tdd

Red-Green-Refactor TDD loop for AI-implemented features. Writes ONE failing test first, verifies red, minimum code to green, then refactors. Blocks AI test-cheating (writing tests AFTER implementation to match code, not spec).

tdd — Red-Green-Refactor for AI-coded features

Trigger

Fire when:

User says "TDD this", "red-green-refactor", "test-first"
New-feature task under [[afk-vs-hitl-tasks]] AFK mode + test infra present

Precondition

Test loop <60s per [[feedback-loops-are-the-ceiling]]. Slow loop? Fix loop FIRST. Never skip TDD to dodge slow tests.

Process — strict order

Step	Action	Verify
1. RED	Write ONE failing test. Smallest observable behavior.	Run. MUST see FAIL with expected error.
2. GREEN	Minimum code to pass. No extras. No untested error handling.	Run. MUST see PASS.
3. REFACTOR	Clean impl + test. Tests stay green. Extract deep modules per [[deep-modules-over-shallow]]. Rename for clarity.	Run. Stays green.
4. LOOP	Next behavior. Back to RED.	—

Anti-cheat — hard rules

MUST see failing test output before writing production code
MUST NOT write multiple tests before first passes
MUST NOT skip refactor — design emerges there
MUST NOT rewrite test to match implementation after the fact

Test written AFTER code = code-shaped test, not spec-shaped. Rejects the point.

Output shape per cycle

RED: <test name> → <expected fail message>
GREEN: <files changed, LOC delta>
REFACTOR: <what got extracted/renamed>

Source

Kent Beck, Test-Driven Development: By Example (1999). AI-coding context: Matt Pocock workshop, 2026-07-03.

tdd