coding-agent

The coding-agent skill is designed for delegating substantial software engineering tasks to external coding agents running as background workers. It is intended for complex development workflows such as feature implementation, pull request reviews, large-scale refactoring efforts, and issue-to-PR execution cycles. Rather than performing code changes directly, the skill coordinates specialized coding agents including Codex, Claude Code, OpenCode, or Pi, ensuring work is executed asynchronously and monitored appropriately. The skill also enforces operational safeguards, notification requirements, and execution constraints to support reliable long-running development tasks.

Key capabilities include launching supported coding agents in background mode, generating worker prompts through temporary files, enforcing agent-specific launch requirements, and requiring explicit completion or failure notifications through a defined messaging route. The skill provides standardized launch procedures for multiple supported agents, guidance for monitoring worker processes, recovery expectations when workers fail or hang, and detailed workflows for extended issue-to-PR automation. It also includes special handling for isolated repositories, scratch repositories for Codex usage, and restrictions that prevent execution in protected OpenClaw-related directories.

This skill is primarily intended for developers, engineering teams, and automation workflows that need to offload substantial coding work to autonomous agents. Typical use cases include implementing features from specifications, reviewing code changes, managing long-running repository tasks, and executing structured development workflows tied to GitHub issues and pull requests. It is not intended for simple code edits, read-only code inspection, or tasks that must remain bound to active OpenClaw state directories.