From josecabralf-skills
Audit existing unit tests and emit one verdict per test (keep / refactor / delete / wrong-layer) with the named smell and a one-line next action. Use when reviewing a PR that touches tests, when tests are described as flaky / brittle / fragile, when asked "is this test worth keeping?", or when a test-suite audit is requested. Filters out tests that should never have existed before naming smells in the survivors. Do not use for authoring new tests, integration tests, e2e tests, property-based tests, or contract tests.
How this skill is triggered — by the user, by Claude, or both
Slash command
/josecabralf-skills:assess-unit-testsThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Per-test verdict driven by a named smell. See `TEST-SMELLS.md` for the catalogue.
Per-test verdict driven by a named smell. See TEST-SMELLS.md for the catalogue.
Confirm the suite is unit tests. If integration / e2e / property-based / contract: STOP — wrong skill.
Both gates:
| Real question | Right tool |
|---|---|
| Holds for all inputs in a class? | Property-based test |
| Matches another team's API? | Contract test |
| Orchestration of collaborators? | Integration test |
| Output stable across versions? | Approval + invariant |
If either gate fails: verdict shortcuts to delete with categorical reason (tests trivial code, belongs as property-based test, etc.). Skip steps 2–3.
Read the test. Match against the catalogue in TEST-SMELLS.md. One named smell — the dominant one. If no smell matches and the assertions reference the public contract: no smell.
Apply the verdict table. Emit:
<test name / location>
Verdict: keep | refactor | delete | wrong-layer
Smell: <name from catalogue, or — if keep>
Next action: <one line, or — if keep>
After all per-test blocks:
testDoFoo) vs. behaviorally (should_…). Method-named tests predict refactor-resistance debt.npx claudepluginhub josecabralf/skillsCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.