| name | specky-test-verifier |
| description | This skill should be used when the user asks to 'verify tests', 'check coverage', 'detect phantom completions', 'check spec drift', or needs guidance on Phase 8 verification. Also trigger on 'sdd test', 'test traceability', 'verification report', or 'gate criteria'. |
Test Verification — Contracted Evidence
Preconditions
The feature must enable tdd. Read these persisted values before running anything:
- framework and property framework
- output directory and trace marker
- coverage threshold
- executable example-test and property bindings
Specky does not generate TODO tests or in-file fake system models. A binding must cover every requirement ID and contain executable, nontrivial assertions.
Workflow
- Run the configured framework and capture machine-readable results.
- Call
sdd_verify_tests with spec_dir, feature_number, and test_results_json.
- Compare
effectiveCoverage to the response’s contracted coverage_threshold and meets_threshold; do not assume 80%, 90%, or 100% globally.
- Call
sdd_verify_tasks with explicit code_paths to identify task evidence and phantom completions.
- Call
sdd_check_sync with explicit code paths to identify requirement drift.
- Re-run EARS validation against the selected feature.
Evidence Rules
- Test names or trace comments carry real
REQ-* IDs.
- Passing status without a requirement trace does not count as coverage.
- Placeholder assertions (
expect(true), assert True, TODO bodies) are rejected by contract validation.
- Property bindings must include their requirement ID in executable code and may not use generated model stubs.
- Every claimed task completion points to implementation evidence in the supplied code paths.
Output
Present coverage by requirement, failing and untested IDs, threshold result, task verification, and drift evidence. The server computes the gate; do not manufacture PASS/FAIL from prose.