asking-codex
Code review, security audits, bug detection, alternative implementations, second opinions via OpenAI Codex. Use when user asks for code review, security analysis, implementation advice, bug detection, code patterns, or wants a second opinion on code. Supports uncommitted changes review. Do not use for architecture design or web searches.
When & Why to Use This Skill
This Claude skill leverages OpenAI Codex to provide expert-level code reviews, security audits, and bug detection. It serves as an intelligent peer reviewer that identifies vulnerabilities, suggests alternative implementations, and offers technical advice on code patterns, including support for reviewing uncommitted local changes to ensure high-quality software delivery.
Use Cases
- Automated Code Review: Analyze code changes for logic errors, style consistency, and adherence to best practices before merging.
- Security Vulnerability Scanning: Conduct deep security audits to identify potential exploits or weak points in the implementation.
- Bug Detection and Troubleshooting: Locate hidden bugs in complex functions and receive actionable suggestions for fixes.
- Implementation Optimization: Get a second opinion on code architecture and suggestions for more efficient or idiomatic alternative implementations.
- Pre-commit Analysis: Review uncommitted local changes to catch errors early in the development lifecycle.
| name | asking-codex |
|---|---|
| description | Code review, security audits, bug detection, alternative implementations, second opinions via OpenAI Codex. Use when user asks for code review, security analysis, implementation advice, bug detection, code patterns, or wants a second opinion on code. Supports uncommitted changes review. Do not use for architecture design or web searches. |
| allowed-tools | Task |
Codex Consultation
Spawn the codex-assistant agent for code-focused questions.
Foreground (blocking)
Task(subagent_type="codex-assistant", prompt="[mode]: <question>")
Background (for context efficiency)
Task(subagent_type="codex-assistant", prompt="[mode]: <question>", run_in_background=true)
Use TaskOutput(task_id="<id>") to retrieve results.
Modes: exec (default), review, plan, implement (add --auto for file changes)
The agent uses Codex MCP tools (mcp__codex__spawn_agent, mcp__codex__spawn_agents_parallel) directly.