What problem does it solve? Substantive implementation tasks often lack explicit acceptance criteria and independent verification, leading to unverified changes, scope creep, and silent model or permission substitutions. This Skill enforces a pre-edit Evaluation-Driven Definition (EDD), deterministic checks, and a fresh read-only verifier so every change is backed by inspectable evidence. ## Core Features & Use Cases - Pre-edit EDD record: Captures required and advisory criteria, exact validation commands, risk boundaries, and evidence targets before any code is edited. - Bounded worker and verifier contracts: Dispatches implementation workers with briefs under 400 tokens and runs a fresh, read-only verifier per round that returns exactly VERIFIED, INCOMPLETE, or BLOCKED. - Round-limited convergence: Defaults to three rounds with an explicit max=N override, escalating with named gaps when the limit is exhausted. - Use Case: Invoke $edd-loop max=5 to implement a multi-file refactor, run the test suite as the deterministic gate, and obtain an independent verifier verdict before handing off. ## Quick Start Ask the agent to run $edd-loop on your implementation task with its scope, acceptance criteria, and validation commands so it builds an EDD, implements, and verifies the result.
