accelint-ts-testing（加速 ts 测试）

Vitest Best Practices

Comprehensive patterns for writing maintainable, effective vitest tests. Focused on expert-level guidance for test organization, clarity, and performance.

NEVER Do When Writing Vitest Tests

• NEVER write tests for files with no behavior - Constants files (just export const X = value), type definition files, GLSL uniform declarations, and pure data files contain no logic to test. Testing expect(MY_CONSTANT).toBe(42) verifies nothing: if the value changes, the test changes with it, providing zero protection. These "tests" waste CI time and create maintenance burden when values change. Test behavior (functions, logic, transformations), not data declarations. If a file exports only types, constants, or data structures with no functions or logic, skip testing it entirely.
• NEVER skip global mock cleanup configuration - Manual cleanup appears safe but creates "action at a distance" failures: a mock in test file A leaks into test file B running 3 files later, causing non-deterministic failures that only appear when tests run in specific orders. These Heisenbugs waste hours in CI debugging. Configure clearMocks: true, mockReset: true, restoreMocks: true in vitest.config.ts once to eliminate this entire class of order-dependent failure.
• NEVER nest describe blocks more than 2 levels deep - Deep nesting creates cognitive overhead and excessive indentation. Put context in test names instead: it('should add item to empty cart') vs describe('when cart is empty', () => describe('addItem',...)).
• NEVER write test descriptions that don't read as sentences - Test descriptions must complete the sentence "it..." in lowercase. Write it('should add item to cart') not it('Add item to cart') or it('It should add item to cart'). The description reads as a sentence when prefixed with "it": "it should add item to cart". Capitalized starts, non-sentence formats like it('addToCart test'), or redundant "It should" break readability and test output consistency. Example-based tests use it('should...') while property-based tests use it('property:...') format.
• NEVER test library internals that the library already tests - Testing expect(array.map(fn)).toEqual(expected) wastes time verifying that Array.prototype.map works correctly. The JavaScript/TypeScript standard library and established third-party libraries are already well-tested. Focus tests on your business logic, not on proving that lodash, React, or the language itself works. If you find yourself testing "does this library function do what it claims?", you're testing the wrong layer. Test how your code uses libraries, not whether libraries work.
• NEVER export internal functions just to test them - Tests should verify behavior through the public API, not reach into implementation details. Exporting private helpers, internal utilities, or implementation functions solely to enable testing is a code smell that indicates either: (1) the public API is insufficient for testing the behavior, or (2) the tests are verifying implementation details instead of behavior. If internal logic is complex enough to warrant dedicated testing, extract it into a separate module with its own public API and test file. Private functions get tested indirectly through the public functions that call them.
• NEVER mock your own pure functions - Mocking internal code makes tests brittle and less valuable. Mock only external dependencies (APIs, databases, third-party libraries). Prefer fakes > stubs > spies > mocks.
• NEVER use loose assertions like toBeTruthy() or toBeDefined() - These assertions pass for multiple distinct values you never intended: toBeTruthy() passes for 1, "false", [], and {} - all semantically different. When refactoring changes getUser() from returning {id: 1} to returning 1, your test still passes but your production code breaks. Loose assertions create false confidence that evaporates in production. toBeTypeOf() is NOT a loose assertion.
• NEVER test implementation details instead of behavior - Tests that verify "function X was called 3 times" create false failures: you optimize code to call X once via memoization, all tests fail, yet the user experience is identical (and faster). These tests actively punish performance improvements and refactoring. Test what users observe (outputs given inputs), not how your code achieves it internally.
• NEVER share mutable state between tests - Tests that depend on execution order or previous test state create flaky, unreliable suites. Each test must be fully independent with fresh setup.
• NEVER use any or skip type checking in test files - When implementation signatures change, tests with as any silently pass while calling functions with wrong arguments. You ship broken code that TypeScript could have caught. Tests are executable documentation: user as any communicates nothing, but createTestUser(Partial<User>) shows exactly what properties matter for this test case.
• NEVER mark test files as complete without running TypeScript type checking - Test files are typically excluded from tsconfig.json compilation paths, so running tsc at the project root won't catch type errors in tests. Type errors in tests cause runtime failures, incorrect test behavior, and false confidence from tests that don't test what they claim. Before marking any test file as "done", you MUST run tsc --noEmit directly against the test file using the project's package manager (npm/pnpm/bun/yarn). For monorepos, cd into the specific package directory first, then run type checking. Fix all type errors before proceeding - never use as any or @ts-ignore to bypass errors.
• NEVER assume TypeScript types prevent runtime errors - TS types are compile-time only and vanish at runtime. Testing only "type-valid" inputs creates a false sense of security. In production, functions receive invalid data from JSON APIs without validation, JSON.parse() results, external libraries, user input, and database records. A function typed as process(data: ValidData) can still receive null, undefined, or malformed objects at runtime. Test defensive programming scenarios: pass null to non-nullable parameters, undefined to required fields, malformed objects to typed parameters. These "type-invalid" tests catch real bugs that TypeScript cannot prevent.
• NEVER write weak properties when stronger ones exist - Property-based tests that only verify "no exception thrown" or "returns a value" provide minimal coverage. When testing encode/decode pairs, verify roundtrip equality (decode(encode(x)) === x), not just that decode succeeds. When testing normalization, verify idempotence (normalize(normalize(x)) === normalize(x)), not just that it returns a string. Weak properties give false confidence: they pass but don't actually validate correctness.

Before Writing Tests, Ask

Apply these expert thinking patterns before implementing tests:

Should This File Be Tested?

• Does this file contain behavior to test? Files that only declare constants, types, or data structures without logic don't need tests. Constants files (export const X = 42), type definition files (type User = {...}), GLSL uniform declarations, configuration objects, and pure data files have no behavior to verify. If the file contains no functions, no logic, no transformations - skip testing it. Test behavior, not data.

Test Isolation and Setup

• Where should cleanup logic live? Think in layers: configuration eliminates entire error classes (mock cleanup in vitest.config.ts), setup files handle project-wide concerns (custom matchers, global mocks), beforeEach handles test-specific state. Each test doing its own mock cleanup is like each function doing its own null checks - it works but misses the point. Push concerns to the highest appropriate layer.
• Does this test depend on previous tests or shared state? Test suites are parallel universes - each test should work identically whether it runs first, last, or alone. State dependency creates "quantum tests" that pass or fail based on execution order. If a test needs data from another test, they're actually one test split artificially.

What to Test

• Am I testing behavior or implementation? Test what users experience (inputs → outputs), not how code achieves it (which functions were called). Implementation tests break during safe refactoring.
• What's the simplest dependency I can use? Real implementation > fake > stub > spy > mock. Each step down this hierarchy adds brittleness. Mock only when using real code is impractical (external APIs, slow operations).

Test Clarity

• Can someone understand this test in 5 seconds? Follow AAA pattern (Arrange, Act, Assert) with clear boundaries. If setup is complex, extract to helper functions with descriptive names.
• Are there multiple variations of the same behavior? Use it.each() for parameterized tests instead of copying test structure. One assertion per concept keeps tests focused.

Performance and Maintenance

• Will this test still be valuable in 6 months? Avoid testing framework internals or trivial operations. Focus on business logic, edge cases, and error handling that actually prevent bugs.
• Is this test fast enough to run on every save? Avoid expensive operations in tests. Use fakes for databases, mock timers for delays, stub external calls. Tests should complete in milliseconds.

What This Skill Covers

Expert guidance on vitest testing patterns:

1. Organization - File placement, naming, describe block structure
2. AAA Pattern - Arrange, Act, Assert for instant clarity
3. Parameterized Tests - Using it.each() to reduce duplication
4. Error Handling - Testing exceptions, edge cases, fault injection
5. Assertions - Strict assertions to catch unintended values
6. Test Doubles - Fakes, stubs, mocks, spies hierarchy and when to use each
7. Async Testing - Promises, async/await, timers, concurrent tests
8. Performance - Fast tests through efficient setup and global config
9. Vitest Features - Coverage, watch mode, setup files, config discovery
10. Snapshot Testing - When snapshots help vs hurt maintainability
11. Property-Based Testing - Using fast-check for stronger coverage with generated inputs

How to Use

This skill uses a progressive disclosure structure to minimize context usage:

1. Start with the Overview (AGENTS.md)

Read AGENTS.md for a concise overview of all rules with one-line summaries and the workflow for discovering existing test configuration.

2. Check for Existing Test Configuration

Before writing tests:

• First check vitest.config.ts for global mock cleanup settings (clearMocks, mockReset, restoreMocks)
• Then search for setup files (test/setup.ts, vitest.setup.ts, etc.) and analyze their configuration
• See the workflow in AGENTS.md

3. Load Specific Rules as Needed

Use these explicit triggers to know when to load each reference file:

MANDATORY Loading (load entire file):

• Writing async tests with promises/timers → async-testing.md
• Working with mocks, stubs, spies, or fakes → test-doubles.md
• Auditing/reviewing existing test files → property-based-testing.md (to identify PBT opportunities)

Load When You See These Patterns:

• Nested describe blocks >2 levels deep → organization.md
• Test files not co-located with implementation → organization.md
• Tests without clear Arrange/Act/Assert structure → aaa-pattern.md
• Duplicate test code with slight variations → parameterized-tests.md
• Missing error case tests or inadequate edge case coverage → error-handling.md
• Loose assertions like toBeTruthy() or toBeDefined() → assertions.md
• Tests running slow (>100ms per test) → performance.md
• Need coverage, watch mode, or vitest-specific features → vitest-features.md
• Considering or reviewing snapshot tests → snapshot-testing.md
• Encode/decode pairs, validators, normalizers, or pure functions → property-based-testing.md
• Code with invariants, mathematical properties, or data transformations → property-based-testing.md
• Existing fast-check or property-based tests → property-based-testing.md

Do NOT Load Unless Specifically Needed:

• Do NOT load performance.md if tests are fast (<50ms)
• Do NOT load snapshot-testing.md unless snapshots are mentioned
• Do NOT load vitest-features.md for basic test writing

4. Apply the Pattern

Each reference file contains:

• ❌ Incorrect examples showing the anti-pattern
• ✅ Correct examples showing the optimal implementation
• Explanations of why the pattern matters

5. Use the Report Template

When this skill is invoked for test code review, use the standardized report format:

Template: assets/output-report-template.md

The report format provides:

• Executive Summary with test quality impact assessment
• Severity levels (Critical, High, Medium, Low) for prioritization
• Impact analysis (test reliability, maintainability, performance, clarity)
• Categorization (Test Organization, Assertions, Test Doubles, Async Testing, Performance)
• Pattern references linking to detailed guidance in references/
• Summary table for tracking all issues

When to use the report template:

• Skill invoked directly via /accelint-ts-testing <path>
• User asks to "review test code" or "audit tests" across file(s), invoking skill implicitly

When NOT to use the report template:

• User asks to "write a test for this function" (direct implementation)
• User asks "what's wrong with this test?" (answer the question)
• User requests specific test fixes (apply fixes directly without formal report)

IMPORTANT: When auditing tests, ALWAYS check for property-based testing opportunities

• Load property-based-testing.md during every audit
• Follow the "Workflow: Test Code Review/Audit" in AGENTS.md
• Check for high-value PBT patterns: encode/decode pairs, normalizers, validators, pure functions, sorting functions
• Include PBT opportunities in the audit report even if no other issues are found

Quick Example

See quick-start.md for a complete before/after example showing how this skill transforms unclear tests into clear, maintainable ones.