harness-evaluator

What problem does it solve?

The harness-evaluator eliminates biased self-review by independently testing running applications against explicit sprint contracts and calibrated grading criteria, ensuring build quality, regression safety, and measurable readiness for release.

Core Features & Use Cases

• Contract Review & Testability Checks: Validates sprint contracts for completeness, measurability, and start commands before coding begins and produces a structured contract-review artifact.
• Sprint QA & Regression Testing: Runs live-app checks, automated test suites, and regression checks against CODEBASE_AUDIT.md for existing projects, producing a detailed .harness/QA_REPORT.md.
• Polish Mode & Calibration: Performs iterative polish evaluations with sub-criteria, drift detection against POLISH_HISTORY.md, and pivot suggestions when stalled; consults embedded calibration anchors and project-specific overrides.
• Integration Requirements: Uses Playwright MCP for web evaluation, consults the frontend-design skill for design grading, and requires Context7 for authoritative API documentation checks.

Quick Start

Run harness-evaluator on the current sprint to produce a calibrated .harness/QA_REPORT.md (or POLISH_REPORT.md in polish mode) and a PASS/FAIL verdict with actionable remediation.