What problem does it solve? Dead code accumulates silently in growing codebases — commented-out blocks, exported symbols nobody calls, and test helpers no test uses — making maintenance harder and reviews noisier. This Skill performs read-only dead-code discovery and reports evidence-backed removal candidates without touching your source files. ## Core Features & Use Cases - Three-Agent Parallel Analysis: Launches parallel agents for commented-out code detection, hierarchical LSP-based unused-symbol analysis (file reachability, then symbol-level, then test helpers), and test-only production code detection. - Precise Reference Classification: Uses LSP operations like findReferences and incomingCalls, counting only actual usage (calls, instantiations, reads, type annotations) while ignoring exports, re-exports, and type-only imports. - Prioritized Evidence Reports: Merges and deduplicates findings into high (unused exports), medium (commented-out blocks), and low (unused test helpers) priorities, each with file:line evidence. - Use Case: Before a cleanup sprint, run the scan on your repository to get a prioritized list of dead-code candidates, then hand confirmed removals to the coding:lint pre-flight workflow. ## Quick Start Scan the src directory for unused code and report removal candidates with file and line evidence.