test-gap-finder
Ranks untested modules by recent churn and size, then proposes concrete test cases without writing tests unasked.
testing
coverage
gap-analysis
repo-hygiene
SKILL.mdskills/test-gap-finder
GitHubtest-gap-finder
Procedure
- Detect test layout:
tests/,__tests__/,*_test.go,*.test.ts,test_*.py,spec/. Detect the runner from config (jest,vitest,pytest,go test,cargo test). - Build the map: list source files (exclude generated, vendored, migrations). For each, find tests that import or name it (grep the module path/name inside test files). Result:
module | test files | test count (rough: count of test functions). - If a coverage report is cheap, run it:
pytest --cov --cov-report=term-missing -q,npx vitest run --coverage,go test -cover ./.... Skip if it takes > 2 min or needs services. - Rank untested/under-tested modules by: recent churn (
git log --since=90.days --format='' --name-only | sort | uniq -c), size (wc -l), and whether they sit on a critical path (auth, payments, data writes, public API). - For the top 5, propose tests: file path, test names, and 2 to 4 concrete cases each (happy path, boundary, failure). Reference the module's real function signatures (read the code).
- Output a table plus the proposals. Ask before writing any test files; if approved, write them following the repo's existing test style and run them.
Rules
- Do not claim a coverage percentage unless a coverage tool produced it.
- Do not propose tests for generated code, vendored code, or trivial getters.
- Proposed cases must be derivable from the code read in this session.
Verify
Maintainers: bash evals/run.sh /tmp/skillkeel-evals test-gap-finder from the plugin root runs this case and grades it; records and transcripts sit in evals/test-gap-finder/.
evals/test-gap-finder/: fixture Python package with auth.py (untested, high churn), utils.py (tested), payments.py (untested); expected: auth.py and payments.py ranked top, utils.py not proposed, ≥2 concrete cases per proposal, no test files written.