Inspect a run
Trace Detail on the Agent Detail Dashboard shows everything recorded for one run of an agent. Use it to find out why a run failed, was blocked or gave an unexpected answer.
Open a run
- Go to Agent Ops → Agent Detail Dashboard and choose an agent.
- Select a row in Recent Runs. To open a run you already know, type its ID in Conversation ID and select Open Run.
The run opens in Trace Detail below the table.
Trace tabs
| Tab | What it shows |
|---|---|
| Overview | The run's status, model, latency, cost and tokens. Prompt Safety shows whether the prompt or its context was flagged or blocked, and why. Error Detail shows the error type, where it happened and whether it can be retried. |
| Mind Map | A map of the run's behaviour, built from its telemetry, with a short assessment. Switch between Map and JSON. You can copy the map's JSON or download the map as a PNG image. |
| Output | Output Preview shows the response, and Tool Path shows the tools the run called. |
| Trace Steps | Agent Trace shows why the run stopped. Runtime Steps lists every step in order with its duration. Execution Flow groups the steps by phase, such as request, tools and completion. |
| Payloads | The Input Payload and Response Payload of the run. |
| Raw Logs | The Raw Trace Logs recorded for the run. Reasoning, system prompts and messages are shown as [redacted]. |
| Critic | The Critic Review of the run, with its AI² score and the Critic Response. |
Export a trace
Select JSON or CSV at the top of Trace Detail to download the trace. The buttons are available once a trace has loaded.
Troubleshooting
- No run selected. Select a row in Recent Runs.
- No execution trace returned yet. or No raw trace log records returned for this run yet. The run may still be in progress. Select Refresh on the dashboard after a moment.
- No critic response recorded yet. This appears only after the post-run critic check writes back to chat history. The critic check hasn't finished for this run. Check again later.