evals-e2e-run

What problem does it solve? After a feature travels through brainstorm, design, plan, and implementation stages, there is no objective measure of whether each artefact was well-made or whether each stage faithfully carried forward the previous one's commitments. This Skill scores that entire chain so quality is measured rather than assumed. ## Core Features & Use Cases - Artefact quality scoring: Judges storm, design, and plan documents against detailed rubrics (testability, scope decisiveness, component decomposition, TDD discipline, and more), producing 0-100 scores. - Stage-to-stage consistency checks: Verifies that the design honors the storm, the plan honors the design, and the implementation commits honor all three, flagging silently dropped or contradicted requirements. - Parallel subagent evaluation: Runs one read-only subagent per eval with a strict output contract, then validates, recomputes, and appends results as JSON entries to ~/.claude/evals/design.json. - Use Case: Right after /feature-implement lands its stage commits, run this Skill to get a score table showing how well each artefact and hand-off performed, plus improvement recommendations for any score below 80. ## Quick Start Ask the assistant to run end-to-end evals on the feature you just implemented so it scores the storm, design, plan, and code consistency and appends the results to the eval log.