fixlisted

# Fix tests Input: `$ARGUMENTS`. A failing test, a failure output, a CI run, or nothing, in which case run the suite and start from what is red. ## When to delegate More than two or three failures, or a refactor across the suite: hand it to the `test-maintainer` agent. It will read a lot of files. One failure the user is watching: do it here. ## The order **1. Reproduce.** `commands.test_one` from the profile. If it passes locally and fails in CI, that gap is the finding. CI has no cached session, real parallelism, a different viewport, a slower machine, possibly a different timezone. **2. Triage.** Use `flake-triage` to work out which of five things it is. Doing this before touching anything is the whole discipline, because four of the five have different fixes and the fifth means not touching the test at all. **3. Fix the cause, once.** If four tests fail the same way there is one fix, usually in a page object or a fixture. Grep before editing, the sibling tests are broken too and nobody has noticed. **4. Prove it.** Run it. If flake was the complaint, run it several times, one green run says nothing about an intermittent failure. Then break it on purpose and confirm it still fails for the right reason. ## Never **A sleep.** Not a small one, not temporarily. If it only passes with one, there is a race and it is usually in the app. **A wider assertion.** Changing `toHaveText` to `toContainText` because it failed removes the check that found something. **Retries.