| name | tdd-workflow |
| description | Red-green-refactor workflow adapted for a 100% branch-coverage trading CLI — narrow regression tests first, no coverage regressions. |
| origin | adapted from ZMB-UZH/omero-docker-extended tdd-workflow |
TDD Workflow
Use this skill whenever you add a feature, fix a bug, or change a behavior that users or tests can observe.
The loop
- Name the contract you're about to change in one sentence ("
shell quits on Ctrl-D without error" / "fetch --batch-size 2000 clamps to the exchange limit"). If you can't name it, you don't know the change yet.
- Write the failing regression test first in
tests/ under the matching file. Target branches, not just happy paths. If the branch is error handling, assert both the failure and the recovery.
- Implement the smallest change that moves the test from red to green. Do not take the opportunity to restructure unrelated code.
- Run targeted tests first (
pytest -q tests/test_<module>.py), then the full suite, then coverage with --fail-under=100.
- Stop at green. Polish and refactoring is a separate commit with its own tests.
Coverage is load-bearing
This repo runs coverage report --fail-under=100 in CI. A new branch with no test assertion is a CI break, not a style issue. When you add code:
- Every new
if, elif, else, except, with, and early return needs a test that exercises it.
- If a branch is genuinely unreachable, prove it — don't guard with
# pragma: no cover unless you also leave a comment explaining why the branch exists.
Tests are documentation
- Prefer fixtures that make the intent obvious (
fake_candles, trained_model_stub) over opaque tuples.
- Assert on public contract (return values, persisted artifacts, printed lines). Avoid asserting on private attribute shape.
- Do not mock
BinanceClient at a lower level than necessary. The existing tests stub at the HTTP seam — follow that pattern.
Don't
- Don't weaken an existing assertion to get a test to pass after your change. If an assertion is wrong, fix it deliberately with an explanation in the commit.
- Don't use
time.sleep in tests. Monkey-patch the clock or inject a sleep hook.
- Don't rely on network in tests. If you need HTTP behavior, stub
requests.Session.request at the seam.
