What problem does it solve? Teams often sense that a codebase has architectural problems but cannot tell which investment would pay off now. This Skill audits a repository or subsystem, gathers evidence from change history, callers, tests, and dependency direction, and ranks bounded architecture candidates so you stop guessing and stop producing generic cleanup lists. ## Core Features & Use Cases - Evidence-backed candidate ranking: Generates candidates such as deepening modules, collapsing pass-through chains, splitting god modules, moving seams, and repairing dependency direction, each with evidence, counterevidence, confidence, and recommendation strength. - Visual HTML report: Produces a self-contained, offline-readable HTML report with before-and-after diagrams, accessible markup, a comparison matrix, and one top recommendation, written to the OS temp directory by default. - Bounded audit scope: Treats the repository as read-only, respects the user's named scope, and stops before designing exact interfaces or editing code. - Use Case: Ask it to audit a multi-module service where every feature touches five files; it traces behavior, identifies a leaked-knowledge candidate as the top recommendation, and delivers a report you can share with the team before committing to a refactor. ## Quick Start Ask the assistant to audit this repository with the improve-codebase-architecture skill and produce a ranked architecture report.
