Testing Strategy Skill | Agent Skills

Testing Strategy

Pyramid: many fast unit tests → fewer integration → thin load/chaos top.
Unit tests mock deps with mock<DepsType>() / mockDeep — never vi.mock for app modules.
Integration tests use real localhost DB only; guardrails must throw on non-localhost URLs.
Name by type: *.test.ts (unit), *.test.int.ts (integration), load/*.js, chaos/*.
Hard to test = design smell (fn-args-deps mocks only what the function uses).
Faker for realistic unique fixtures so tests parallelize without cleanup.
Before long setup/Prisma/Faker examples, read the matching resource below.

Pick the layer by the question:
- Logic / injectable deps? → Unit (mock deps, ms)
- Crosses real boundary (DB, FS)? → Integration (.test.int.ts, localhost)
- Concurrent traffic / SLOs? → Load (performance-testing)
- Survive dependency failure? → Chaos (performance-testing)
Unit: typed mocks from the deps interface; assert Results and values (writing-tests).
Integration: real DB, seed unique Faker data, query only that data.
Ensure vitest.setup.ts enforces localhost DB + vi.clearAllMocks() between tests.
Prisma fluent API: mockDeep<PrismaClient>() (+ $transaction callback helper).
Load/chaos: critical paths only — see performance-testing.

references/examples.md — unit/integration/Prisma/Faker patterns. Read when implementing a layer.
references/setup.md — file naming, guardrails, vitest config, load smoke sketch. Read when scaffolding the suite.
references/rationalizations.md — excuse→reality and red flags. Read when tempted to skip the pyramid.

Not the red-green-refactor loop (tdd-workflow) or assertion craft (writing-tests).
Deep k6 profiles and Toxiproxy live in performance-testing.
Adjacent: fn-args-deps, result-types.