[siddhant573]

Summary

Adds @testmuai/e2e-evidence under integrations/e2e: a reporter for the e2e framework that writes each run as a sealed .evidence pack, validated at L1, using e2e's public reporters API.

import { evidence } from '@testmuai/e2e-evidence';

export default {
  targets: [{ engine: web(), app: { url: 'http://localhost:3000' } }],
  reporters: ['list', evidence()],
} satisfies E2EConfig;

The e2e maintainers prefer this as a standalone package over a built-in reporter (e2e-inc/e2e#754), so it lives here, next to the format library it seals with.

• One pack per run in .e2e/evidence/, named in the terminal summary. Options: outDir, profile (L0/L1), enabled. E2E_EVIDENCE=0 turns it off for a run. An unknown option throws.
• Built from e2e's report-1 document and the artifacts it links: readable test names (title path), step summaries and durations, failed (only ASSERTION_FAILED) vs broken, the failure record with expected/actual, locator, URL, frame and screen text, agent steps with their turns, and shareable traces.
• Copies only artifacts whose redaction is complete or not-required, never the test source (the definition is the test's identity from the report).
• It never changes a run's outcome: a pack it cannot write, or one that does not validate, is a summary line.
• Its own package.json (ESM, Node 22.12+, peer e2e >=0.17.0 <1) and its own CI job. The root package's build, typecheck, and tests do not include it.

Not yet, because e2e's report does not carry them: the value a step typed or picked, where a step acted (the viewer's cursor and element box), and a frame on every passed step (the validator reports those as warnings). The README says so; an e2e PR adding step detail to the report is in progress.

Publishing is not wired: release.yml publishes the root package only. This needs its own release step or a manual npm publish from integrations/e2e, to decide in review.

Verified

Locally, with the packed tarball installed in a fresh project next to the published e2e@0.17.0 and @e2e-dev/web@0.12.0, against a one-page login app, one passing test and one failing on purpose:

   ✓ signs in with a known user 831ms
   × welcomes an unknown user (fails on purpose) 5.25s
      Tests  1 failed | 1 passed (2)
   Evidence  .e2e/evidence/01a10bfd-8b2d-7735-b776-83a7f5adb0ae.evidence

$ npx evidence validate .e2e/evidence/*.evidence --profile L1
! tests/t-bb3729ed188c3ecf/steps/1-0-0: step folder "1-0-0" has no screenshot.<ext> [l1.steps.screenshot_missing]
…
ok valid (profile L1, evidence 0.1, status finalized)

The failing step's failure.yaml has expected: text "Welcome, grace", actual: text "Invalid credentials", the locator, the URL, screenshot.png, and screen.txt. With E2E_EVIDENCE=0 no pack was written.

npm ci && npm run typecheck && npm run build && npm test in integrations/e2e: 33 tests pass. Root npx tsc --noEmit is unchanged.