Each failure is diagnosed from the full recording

Each failed test now gets its own read-only investigation after the run, on the machine that ran it. The agent reads the run's full recording: the complete error, logs, network traffic, the trace, the code as it ran, and earlier runs. It can compare its failure with every other failure in the run.

When the same failure was seen before, the investigation starts from what was concluded then, and checks whether this run shows the same cause. A failure a person already judged keeps that verdict.

• Runs with or without --auto-heal; auto-heal acts on these diagnoses
• Every claim in a diagnosis cites something recorded
• donobu evidence gives the same read-only queries to you or your own coding agent
• donobu diagnose --from <test-results> runs the investigation on a downloaded CI artifact
• Failure reasons now cover a stall in your app's own backend and an overlay that covers the page

Available for accounts with heal agent access. Other accounts keep the classic triage.