Read a deep-dive trace to see exactly what an agent did

What a trace contains

• The plan the agent made.
• Each tool call with its arguments and result.
• Reflection and verification steps.
• Tokens, latency, and cost per step.
• The run id, so you can link a trace to an audit entry.

Open a trace

1. From the live log or history, click a run’s "trace" link.
2. The Deep Dive view opens with the full sequence.
3. Scroll through the steps in order.

What to look for

A tool call that returned an unexpected result, a plan that was too broad, or a verify step that did not actually verify. The trace is where "why did it do that" gets answered.

Use it to tune

Found a step that keeps failing? Tighten the capabilities or shorten the system prompt. The trace tells you what changed, not just that something went wrong.