| name | run-e2e |
| description | Bring up infra and run the tix end-to-end suites — `@tix/api-e2e` (cross-service integration over docker compose) and/or `@tix/web-e2e` (Playwright browser flows over testcontainers). Handles infra startup, picks the right suite, and surfaces results + the Playwright report. Use when the user asks to run e2e tests, "run the integration suite", "run the browser tests", or invokes `/run-e2e`. |
| disable-model-invocation | true |
Run the e2e suites
Two independent tiers (see each app's CLAUDE.md for internals):
| Suite | Command | Infra |
|---|
@tix/api-e2e | pnpm e2e | needs docker compose up -d first |
@tix/web-e2e | pnpm -F @tix/web-e2e test:e2e | self-contained (testcontainers + Docker daemon) |
Steps
-
Pick the suite from the user's ask. "integration / saga / events" → api-e2e. "browser / buyer / seller / UI" → web-e2e. "all e2e" → both, api first.
-
Confirm Docker is up. Run docker ps. If it errors, stop and tell the user to start Docker Desktop — both suites need the daemon.
-
For api-e2e, bring up infra first:
docker compose up -d
pnpm e2e
pnpm e2e builds packages then runs the suite. It does not start infra itself; skipping step-3a is the #1 cause of a hang at startup.
-
For web-e2e, no compose needed:
pnpm -F @tix/web-e2e test:e2e
The buyer spec skips itself unless STRIPE_TEST_KEY (test-mode secret) is set. If the user wants the buyer flow covered, check that env var is present and warn if not — a "skipped" buyer test is not a passing buyer test.
-
Report. Summarize pass/fail. On a web-e2e failure, point the user at the HTML report (apps/web-e2e/playwright-report/) and the retained trace/video/screenshot in apps/web-e2e/test-results/ — these are Read-denied for the agent, so the user opens them locally. For deeper interactive diagnosis, the Playwright MCP can drive a browser against a running harness.
Notes
- Leave infra running between iterations; tear down with
docker compose down only when done (api-e2e migrations drop/recreate schemas each run, so a warm cluster is fine to reuse).
- This skill is user-invoked only — it starts containers and long-running suites, so it shouldn't fire as a side effect of model reasoning.
