Skip to content

Getting Started

m4bwav edited this page Nov 21, 2026 · 1 revision

Ask the agent in plain words. The skill triggers on requests such as:

  • "what should I test in this repo"
  • "run the tests and tell me what failed"
  • "is this test flaky"
  • "are my tests any good"
  • "coverage on my diff"
  • "add tests before I refactor this"
  • "test my Unity game"

1. What to test first

The skill runs tw.py plan and reads the top ten files. On the shop fixture in the repository:

python scripts/tw.py plan evals/fixtures/shop --top 5
  50  NO TEST  payment.py  (money or economy)
  30  NO TEST  models.py  (money or economy)
  30  NO TEST  save.py  (saves and persistence)

Only three files are listed although five were asked for: utils/strings.py matched no risk pattern, and files that score zero are left out. For each file the agent writes the failure it fears in one line, then the smallest test that catches it. See How the plan ranks files.

2. Run and read

python scripts/tw.py run evals/fixtures/red

The runner's own output comes first; the summary is the last lines:

junit: 2 tests, 1 passed, 1 failed, 0 skipped, 0.0s
FAIL test_cart.test_total_discount
  assert 0 == 180

The exit code is 1 because a test failed.

3. Are the tests any good

python scripts/tw.py lint evals/fixtures/weak
evals\fixtures\weak\test_score.py:6: assert True can never fail
evals\fixtures\weak\test_score.py:15: compares a value with itself
evals\fixtures\weak\test_score.py:8: asserts only on mock calls; checks wiring, not behaviour
3 finding(s)

Paths print with the platform separator; this run was on Windows. The skill then asks for mutation testing on the changed code (Stryker.NET, StrykerJS, mutmut, or hand mutants for Unity and Luau), and for each new test to be watched failing once against broken code.

Clone this wiki locally