, error paths"

Red Flags (When to ALWAYS use Codex)

  • Authentication/authorization code
  • Data validation logic
  • SQL query construction
  • File system operations
  • Cryptographic operations
  • Regular expressions
  • Parsing user input
  • Rate limiting/throttling

Benefits

Catch blind spots: Different LLM, different perspective ✅ No blocking: Suggestion only, no workflow disruption ✅ Fast: 2-10 seconds for review ✅ Cost-effective: Single API call for entire changeset ✅ Learning: See alternative approaches and patterns


Remember: Codex review is advisory. Use your engineering judgment for final decisions.