Test-Driven Development (TDD)

Write the test first. It forces precise specifications and gives a free regression net.

Red → Green → Refactor

1. Red. Write ONE small test for the behavior you want. Run it. It must FAIL (and fail for the right reason — an assertion, not a compile error in unrelated code).
2. Green. Write the minimum code to make that test pass. No more, no less. Run it.
3. Refactor. Clean up duplication while keeping tests green. Re-run after each change.

Repeat with the next smallest increment.

Rules of thumb

• Test behavior, not implementation. Prefer public API; don't poke private internals.
• One assertion concept per test. If "and" appears in the test name, consider splitting.
• Name tests like it("returns X when Y") / should do Z when ….
• If a bug fix, write the test that reproduces the bug FIRST (see systematic-debugging).
• Don't write tests that depend on wall-clock, random, or network unless you stub them.

Running tests

• Match the project's framework and command (bun test, pytest, go test, …).
• For Bun projects: bun test (per AGENTS.md, Bun is the only JS runner).

When TDD is overkill

• Throwaway exploration / spikes — write the test after you know the shape.
• Pure UI tweaks with no logic — a visual check is fine.

TDD pays off most for: parsers, state machines, financial/permission logic, and anything with many edge cases.