Direct answer: Flaky Playwright or Cypress test repair begins by reproducing the failure and classifying whether the cause is the test, product or environment. The repair should remove unsafe timing assumptions, preserve real product failures and prove the agreed suite across repeated clean runs. Retries may collect evidence, but they must not define success.
Classify the failure before editing the test
Capture the trace, screenshot, network or console evidence needed to distinguish a selector or timing problem from a real application defect or an unavailable dependency.
Use observable conditions instead of fixed delays
Wait for the user-visible state or network result that proves the next action is safe. Keep assertions tied to business outcomes, and do not weaken them merely to turn a red run green.
Use repeat runs as acceptance evidence
Run the agreed scope from a clean state enough times to expose recurrence, record the environment and retain genuine failures. A passed repeated suite proves only the tested scope, not every product path.
Sources reviewed
These sources support the category wording or external method referenced above. SystemLineage’s workflow and limits are stated separately.
Review Playwright and Cypress Test Repair, then decide which records, limits and setup apply to your work.
See Playwright and Cypress Test Repair