How it gets overlooked
The agent comes back online, so everything looks fine. Then it starts over again. Repeated restarts can hide behind a process that appears healthy whenever you happen to look at it.
An unfinished task survives in your to do list while the runtime keeps relaunching. Without the sequence of session events, each restart looks like a fresh beginning. Together, they suggest a reliability problem that deserves investigation.
What ClawMetry detects
ClawMetry counts recorded session starts and restarts within a bounded window. When enough accumulate, it raises a repeated restart finding. The evidence includes the count and whether the comparison used timestamps or the event window.
The difference that changes the finding
Triggering example
2 starts in one minute
Warning finding
Quiet comparison
A single start
No finding for this detector.
The checked example records two starts one minute apart and produces a warning. A single start stays quiet. The default restart window is fifteen minutes, and the finding reports the threshold used for that evaluation.
Inspect the detector result
{
"kind": "crashed",
"severity": "warning",
"evidence": {
"restarts": 2,
"threshold": 2,
"window_sec": 900,
"threshold_source": "static",
"observed": "session.started/restarted events counted by timestamps",
"span_sec": 60
}
}How the example was checked
These examples evaluate the published detector with authored event data or disposable configuration files. The videos illustrate those behaviors. They are not recordings of live agents or the product interface. No command in the examples was executed.
The result establishes behavior for these inputs. It does not establish runtime ingestion, prevention or a real compromise. Inspect the pinned source contract.
What to check next
Check the runtime logs for exit reasons and startup errors. Look at resource pressure and the last action before the restart. Once the cause is addressed, confirm that the agent stays running and returns to the original task.
- Read runtime exit logs
- Inspect the preceding action
- Confirm the session stays up
What this signal establishes
Repeated start events are a reliability signal. They do not prove a crash cause, and an intentional restart can contribute to the count.