TDD (Claude Code skill)
You invoke /tdd on Claude Code to enforce red-green-refactor as vertical tracer bullets: one test, one implementation, repeat, no batched RED of 5 tests at once. The skill explicitly bans 'horizontal slicing' (write all tests then all code) which produces tests that validate shape, not behaviour. Distributed by Matt Pocock in his skills repo (github.com/mattpocock/skills).
Strengths
- Bans horizontal slicing, vertical tracer bullets, one test → one impl → repeat
- Forces the design conversation: 'what should the public interface look like?' before coding
- Forbids refactor while RED, simple rule that prevents 80% of TDD sessions going off the rails
- Explicit refs to deep-modules.md, mocking.md, refactoring.md, not a sheet, a mini-manual
Limitations
- Not magic: if the stack doesn't have a proper test seam, /tdd won't create one for you
- Requires prioritising tested behaviours, 'you can't test everything' (the author insists, which is unusual)
- No long-term skill API stability guarantee (distributed outside official Anthropic registry)
Best for
- Devs doing TDD but ending up with tests that break on every refactor
- Tech leads who want to industrialise a shared TDD method in an agent-first team
- Devs who want Claude Code to write tests describing the system instead of testing its structure