Task detail video

[00:00] Every task has its own detail page. The header shows the task title and the task ID, with a Copy button and a shortcut to test the task in the Eval Playground. Context usage shows the model, the latest context tokens, total tokens, estimated cost, and number of turns.

[00:16] The token breakdown splits usage into input, output, cache read, and cache write. Below it, the per-turn usage chart plots how the context window grows turn by turn, and it makes spikes easy to spot. Hover to read each series.

[00:29] Session metadata records the collector, the repository, the branch, the commit, and whether the working tree was dirty. This tells you exactly which code state the agent was working on. Sensitive paths are blurred in this video.

[00:41] Summary tiles show the number of retries, when the task started, when it completed, and the estimated cost. Click the retries tile to filter the timeline down to retried events.

[00:52] The event timeline lists every event in order, from session start through test failures and edits. Each event shows the model, token counts for that turn, and its cost.

[01:01] With the retry filter on, only retried events remain. Each one is tagged with the retry type and shows the tokens it consumed.

[01:08] Scroll through the retries to see how the agent kept rewriting the same edit, and how much each attempt cost.

[01:15] Select Show change details on an event to see the code before and after the edit. The code is blurred here for privacy.