Preserve failure evidence

When tests fail, the default behavior stops the job. Steps after the failing step never run. If your screenshot upload comes after npm test, you get a red run and no picture of what broke.

That sends people rerunning jobs blindly. Reruns cost minutes and sometimes hide intermittent failures.

Upload steps should use explicit conditions. if: failure() uploads only on failure. if: always() also runs on cancel unless you add extra guards.

Upload evidence even on failure

Use if: always() on upload steps that must run when tests fail:

- name: Run browser tests
  run: npm run test:e2e

- name: Upload Playwright report
  if: always()
  uses: actions/upload-artifact@v4
  with:
    name: playwright-report
    path: playwright-report/
    retention-days: 7

always() runs unless the job was cancelled. You get HTML reports and screenshots from the failed run.

Be careful: always() also runs on success. That is fine for test reports. It is not fine for deploy steps.

Keep evidence focused and redacted

Do not dump every environment variable into a public artifact. Secrets can leak through build logs, .env copies, or HAR files.

Collect what a developer needs:

• junit or HTML test report
• one failing screenshot
• trimmed server log

Set retention-days so old failures do not fill storage forever.

Diagnose without a local rerun

Force a browser test to fail on purpose. Open the workflow run page, download the artifact, and find the failing selector from the report alone.

If a teammate can do that without you on a call, your evidence step works.

Large video artifacts slow downloads. Prefer one PNG and a short HTML report for pull request debugging.

Try this on your own project: break one assertion, confirm the artifact still uploads, then fix the test.