Every run is a Playwright test

A run used to end in a report. Now it also ends in a test you own. How it works.

• The Playwright test panel. On a finished run, linked from next to its id: view the code, copy it or download it. The file is a plain @playwright/test spec: one test.step per browser step the agent took, in order, named after the step.
• Replay, with no model. Replay runs the same steps in a fresh browser on the test runner and shows each one passed, failed or skipped, in seconds. It stops at the first step that breaks and shows the error with a screenshot. No tokens, and it does not count as a run.
• Only what worked. Steps that errored are left out, so the file replays the path that reached the verdict, not the dead ends on the way.
• Runs without ZeusQA. The computer helpers the agent used are written into the file on top of Playwright, so it needs nothing but @playwright/test. Run it with npx playwright test; when the app stops doing what the run saw, the step that broke fails.
• In the pull request. The ZeusQA comment links the test for that run, so the check that reviewed a change can stay in the repository as a regression test.