What problem does it solve? Codebases accumulate shallow wrapper modules that spread logic across many files, making them hard to test and reason about. This Skill finds that architectural friction and produces scored, structured refactor proposals that downstream skills can consume programmatically. ## Core Features & Use Cases - Numeric Scoring: Scores each candidate module on depth, leverage, and locality (1-5 each) plus a deletion-test verdict (vanish, concentrate, redistribute). - Dependency Classification: Categorizes dependencies as in-process, local-substitutable, remote-owned, or true-external to determine the right test strategy and seam discipline. - Structured Proposals: Emits YAML proposal payloads with current/target scores, planned adapters, and test changes, consumable by surgeon, brainstorm, audit, and review skills. - Use Case: Point it at src/auth/ where login, refresh, and logout are three thin modules; it scores the cluster, proposes a deepened AuthPort with production and in-memory adapters, and hands the payload to a refactor skill. ## Quick Start Ask the agent to run improve-architecture on a target module path such as src/auth/ to get scored deepening candidates and a recommendation.
