Skip to main content
When a run doesn’t behave as expected, start with the lines the component writes at the top of its job log, then the report.

Read the plan

The component’s first real step, Resolve the verification plan, logs:
  • the mode and why it was picked, for example: auto, the repository is public, so the platform can check it out
  • the target: a deployed URL, or the port of an app on the runner
  • the timeout: the job needs, as “This configuration needs timeout: of at least 85m on the job that runs it.” Set at least that much on the job. When the job’s own timeout is lower, the component also warns
With the plan, the component warns about inputs that don’t apply to the mode it runs in, and about conflicts such as a port in app_url that differs from app_port. A warning means a setting was ignored, not that the run failed. Errors and warnings start with ERROR: and WARNING:.

Verbose logging

Set verbose: 'true' to log each tool call with its arguments, the tool responses, the prompt and the verdict details:

Common problems


Look at a run from your machine

With the IronBee CLI signed in to the same account, check a run on IronBee by its IRONBEE_JOB_ID:
See Verification jobs.

Session logs from your runner

When the verification ran on your runner, the job’s evidence artifacts include the session’s own logs. Each agent of the session has its own folder, sessions/<id>/agents/<agentId>/, and the main agent’s is agents/main/:
  • verdict.json holds that agent’s latest verdict.
  • actions.jsonl is its full event log.
If the agent stopped early, the report shows a banner with a collapsible block: the reason, the turns, errors and any blocked tool calls.

What’s next?

Evidence

Where the recording, screenshots, logs and report are.

Configuration

Every input and output, with defaults.