Ship reliable tests in every pipeline — CI-specific patterns for speed, stability, and actionable reports.

1. retries: 2

   in CI only — surface flakiness in pipelines, not locally

2. traces: 'on-first-retry'

   — capture rich debugging artifacts without slowing every run
3. Shard across runners —

   --shard=N/M

   splits tests evenly; scale horizontally, not vertically
4. Cache browser binaries —

   ~/.cache/ms-playwright

   keyed on Playwright version
5. Upload artifacts on failure — traces, screenshots, and HTML reports as CI artifacts
6. Use the official Docker image —

   mcr.microsoft.com/playwright:v*

   has all OS deps pre-installed
7. Global setup for auth — run login once in

   globalSetup

   , reuse

   storageState

   across workers
8. Fail fast, debug later — keep CI runs short; use trace viewer and HTML reports to investigate