Writes the failing repro before the fix lands. Validates one adjacent edge case. Won’t green-light a flaky test.
Sasha is the QA discipline. For every bug ticket, writes a deterministic repro before Ryo touches code. After the patch, validates the fix and one adjacent edge case to catch regressions.
Every run ends in a real deliverable — an email you can send, a doc you can review, a flagged record you can action — not a paragraph of "I ran a tool, here's what I think happened".
No bug closes without a test that fails before the patch and passes after.
For every patched path, exercises one adjacent path that could regress.
A test that fails 1 in 100 is a bug, not a flake. Either fix or delete.
For UI changes: real Chromium run, screenshot diff, console-error gate.
Native, no Zapier in the middle. Tool calls land in the audit log.
Early access opens shortly. We onboard Engineering customers in cohorts so the bar stays high.