Every prompt, every diff, one timeline
Execution trace viewer 2.0 puts the full run on a single scrollable timeline — with deep links to any step.
By Platform team
Trusting an autonomous system requires being able to inspect it. Not just the output — the reasoning. Why did the agent make that edit? What did it see when it decided to replan? Which tool call produced that unexpected result?
Execution trace viewer 2.0 answers those questions on a single timeline. Every prompt, every tool call, every diff, every test result, every replan — in the order they happened, with timestamps and durations. You can follow the exact path the agent took from task start to PR open.
The most useful addition is deep links. Any step in a trace has a stable URL you can share with a teammate or paste into a ticket. When a run produces an unexpected result and you need a second opinion, you send the link. No screenshots, no log exports, no context-setting — the full trace is already there.
We rebuilt the viewer's data model to handle long-running tasks without pagination. A complex multi-agent run that spans hundreds of steps loads as a single scrollable timeline. Filtering by agent, by step type, or by time range narrows it without losing the context of the full run.