Playwright E2E — QA Execution, Data Lifecycle, and Evidence

This skill executes browser scenarios; it does not invent their product behavior. QA design remains the source of case IDs, priority, preconditions, synthetic dataset, and observable outcome.

1. Bounded preflight

Run node scripts/detect-playwright-profile.js <repository-root> once before changing or executing a test. It reads the package/configuration files and at most 12 nearby Playwright tests, returning the configured command, projects, data-fixture candidates, selector conventions, and every inspected file. Save that JSON unchanged in the run report.

Use the profile rather than rescanning the suite. Its fixture result proves only that seed/cleanup/absence symbols exist; read the selected helper and execute the absence probe before treating the lifecycle as safe. If the Playwright dependency/configuration, command, or data lifecycle cannot be proven, stop as BLOCKED; never substitute another browser runner.

2. Run-owned environment

PW-0 (One owner) — MUST. Run the companion e2e-runner infrastructure profile first. Compose is evidence only. A verification run uses Node Testcontainers owned by one setup process for the full browser run. Prefer Playwright globalSetup returning its teardown function for container lifetime; setup-project dependencies may seed data but must not be the sole container owner. Write environment.json with container/image IDs, mapped endpoints, app URL, property bindings, and external stubs. Tests consume that manifest through a custom fixture.

If the repository has Playwright but lacks Node Testcontainers, an explicit request for test infrastructure authorizes the smallest test-only dev dependency and setup/teardown fixture. Start the application as a run-owned test process/container after infrastructure is ready. reuseExistingServer, an existing storageState, a non-local baseURL, or an app that cannot receive mapped endpoints is UNTRUSTED_ENVIRONMENT and cannot PASS.

3. QA handoff contract

Each executable case must contain:

Missing Seed owner or Cleanup owner is a refusal to execute, not a reason to use a broad database delete. UI text selectors are allowed only when the repository has no stronger stable semantic locator.

4. Data lifecycle — mandatory

PW-1 (Run namespace) — MUST. Generate one run ID such as pw-<timestamp>-<short-random> and scope every created user, order, file, key, and search document to it. Use synthetic data only.

PW-2 (Manifest before action) — MUST. Before the UI action, write the created resources to an in-memory test manifest and the per-case result: {kind, id, seedOwner, cleanupOwner}. Do not record secrets, cookies, access tokens, or PII.

PW-3 (Finally cleanup) — MUST. Register cleanup before the first mutating UI action and run it in Playwright test.afterEach / try...finally, whether the case passes, fails, times out, or is aborted. Only delete resources in this run's manifest; database-wide reset, volume removal, DELETE FROM <table> without an ID predicate, and cleanup of adopted/shared resources are forbidden.

PW-4 (Cleanup verification) — MUST. After cleanup, use the repository-provided absence probe (test API, fixture, or scoped query) to prove each manifest resource is gone. Cleanup verification failure makes the case FAIL even when the UI assertions passed.

Containers are run-scoped; data namespaces are case-scoped and cleaned before the next case. Final teardown stops the app, containers, and network in reverse order and audits run labels for zero orphans. Teardown/orphan failure makes the run FAIL.

5. Browser execution

PW-5 (Verified project configuration) — MUST. Preserve repository browser projects, locators, workers, and timeouts, but force retries to zero and use only the run-owned environment manifest for baseURL/auth. Do not reuse an existing server or storage state. Every stateful/external endpoint maps to a run-owned container/stub.

PW-6 (One execution) — MUST. Execute each selected case once. Do not enable retries to hide flaky behavior. Use Playwright's auto-waiting and explicit assertions; fixed sleeps are forbidden. Eventual UI state must use a bounded expectation timeout from the repository configuration.

PW-7 (Evidence) — MUST. Capture trace, screenshot, and video on failure using the existing Playwright config. Mask or omit sensitive values before writing reports. Never save authenticated storage state, secrets, or raw PII in the report directory.

PW-7.1 (Evidence mode) — MUST when e2e-evidence-report requests it. Record video and a screenshot for every case, not only failures, by running with the run-owned overlay that skill generates outside the repository (--config <overlay>). The overlay imports the repository config unchanged; do not edit the repository config to get recordings. PW-5, PW-6 and the masking rules above still apply, and evidence-mode artifacts stay in that run's temporary evidence directory.

6. Report and terminal result

Write only under test-report/browser-e2e/{run-id}/:

{run-id}.json       # case status, timing, data manifest, cleanup evidence, artifact paths
{run-id}.md         # concise human summary
artifacts/<case-id>/ # Playwright trace/screenshot/video only when generated

For every case report: PASS, FAIL, BLOCKED, or NOT_RUN; the QA case ID; environment; seed/cleanup owners; created resource count; cleanup result; cleanup-verification result; and artifact paths. Print a compact summary by priority and exit non-zero for FAIL or BLOCKED P0 cases.

Checklist

• Existing Playwright setup and safe fixture proven; any added dependency is test-only and explicitly authorized
• Node Testcontainers owner holds infra/app through browser teardown; Compose/existing servers are not used
• Bounded preflight profile saved and reused; no suite-wide rediscovery
• QA handoff has case ID, synthetic data, seed owner, cleanup owner, and visible assertion
• Every created resource is run-scoped and recorded before mutation
• Cleanup runs in finally semantics and is verified case by case
• Playwright runs once with semantic locators and bounded assertions
• Failure evidence is preserved without secrets or PII
• Per-case report records result, data lifecycle, and artifacts