codex-code-review_skill

Overview

This skill uses the Codex CLI to run AI-powered code reviews locally, in CI, and on GitHub pull requests. It helps catch bugs, security vulnerabilities, performance issues, and maintainability problems before merge. The integration supports interactive CLI sessions, focused review prompts, and automated CI/GitHub workflows.

How this skill works

The skill invokes the Codex CLI to analyze uncommitted changes, branch diffs, individual commits, or full pull requests. It can run presets (security, performance, etc.), accept custom instructions, and emit structured JSON findings for CI or automated commenting on GitHub. You can configure review guidelines and severity in a repository-level AGENTS.md and use a GitHub Action for continuous reviews.

When to use it

• Run pre-commit checks on staged or unstaged files
• Review a branch against base before opening a pull request
• Approve or audit a specific commit or merged change
• Trigger focused reviews (security, accessibility, performance)
• Add automated reviews to CI for every pull request

Best practices

• Add an AGENTS.md with project-specific review rules and exclusions
• Run both uncommitted and branch-comparison reviews for full coverage
• Use structured output (JSON schema) for CI parsing and metrics
• Raise severity thresholds to reduce noise and ignore generated files
• Select the strongest review model for high-stakes security reviews

Example use cases

• Pre-commit quick check: run codex → /review → "Review uncommitted changes" to catch immediate issues
• Security-focused PR review: comment @codex review for security regressions to get prioritized findings
• CI integration: add openai/codex-action to GitHub Actions to produce review-findings.json and post inline comments
• Performance tuning: run a custom review instructing Codex to look for unnecessary re-renders and heavy bundle imports in React
• Post-refactor verification: compare against main before and after refactor to verify issue count decreased