Auto-heal used to refresh cached AI steps and try again. Now, when tests fail in CI, Donobu puts an AI coding agent on the failures. The agent works out why each test failed, fixes the tests that a test change can fix, and Donobu re-runs them. The agent then opens a pull request, and Donobu adds the results it confirmed.
The agent never grades its own work: only a test that passes Donobu's re-run counts as fixed. When a failure is a real product bug, the agent leaves the test alone and explains what a person should look at.
- Same command as before:
donobu test --auto-heal - When a fix touches shared code, the dependent tests are re-run too, and the pull request says exactly what was re-run
- Re-running updates the same pull request; it never opens a second one
- The pull request flags any change that weakens a test, such as a skipped test or a removed check
- A reviewer's commits on the fix branch are never overwritten
- If the agent is unavailable, the classic auto-heal runs instead
Available for accounts with heal agent access. Ask us to turn it on.