Managing skills with skillset¶

Keep skills in topic repos, not loose in ~/.claude/skills/ — a jira-skills repo, a cloudflare-skills repo, your own personal grab-bag. ~/.claude/skills/ is then just symlinks into those repos, never edited directly. This is the starting point, not something to graduate into later.

Managing those symlinks by hand — cloning, finding the right subfolder, linking, remembering to pull updates — is what skillset is for. It clones/caches skill repos, links the skills you want, and searches across everything cached without you tracking which repo a skill came from.

Installing¶

skillset add vivainio/agent-skills          # clone, pick skills interactively
skillset add vivainio/agent-skills -s zaira # install just one

Installs to the project scope if a skillset.yaml exists in the current directory or a parent; otherwise falls back to ~/.claude/skills/. -g forces global.

Searching¶

skillset search jenkins
skillset search doc-%     # glob

Searches everything cached locally, offline — which means you can stock up on repos before you know which skill you want:

skillset add vivainio/agent-skills --fetch   # clone/cache only, link nothing
skillset add some-org/other-skills --fetch

Then later, once you remember you need something:

skillset search invoice

turns up matches across every repo you've fetched, without you having to recall which one it was in.

Removing¶

skillset remove zaira                    # unlink one skill
skillset remove vivainio/agent-skills    # drop the whole repo

Declaring the set in skillset.yaml¶

For a reproducible setup across machines or a team, declare it instead of installing ad hoc:

skills:
  vivainio/agent-skills:
    enabled: ["*"]

  vivainio/jira-skills:
    enabled: [mycompany-jira-process]
    disabled: [experimental-jira-thing]

skillset update    # pulls each repo, links enabled, unlinks disabled

skillset init creates this file (-g for the global one). Check it into the repo and a fresh clone gets the same skills back with one skillset update.