claude-codex

Claude Codex is a Claude Code plugin that automates code review through sequential multi-AI orchestration. Instead of relying on a single AI reviewer, it routes code through three independent reviewers—Claude Sonnet for obvious bugs and style issues, Claude Opus for architectural and subtle bugs, and Codex for a final independent perspective. The plugin includes specialized agents for requirements gathering, planning, implementation, and code review, each checking for OWASP Top 10 vulnerabilities, error handling, and code quality. It supports both feature development pipelines and bug-fix workflows with dual root-cause analysis, ensuring code doesn't proceed until all reviewers approve.

1. **Configure the Pipeline**: Set up your `.claude-codex.yml` file in the project root with rules for tools (e.g., ESLint, Snyk), ignored files, and coding standards. Example: ```yaml tools: eslint: true snyk: true prettier: true ignore: - "*.min.js" - "*.lock" standards: - "eslint-config-airbnb" - "OWASP Top 10" ``` 2. **Trigger the Automation**: Use the GitHub Action or CLI command to start the pipeline. For GitHub: ```yaml - name: Run Claude Codex Review uses: claude-codex/action@v1 with: pr_number: ${{ github.event.pull_request.number }} repo_url: ${{ github.repository }} ``` 3. **Review the Report**: The AI will post a consolidated review as a PR comment. Use the `claude-codex dashboard` CLI tool to inspect detailed logs if needed. 4. **Iterate**: Address the feedback in new commits. The pipeline will re-run automatically on push. Use `claude-codex rerun --pr 42` to manually trigger a re-review. 5. **Customize Rules**: Adjust the `.claude-codex.yml` file to match your project’s needs (e.g., add custom linting rules or security checks). **Tips:** - Use `claude-codex init` to scaffold the configuration file for your project. - For large PRs, split reviews into smaller chunks by reviewing files in batches. - Enable `fail_fast: true` in the config to block merges until critical issues are resolved.