Playwright Visual Testing

Playwright Visual Testing

Trigger On

• the user asks for pixel, screenshot, visual, or UI regression testing with Playwright
• a .NET repo needs visual baselines for ASP.NET Core, Blazor, WebAssembly, static pages, or generated frontend assets
• GitHub Actions should run Playwright screenshots and expose expected, actual, and diff artifacts
• tests fail with screenshot mismatches, noisy baselines, or unstable visual snapshots

Do Not Use For

• pure .NET unit or integration tests without a browser surface
• accessibility, SEO, PWA, or security-header audits; route those to webhint
• browser debugging or live DOM inspection; route that to chrome-devtools-mcp
• JavaScript, TypeScript, CSS, or HTML linting; route those to biome, eslint, stylelint, or htmlhint

Load References

• Read CI and snapshot patterns when adding a new visual test suite, wiring GitHub Actions, choosing between Playwright snapshots and a standalone Pixelmatch script, or stabilizing screenshot diffs.

Current Upstream Notes

• The August 2026 Playwright CI and visual-comparison docs still require browser dependencies to be installed explicitly in CI and warn that screenshot rendering varies by host OS, browser build, fonts, headless mode, and hardware. Generate and review baselines in the same environment used for comparison.
• The CI guide recommends against caching browser binaries by default: restoring them often costs as much as downloading, and OS dependencies still need an explicit install. If a runner must cache browsers, key it by the exact Playwright version and keep dependency installation in the job.
• Current CI examples use actions/checkout@v6, actions/setup-node@v6, and actions/upload-artifact@v5; use a full checkout only when --only-changed needs the pull-request base ref.
• Keep Playwright parallel by default. Do not set workers: 1 merely because CI or screenshots are involved. Isolate test data and browser contexts, enable fullyParallel when tests are independent, and shard large suites across CI jobs. Reduce concurrency only for the smallest tests that destructively change the same external state.
• Playwright v1.62.1 fixes TypeScript configuration resolution regressions, accessibility snapshots that dropped names or image-style actionable elements, and branded primitive arguments passed to page.evaluate(). Re-run config discovery, accessibility snapshots, and TypeScript compile checks before accepting new visual baselines.
• Keep --update-snapshots as an intentional local review action. Pull-request CI should retain expected, actual, diff, trace, and report artifacts instead of silently accepting a new baseline.

Workflow

flowchart TD
    A["Need visual regression coverage"] --> B{"Uses Playwright Test"}
    B -->|"Yes"| C["Prefer expect(page).toHaveScreenshot"]
    B -->|"No or custom compare needed"| D["Capture page.screenshot output"]
    D --> E["Compare with pixelmatch and pngjs"]
    C --> F["Stabilize viewport, data, animation, and volatile regions"]
    E --> F
    F --> G["Commit reviewed baselines"]
    G --> H["Run in CI and upload reports or image diffs"]
    H --> I["Triage expected, actual, and diff before changing thresholds"]

1. Inspect the current browser-test surface:
   • nearest AGENTS.md
   • package.json, lockfile, Playwright config, test folders, and CI workflows
   • how the app starts locally: dotnet run, Aspire AppHost, static preview, or frontend dev server
2. Choose the comparison path deliberately:
   • default to Playwright Test expect(page).toHaveScreenshot() when the repo can use Playwright Test snapshots
   • use page.screenshot() plus a standalone Pixelmatch script only when the repo needs article-style central screenshots/baseline, screenshots/actual, and screenshots/diff folders, non-Playwright image inputs, or custom reporting outside Playwright Test
3. Make screenshots deterministic before tuning thresholds:
   • fix viewport, browser project, locale/time zone, color scheme, and device scale factor
   • use stable test data and wait for the app-specific ready state
   • disable animations or use Playwright screenshot options for animations
   • mask or hide volatile regions such as ads, time, avatars, random IDs, spinners, and third-party iframes
4. Keep baseline updates explicit:
   • generate missing baselines once, review them, and commit them
   • update intended Playwright snapshots with npx playwright test --update-snapshots
   • do not auto-create or auto-update baselines in pull-request CI
5. Wire CI for repeatability:
   • use npm ci, then npx playwright install --with-deps, then the focused Playwright command
   • preserve Playwright's parallel workers; use fullyParallel for isolated tests and CI sharding for large suites
   • constrain only a narrow destructive shared-state collision, never the whole visual suite for generic stability
   • optionally run npx playwright test --only-changed=origin/$GITHUB_BASE_REF first on pull requests for faster feedback, but always follow it with the full suite because changed-test selection is heuristic
   • use the same OS, browser build, fonts, headless mode, and rendering environment that produced the committed baselines; an official Playwright container is useful when host drift keeps changing pixels
   • upload the Playwright HTML report and test-results/, or upload screenshots/baseline, screenshots/actual, and screenshots/diff for a custom Pixelmatch flow
6. Triage failures from artifacts:
   • inspect expected, actual, and diff images together
   • classify the mismatch as intentional design change, rendering nondeterminism, app bug, or baseline drift
   • fix nondeterminism before increasing maxDiffPixels, maxDiffPixelRatio, or Pixelmatch mismatch thresholds