Heal pull requests flag changes that weaken a test

The heal agent can re-run transient failures without editing them, and every heal pull request flags removed checks, raised timeouts, loosened bounds and guarded steps.

A fix that makes a test pass is not always a good fix. A heal pull request now points out every change that could make a test weaker, so a reviewer can look at those lines first.

The heal agent can also re-run a failure it judges transient, such as a slow page or a service restart, without changing the test.

  • Flags for a removed check, an edited check, a raised timeout or retry count, a loosened bound, and a step that now runs only under a condition
  • Flags for a value the test used to read from the app and now writes itself
  • Flags are advisory: they never block a fix
  • A raised timeout counts as a repair only when the evidence shows the wait was too short
  • Tests that failed and then passed on a Playwright retry now get a treatment plan too