Test your web app in a real browser, and prove each fix

Claude Code opens your app in a real browser at phone and computer size, finds what people would hit, and writes a test for each problem. A check proves each test failed before the fix and passes after it.

Our run, 2026-09-28. Claude used the refund page in a real browser and found both planted problems. Its 2 tests for them failed on the page as it was and pass after its fix, unchanged. 20 of 20 working-flow tests passed both times.