Skip to content

Stall detection

Watchdogs distinguish a task producing no output at all from one that's genuinely just taking a long time, without prematurely killing healthy work.

AgentDock Pro

Stall detection is part of AIgency, available only in AgentDock Pro.

A task that's silent isn't necessarily stuck — a real agent turn can legitimately take a while. AIgency uses a few distinct timing thresholds to tell the difference rather than applying one blanket timeout.

The thresholds

ConditionTimeout
No output at all after starting90 seconds
No genuine content (only heartbeat-style activity)12 minutes
Mid-run silence after content has already started15 minutes

Note

Crossing the mid-run silence threshold flags a task as a possible stall — it does not automatically fail the task. This is a signal for you to check the raw terminal drawer or technical view, not an automatic kill switch.

Why not just one timeout

A single fixed timeout would either kill genuinely slow-but-healthy work or let a truly dead process sit unnoticed for far too long. Separating "nothing has ever happened" from "it was working and then went quiet" lets AIgency flag the second case as suspicious much sooner than the first, without treating either as an automatic failure.