콘텐츠로 이동

Create and load skills

이 콘텐츠는 아직 번역되지 않았습니다.

A skill is a directory centered on SKILL.md. It tells an agent when and how to perform a bounded kind of work; it is not executable proof by itself.

Use project scope under .xcsh/skills/ for repository-specific instructions and user scope under the xcsh data directory for personal skills. Project instructions should be reviewable with the repository.

Start xcsh with --skills <GLOB> to narrow discovery, invoke a prompt that matches the trigger, and inspect the tools and output. Use --no-skills to confirm behavior without it.

Give SKILL.md a narrow trigger, explicit prerequisites, ordered actions, and a verifiable stopping condition. Keep executable helpers inside the skill directory and invoke them through their documented interface rather than embedding machine-specific paths. Test positive and negative prompts with only that skill selected, inspect which instructions were loaded, then remove the temporary skill and confirm the trigger no longer matches.