← Back to Operations and recovery
OpenClaw operations

Stalled jobs, runtime drift, and recovery

Not all stalls are the same. This page helps you classify the stall correctly so the recovery move is honest and effective.

Recovery doctrineDetect the stall class, fix the minimum blocker, restart immediately, then verify movement resumed.
Stall classes

The main ways work goes false-active

Follow-through stallWork was claimed or started, but the real next artifact never landed.
Handoff stallOne stage ended, but the next owner never actually picked it up.
Runtime stallThe execution context, process, session, or service itself is unhealthy or missing.
Wrong-problem driftAdjacent cleanup continued while the requested outcome stopped moving.
Examples

Examples of correct recovery moves

Draft exists but queue still says draftingTreat it as a false-active state and advance the next real owner or validation step.
No process, no session, no artifact growthTreat it as an actual stall and restart or reroute the stage immediately.
Browser or gateway flaked mid-QARepair the runtime layer, then rerun the blocked verification step.
Polish work continued while the required deliverable stalledCut back to the core outcome and resume the main line.
Open deeper reference links for this page