tddlisted
Install: claude install-skill t8jwscc7qy-sketch/foma-skills
Write a failing test, see it fail for the right reason, add the minimum code to pass it, repeat one behaviour at a time.
Path: one small behaviour on a named function → steps 1, 3–5 once; several behaviours, a bug fix or an unclear interface → the full loop. Say which path in one line.
## Steps
1. Test command: `## Commands` in the project's CLAUDE.md. If missing, detect it (package.json, Makefile, pyproject.toml, Cargo.toml); if none, say so and ask how to run tests.
2. Agree the seam: the public interface to test at, ideally one. If the user named the function or interface, that is the seam; state it and go on. Otherwise propose one, with a line on what it catches and misses, and ask.
3. Pick the next behaviour. Write one test for it through that seam. Take the expected value from the spec or a hand calculation, never from the code under test.
4. Run that test. It must fail (not error) because the behaviour is missing. Passes or fails oddly: fix the test.
5. Write the least code that passes. Run it, then the rest of the suite.
6. Repeat from 3 for the next behaviour. Bug fix: the first test reproduces the bug.
7. Refactoring happens after green, as a separate step, not inside the loop.
## Rules
- One slice at a time: one test, then its code. Never write all tests first, never code ahead of a test.
- Never edit, skip or weaken a test to make it pass; fix the code.
- Test behaviour through the public interface, not private methods or internals; a refactor must not break th