list-skills

Bucket: meta · Invocation: manual

Returns one reference card: every installed skill, how to invoke it on the current host, and what it does.

The card is built by reading the installed skills themselves rather than a list held in the skill, so it reflects what is actually present — including a partial or outdated installation. Invocation comes from the same frontmatter that gates it, so explicit and automatic skills never appear under the wrong heading. An automatic workflow may still document explicit modes in its own instructions.

Invocation syntax differs per host: /agent-os:<name> through the Claude plugin, /<name> for a direct Claude install, and $<name> on Codex. The card names which form it is showing.

It closes with entry points for a developer who has not named a skill: coherent planning starts at plan-work, delivery structure with material choices at shape-work, and a decision-complete unit at deliver-work. Automatic skills need no entry point for an implicit trigger. It is read-only: it never runs, stages, or offers to run what it lists. Choosing the next action is dispatch-next; describing work already underway is explain-work.