code-modularization-evaluator — Claude Code skill: score your coupling

What code-modularization-evaluator does

The skill reads your code along three axes: integration strength (what components actually know about each other), distance (how far that knowledge travels) and volatility (how often the area changes). Strength runs from intrusive down to contract, distance from methods in one class up to services owned by different teams. The rule it applies: strong coupling belongs close together, weak coupling can sit far apart, and low-volatility code tolerates a mismatch.

A checklist flags the worst cases first — direct database access into another service, reflection on private fields, business logic copied across services, microservices that need synchronized deploys, CBO above 14, circular dependencies. Then comes connascence, static and dynamic, with a fix per type: magic values become named constants, positional parameters become a parameter object, a shared algorithm moves to one place and gets referenced by name.

The model comes from Vlad Khononov's writing on balancing coupling in software design, and the skill points to references/connascence-types.md for worked examples. It installs into Claude, Codex, Cursor, GitHub Copilot, Gemini CLI and OpenCode, each with its own command. On a small codebase with a couple of modules the dependency graph comes back nearly empty, so the analysis has little to chew on.