Full Coverage — Test Generation

Generate the right tests for every layer of the testing pyramid: unit, integration, API, and E2E. The skill analyzes the code, selects the correct layers, and implements each one following strict guidelines.

Testing Pyramid

        ╔═══════════════════════╗
        ║    E2E Tests          ║  Playwright browser · user journeys
        ╠═══════════════════════╣
        ║  API Tests            ║  Playwright HTTP · HTTP contracts
        ╠═══════════════════════╣
        ║Integration            ║  Vitest + Prisma · every branch with real DB
        ╠═══════════════════════╣
        ║  Unit Tests           ║  Vitest · isolated logic, no I/O
        ╚═══════════════════════╝

Layer Responsibility Matrix

Each concern is owned by exactly one layer. Do not duplicate responsibilities.

Step 1: Analyze the Code

Before selecting layers or writing any tests, read the target code.

Read these files:

• Zod schemas / DTOs (<domain>.dto.ts)
• Service class (<domain>.service.ts)
• Route handlers (app/api/<path>/route.ts)
• Utility functions (<domain>.utils.ts)
• DB schema (prisma/schema.prisma)
• Error definitions (src/lib/server/errors.ts)
• External service clients (if any)
• UI components / pages (if any — look for data-testid attributes)

Build an inventory:

• Schemas found: list field names and rules
• Service methods found: list method names and their dependencies
• HTTP endpoints found: list METHOD /path
• Utility functions found: list exported functions
• External service calls found: list service names
• UI pages found: list page URLs

Write a one-paragraph summary: “This module has X schemas, Y service methods, Z endpoints… Recommended layers: [list].”

Step 2: Select Layers

Apply this decision tree after completing Step 1:

Does the code have Zod schemas or pure utility functions?
  YES → Unit tests required (always the baseline layer)

Does the code have service methods with branches (if/else, switch, throw)?
  YES → Unit tests (mocked) + Integration tests (real DB) required
  Why: unit tests prove logic; integration tests prove wiring with real DB.
        Together they are not redundant — they cover different failure modes.

Does the code expose HTTP endpoints?
  YES → API tests required (one spec file per endpoint)

Does the code have browser UI with data-testid attributes?
  YES → E2E tests required (one spec per feature area)

Common combinations:

• Pure utility library → Unit only
• Backend service, no UI → Unit + Integration + API (if endpoints exist)
• Full-stack feature → All four layers

State explicitly which layers are selected and which are skipped with a reason: “Skipping E2E: no browser UI found in this module.”

Step 3: Generate Tests, Layer by Layer

Work bottom-up. Generate unit tests first, then integration, then API, then E2E. Complete and self-validate each layer before moving to the next.

If Unit tests are selected:

Read references/unit-testing.md before writing any code.

Key steps:

1. Read schemas, service, utils, errors
2. Map every field × rule and every service branch as a case tree (write as comments first)
3. Create schema.helper.ts and <domain>.unit-factory.ts
4. Write schema tests (one describe per field, one it() per rule)
5. Write service tests (all branches mocked, exact mock argument verification)
6. Write utility tests (if applicable)
7. Run: vitest run src/modules/<domain>/test/unit/
8. Run self-validation checklist from the reference file

If Integration tests are selected:

Read references/integration-testing.md before writing any code.

Key steps:

1. Read source + auth implementation + error response shape (determines assertion format)
2. List every execution path as comments inside describe blocks
3. Create DatabaseHelper, AuthHelper, fixtures, api-utils
4. Write handler tests (1–2 validation wiring tests per endpoint — no exhaustive Zod rules)
5. Write service layer tests (direct calls, real DB)
6. Run: yarn test:integration (or equivalent)
7. Run self-validation checklist from the reference file

If API tests are selected:

Read references/api-testing.md before writing any code.

Key steps:

1. Read route handlers, DTOs, error definitions
2. Verify/create response helpers for all error codes
3. Create test-owned factory (never import backend DTOs)
4. Create API utilities (one function per endpoint)
5. Create cleanup utilities
6. Write test specs: plan scenarios as comments first, then implement
7. Run: npx playwright test --project=api
8. Run self-validation checklist from the reference file

If E2E tests are selected:

Read references/e2e-testing.md before writing any code.

Key steps:

1. Read PRD, page components (data-testid attributes), existing fixtures
2. Identify user flows (one spec file per feature area)
3. Create or update Page Object Models (locators via data-testid only)
4. Create test data factory with counter + timestamp + random pattern
5. Create API-based setup helper and cleanup helper
6. Write test specs (API setup, inject auth, navigate, UI interaction, assert)
7. Run: npx playwright test --project=chromium
8. Run self-validation checklist from the reference file

Universal Patterns

These apply at every layer. Do not repeat in reference files.

Unique test data

let counter = 0;
function uniqueSuffix(): string {
  counter++;
  return `${counter}-${Date.now()}-${Math.floor(Math.random() * 100000)}`;
}
// Usage: `test-user-${uniqueSuffix()}@test.com`

Why: parallel test runs share a database. Non-unique data causes false failures when one test’s cleanup deletes another test’s records.

Arrange-Act-Assert

Every test follows this exact structure:

// Arrange — set up state
const cookie = await authenticateUser(request);
const dto = generateCreateNoteDto();

// Act — execute the behavior
const response = await executePostNoteRequest(request, dto, cookie);

// Assert — verify the result
expect201Created(response);
expect(response.data.title).toBe(dto.title);

Cleanup in afterEach

// ✓ Correct — runs even when test throws
test.afterEach(async ({ request }) => {
  for (const cookie of cookiesToCleanup) {
    await cleanupNotes(request, cookie);
    await cleanupUser(request, cookie);
  }
  cookiesToCleanup.length = 0;
});

// ✗ Wrong — skipped when test throws before reaching cleanup
test("...", async ({ request }) => {
  const cookie = await authenticateUser(request);
  // ...
  await cleanupUser(request, cookie); // This line may never run
});

No shared state between tests

• No beforeAll for shared data
• No shared auth sessions
• Each test creates its own user and test entities

No static waits

// ✗ Forbidden at all layers
await new Promise(resolve => setTimeout(resolve, 1000));
await page.waitForTimeout(5000);

// ✓ Use explicit conditions instead
// API/Integration: poll with 400ms interval, 15s timeout
// E2E: locator.waitFor(), page.waitForURL(), expect(locator).toBeVisible()

No snapshot tests

.toMatchSnapshot() and .toMatchInlineSnapshot() are forbidden at all layers. Snapshots hide intent and produce opaque diffs.

No hardcoded URLs

Always use relative paths. Playwright’s baseURL from config handles environment switching.

Reference File Guide

When generating tests for a layer, read the corresponding reference file first. Each file includes the full workflow, verbatim templates, and a self-validation checklist.

Cross-Layer Checklist

After all layers are complete, verify the pyramid is correctly shaped:

No layer duplication:

• Unit tests cover Zod field validation exhaustively — integration/API/E2E do not
• Integration tests have only 1–2 validation wiring tests per endpoint
• API tests have only 1 validation test per endpoint
• E2E tests cover user journeys and navigation — no field validation

Pyramid shape (count tests per layer):

• Unit tests are the largest group (60–70% of total)
• Integration tests are the second largest (20–25%)
• API tests are smaller (10–15%)
• E2E tests are the fewest (5–10%)

All self-validation checklists were run:

• Unit layer checklist ✓
• Integration layer checklist ✓
• API layer checklist ✓
• E2E layer checklist ✓

Cross-Layer Examples

The same “notes” domain — showing exactly how coverage is divided, not duplicated.

Testing a title field (min 1, max 255, required)

Testing a notFound service branch

Testing an authorization rule

Completion

After all layers are generated, provide:

File inventory per layer: