Vitest Testing Skill | Agent Skills

Vitest Testing

Vitest is a modern test runner designed for Vite projects. It's fast, ESM-native, and provides a Jest-compatible API with better TypeScript support and instant HMR-powered watch mode.

When to Use This Skill

| Use this skill when... | Use another skill instead when... | |------------------------|----------------------------------| | Setting up or configuring Vitest | Writing E2E browser tests (use playwright-testing) | | Writing unit/integration tests in TS/JS | Testing Python code (use python-testing) | | Migrating from Jest to Vitest | Analyzing test quality (use test-quality-analysis) | | Configuring coverage thresholds | Generating property-based tests (use property-based-testing) | | Using mocks, spies, or fake timers | Validating test effectiveness (use mutation-testing) |

Core Expertise

Vite-native: Reuses Vite config, transforms, and plugins
Fast: Instant feedback with HMR-powered watch mode
Jest-compatible: Drop-in replacement with similar API
TypeScript: First-class TypeScript support
ESM: Native ESM support, no transpilation needed
jsdom DOM tests: and the assertions that silently lie under it (see below)

Installation

bun add --dev vitest
bun add --dev @vitest/coverage-v8      # Coverage (recommended)
bun add --dev happy-dom                # DOM testing (optional)
bunx vitest --version                  # Verify

Configuration (vitest.config.ts)

import { defineConfig } from 'vitest/config';

export default defineConfig({
  test: {
    globals: true,
    environment: 'node',
  },
});

Essential Commands

bunx vitest                            # Watch mode (default)
bunx vitest run                        # Run once (CI mode)
bunx vitest --coverage                 # With coverage
bunx vitest src/utils.test.ts          # Specific file
bunx vitest -t "should add numbers"    # Filter by name
bunx vitest related src/utils.ts       # Related tests
bunx vitest -u                         # Update snapshots
bunx vitest bench                      # Benchmarks
bunx vitest --ui                       # UI mode

Writing Tests

Basic Test Structure

import { describe, it, expect } from 'vitest';
import { add, multiply } from './math';

describe('math utils', () => {
  it('should add two numbers', () => {
    expect(add(2, 3)).toBe(5);
  });

  it('should multiply two numbers', () => {
    expect(multiply(2, 3)).toBe(6);
  });
});

Key Assertions

| Assertion | Description | |-----------|-------------| | toBe(value) | Strict equality | | toEqual(value) | Deep equality | | toStrictEqual(value) | Deep strict equality | | toBeTruthy() / toBeFalsy() | Truthiness | | toBeNull() / toBeUndefined() | Null checks | | toBeGreaterThan(n) / toBeLessThan(n) | Numeric comparison | | toBeCloseTo(n) | Float comparison | | toMatch(regex) / toContain(str) | String matching | | toHaveLength(n) | Array/string length | | toHaveProperty(key) | Object property | | toMatchObject(obj) | Partial object match | | toThrow(msg) | Error throwing |

Async Tests

test('async test', async () => {
  const data = await fetchData();
  expect(data).toBe('expected');
});

test('promise resolves', async () => {
  await expect(fetchData()).resolves.toBe('expected');
});

test('promise rejects', async () => {
  await expect(fetchBadData()).rejects.toThrow('error');
});

DOM tests under jsdom — assertions that lie

environment: 'jsdom' gives you a DOM without a layout engine, and its CSS parser is narrower than a browser's. Several natural-looking assertions are therefore vacuous: they pass against the very bug they were written to catch, and read as coverage so nobody looks again.

| Trap | Why it passes against the bug | Assert instead | |---|---|---| | el.style.overflowY for a style set by a stylesheet | Inline style is empty; the declaration lives in a class rule | getComputedStyle(el).overflowY — jsdom does resolve injected <style> rules | | getComputedStyle(el).width for min() / calc() values | jsdom's parser silently drops the whole declaration, reporting ""/0 either way | the stylesheet source text, or defer to a real browser | | Anything about size or position | getBoundingClientRect() is all zeros; there is no layout | a real-browser tier | | Asserting right after clicking something that renders async | The panel is still empty, so "no bad element found" is trivially true | flush (await new Promise(r => setTimeout(r, 0))), then assert the container is non-empty before the real check |

Also: Element.prototype.scrollIntoView does not exist in jsdom, so any code path that centres an element throws on mount. Stub it (Element.prototype.scrollIntoView = () => {}) — that is a harness gap, not a behaviour change.

And never write a conditional assertion:

// Passes silently in exactly the case it was meant to catch — a renamed class.
if (found.length === 1) expect(found[0].textContent).toMatch(/x/);

// Assert unconditionally.
expect(found).toHaveLength(1);
expect(found[0].textContent).toMatch(/x/);

Make it fail on purpose

A regression test that has never failed has not been shown to test anything. Before trusting one, force red and read the message:

Fix in this package → revert the fix in place, confirm red for the right reason, restore.
Cross-package suite → re-pin the dependency to the release before the fix, confirm red, restore the pin.

Record the observed failure output in the PR body. "It goes red" is a claim; the message is the evidence.

Loading a sibling package's source for a real integration test

When a defect is a property of two packages together, testing each against a stand-in keeps both green while the pair is broken. To load a sibling's real source:

depend on it pinned to a release tag (e.g. a git dependency), so the suite tests a published artifact rather than a moving branch;
add it to server.deps.inline — vitest externalizes node_modules by default and would hand Node raw TypeScript:

export default defineConfig({
  test: { server: { deps: { inline: [/sibling-package/] } } },
});

export the host's real entry point, so the suite drives the actual code path rather than a per-unit seam where the bug cannot appear.

Mocking (Essential Patterns)

import { vi, test, expect } from 'vitest';

// Mock function
const mockFn = vi.fn();
mockFn.mockReturnValue(42);

// Mock module
vi.mock('./api', () => ({
  fetchUser: vi.fn(() => Promise.resolve({ id: 1, name: 'John' })),
}));

// Mock timers
vi.useFakeTimers();
vi.advanceTimersByTime(1000);
vi.restoreAllMocks();

// Spy on method
const spy = vi.spyOn(object, 'method');

Snapshot Testing

test('snapshot test', () => {
  expect(data).toMatchSnapshot();
});

test('inline snapshot', () => {
  expect(result).toMatchInlineSnapshot('5');
});
// Update snapshots: bunx vitest -u

Coverage

bun add --dev @vitest/coverage-v8
bunx vitest --coverage

Key config options: provider, reporter, include, exclude, thresholds.

Agentic Optimizations

| Context | Command | |---------|---------| | Quick test | bunx vitest --reporter=dot --bail=1 | | CI test | bunx vitest run --reporter=junit | | Coverage check | bunx vitest --coverage --reporter=dot | | Single file | bunx vitest run src/utils.test.ts --reporter=dot | | Failed only | bunx vitest --changed --bail=1 |

For detailed examples, advanced patterns, and best practices, see REFERENCE.md.

References

Official docs: https://vitest.dev
Configuration: https://vitest.dev/config/
API reference: https://vitest.dev/api/
Migration from Jest: https://vitest.dev/guide/migration.html
Coverage: https://vitest.dev/guide/coverage.html

Agent Skills: Vitest Testing

Install this agent skill to your local

Skill Files