Self-healing Selenium tests with Claude

Automated tests usually break for a boring reason: a button or field moved, even though the feature still works fine. Normally that means a red, failing build until someone manually updates the test.

Our framework spots that kind of failure and asks an AI model to find the button or field's new location, then carries on the test automatically. In the verified run, it fixed a broken test on its own with 98% confidence, and all three demo tests passed.

It can also write new test cases from a plain description of a feature. The code is open source on GitHub.

The point is not to hide failures. A healed locator is still a sign the page changed, so the fix should be reviewed and written back into the test. The verified run is one example on our own demo tests, not a promise for every suite. To try the idea without cloning the repo, the self-healing locators demo on this site runs the same kind of lookup live: paste a broken locator and the page's HTML, and get a replacement with a confidence score.

Written by Rushi Prajapati