test-driven-development

name	test-driven-development
description	Enforces RED-GREEN-REFACTOR test-driven development. Use when implementing features, fixing bugs, or writing any code. Triggers on phrases like "implement this", "add a feature", "write code for", "fix this bug", "TDD", "write tests first". Never write implementation code before a failing test exists — always enforce this order regardless of how the request is phrased.

Test-Driven Development

Announce: "I'm using the test-driven-development skill. I'll write a failing test before any implementation."

The Cycle

RED → write failing test → verify it fails for the RIGHT reason
GREEN → write minimal code to pass → verify it passes
REFACTOR → clean up → keep all tests green
REPEAT

RED Phase

Write ONE minimal test that captures the desired behavior. Show it to the user. The test must:

• Test a single behavior
• Have a descriptive name (no "and" — if you need "and", split it)
• Fail because the code doesn't exist yet, not because of a syntax error

# Example
def test_retries_failed_operation_three_times():
    attempts = []
    def flaky():
        attempts.append(1)
        if len(attempts) < 3:
            raise RuntimeError("not yet")
        return "ok"
    assert retry(flaky) == "ok"
    assert len(attempts) == 3

Show the expected failure output. Ask the user to run it and confirm it fails.

GREEN Phase

Write the minimal code that makes the test pass. No extras. No "while I'm here" improvements. If the test passes, confirm with the user before moving on.

REFACTOR Phase

Clean up duplication, naming, structure — without adding behavior. Run tests after each change. If anything goes red, revert immediately.

Rules

• Test fails for wrong reason? Fix the test, not by adding code — go back to RED.
• Bug found? Write a failing test that reproduces it first. Never fix bugs without a test.
• Code written before a test? Delete it. Start over. The sunk cost is already gone.
• "I'll write tests after"? No. Tests-after prove nothing — you already know it works.

Checklist Before Moving to Next Task

• Test was written and confirmed failing before implementation
• Implementation is minimal (nothing beyond what the test requires)
• All tests passing
• Code committed