monitorslisted

# Monitors Skill Durable, event-triggered watchers. **Routines fire on a *clock*; monitors fire on a *change*.** A monitor watches a source, decides whether it changed, and fires an action — spawn an agent, kick a routine, or notify. It runs in the same background daemon as routines, so it survives restarts and is fleet-schedulable. Think of it as the **cross-agent layer**: agents watching sources — including the fleet and other agents — and firing agents in response. ## The model: source → condition → action A monitor is three parts: - **Source** — what to watch (a command's output, an HTTP endpoint, a file, a fleet device, a webhook). - **Condition** — what counts as a fire (any change, a regex match, every tick), deduped by a native state store so it stays silent until something *actually* changes. - **Action** — what to do (run an agent with the event in its prompt, kick a routine, notify, POST a webhook). ## When to reach for this (vs. siblings) | You want… | Use | |---|---| | Run on a schedule (9am weekdays, hourly) | `agents routines` (clock) | | Run when a *condition changes* (CI goes red, cert issues, box goes down) | **`agents monitors`** (change) | | Nudge a *stalled agent session* to continue | `agents watchdog` | ## Quickstart ```bash # CI goes red -> a Claude agent triages it (poll a command, diff, match a pattern) agents monitors add ci-red \ --poll 'gh pr checks 1249 --json name,bucket' 30s --match fail \ --run claude --prompt 'CI failed: {event}.