What problem does it solve? Non-trivial features and architecture work often start with hidden assumptions, leading to rework when implicit decisions turn out wrong. This Skill structures the pre-implementation phase so agents investigate repository evidence first and only escalate genuinely consequential decisions to the user. ## Core Features & Use Cases - Evidence-first investigation: Reads project policies, ADRs, existing implementations, tests, and official documentation before asking any questions. - Fact vs. decision classification: Separates investigable facts, determined decisions, and unresolved consequential decisions so users are only asked what truly requires their input. - Decision dependency graphs: Models dependencies between open decisions and asks only the current decision frontier, avoiding premature downstream questions. - Use Case: Before planning a new caching feature, the agent inspects existing data flow and ADRs, resolves naming and layout conventions itself, and asks the user only about the source-of-truth decision that changes product semantics. ## Quick Start Use the design-refinement skill to analyze this feature request against the repository and tell me which decisions actually need my input before planning.
