cd /news/developer-tools/unit-test-ai-guide-zero-hallucinatio… · home topics developer-tools article
[ARTICLE · art-33873] src=dev.to ↗ pub= topic=developer-tools verified=true sentiment=· neutral

Unit Test AI Guide — Zero Hallucination, Cross-Stack Standard

A developer outlines a deterministic, zero-hallucination approach to AI-generated unit tests across Node.js, React, Python, Angular, and Laravel stacks. The guide mandates a single testing library per stack—Jest for Node.js and Angular, Vitest for React, pytest for Python, and Pest for Laravel—and relies on Cursor's .cursor/rules system to inject persistent, project-scoped instructions into every AI interaction, preventing hallucination at the source.

read14 min views1 publishedJun 19, 2026

Focus:Unit Tests ONLY — no integration, no E2E

Stacks:Node.js (NestJS/Express) · React.js · Python · Angular · Laravel

Goal:AI generates unit tests consistently, deterministically, without hallucination

IDE:Cursor (Primary) + Claude (Secondary)

Do not mix libraries. Pick one per stack, configure it fully, never deviate.

| Stack | Library | Why This One |

|---|---|---|

| Node.js / NestJS / Express | Jest

| Native DI mocking, @nestjs/testing

built around it, widest ecosystem |

| React.js | Vitest

  • @testing-library/react

| Native Vite/ESM support, Jest-compatible API, 3–10x faster |

| Python | pytest

| De facto standard, fixture system eliminates boilerplate, best plugin ecosystem |

| Angular | Jest

(replace Karma) | Karma is deprecated in Angular 17+; Jest is the official migration target |

| Laravel | Pest

| Modern syntax, built on PHPUnit, higher signal-to-noise ratio |

Rule:If someone suggests a second library for the same stack, reject it. One library per stack, configured once, followed always.

| Capability | Cursor | VS Code + Copilot | WebStorm |

|---|---|---|---|

| Project-level AI rules | ✅ .cursor/rules/

| ❌ | ❌ |

| Codebase-aware context | ✅ @codebase

| Partial | Partial |

| Run terminal + read output | ✅ Composer | ❌ | ❌ |

| Multi-file generation | ✅ Agent mode | Limited | ❌ |

| Custom instructions per filetype | ✅ | ❌ | ❌ |

| MCP server integration | ✅ | ❌ | ❌ |

Cursor's .cursor/rules/

system is the only IDE-native mechanism that injects persistent, project-scoped instructions into every AI interaction — this is what prevents hallucination at the source.

project-root/

├── .cursor/

│   └── rules/

│       ├── unit-test-global.mdc       ← applies to all files

│       ├── unit-test-nestjs.mdc       ← applies to *.service.ts, *.guard.ts

│       ├── unit-test-react.mdc        ← applies to *.tsx, *.component.tsx

│       ├── unit-test-python.mdc       ← applies to *.py

│       ├── unit-test-angular.mdc      ← applies to *.component.ts, *.service.ts

│       └── unit-test-laravel.mdc      ← applies to *Service.php, *Model.php

├── CLAUDE.md                          ← Claude project memory file

└── ...

These files are injected into every AI prompt automatically when working on matching files.

File: .cursor/rules/unit-test-global.mdc

---

description: Global unit test rules — applies to all files in this project

globs: ["**/*.spec.ts", "**/*.test.ts", "**/*.spec.tsx", "**/*_test.py", "**/Test*.php"]

alwaysApply: true

---


## What is a unit test here

- Tests ONE class or function in complete isolation

- ALL external dependencies are mocked — no real DB, no real HTTP, no real file system

- Each test runs independently — no shared mutable state between tests

- Runs in < 100ms

## Structure: AAA — Always, Without Exception

Every test MUST have these three sections, with comments:

// Arrange — set up inputs, mocks, expected values

// Act — call the single function/method under test

// Assert — verify exactly one outcome

## Naming Convention — Mandatory

- File: `[module-name].spec.ts` or `[module-name].unit.spec.ts`

- Describe block: exact class or function name

- It/test block: "should [expected behavior] when [specific condition]"

  GOOD: "should throw NotFoundException when user does not exist"

  BAD:  "user not found", "test 1", "works correctly"

## What to test per function — minimum coverage

1. Happy path — valid input → expected output

2. Null/undefined input — how does it fail safely

3. Empty input — empty string, empty array, zero

4. Error path — when dependency throws, what happens

5. Boundary — max length, negative numbers, boolean edge

## What NOT to include in unit tests

- Database queries (mock the repository)

- HTTP calls (mock the service/axios/fetch)

- File system operations (mock fs)

- Timer behavior (use jest.useFakeTimers)

- Random values (mock Math.random or faker with fixed seed)

## AI Instruction

When asked to generate a unit test:

1. READ the function signature and types first

2. IDENTIFY all dependencies (constructor args, imported modules)

3. MOCK every dependency — never use real implementations

4. Generate minimum 4 test cases per method (happy, null, error, edge)

5. NEVER import from test doubles or assume what isn't in the source file

6. If you are unsure of a type — ASK, do not assume

File: .cursor/rules/unit-test-nestjs.mdc

---

description: NestJS unit test rules

globs: ["src/**/*.service.ts", "src/**/*.guard.ts", "src/**/*.interceptor.ts", "src/**/*.pipe.ts"]

---


## Library: Jest + @nestjs/testing + jest-mock-extended

## Always use Test.createTestingModule

typescript

const module = await Test.createTestingModule({

providers: [

SubjectService,

{ provide: DependencyService, useValue: mockDependency }

]

}).compile();

## Mock Pattern — jest-mock-extended

typescript

import { createMock } from '@golevelup/ts-jest';

// OR

import { mock } from 'jest-mock-extended';

const mockUserRepo = mock();

## Never

- Never use `new SubjectService()` directly — always use TestingModule

- Never let `useValue` contain real implementations

- Never test more than one service per describe block

## Repository Mock Template

typescript

const mockRepo = {

findOne: jest.fn(),

save: jest.fn(),

delete: jest.fn(),

findAll: jest.fn(),

update: jest.fn(),

};

## Required imports for every NestJS spec file

typescript

import { Test, TestingModule } from '@nestjs/testing';

import { NotFoundException, BadRequestException } from '@nestjs/common';

markdown

File: .cursor/rules/unit-test-react.mdc

---

description: React unit test rules

globs: ["src/**/*.tsx", "src/**/*.component.tsx", "src/hooks/**/*.ts"]

---


## Library: Vitest + @testing-library/react + @testing-library/user-event

## Required setup in vitest.config.ts

typescript

export default defineConfig({

test: {

environment: 'jsdom',

globals: true,

setupFiles: ['./src/test/setup.ts']

}

})

## setup.ts

typescript

import '@testing-library/jest-dom';

## Component test structure

typescript

import { render, screen } from '@testing-library/react';

import userEvent from '@testing-library/user-event';

import { vi } from 'vitest';

## Mocking rules for React

- Mock ALL hooks that call APIs: `vi.mock('../hooks/useUsers')`

- Mock ALL context providers — wrap with test-specific providers

- Mock router: use `MemoryRouter` from react-router-dom

- NEVER mock internal state (useState) — test behavior, not implementation

## Query priority (RTL best practice — mandatory)

1. getByRole — prefer always

2. getByLabelText — for forms

3. getByText — for content

4. getByTestId — LAST RESORT only, requires data-testid attribute

## User event — always use userEvent, never fireEvent

typescript

const user = userEvent.setup();

await user.click(button);

await user.type(input, 'value');

## Custom hook testing

typescript

import { renderHook, act } from '@testing-library/react';

const { result } = renderHook(() => useMyHook());

act(() => result.current.doSomething());

expect(result.current.value).toBe('expected');

typescript

File: .cursor/rules/unit-test-python.mdc

---

description: Python unit test rules

globs: ["**/*.py", "!**/*migrations*"]

---


## Library: pytest + pytest-mock + factory-boy

## File naming

- Source: `app/services/user_service.py`

- Test:   `tests/unit/test_user_service.py`

- ALWAYS mirror the source directory structure under tests/unit/

## Class under test — always use dependency injection

python

class UserService:

def __init__(self, repo: UserRepository):

    self.repo = repo

class UserService:

def __init__(self):

    self.repo = UserRepository()  # can't mock this
## Mock pattern — pytest-mock (mocker fixture)

python

def test_raises_when_user_not_found(mocker):

mock_repo = mocker.Mock()

mock_repo.find_by_id.return_value = None

service = UserService(mock_repo)

with pytest.raises(UserNotFoundException):

    service.get_user("missing-id")
## Async tests — pytest-asyncio

python

import pytest

@pytest.mark.asyncio

async def test_async_service(mocker):

mock_repo = mocker.AsyncMock()

...
## Fixture pattern — for shared setup

python

@pytest.fixture

def user_service(mocker):

repo = mocker.Mock()

return UserService(repo), repo

def test_get_user_happy_path(user_service):

service, repo = user_service

repo.find_by_id.return_value = User(id="1", email="a@b.com")

result = service.get_user("1")

assert result.email == "a@b.com"
## Naming

- File: `test_[module_name].py`

- Function: `test_[expected_behavior]_when_[condition]`

markdown

File: .cursor/rules/unit-test-angular.mdc

---

description: Angular unit test rules

globs: ["src/app/**/*.component.ts", "src/app/**/*.service.ts", "src/app/**/*.pipe.ts", "src/app/**/*.guard.ts"]

---


## Library: Jest (NOT Karma — Karma is deprecated Angular 17+)

## Migration from Karma to Jest (one-time)

bash

ng add @angular-builders/jest

## TestBed configuration — always minimal

typescript

await TestBed.configureTestingModule({

imports: [ComponentUnderTest], // standalone components

providers: [

{ provide: UserService, useValue: mockUserService }

]

}).compileComponents();

## Service mock pattern

typescript

const mockUserService = {

getUser: jest.fn(),

createUser: jest.fn(),

};

## Component test — check DOM behavior, not internal state

typescript

fixture.detectChanges(); // trigger ngOnInit

const button = fixture.debugElement.query(By.css('[data-testid="submit"]'));

button.nativeElement.click();

fixture.detectChanges();

expect(fixture.debugElement.query(By.css('.error-msg'))).toBeTruthy();

## Pipe test — pure function, no TestBed needed

typescript

it('should transform date correctly', () => {

const pipe = new DateFormatPipe();

expect(pipe.transform(new Date('2024-01-01'))).toBe('Jan 1, 2024');

});

## Guard/resolver test — inject and call directly

typescript

const guard = TestBed.inject(AuthGuard);

const result = await guard.canActivate(mockRoute, mockState);

expect(result).toBe(false);

python

File: .cursor/rules/unit-test-laravel.mdc

---

description: Laravel unit test rules

globs: ["app/Services/**/*.php", "app/Models/**/*.php", "app/Http/Requests/**/*.php", "app/Actions/**/*.php"]

---


## Library: Pest (NOT PHPUnit directly — Pest wraps it with better syntax)

## File location

- Source: `app/Services/UserService.php`

- Test:   `tests/Unit/Services/UserServiceTest.php`

## Class under test — use constructor injection

php

class UserService {

public function __construct(

    private readonly UserRepositoryInterface $repo

) {}

}

## Mock pattern — Mockery (included with Pest/PHPUnit)

php

it('throws exception when user not found', function () {

// Arrange

$repo = Mockery::mock(UserRepositoryInterface::class);

$repo->shouldReceive('findById')->once()->andReturn(null);

$service = new UserService($repo);

// Act & Assert

expect(fn () => $service->getUser('abc'))

    ->toThrow(UserNotFoundException::class);

});

## Data provider pattern (Pest datasets)

php

it('validates email format', function (string $email, bool $valid) {

expect(validateEmail($email))->toBe($valid);

})->with([

['valid@email.com', true],

['notanemail', false],

['', false],

['@nodomain.com', false],

]);

## What NOT to do in unit tests

- Never call $this->get() or $this->post() — that is feature testing

- Never use RefreshDatabase — that is feature/integration testing

- Never resolve from service container — inject directly

markdown

Place this at project root. Claude reads it on every session automatically.

File: CLAUDE.md



## Stack

- Backend: NestJS + TypeScript

- Frontend: React + Vite + TypeScript

- Testing: Jest (NestJS) + Vitest (React) + Pest (Laravel) + pytest (Python)

## Unit Test Rules — NON-NEGOTIABLE

### When asked to generate a unit test:

1. NEVER generate integration or E2E tests unless explicitly asked

2. ALWAYS mock every external dependency

3. ALWAYS follow AAA (Arrange-Act-Assert) with comments

4. ALWAYS generate minimum 4 cases: happy path, null/empty, error thrown, edge case

5. NEVER use `any` type in TypeScript tests

6. NEVER leave `TODO` comments in generated tests

### Naming

- NestJS: `[name].service.spec.ts` inside same directory as source

- React: `[ComponentName].test.tsx` inside `__tests__/` or same directory

- Python: `test_[module].py` under `tests/unit/`

- Laravel: `[Name]Test.php` under `tests/Unit/`

### Test should describe BEHAVIOR, not implementation

GOOD: "should return empty array when no users match search"

BAD: "test getUserByFilter"

### If you are unsure of the correct mock structure:

- Ask what the dependency interface looks like

- Do NOT invent method names that don't exist in the source

### Coverage target: 80% lines, 75% branches per module

## Project Structure Reference

src/

  modules/

    users/

      users.service.ts

      users.service.spec.ts   ← unit test lives here

      users.repository.ts

      users.repository.spec.ts

MCP (Model Context Protocol) servers extend AI capabilities. For unit testing, these are the relevant ones:

Cursor already has filesystem access. No extra MCP needed to read source files and generate tests. The .cursor/rules/

system handles this.

1. @executeautomation/playwright-mcp-server

Scope: E2E only — not relevant for unit tests

Skip for this use case

2. @modelcontextprotocol/server-filesystem

Gives Claude access to project files

Use in Claude Desktop to read source → generate tests

Install: configured in claude_desktop_config.json

3. Custom Testing MCP (Build This — Highest Value)

Build a lightweight MCP server that:

Reads a source file

Extracts function signatures + types

Returns structured JSON to Claude

Claude generates tests from structure, not guesswork

// mcp-test-generator/src/index.ts

import { Server } from '@modelcontextprotocol/sdk/server/index.js';

import { StdioServerTransport } from '@modelcontextprotocol/sdk/server/stdio.js';

import * as ts from 'typescript';

const server = new Server({ name: 'test-generator', version: '1.0.0' }, {

  capabilities: { tools: {} }

});

server.setRequestHandler('tools/call', async (req) => {

  if (req.params.name === 'extract_signatures') {

    const { filePath } = req.params.arguments;

    const signatures = extractFunctionSignatures(filePath); // parse AST

    return { content: [{ type: 'text', text: JSON.stringify(signatures) }] };

  }

});

function extractFunctionSignatures(filePath: string) {

  // Use TypeScript compiler API to extract:

  // - class name

  // - method names

  // - parameter types

  // - return types

  // - injected dependencies (constructor params)

  const program = ts.createProgram([filePath], {});

  const sourceFile = program.getSourceFile(filePath);

  // ... AST traversal

  return { className, methods, dependencies };

}

Why this matters: When Claude receives typed signatures instead of raw source code, it cannot hallucinate — it generates tests from concrete types, not guesses.

// ~/Library/Application Support/Claude/claude_desktop_config.json (macOS)

// %APPDATA%\Claude\claude_desktop_config.json (Windows)

{

  "mcpServers": {

    "filesystem": {

      "command": "npx",

      "args": ["-y", "@modelcontextprotocol/server-filesystem", "/path/to/your/project"]

    },

    "test-generator": {

      "command": "node",

      "args": ["/path/to/your/mcp-test-generator/dist/index.js"]

    }

  }

}
// .cursor/mcp.json (project-level)

{

  "mcpServers": {

    "test-generator": {

      "command": "node",

      "args": ["./tools/mcp-test-generator/dist/index.js"]

    }

  }

}

The root cause of AI hallucination in tests: AI invents method names, wrong mock structures, non-existent imports. These prompts eliminate that.

Generate a unit test file for this [NestJS service / React component / Python class / Angular service / Laravel service].

Rules:

- Library: [Jest / Vitest / pytest / Jest / Pest]

- Test only THIS file — no integration

- Mock ALL dependencies listed in the constructor/props

- Follow AAA pattern with comments

- Minimum 4 test cases per public method

- Use ONLY methods/properties that exist in this source file

- Do NOT import anything not shown in the source

- Naming: "should [behavior] when [condition]"
Read [filename].

Extract:

1. Class/function name

2. All public methods with parameter types and return types

3. All constructor dependencies (for mocking)

Then generate the unit test file following CLAUDE.md rules.

Do NOT assume any method exists unless you can see it in the source.
For each file in [directory], generate a corresponding .spec.ts file.

Process one file at a time.

For each file:

  1. Read the source

  2. List the public methods you found (confirm before generating)

  3. Generate the test file

  4. Show the test file path

Do not proceed to the next file until the current one is confirmed.

All stacks follow the same proximity principle: test file lives next to source file.

src/

  modules/

    users/

      users.service.ts

      users.service.spec.ts        ← unit test

      users.repository.ts

      users.repository.spec.ts     ← unit test

      users.guard.ts

      users.guard.spec.ts          ← unit test
src/

  components/

    UserCard/

      UserCard.tsx

      UserCard.test.tsx            ← unit test

      UserCard.stories.tsx         ← storybook (optional)

  hooks/

    useUserData.ts

    useUserData.test.ts            ← unit test
app/

  services/

    user_service.py

tests/

  unit/

    services/

      test_user_service.py        ← mirrors app/ structure
app/

  Services/

    UserService.php

tests/

  Unit/

    Services/

      UserServiceTest.php         ← mirrors app/ structure

All of this means nothing without enforcement. The pipeline is the final gate.

name: Unit Tests

on: [push, pull_request]

jobs:

  nestjs-unit:

    runs-on: ubuntu-latest

    steps:

      - uses: actions/checkout@v4

      - uses: actions/setup-node@v4

        with: { node-version: '20' }

      - run: npm ci

      - run: npm run test:unit -- --coverage --coverageThreshold='{"global":{"lines":80}}'

  react-unit:

    runs-on: ubuntu-latest

    steps:

      - uses: actions/checkout@v4

      - uses: actions/setup-node@v4

        with: { node-version: '20' }

      - run: npm ci

      - run: npx vitest run --coverage

  python-unit:

    runs-on: ubuntu-latest

    steps:

      - uses: actions/checkout@v4

      - uses: actions/setup-python@v5

        with: { python-version: '3.12' }

      - run: pip install -r requirements-dev.txt

      - run: pytest tests/unit/ --cov=app --cov-fail-under=80

  laravel-unit:

    runs-on: ubuntu-latest

    steps:

      - uses: actions/checkout@v4

      - uses: shivammathur/setup-php@v2

        with: { php-version: '8.3' }

      - run: composer install

      - run: ./vendor/bin/pest --coverage --min=80
// package.json

{

  "lint-staged": {

    "src/**/*.service.ts": ["jest --findRelatedTests --passWithNoTests"],

    "src/**/*.tsx": ["vitest related --run"]

  }

}
□ Install test library (one per stack — see Part 1)

□ Configure jest.config.ts / vitest.config.ts / pytest.ini / pest.php

□ Create .cursor/rules/ directory with all rule files from Part 3

□ Create CLAUDE.md at project root from Part 4

□ Add coverage thresholds to config

□ Add pre-commit hooks (Husky for Node, pre-commit for Python)

□ Add CI workflow from Part 8

□ Create test factory files (one per major entity)

□ Write first failing test → implement → green → commit
□ Add test library without touching existing code

□ Add .cursor/rules/ + CLAUDE.md

□ Run coverage on existing code → document current baseline

□ Pick 3 critical services → generate tests → set as new baseline

□ Add CI with threshold = current baseline (not ideal — current)

□ Raise threshold by 5% per sprint
npm install --save-dev jest ts-jest @types/jest @nestjs/testing jest-mock-extended @golevelup/ts-jest

npx ts-jest config:init
npm install --save-dev vitest @vitest/coverage-v8 @testing-library/react @testing-library/user-event @testing-library/jest-dom jsdom
pip install pytest pytest-cov pytest-mock pytest-asyncio factory-boy
ng add @angular-builders/jest

npm install --save-dev jest jest-preset-angular @types/jest
composer require pestphp/pest pestphp/pest-plugin-laravel --dev

./vendor/bin/pest --init

Last updated: 2025 — verify library major versions before adopting

── more in #developer-tools 4 stories · sorted by recency
── more on @cursor 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/unit-test-ai-guide-z…] indexed:0 read:14min 2026-06-19 ·