← ClaudeAtlas

smoke-testlisted

Fast critical-path sanity check: app boots, core path works, key endpoint responds. Run after integrating a change and again after deploy or merge, not as a full user journey. Use for smoke test, sanity check, critical path verification, does it still work, or a post-deploy check.
alex-macra/ai-skills-assembly · ★ 0 · Testing & QA · score 76
Install: claude install-skill alex-macra/ai-skills-assembly
# Smoke test A smoke test answers one question in seconds, not minutes: **is the thing fundamentally alive?** It runs after you integrate a change and again after deploy/merge. If it fails, you stop - you do not proceed to ship. ## The critical path - and only the critical path Cover the few things whose failure means "nothing works": - The app **boots** with no fatal error. - The **primary surface** loads (home screen renders / process starts / scene loads). - A **health/liveness** signal is green. - **One key end-to-end action** succeeds (the single most important thing the product does). That's it. If you're writing the tenth assertion, you've drifted into `e2e-qa` territory - pull back. ## Keep it fast and deterministic - Seconds, not minutes. A smoke test you skip because it's slow protects nothing. - No flaky waits, no giant fixtures, no dependence on third-party services you can't stub. - Deterministic pass/fail with a clear signal. A "probably fine" smoke test is worthless. ## By stack | Stack | Smoke | |---|---| | **Godot** | `godot --headless --quit` (boots, no parse errors) + one `--script tests/<x>_check.gd` domain path | | **Web** | Preview/dev server up → home route returns 200 → one key API call succeeds | | **Backend / API** | `/health` returns 200 → one representative endpoint returns the expected shape | | **CLI** | Binary runs `--help` / a trivial real command → exit 0 | ## Post-deploy smoke (the second run) - Hit the **deployed** URL, not localh