How it gets overlooked
Your agent has opened another file. And another. The session looks busy, but the fix still has not appeared. When you are watching multiple agents, a stream of activity can hide a task that is not advancing.
Imagine asking for a small code change. The agent reads twenty five files without making an edit or finishing the turn. You could discover that by scrolling through the session. ClawMetry gives you a specific signal to investigate instead.
What ClawMetry detects
The no progress detector counts tool calls since the latest user turn or completion marker. If the count reaches its threshold without a recognized write or edit, it raises a warning. The finding includes the observed call count, the threshold and zero recorded writes.
The difference that changes the finding
Triggering example
25 reads, zero edits
Warning finding
Quiet comparison
24 reads, then an edit
No finding for this detector.
The checked example has twenty five read calls and raises a warning. Replace one of those calls with an edit, and this detector stays quiet. A fresh user turn resets the window. Where a runtime never exposes writes in its learned sample, ClawMetry can disable this signal rather than flag every session.
Inspect the detector result
{
"kind": "no_progress",
"severity": "warning",
"evidence": {
"tool_calls": 25,
"writes": 0,
"threshold": 20,
"threshold_source": "static",
"baseline": null
}
}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
Start with the task. Research may correctly involve many reads. A requested code fix may need intervention. Check what the agent is trying to learn, clarify the stopping point, and look for the first meaningful change.
- Check the task type
- Inspect the latest activity
- Define a useful stopping point
What this signal establishes
No recorded edit is not proof of wasted work. Read-only research can be legitimate, and thresholds depend on available telemetry.