harness-evaluator-rubric

What problem does it solve? It prevents self-judged, evidence-free verdicts in agent workflows by producing a structured scoring rubric grounded in the active contract and measured gate reports before any accept or revise decision. ## Core Features & Use Cases - Contract-driven rubric: Loads the sprint contract, feature spec acceptance criteria, or approved PREVC plan to derive judgement rows specific to the task. - Measured vs judgement rows: Copies measured metrics verbatim from agent-os/quality-thresholds.json and scores them from the newest gate report, avoiding drifting duplicate thresholds. - Scoring discipline: Applies a 1-5 scale with accept, block, and revise rules, including unavailable scoring for missing tools and a Revise cap for any red measured row. - Use Case: After Execute and Validate complete in a PREVC sprint, run this skill to produce a rubric with evidence that is handed to the Judge phase instead of self-approving the work. ## Quick Start Build the evaluator rubric for the current sprint using the contract in docs/harness/sprint-contract.md and the newest gate report under docs/harness/quality.