Skip to content

Anti-Patterns

Decision

Ranked by how often they cause flake or wasted CI minutes:

  1. page.waitForTimeout(2000) as the default wait — always wrong outside debugging. Replace with web-first assertion or waitForResponse
  2. Tests that depend on each other's state — a test that only passes when the previous one ran is a bug. Each test creates its own state and cleans up. test.describe.serial is a smell
  3. Brittle CSS selectors instead of role/label/test-id — .btn.btn-primary.mt-3 > span:nth-child(2) breaks on the next theme refactor
  4. Missing await on async calls — page.click(...) returns a Promise; without await, the next line races. Turn on eslint-plugin-playwright
  5. Asserting implementation details — await expect(page.locator('.has-loaded')).toBeVisible() ties to CSS-class side effect. Prefer user-visible: await expect(page.getByText('5 results')).toBeVisible()
  6. Test files >500 lines / mega-test asserting 20 things — split per journey; use test.step() to keep readable
  7. expect(true).toBe(false) for sad-path — use throw new Error('Unexpected branch') or test.fail()
  8. Overusing page.evaluate() — bypasses auto-wait, breaks trace viewer's locator picker, Selenium-porting tell. Most calls have a Playwright API
  9. Using isVisible() / textContent() for assertions — return immediately, no auto-retry. Always expect(locator)
  10. Disabling fullyParallel to "stabilize" — hides flake; right answer is locator/auth/state isolation
  11. Committing playwright/.auth/*.json — leaks session cookies; gitignore
  12. Trace viewer only for failures — open them on slow CI runs to find race conditions in green tests
  13. .first() instead of .filter() — hides "two elements match" bugs. Pair .toHaveCount(1) if uniqueness matters
  14. Mixing VR (toHaveScreenshot) and functional E2E in the same test — different stability requirements; keep separate

See Also