goal-modelisted
Install: claude install-skill search-atlas-group/amm-founding-circle
# goal-mode
**The problem this solves:** you hand your agent a real piece of work — "get this article publish-ready", "keep fixing until every check passes", "clean up all these files" — and then you sit there nudging it. It does one step, stops, and waits. You type "keep going." It does another step, stops, waits. You type "continue." An hour of your day disappears into being the button that keeps the agent alive. That's not an agent working for you — that's you working for the agent.
`goal-mode` kills that loop. You give the agent **the finish line, not the next step.** It keeps working toward that finish line on its own — no "what should I do next?" check-ins — and it **pings your computer** two ways:
1. **When something meaningful changes** — a step finishes, a test flips, it hits a blocker. An instant desktop notification, so you find out the moment it matters.
2. **On a steady heartbeat** (about every 5 minutes) — a quiet "still working" ping, so you know it's alive even when nothing newsworthy happened.
Then when the finish line is reached, it pings you one last time — "done" — and stops. You get your attention back and spend it only when the work actually needs you.
> This is the "stop watching, get pinged instead" half of an always-on setup. Its close sibling is `host-your-agent`, which gets a job running *while your machine sleeps* on a schedule. Use `goal-mode` when you're at your machine and want a long job to finish itself without you nudging it; use `host-y