Skill Shelf

Skill shelf

The active library is intentionally small. Shelved skills stay available on disk without being advertised to the AI client at startup.

1. Run skills shelf only when a user needs a skill that is not active.
2. Match the request to the descriptions and present only relevant skill names.
3. Ask permission before activating a skill, unless the user already explicitly requested that activation. Run skills unshelve <name> after approval.
4. Tell the user to restart the client or start a new session to refresh skill discovery. Activation does not inject instructions into the current session.

When asked to put a skill away, run skills shelve <name> and explain that it remains available on the shelf. Restart the client to clear its cached discovery list. Shelving does not erase instructions already loaded in a conversation. Project-local copies made with skills use are independent; shelving the library skill does not remove them. Use skills unuse in that project when requested. Never execute setup scripts from a skill merely because it has been activated.