jellydn/my-ai-tools/tdd

Test-Driven Development (TDD)

Guides you through the complete TDD workflow with Red-Green-Refactor cycle.

Usage

/tdd <ACTION> [ARGUMENTS]

Actions

• start <FEATURE> - Initialize TDD session for a feature
• red <TEST_NAME> - Create failing test (Red phase)
• green - Run tests and implement code (Green phase)
• refactor - Guide refactoring process (Refactor phase)
• cycle <FEATURE> - Run complete Red-Green-Refactor cycle
• watch - Start test watcher for continuous feedback
• status - Show current test status and next steps

TDD Principles

The Red-Green-Refactor Cycle

1. Red: Write a failing test that defines desired behavior
2. Green: Write minimal code to make the test pass
3. Refactor: Improve code quality while keeping tests green

Best Practices

• Write tests first - Tests define the interface and behavior
• Small steps - Make tiny, incremental changes
• Fast feedback - Run tests frequently for immediate validation
• Clean code - Refactor regularly to maintain quality
• One concept per test - Keep tests focused and atomic
• AAA Pattern - Structure tests as Arrange, Act, Assert
• Black-box testing - Test only public methods and behavior, not implementation details
• Test at an agreed seam - Before writing a test, identify the public interface that exposes the behavior and prefer the highest existing seam
• Independent expectations - Expected values come from a literal, worked example, or spec; do not recompute them with the same algorithm as the implementation
• Vertical slices - Write one meaningful test, implement the minimum, then repeat; avoid writing an entire imagined test suite before learning from the first cycle

Anti-Patterns

• Implementation-coupled tests mock private collaborators or inspect internal state. They fail during harmless refactors and provide weak behavioral confidence.
• Tautological tests derive the expected value with the same logic as the code under test, so both can be wrong together.
• Horizontal slicing writes all tests first and all implementation later. Prefer tracer-bullet slices that exercise one user-visible behavior at a time.
• Unconfirmed seams test a low-level helper only because it is easy to reach. If the behavior matters at a higher interface, test there instead or document why the lower seam is the correct one.

Before the first Red phase, discover the repository's test runner and command from its manifest, task runner, docs, or CI. Do not assume TypeScript, Vitest, npm, or pnpm. Record the exact command used for the red failure and why it failed; a syntax/import error is not a valid Red state.

Check that tests can detect mistakes

For risky behavior, name a plausible wrong implementation and choose an input whose expected result differs from it. A test that still passes when the relevant result or side effect is removed does not protect that behavior. Strengthen the observable assertion, not checks of private calls. If no honest automated test is practical, record the missing seam and use a repeatable manual check; do not claim automated regression coverage.

If the project already uses mutation testing, run it on the affected scope when useful. Inspect uncovered paths and surviving mutations against the contract. Add a test for a meaningful missed behavior; do not chase equivalent mutations or a perfect score, install a new tool by default, or delete code solely because coverage did not exercise it.

Process

For "start <FEATURE>":

1. Initialize TDD session for the feature
2. Create or identify target source file
3. Create corresponding test file if it doesn't exist
4. Explain the feature requirements and acceptance criteria

For "red <TEST_NAME>":

1. Create a failing test that describes the desired behavior
2. Ensure test fails for the right reason (not syntax errors)
3. Run tests to confirm red state
4. Explain what the test is validating

For "green":

1. Implement minimal code to make failing tests pass
2. Focus on making tests pass, not perfect code
3. Run tests to confirm green state
4. Write just enough code to pass the test

For "refactor":

1. Improve code quality while maintaining green tests
2. Remove duplication and improve design
3. Run tests continuously during refactoring
4. Make one refactoring change at a time

Test Template

A test template is available at $SKILL_PATH/templates/test-template.md:

import { describe, it, expect } from "vitest";
import { functionName } from "./module";
describe("functionName", () => {
	it("should return formatted output when given valid input", () => {
		// Arrange - Setup test scenario
		const input = "test input";
		const expectedOutput = "expected output";
		// Act - Execute the unit under test
		const result = functionName(input);
		// Assert - Verify expected outcome
		expect(result).toBe(expectedOutput);
	});
});

Common Commands

• Run tests: npm test or pnpm test
• Watch mode: npm test --watch
• With coverage: npm test --coverage

Attribution

Test-strength guidance is adapted from Robert C. Martin's `ub-mode` — TDD, Test Behavior, and Mutator.