How it gets overlooked
The build failed, but the agent already moved on to writing a summary. Did it recover, or did an important failure disappear between two steps? The next action is often where a useful review should begin.
In a busy session, an error can scroll past just before a different tool call. The agent may have a valid recovery strategy. Or it may be proceeding without handling the failure. Either way, you need the surrounding sequence to decide.
What ClawMetry detects
ClawMetry flags a failed tool result followed by continued activity without a recognized retry or acknowledgement. It creates an informational action discrepancy finding. That lower severity matters: this is a prompt to inspect the transition, not a verdict about the agent intent.
The difference that changes the finding
Triggering example
Failure, different tool
Info finding
Quiet comparison
Failure, same-tool retry
No finding for this detector.
The checked example fails a shell build and then calls a different tool. It raises an informational finding. A failure followed by another call to the same tool is treated as a retry and stays quiet for this detector.
Inspect the detector result
{
"kind": "action_discrepancy",
"severity": "info",
"evidence": {
"failed_tool": "Bash",
"continued_as": "tool_call"
}
}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
Compare the failed step with what followed. Was the new action a recovery path? Was the error acknowledged elsewhere? Before relying on a completion report, verify the underlying result that the failed operation was meant to produce.
- Inspect the failed result
- Check the recovery path
- Verify the claimed outcome
What this signal establishes
The detector does not verify the final completion claim. Legitimate recovery through another tool can also match.