Reliability

Auto-restart

An agent you started with /agent up that dies is restarted automatically. Check the current state with /agent status.

If an agent crashes repeatedly in a short window, auto-restart pauses for a while to avoid an endless loop — you get a notification in chat saying it has paused and when it will try again.

Sleep / wake

An idle agent can be put to sleep to save resources; when a new message arrives for it, the agent wakes up automatically.

Hang detection

If an agent stays silent longer than hang_threshold_minutes while a message is waiting on it, you get a hang alert in chat naming the agent, how many minutes it has been silent, and who is waiting. The agent is not restarted automatically for a hang — the alert is just so you can decide what to do next.

Watchdog alerts

Triggered by idleness, not runtime: a working agent is never nudged, no matter how long the task takes.

Sender	To chat?	Meaning
watchdog-nudge:<agent>:<sender>	no	An internal reminder the agent itself receives when it leaves a message unanswered
watchdog-alert:<agent>	yes	Sent to you when the agent still hasn’t answered — run /agent status

Orphan process alert

If one of an agent’s processes gets left behind unmanaged (for example after a session closes abnormally), you get an alert in chat naming the agent, why it’s considered orphaned, its process id, how much RAM it’s using, and the time it started, along with a suggested command to stop it.

Settings

/agent set <name> <field> <value>:

Field	Meaning	Default
autoboot	Whether this agent starts automatically when you run /agent up with no name (0 / 1)	1
hang_threshold_minutes	Minutes of silence before you get a hang alert	20
auto_sleep_enabled	Whether an idle agent is allowed to sleep automatically (0 / 1)	0
auto_sleep_minutes	Idle minutes before an agent (with auto sleep on) goes to sleep	45