Skip to lesson
Exit
Checkpoints, queued messages and run control1 / 2

2 min lesson

Restore one bad turn

Restore the checkpoint before one bad request, rerun the check and inspect the remaining diff.

Step 1 of 2

Make a small change in two requests. Run its check after the first request. In the second request, add recency sorting. When that edit breaks the check, preview the checkpoint before the request, restore it and rerun the check. Inspect the remaining diff to confirm that the first change stayed and the sorting edit is gone. Use Git for permanent history.

Restore from the bad prompt's ↺

0:31 · narrated
Read this demo as text
  1. A second prompt sent the agent into the file searchFilter.ts, and a search test now fails. You don't debug forward through a wrong turn. You restore to just before it.
  2. Click the restore arrow on the bad prompt. Not the first one. The component and hook survive; only the filter edit is gone. And the bad prompt returns to the composer for a rethink.
Primary path: restore from the bad prompt's ↺ without losing the good work.

Practice next: Now run the walkthrough: restore from the bad prompt's checkpoint and keep the good diff intact.

Simulated Cursor 3.12 (macOS, light) — beta educational reconstruction, not the real product.