Skip to main content

Session Trace

The session trace is a complete log of the agent’s conversation during the eval. It shows every thought, action, and observation.

What’s included

  • Agent thoughts — the agent’s reasoning process
  • Tool calls — API calls, browser actions, file operations
  • Observations — results the agent received
  • Errors — any errors encountered and how the agent handled them
  • Screenshots — browser snapshots showing what the agent saw

Using the trace

The trace is your best tool for understanding why a score is what it is. Look for:
  • Where the agent got stuck — repeated attempts at the same step
  • What confused the agent — misunderstanding docs or error messages
  • What worked well — smooth flows the agent navigated easily

Accessing traces

  • Dashboard — click any report, then click “Session Trace”
  • MCP — use the get_session_trace tool with the report slug