test-write

name	test-write
description	Use when backfilling tests on existing code. Detects test framework, writes behavior-focused regression tests via public interfaces. Modes: targeted (files) or diff-based (changed).

Test Write

Generate regression tests for existing code. Detect the test framework, analyze behavior through public interfaces, write tests, and run them.

Modes

Targeted Mode

When the user specifies files or modules:

/test-write src/actions/createOrder.ts
/test-write src/utils/

Analyze the specified code and generate tests for its public interface.

Diff-Based Mode

When no target is specified:

/test-write

Read git diff (staged + unstaged) to find changed files. Generate tests for the modified behaviors.

If the working tree is clean, check the last commit with git diff HEAD~1.

Process

1. Detect Test Framework

These commands run automatically when the skill loads — output replaces each line below:

• Test config files: !ls vitest.config.* jest.config.* pytest.ini pyproject.toml 2>/dev/null || true
• Test directories: !ls -d __tests__ tests test src/__tests__ 2>/dev/null || true
• Test scripts in package.json: !cat package.json 2>/dev/null | grep -E '"test"' || true
• Unstaged changes: !git diff --stat
• Staged changes: !git diff --cached --stat

Look for existing setup in this order:

1. Test config files (vitest.config.*, jest.config.*, pytest.ini, etc.) — see pre-run output
2. package.json test scripts — see pre-run output
3. Existing test files for patterns and conventions
4. If nothing found, see Scaffolding section below

Also detect:

• Where tests live — see pre-run directory output
• Naming conventions used in existing tests
• Import patterns and assertion style

The git diff pre-run feeds Diff-Based Mode — if the user gave no target, use the changed files from the diff output.

2. Analyze the Code

Read the target files and identify:

• Public interfaces — exports, public methods, API endpoints
• Key behaviors — what does this code do from a caller's perspective?
• Inputs and outputs — parameters, return values, thrown errors
• Edge cases — nulls, empty collections, boundary values
• Side effects — if any exist at system boundaries

Do NOT test:

• Private/internal functions
• Implementation details
• Code you'd have to mock extensively to isolate

3. Write Tests

For each identified behavior, write a test that:

• Describes WHAT the code does, not HOW
• Uses the public interface only
• Would survive an internal refactor
• Has a clear, descriptive name

Follow the project's existing conventions for:

• File location and naming
• Import style
• Assertion library
• Test organization (describe blocks, etc.)

If no conventions exist, use sensible defaults for the detected framework.

4. Run Tests

Execute the tests immediately after writing them.

• All pass: Report what was covered
• Some fail: This is a discovered bug, not a test error
  • Report the failure and what it reveals about the code
  • Do NOT silently adjust the test to make it pass
  • Ask the user: is this a bug to fix, or should the test expectation change?

5. Report

Tests written: 5
  src/actions/createOrder.test.ts — 3 tests (validation, calculation, error handling)
  src/utils/formatCurrency.test.ts — 2 tests (formatting, locale handling)
Passed: 5
Failed: 0
Not covered: createOrder webhook side effects (requires external service mock)

Scaffolding

If no test framework is detected:

1. Identify the project language and runtime
2. Suggest a minimal framework (Vitest for Node/TS, pytest for Python, etc.)
3. Install it and create a minimal config
4. Add a test script to package.json (for Node projects)

Ask the user before installing dependencies.

Principles

• Behavior over implementation — test what the code does through its public API
• One behavior per test — each test verifies one thing
• Readable names — test names are documentation: "calculates total from line items"
• No mocking internals — only mock at system boundaries (external APIs, databases)
• Discovered bugs are valuable — a failing test means the skill found something