core-evidence-based-qa

# Evidence-Based QA No claim about software quality is accepted without a verifiable test signal. "Should work" is not evidence. "I checked it manually" is not a finding. "Tests pass" without citing which tests, on which branch, against which commit, is a self-report — and self-reports are the single largest source of phantom QA sign-offs in multi-agent systems. This skill encodes the reasoning discipline that sits **behind** the CI coverage gate and `mutation-gate.yml`. The mechanical gates enforce thresholds; this skill teaches why those thresholds exist, what they fail to catch, and how a QA agent produces a sign-off that survives adversarial scrutiny. --- ## What This Skill Is (and isn't) **Is:** A reasoning doctrine. A set of hard rules for interpreting test output, classification heuristics for regression types, and a WRONG/CORRECT vocabulary for common QA errors. It applies whenever an agent evaluates whether a change is correct, safe to merge, or genuinely fixed. **Is not:** A test-runner manual. It does not prescribe which test framework to use, how to structure test files, or how to configure CI pipelines. That lives in `core/testing-strategy`. This skill is about the *judgment layer* that interprets the output of whatever runner {{PROJECT_NAME}} uses. **Is not:** A replacement for the mechanical gates. The CI coverage gate, `mutation-gate.yml`, and the hook-enforced test count are the load-bearing enforcement. This skill makes those gates interpretable. When...