What problem does it solve?
This Skill enables Claude to delegate coding tasks to an OpenAI Codex CLI session so you can prototype, debug, and review code with a secondary model while your primary session remains the final authority. It solves the friction of running deterministic CLI-based code tools, capturing structured outputs, and continuing multi-turn workflows across separate agent sessions.
Core Features & Use Cases
- Multi-turn sessions: Preserve and resume context using SESSION_ID for continued interactions.
- Sandboxed execution: Choose between read-only, workspace-write, and danger-full-access sandboxes to limit side effects.
- Structured JSON output: Returns success, SESSION_ID, agent_messages, and optional full traces for programmatic consumption.
- Image and cross-platform support: Attach images to prompts and handle Windows path/escaping automatically.
- Use Cases: Prototype algorithm implementations, generate unified diffs for changes, debug runtime errors with full trace output, or ask Codex to write unit tests and review fixes.
Quick Start
Ask the skill to run Codex CLI to analyze the authentication flow in /path/to/project and return a JSON summary including a SESSION_ID.