What problem does it solve? Per-diff code reviews and exact-clone linters structurally miss whole-graph defects: dead paths reachable only from other dead code, the same job implemented twice in textually different code, and values produced but never consumed. This Skill runs a deliberate, repeatable discovery pass that surfaces those findings before they calcify. ## Core Features & Use Cases - Four analysis lenses: whole-program reachability for dead paths, the clone detector's known-duplication catalogue, a convergence-scoring pass against a code-health rubric, and a producer-to-consumer data-flow trace for dropped values and orphan stages. - Ownership triage: every reachability hit is checked against tracked issues, decisions, and coupled specifications before any removal is proposed, preventing deletions that contradict recorded decisions. - Scorecard and issue candidates: emits a Strong/Weak/Missing scorecard with location evidence plus ranked, scoped tracked-issue candidates — it never edits code. - Use Case: Before a large refactor of a fast-moving Go codebase, run the sweep with deadcode (tests as roots) and the dupl linter to size the duplication the refactor must absorb and file each finding as its own issue. ## Quick Start Run a structural sweep over this repository and report dead paths, convergent duplication, and unconsumed data flow as a scorecard with issue candidates.