Testing

Testing Skill

Test Strategy Layers

Unit Tests (pytest/vitest)     — Fast, isolated, run on every save
Integration Tests (pytest)     — API endpoints + DB, run on commit
E2E Tests (Playwright)         — Full user flows, run before merge

Test Case Design

From each acceptance criterion, generate 1-3 test cases:

Criterion: "Upload accepts PDF files up to 50MB"
→ TC-001: Upload 1MB PDF — succeeds
→ TC-002: Upload 50MB PDF — succeeds (boundary)
→ TC-003: Upload 51MB PDF — fails with size error

For detailed test data patterns, see references/test-data.md. For Playwright patterns, see references/playwright.md. For test case templates, see examples/test-case-template.md.

Output Structure

specs/test_artefacts/
├── test-plan.md              # Overall test strategy
├── test-cases.md             # All test cases with expected results
├── test-data/
│   └── fixtures.json         # Shared test data
└── e2e/
    ├── playwright.config.ts
    ├── flows/
    │   └── *.spec.ts
    └── helpers/
        └── test-utils.ts

Gotchas

• CSS class selectors in Playwright. Never use them. Use getByRole(), getByLabel(), getByText(), or getByTestId() (last resort).
• Tests depending on other tests' state. Every test must be independent. If test B fails when test A doesn't run first, the tests are coupled.
• Random data in tests. Tests must be deterministic — same input, same output. No random.choice() or Date.now() in test data.
• Missing boundary tests. If a criterion says "up to 50MB," there must be a test at exactly 50MB AND one at 51MB.
• Playwright timeouts on CI. Default 30s is often too short. Use toBeVisible({ timeout: 10000 }) for elements that depend on network requests.
• Forgetting empty state tests. Every list/table component needs a test with zero items.
• Testing implementation details. Don't assert on internal state or method calls — assert on observable behavior.

Verification

uv run pytest -x -q --cov=src --cov-report=term-missing
npm test -- --coverage
npx playwright test
npx playwright show-report