Replay Review

The replay page turns raw agent activity into a compact review artifact.

Timeline

The timeline shows normalized events in timestamp order. Scrub to move through the session, play it like a video, or jump directly to important moments.

Colored segments help separate execution, editing, testing, review, and error-heavy phases.

Terminal output

Terminal frames show commands and output previews for events such as terminal_command, terminal_output, and command failures.

Use terminal frames to answer:

• What command did the agent run?
• Did it fail?
• Did the agent retry the same command?

Diffs

Diff frames show file changes for events such as file_diff and edit failures. Review these frames when a marker points to a large change, risky file, or failed edit.

Test results

Test-related events appear in the same timeline as commands and edits. A reviewer can see when tests failed, whether the agent changed code afterward, and whether a later run passed.

Chapters

Chapters group the session into meaningful phases such as setup, implementation, debugging, and verification. Use chapters to scan a long session before opening detailed events.

Notes

Session notes summarize what happened and what remains risky. Notes are not a replacement for review; they are a fast entry point into the replay.

Needs Review markers

Needs Review markers point to moments that deserve human attention. A marker can be created for a large diff, a failing test, repeated retries, tool failures, risky config changes, or low-progress loops.

Needs Review does not automatically mean the agent was wrong. It means the event is important enough to inspect.