Iron Law TDD

Core Principle

Tests MUST be written first. Implementation MUST NOT proceed without failing tests.

This skill enforces Test-Driven Development as a non-negotiable practice during all code implementation. The Iron Law exists because AI coding tools can easily generate code that "should work" but doesn't handle edge cases or integration complexities.

When to Use

Apply this skill during:

• Feature implementation
• Bug fixes
• Refactoring
• API development
• Database operations
• Any code changes that affect behavior

The 11-Step Cycle

Follow this cycle for every implementation task, no exceptions:

1. Write the Test First

Before any implementation code, write a test that:

• Defines the desired behavior clearly
• Will fail initially (proves test is actually running)
• Tests the interface, not implementation details
• Covers the happy path first

2. Run the Test - Watch It Fail

Execute the test suite and verify:

• The new test fails for the right reason
• Failure message is clear and actionable
• No other tests are broken

Why this matters: Confirms the test is actually running and detecting the missing functionality.

3. Write Minimal Implementation

Write only enough code to make the failing test pass:

• Implement the simplest solution
• Do not add features not covered by tests
• Avoid premature optimization
• Focus on making the test green

4. Run Tests - Achieve Green

Execute the full test suite:

• The new test must pass
• All existing tests must still pass
• No tests skipped or ignored

5. Add Edge Case Tests

Now expand test coverage for:

• Error conditions and exceptions
• Boundary values
• Null/undefined inputs
• Empty collections
• Invalid states

6. Update Implementation for Edge Cases

Modify implementation to handle all edge cases:

• Add validation
• Implement error handling
• Handle boundary conditions
• Ensure robustness

7. Run Full Test Suite

Execute complete test suite including:

• Unit tests
• Integration tests (if applicable)
• All tests must pass

8. Refactor (Optional)

If code is working but not clean:

• Improve readability
• Extract functions/methods
• Remove duplication
• Simplify complex logic

Important: Run tests after each refactoring step.

9. Verify Test Coverage

Check that tests cover:

• All code paths
• Error handling
• Edge cases
• Integration points

Target coverage: 80% minimum, 95% for critical code.

10. Review Test Quality

Ensure tests are:

• Independent (no shared state)
• Deterministic (same result every time)
• Fast (< 1 second per unit test)
• Clear (test name describes what's tested)
• Maintainable (easy to update)

11. Documentation Check

Update documentation if:

• Public API changed
• Behavior changed
• New features added
• Configuration options added

Rejected Rationalizations

Common excuses and why they're wrong:

"It's too simple to test"

❌ Excuse: "This function is so simple, it doesn't need a test." ✅ Reality: Simple functions break too. Write the test. It takes 2 minutes.

"I'll add tests later"

❌ Excuse: "Let me get it working first, then I'll add tests." ✅ Reality: "Later" never comes. Tests written after implementation are biased by the implementation and miss edge cases.

"The tests are slowing me down"

❌ Excuse: "Writing tests first is taking too long." ✅ Reality: Tests save time by catching bugs immediately instead of during debugging sessions or production incidents.

"This is just a prototype"

❌ Excuse: "This is prototype code, tests aren't necessary yet." ✅ Reality: Prototypes become production code. Either write tests now or rewrite it later with tests.

"I'm just trying something quickly"

❌ Excuse: "Let me experiment without tests to see if this approach works." ✅ Reality: Acceptable for throwaway spike code that won't be committed. Any code that stays must have tests.

"The code is obviously correct"

❌ Excuse: "This logic is straightforward, clearly correct by inspection." ✅ Reality: "Obviously correct" code has bugs. Integration with the rest of the system creates unexpected interactions.

Verification Commands

Use these commands to verify TDD compliance:

Run Tests

# Node/JavaScript
npm test
npm run test:coverage

# Python
pytest
pytest --cov=src

# Go
go test ./...
go test -cover ./...

# Rust
cargo test
cargo test --all-features

Check Coverage

# Node/JavaScript
npm run coverage
npx jest --coverage

# Python
pytest --cov=src --cov-report=html

# Go
go test -coverprofile=coverage.out ./...
go tool cover -html=coverage.out

# Rust
cargo tarpaulin --out Html

Watch Mode (Continuous Testing)

# Node/JavaScript
npm run test:watch
npx jest --watch

# Python
pytest-watch

# Rust
cargo watch -x test

Test Organization

File Structure

src/
├── feature.ts          # Implementation
└── feature.test.ts     # Tests

# Or separate directories
src/
tests/

Test Naming

// Good: Descriptive test names
describe('UserAuthentication', () => {
  it('should return token when credentials are valid', () => {})
  it('should throw error when password is incorrect', () => {})
  it('should handle missing username gracefully', () => {})
})

// Bad: Vague test names
describe('auth', () => {
  it('works', () => {})
  it('fails', () => {})
})

Test Structure (Arrange-Act-Assert)

it('should calculate discount correctly', () => {
  // Arrange: Set up test data
  const price = 100
  const discountPercent = 20

  // Act: Execute the function
  const result = calculateDiscount(price, discountPercent)

  // Assert: Verify the result
  expect(result).toBe(80)
})

What Makes a Good Test

Characteristics of Good Tests

FIRST principles:

• Fast: Tests run quickly (unit tests < 1 second)
• Independent: Tests don't depend on each other
• Repeatable: Same result every run
• Self-validating: Clear pass/fail (no manual inspection)
• Timely: Written before implementation code

Test Coverage Goals

Minimum coverage: 80% of lines Production code: 90-95% of lines Critical paths: 100% coverage

What to test:

• Public API/interfaces
• Business logic
• Error handling
• Edge cases and boundaries
• Integration points

What NOT to test:

• Private implementation details
• External libraries (trust their tests)
• Trivial getters/setters
• Framework boilerplate

Common Pitfalls

Testing Implementation Details

❌ Bad:

// Test checks internal variable
expect(service._internalCache).toHaveLength(1)

✅ Good:

// Test checks observable behavior
expect(service.getUser(id)).toEqual(expectedUser)

Tests That Don't Test

❌ Bad:

it('should work', () => {
  const result = doSomething()
  // No assertion!
})

✅ Good:

it('should return calculated total', () => {
  const result = calculateTotal([10, 20, 30])
  expect(result).toBe(60)
})

Shared State Between Tests

❌ Bad:

let sharedUser // State shared between tests
it('creates user', () => {
  sharedUser = createUser()
})
it('updates user', () => {
  updateUser(sharedUser) // Depends on previous test
})

✅ Good:

it('creates user', () => {
  const user = createUser()
  expect(user).toBeDefined()
})
it('updates user', () => {
  const user = createUser() // Independent setup
  updateUser(user)
  expect(user.updated).toBe(true)
})

Integration with Workflow

This TDD skill integrates with the overall adaptive workflow:

Phase 4: Implementation

• Step 4A: Write tests FIRST (this skill enforces this)
• Step 4B: Implement to make tests pass
• Step 4C: Parallel execution with TDD in each worktree
• Step 4.5: Verification requires all tests passing

Phase 4.5: Verification

• Evidence-based verification skill checks test results
• No "should work" claims accepted
• Tests must actually run and pass

Phase 6: Knowledge Capture

• Patterns in test organization captured in solutions/
• Effective testing approaches stored in learned/effective-prompts/

Language-Specific Guidance

JavaScript/TypeScript

Frameworks: Jest, Vitest, Mocha + Chai Coverage: nyc, c8, Jest built-in Watch mode: npm run test:watch

Python

Frameworks: pytest, unittest Coverage: pytest-cov, coverage.py Watch mode: pytest-watch

Go

Framework: Built-in testing package Coverage: go test -cover Table tests: Use for multiple test cases

Rust

Framework: Built-in test framework Coverage: cargo-tarpaulin Doc tests: Test code examples in documentation

Enforcement

This skill is enforced through:

1. Phase 4A mandate: Tests written before implementation
2. Verification gate: All tests must pass before proceeding
3. Review criteria: Test quality checked in code review
4. CI/CD gates: Tests run automatically, block merge if failing
5. Metrics tracking: Test coverage tracked in session metrics

Exceptions

The ONLY acceptable exceptions to TDD:

1. Spike code: Exploratory code that will be thrown away
2. External constraints: Third-party code without test infrastructure
3. User explicitly opts out: User acknowledges risk and accepts reduced quality

All exceptions must be:

• Explicitly documented
• Time-limited
• Converted to tested code before production

Summary

Remember:

• Tests first, always
• Watch tests fail, then pass
• Cover edge cases
• Run full suite
• No rationalizations accepted
• TDD saves time by preventing bugs

The Iron Law exists because it works. Follow it.

References

• .mycelium/ directory structure
• Session state docs