← Back to Programming and automationOpenClaw automations
Automation debugging and recovery
When automation stops moving, the question is not whether to narrate it. The question is which recovery move fits the actual failure.
Recovery orderVerify the stall, classify it, fix the minimum blocker, then confirm output resumed.
Recovery choices
Retry, reroute, restart, or repair?
RetryUse when the workflow is sound and the failure looks transient.
RerouteUse when the work landed in the wrong context, session shape, or owner.
RestartUse when the stage is blocked and no real next artifact landed.
RepairUse when a narrow defect can be fixed without broad rebuild drift.
Examples
Common failure classes and the right move
Subagent died after producing nothingRestart or reroute the stage, then verify the expected artifact appears.
Task is “active” but outputs are staleTreat the active label as suspicious and inspect the real artifact chain.
One config or path defect broke the lineRepair the narrow defect first instead of rebuilding the whole automation.
Open deeper reference links for this page