agentic-dev-template/docs/guides/testing-strategy.md

Testing Strategy

A layered approach: direct factory injection + colocated unit tests + Playwright e2e.

For the how of TDD (red-green-refactor cycle, when to mock, what NOT to test), see tdd-workflow.md. This document covers test placement and infrastructure.

For the full R25 / R26 / R27 / R28 patterns with worked examples, see tdd-workflow.md §4 (mock decision tree).

Related ADRs: ADR-012 — Clean Architecture conformance; ADR-013 — input/output unification + presenter + error middleware.

Test placement

Level	Location	Tool	Example
Unit (colocated)	packages/<feature>/src/entities/models/<entity>.test.ts	Vitest	Schema validation, type guards
Use case	packages/<feature>/src/application/use-cases/<name>.use-case.test.ts	Vitest	Direct factory injection + R25 output-validation
Controller	packages/<feature>/src/interface-adapters/controllers/<name>.controller.test.ts	Vitest	Direct factory injection + R10 input validation + R27/R28 view shape
Repository contract	packages/<feature>/src/infrastructure/repositories/<impl>.repository.test.ts	Vitest + contract suite	Interface conformance
Router (integration)	packages/<feature>/src/integrations/api/router.test.ts	Vitest + container	R26 domain → TRPCError mapping
Feature level	packages/<feature>/tests/<name>.feature.test.ts	Vitest	Cross-layer integration via direct injection
E2E (app)	apps/web-next/e2e/<name>.spec.ts	Playwright	Full user flow across frontend + backend

Colocated vs feature-level: Colocated tests (*.test.ts next to source) test isolated units. Feature-level tests (tests/ folder) wire the full chain via direct injection and test interactions between layers.

Per-feature DI in tests

Default: direct factory injection (use-case + controller tests)

Use cases and controllers are factory functions. Tests construct mock repositories directly and pass them in — no container, no rebinding helpers.

// packages/blog/src/application/use-cases/get-articles.use-case.test.ts
import { describe, it, expect } from "vitest";
import { getArticlesUseCase } from "@/application/use-cases/get-articles.use-case";
import { MockArticlesRepository } from "@/infrastructure/repositories/articles.repository.mock";
import { articleFactory } from "@/__factories__/article.factory";

describe("getArticlesUseCase", () => {
  it("filters by status", async () => {
    const repo = new MockArticlesRepository();
    articleFactory.reset();
    await repo.createArticle(articleFactory.build({ id: "1", status: "draft" }));
    await repo.createArticle(articleFactory.build({ id: "2", status: "published" }));

    const useCase = getArticlesUseCase(repo);   // direct injection
    const result = await useCase({ status: "published" });
    expect(result).toHaveLength(1);
  });
});

Controller tests follow the same pattern — construct the use case with mocks, then construct the controller with that use case:

// packages/auth/src/interface-adapters/controllers/sign-in.controller.test.ts
import { signInController } from "@/interface-adapters/controllers/sign-in.controller";
import { signInUseCase } from "@/application/use-cases/sign-in.use-case";
import { MockUsersRepository } from "@/infrastructure/repositories/users.repository.mock";
import { MockAuthenticationService } from "@/infrastructure/services/authentication.service.mock";

describe("signInController", () => {
  it("returns a cookie on successful sign-in", async () => {
    const users = new MockUsersRepository([]);
    const auth = new MockAuthenticationService(users);
    const useCase = signInUseCase(users, auth);
    const controller = signInController(useCase);   // direct injection

    const result = await controller({ username: "alice", password: "testpassword" });
    expect(result.name).toBe("session");           // presenter view shape (R27)
    expect(result.value).toBeTruthy();
  });
});

There is no getTestContainer() or rebindRepository() helper in use-case or controller tests. Those helpers are no longer needed.

Router-level tests: container rebind

The tRPC router resolves controllers from the feature's DI container (a singleton). Router tests must reload the container state around each test:

// packages/blog/src/integrations/api/router.test.ts
import { beforeEach, afterEach, describe, it, expect } from "vitest";
import { TRPCError } from "@trpc/server";
import { blogContainer } from "@/di/container";
import { BlogModule } from "@/di/module";
import { blogRouter } from "@/integrations/api/router";

describe("blogRouter (R26 error mapping)", () => {
  beforeEach(() => {
    blogContainer.unbindAll();
    blogContainer.load(BlogModule);   // loads default mock bindings
  });

  afterEach(() => {
    blogContainer.unbindAll();
  });

  it("translates ArticleNotFoundError → NOT_FOUND", async () => {
    const caller = blogRouter.createCaller({});
    await expect(caller.articleBySlug({ slug: "missing" }))
      .rejects.toSatisfy((e: unknown) =>
        e instanceof TRPCError && e.code === "NOT_FOUND"
      );
  });
});

When a test needs a specific repo behaviour at the router level, bind it directly:

beforeEach(() => {
  blogContainer.unbindAll();
  blogContainer.bind(BLOG_SYMBOLS.IArticlesRepository).toConstantValue(new AlwaysEmptyRepo());
});

This is the only place container-level rebinding belongs. Use-case and controller tests never touch the container.

Mocking Payload in feature tests

Option 1: Direct factory injection of mock repos (default)

Construct a MockXRepository and pass it directly to the factory:

const repo = new MockArticlesRepository();
// seed data
await repo.createArticle(articleFactory.build({ status: "published" }));

const useCase = getArticlesUseCase(repo);
const result = await useCase({});
expect(result).toHaveLength(1);

The mock repository satisfies the repository interface without touching Payload at all.

Option 2: Router-level — use the feature's DI container

When testing the tRPC router (R26 error-mapping, procedure wiring), bind a mock or stub repository through the feature container in beforeEach:

beforeEach(() => {
  blogContainer.unbindAll();
  blogContainer.bind(BLOG_SYMBOLS.IArticlesRepository)
    .toConstantValue(new MockArticlesRepository());
});

For infrastructure tests that exercise the Payload repository implementation directly, mock the payload module:

import { vi } from "vitest";
vi.mock("payload", () => ({ getPayload: vi.fn() }));

Then provide a stub via stubPayloadConfig from @repo/core-testing/payload (see tdd-workflow.md §4 for the full contract-suite pattern).

Test obligations per layer (Plan 9)

Layer	Test type	Required by spec	Example
Entity	Schema validation	—	articleSchema.safeParse(...) → accepts valid / rejects invalid
Use case (input)	Behavior	—	factory injection; assert result shape and filtering
Use case (output, R25)	Runtime guarantee	R25	inject malformed mock output → .rejects.toBeInstanceOf(ZodError)
Controller (input, R10)	Validation	R10	controller({} as unknown) → .rejects.toBeInstanceOf(InputParseError)
Controller (presenter, R27/R28)	View shape	R27/R28 (when reshaping)	expect(result.name).toBe("session") (not result.session)
Repository contract	Interface conformance	—	run defineContractSuite against mock + real impl
Router (R26)	Domain → TRPCError	R26	xRouter.createCaller({}).x(...) → assert TRPCError.code
Feature-level (tests/)	Cross-layer integration	—	wire the chain via direct injection (no container)
E2E	Full user flow	—	Playwright

R25 — Every non-void use case ends with xOutputSchema.parse(result). The R25 test proves this: inject a mock that returns a structurally invalid object and assert ZodError propagates.

R26 — Every feature has procedures.ts with a defineErrorMiddleware error map. The R26 test calls the tRPC procedure through router.createCaller({}) and asserts the correct TRPCError.code (e.g. NOT_FOUND, BAD_REQUEST, UNAUTHORIZED).

R27/R28 — Controllers that reshape the use-case output (e.g. auth controllers that return a cookie instead of the full session object) must have a test asserting the view shape — not the raw use-case output.

Vitest setup per package

Each feature package has vitest.config.ts:

import { defineConfig } from "vitest/config";
import path from "path";

export default defineConfig({
  test: {
    environment: "node",
    globals: true,
    include: ["src/**/*.test.ts", "tests/**/*.test.ts"],
    setupFiles: [],
  },
  resolve: {
    alias: {
      "@": path.resolve(__dirname, "./src"),
    },
  },
});

The @/ alias resolves to src/ — use it in test files only to import from the feature: import { getArticlesUseCase } from "@/application/use-cases/get-articles.use-case". Source files (src/) must use relative imports (../../repositories/...), not the @/ alias.

Also set "rootDir": "." in the package's tsconfig.json so TypeScript finds both src/ and test files that sit at the package root.

Playwright setup (apps)

Each app has playwright.config.ts:

import { defineConfig, devices } from "@playwright/test";

export default defineConfig({
  testDir: "./e2e",
  fullyParallel: true,
  forbidOnly: !!process.env.CI,
  retries: process.env.CI ? 2 : 0,
  workers: process.env.CI ? 1 : undefined,
  reporter: "list",
  use: {
    baseURL: "http://localhost:3000",
    trace: "on-first-retry",
  },
  projects: [
    {
      name: "chromium",
      use: { ...devices["Desktop Chrome"] },
    },
  ],
  webServer: {
    command: "pnpm dev",
    url: "http://localhost:3000",
    reuseExistingServer: !process.env.CI,
    timeout: 60_000,
  },
});

The webServer block auto-starts the dev server before tests. Set reuseExistingServer: true locally to reuse a running dev server; CI always starts fresh.

Smoke spec example

apps/web-next/e2e/home.spec.ts:

import { test, expect } from "@playwright/test";

test("home page renders site name + nav", async ({ page }) => {
  await page.goto("/");
  await expect(page.locator("h1").first()).toBeVisible();
  await expect(page.locator("nav a").first()).toBeVisible();
});

Running tests

# All tests (unit + feature + e2e)
pnpm test

# Just unit tests
pnpm test --filter "@repo/blog" -- src/

# Just feature tests
pnpm test --filter "@repo/blog" -- tests/

# Just e2e
pnpm test:e2e

# E2E with UI
pnpm test:e2e -- --ui

CI integration

Root package.json:

{
  "scripts": {
    "test": "turbo run test",
    "test:e2e": "turbo run test:e2e"
  }
}

Root turbo.json:

{
  "tasks": {
    "test": {
      "outputs": ["coverage/**"],
      "cache": false
    },
    "test:e2e": {
      "dependsOn": ["^build"],
      "cache": false
    }
  }
}

Key principles

1. Colocated unit tests validate single functions/classes in isolation
2. Feature-level tests exercise the full feature with mocked repos (direct injection, no container)
3. E2E tests prove the app works end-to-end (minimal smoke specs initially)
4. Per-feature containers are used at the router level; use-case + controller tests inject mocks directly into the factory (no container)

R49 / R50 — Instrumentation testing (Plan 10)

R49 — No real Sentry in tests. The core-testing/setup/no-sentry.ts guard mocks @sentry/nextjs, @sentry/node, and @sentry/react at the module level, so any code that imports them gets a no-op surface during vitest runs. Tests that want to assert specific Sentry SDK calls add their own vi.mock(...) per file.

R50 — Repository contracts assert span shape. Every __contracts__/<x>-repository.contract.ts includes a span emission (R50) describe block enumerating one assertion per public method. Suites run against both mock and real (Payload-backed) implementations, ensuring span emission stays in sync. Wire the recording tracer at the call site:

const tracer = new RecordingTracer();
articlesRepositoryContract.run(
  () => new MockArticlesRepository(tracer),
  { tracer: () => tracer },
);

Capture vs span assertions:

• RecordingTracer.spans — every span emitted with { name, op, attributes, status, durationMs }.
• RecordingLogger.captures — every captureException / captureMessage call.
• RecordingLogger.breadcrumbs — every breadcrumb added.
• RecordingLogger.users — every setUser call (history).

Test cleanup: call tracer.reset() and logger.reset() in beforeEach if the test creates one shared instance across multiple cases. 5. Mock repos are the default; only use real Payload in dedicated infrastructure tests