ava-watcher

Write a Python loop that checks your condition and calls ava.agents.send_message when it fires, then launch it with ava.watcher.launch(code, timeout, name). The watcher runs independently and wakes you with a message, so you can idle instead of polling.

Call ava.watcher.cron with a cron expression, message, and timezone. Schedules without an explicit end_time expire after 7 days; re-registering the same expression and timezone replaces the old watcher instead of stacking a duplicate.

Every watcher has a bounded lifetime so a forgotten one cannot run forever. When the timeout elapses the watcher stops and sends an exit notice with its exit code and output tail, so you are never silently un-watched.

Use the reference watcher watch_idle.py, which subscribes to the Redis event stream and notifies you when the target agent reports an idling status. If the stream dies, it falls back to polling the agents table in PostgreSQL.

Each send_message wakes the agent for a full turn, so repeated identical messages waste turns. Keep the last-seen state, compare on every poll, and only message on transitions you care about; use the automatic exit notice instead of heartbeat messages.

Time watchers do not survive a host reboot and are not automatically rebuilt. Re-arm any watcher you still need after a restart; if a watcher was reclaimed while you were terminated, the reaper's notice tells you to re-register.