Agent skills

If you work with a coding agent — Claude Code or anything else that reads agent skills — the loopkeep skill teaches it to operate loopkeep alongside you: investigate runs, author workflows, tune policies, and manage the daemon, all through the same lk CLI you use.

Install

npx skills add loopkeep/skills

The skill lives at github.com/loopkeep/skills.

What it teaches the agent

• Investigating runs — why a run stalled, failed, or got blocked, working from run ids (run_000123), lk runs, and the inbox.
• Authoring workflows — writing and editing the markdown files in .loopkeep/workflows/ with correct frontmatter.
• Tuning policies — adjusting rules, explaining verdicts with lk policy explain, and verifying changes with lk policy test.
• Managing the daemon — status, start/stop, workspaces.

The safety rules it carries

The skill isn't just command syntax — it encodes how to operate loopkeep without undermining it:

• Never weaken a policy to unblock an action without saying what is being relaxed and why; prefer the narrowest rule and verify with lk policy test.
• Treat the event log as read-only evidence.
• Don't hand-edit loopkeep's state files; go through lk commands.
• Warn before restarting the daemon while runs are in flight.

These mirror loopkeep's own built-in floor — the skill makes your interactive agent a well-behaved operator of the same machinery your resident workflows run under.