host-your-agentlisted

# host-your-agent **The problem this solves:** your agent only works when you're sitting there. You open a terminal, run a script, watch it, and if you close the laptop or it dies, the work is gone. That's not an always-on system — that's a manual tool you happen to be running with AI. This skill gets your agent **off your laptop**: it runs on a schedule whether you're watching or not, and it never loses work if something breaks at 3am. It does two things: 1. **Auto-save hook** — every time your agent finishes a step or a session, it quietly saves a snapshot (a git commit) of its work. So an unattended run always leaves a rollback trail. Nothing is lost, even if the machine restarts mid-run. 2. **Scheduled unattended runner** — a job you define runs on a schedule (say, 1am every night), pointed at *your own* work — your repo, your clients, your inbox — and leaves a finished artifact you read over coffee. On your Mac it uses launchd, on Windows it uses Task Scheduler, on Linux it uses cron. You don't have to remember to start it. It bounds every run (a time box, read-only unless you say otherwise) and it degrades gracefully — if a model or provider is down, you still get a report of what it *could* check, never a silent failure. > This skill is the **hands** of always-on. Its safety rules — the time box, the worker caps, read-only-by-default, the failure ledger — come from the `night-shift` skill, which is the *contract* for unattended work. Read `night-shift` once; it's