| name | fetch-repo-skill |
| description | Looks up which Claude skills and kb/context docs apply to a named repository and returns the paths. User-invoked at the start of a repo session. Companion to fetch-repo-path. Use when the user says "/fetch-repo-skill <repo>", "load skills for <repo>", or "what skills apply to <repo>?".
|
Fetch Repo Skill
Companion to fetch-repo-path. Given a repo name, reads the registry at kb/context/registry/repo-skills-registry.md and returns the paths to the skills and context docs registered for that repo.
When to use
- User invokes
/fetch-repo-skill <repo_name> at the start of a repo session
- User asks "what skills apply when working on ?"
- User asks "load context for "
When NOT to use
- The user has not specified a repo
- The repo is not registered in
repo-skills-registry.md (in that case, suggest registering it)
References
kb/context/registry/repo-skills-registry.md — the registry mapping repo → {skills, context_docs}
kb/context/registry/repos.md — the canonical path registry (use fetch-repo-path to resolve actual filesystem paths)
Steps
- Read
kb/context/registry/repo-skills-registry.md
- Locate the entry for the requested repo by name
- If not found: report the available registered repos and suggest registering the new one (point to the "How to register a new repo" section of the registry)
- If found: return a structured result with two lists:
skills: the names of skills under ~/.claude/skills/ that should be active
context_docs: the kb/context files that may be relevant; the lead agent reads them on demand (lazy)
Output format
Repo: <name>
Path: <from repos.md>
Skills:
- <skill1> (path: ~/.claude/skills/<skill1>/)
- <skill2>
...
Context docs (read on demand, do not pre-load all):
- kb/context/<doc1>.md
- kb/context/<doc2>.md
...
The lead agent then chooses which context docs to Read based on the current task — do not pre-load all of them, that would bloat context for nothing.
Design note
This skill returns paths only, not file contents. Lazy loading on demand is intentional. If a future task needs broader context, the lead agent reads the specific files at that point.