What problem does it solve? Evaluating whether a spec-driven development (SDD) implementation actually fulfills its PRD is subjective and irreproducible across evaluators. This Skill decomposes every acceptance criterion into atomic binary checks backed by file:line evidence, producing a single comparable final grade that stays stable across runs, evaluators, and frameworks. ## Core Features & Use Cases - Binary checklist scoring: Decomposes each acceptance criterion into MET/UNMET implementation (I) and test (T) checks with mandatory file:line evidence, then rolls up to AC, story, and final grades weighted by P0/P1/P2 priority. - Framework vs harness separation: Grades requirement extraction (Elicitation E), scope adherence (S), robustness (R), engineering gates (G), and test distribution (D) as independent dimensions reported beside the final grade. - Reproducibility controls: Frozen AC baselines, k=3 self-consistency voting, script-computed roll-ups, and calibration anchors keep the same PRD plus commit yielding the same grade. - Use Case: Benchmark three SDD frameworks against one PRD by freezing the baseline once, running plan/implement/evaluate sessions per framework, and comparing the resulting Final scores. ## Quick Start Ask the agent to run spec-driven-eval to grade the implementation on the current branch against the PRD, reusing the frozen baseline and producing a timestamped evaluation report.
