Typescript Test Writing

Writing TypeScript Tests in llxprt-code

Distilled from dev-docs/RULES.md. When in doubt, read RULES.md in full — it is the source of truth for development guidelines.

Core principle: TDD is mandatory

Every line of production code must be written in response to a failing test. No exceptions.

Red-Green-Refactor, followed strictly:

1. RED: Write a failing test for the next small behavior.
2. GREEN: Write ONLY enough code to make the test pass.
3. REFACTOR: Only if it improves clarity.
4. COMMIT: Feature + tests together; refactoring separately.

Stack rules (non-negotiable)

• Bun + bun:test ONLY. Never create Vitest or Node test suites, and never add new .js files — everything is TypeScript run with bun.

• TypeScript strict mode: no any (use unknown with type guards), no type assertions (use type predicates), explicit return types.

• Import pattern (bun:test re-exports vi, mock, and Mock):

  import {
    describe,
    it,
    expect,
    beforeEach,
    afterEach,
    vi,
    type Mock,
  } from 'bun:test';
  

• Prefer immutable data in tests and fixtures ({ ...cart, items: [...cart.items, item] }); never mutate shared fixtures between tests.

Test behavior, not implementation

✅ Test:

• Public API behavior
• Input → Output transformations
• Edge cases and error conditions
• Integration between units
• Schema validation

❌ Never test:

• Implementation details
• Private methods
• Third-party libraries
• Mock interactions

❌ Never enshrine bugs as specification: do not write a passing test that asserts incorrect behavior, even if that is what the code currently does. If you discover a bug while writing tests: (1) file an issue or ask the user, (2) write a failing test that asserts the CORRECT behavior, (3) fix the production code so the test passes. A suite that encodes bugs as passing tests is worse than no tests — it actively prevents future fixes.

Test structure

• describe: feature/component name.
• it: specific behavior in plain English.
• Arrange-Act-Assert: clear sections; one behavior per test.
• DRY setup: never copy-paste identical beforeEach/afterEach boilerplate (e.g. temp-dir creation, cleanup) across describe blocks. Extract a shared helper that wires the lifecycle hooks (e.g. a useTempDir() helper that registers beforeEach/afterEach internally and returns a lazy accessor). Repeating 5+ lines of identical setup in N describe blocks is a maintenance hazard; one line of shared setup per describe block is the target.

File conventions

• Tests are TypeScript, co-located with the code under test.
• *.test.ts is the dominant pattern in this repo; *.spec.ts and __tests__/ directories are also in use — match the existing pattern of the neighboring tests before creating a new file.
• File names: kebab-case.
• Run a single file with bun test path/to/file.test.ts; the full suite is npm run test.

Mock hygiene: no mock theater

The fundamental rule: you cannot test a component by mocking that component.

Mock decision tree

Is it the component you're testing?
├─ Yes → [ERROR] NEVER MOCK IT
└─ No → Is it doing the core work being tested?
    ├─ Yes → [ERROR] DON'T MOCK IT
    └─ No → Is it infrastructure (FS, network, DB)?
        ├─ Yes → [OK] OK to mock
        └─ No → Is it completely unrelated to the test?
            ├─ Yes → [OK] OK to mock
            └─ No → WARNING: Probably shouldn't mock

Prohibited mock patterns

1. Self-mocking — mocking the component under test. If you mock EmojiFilter to test EmojiFilter, you are testing the mock, not the component.
2. Direct-value mock — a mock that returns exactly the output the test expects. The test is worthless: it can never fail for the right reason.
3. Mock verification — asserting a mock was called (expect(mockFn).toHaveBeenCalledWith(...)) proves nothing about real code.
4. Mirror-echo assertion — configuring a stub with a literal and asserting that same literal back (.mockReturnValue('JIT_MARKER') … expect(policy.marker).toBe('JIT_MARKER')). Unless producing that literal IS the behavior under test, the test asserts its own configuration and can never fail for the right reason. Assert a derived property instead — ordering, precedence, counts, aggregation — with literals on the input side only.

Also watch for: the mock chain (A calls MockB calls MockC — no real code tested) and mocks with complex implementations (if the mock has the logic, test the real code instead).

Allowed mock patterns

1. Infrastructure mocking — mock filesystem, network, databases (not business logic); instantiate the REAL component; assert real transformations through the real code paths.
2. Irrelevant service mocking — mock services unrelated to the behavior under test (auth, logging), while the component under test is real.
3. Test data builders — build input data; never mock behavior.

The litmus test

After writing a test, all four answers must be the "good" one:

1. If I delete the real implementation, will this test fail? (If NO: worthless)
2. If I break the real implementation, will this test catch it? (If NO: worthless)
3. Am I testing my mock or my code? (If MOCK: worthless)
4. Could I replace the component with return 'expected' and pass? (If YES: worthless)

Before submitting

• All tests pass (npm run test).
• No TypeScript errors (npm run typecheck).
• No linting warnings (npm run lint).
• No console.logs or debug code left behind.
• Every production line you touched is covered by behavior tests.
• Self-check files you touched with the AST test-audit scanner: bun scripts/test-audit/scan.ts — no new MOCK_MIRROR, ALWAYS_TRUE, SELF_CONFIRMING, or NO_ASSERT findings may appear on them. The scanner runs over the full corpus and writes findings.tsv + file-stats.tsv to the output dir (defaults to tmp/test-audit). To compare against the baseline, run it once on main with a custom output dir, then on your branch with a different output dir, and diff the TSVs: bun scripts/test-audit/scan.ts tmp/scan-main vs bun scripts/test-audit/scan.ts tmp/scan-branch, then diff tmp/scan-main/findings.tsv tmp/scan-branch/findings.tsv. Note: the scanner compares exact stable literals (not substrings) between mock configurations and assertion arguments. A MOCK_MIRROR finding on a test that verifies a transformation (e.g., asserting 'A B' where the stub returns 'A' and 'B' separately) is a false positive — the assertion literal does not exactly match any single mock literal. Inspect the exact positive equality that triggered the finding before dismissing it. Disjoint-source and leakage assertions (e.g., not.toContain('X')) are not mirror echoes and will not trigger MOCK_MIRROR.