What problem does it solve? Complex, subjective, or UI-heavy tasks often suffer from quality drift and missed requirements when a single agent both writes and reviews its own work. This Skill enforces explicit role separation so planning, implementation, and evaluation happen as distinct phases with a written contract. ## Core Features & Use Cases - Sprint Contract First: Requires filling docs/harness/sprint-contract.md with scope in/out, acceptance criteria, and a verification plan before any code is written. - Three Sequential Roles: Planner defines scope, Generator implements only contracted scope, and Evaluator judges against acceptance criteria using the harness-evaluator-rubric without self-approval. - Scope Discipline: Adjacent discoveries are logged to feature_list.json as not_started instead of being implemented, and scope changes require contract updates and approval. - Use Case: When building a multi-file UI feature where previous attempts missed requirements, invoke this Skill to produce a sprint contract, implement strictly within scope, then evaluate against observable acceptance criteria with an evidence log. ## Quick Start Invoke the harness-role-separation skill to plan, implement, and evaluate my UI feature using a sprint contract with separate planner, generator, and evaluator phases.
