UnrollWalkthrough transcript
A text version of the one-minute walkthrough on the home page. It was recorded with the extension’s own viewer and three made-up example files. The window frame, raw editor and context menu are drawn for the recording; everything inside the preview is what Unroll displays.
What happens
| Time | On screen |
|---|---|
| 0:00 | A Claude Code session is open as raw JSONL. Every line starts with the same envelope fields, so the conversation is hard to follow. The file is opened with Unroll: Open as Agent Trace and appears as a conversation: the request to fix a failing checkout test, the assistant’s reasoning, a Bash call and its test output. |
| 0:10 | Selecting result at line 4 on the npm test -- cart call shows the call and its failing result together. Clear returns to the whole session. |
| 0:16 | The assistant’s explanation of the bug switches to its original JSON record, with fields such as parentUuid, gitBranch, message and usage. Its line button opens line 9 of the file in a split editor beside the preview. |
| 0:28 | Searching for applyCoupon finds the Grep call, its matching lines and the file the assistant read, with every match highlighted. |
| 0:33 | With Follow latest on, two new records are added to the file: the passing test run and the final summary, which includes a before/after table and the one-line fix. |
| 0:41 | A dataset of eight agent runs opens in another tab. The list shows each run’s outcome. Adding the condition resolved is false narrows it to the three failed runs, and selecting the most expensive one shows its model, cost and turn count above its conversation. |
The tool calls shown are recorded data. Unroll doesn’t run them.
Watch the walkthrough or download the Claude Code session, agent runs dataset and chat dataset.