Detectors

The rules Tracez runs over every recorded step of your sessions. Each finding on the Findings page came from one of these. Found by rule, no model.

Loop

found by rule, no model

The same tool call, with the same command or on the same file, fails again and again with no success of that call in between. Retrying with different inputs is a different call, so it is not a loop.

Trips at: 3 failures in a row of one call
Same call: same tool and command, or same tool and file
Recovered: when that call later succeeds
Cost shown: tokens recorded on the failing steps, est. $ at list prices

Stall

found by rule, no model

The session stopped mid-work: its last step is a tool call that never got a result, or the agent’s turn ended waiting on a tool, and nothing newer has arrived.

Trips at: 15 minutes with nothing newer
Mid-work: a tool call with no result, or a turn that stopped on a tool call
Status: always open: a session that records something newer is no longer stalled

Both rules are counted from recorded steps, the same ones the Tracez session view shows. Estimated $ = recorded token counts × list prices for the model family; unpriced models are left out.

Stop paying for agents that are stuck.

Detectors

Loop

Stall