codex-cli

What This Skill Does

The codex-cli skill empowers your OpenClaw agent to leverage the OpenAI Codex CLI directly within your local development environment. It bridges the gap between high-level task delegation and granular file manipulation. By acting as a sub-agent, Codex CLI gains the ability to navigate your filesystem, analyze code, execute terminal commands, and perform autonomous refactoring or debugging. It serves as an interface between your requirements and the raw power of the GPT-5-Codex model, allowing the agent to perform complex engineering tasks that would otherwise require manual intervention.

Installation

To integrate this skill into your OpenClaw environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/adamsardo/codex-sub-agents. Once installed, ensure you have the Codex CLI package installed globally via npm i -g @openai/codex. Authentication is mandatory; you can trigger the OAuth browser flow using codex login or provide your OPENAI_API_KEY directly. Verify your installation by running codex login status to confirm your session is active.

Use Cases

This skill is designed for scenarios where coding logic and workspace state are inextricably linked. Use it when you need to refactor a legacy module, fix broken CI pipelines, or implement features based on existing codebase patterns. It is particularly effective for large-scale code reviews, where the agent can analyze git diffs or perform structural impact analysis across multiple files. Because it supports non-interactive execution, it is ideal for automated scripting during CI/CD workflows.

Example Prompts

1. "Codex, please review the current diff in the feature-login-page branch and identify any potential security vulnerabilities in the authentication logic."
2. "Fix the CI failure in the /src/components directory by updating the prop types and re-running the build command to verify the fix."
3. "Refactor the database connection helper in /lib/db.ts to use the new connection pooling pattern, ensuring all usages are updated accordingly."

Tips & Limitations

Always be mindful of the approval mode set within the TUI. While 'Auto' is convenient, 'Read Only' is safer for initial explorations. If you find the context window filling up, use the /compact command to maintain agent performance. Note that the skill operates with filesystem access; always review code changes generated by the tool before committing, especially when running in 'Full Access' mode, which allows external network calls.