clean-code

Invoke the skill explicitly (for example /clean-code) or preload it in a subagent's skills field, then ask for a review. It returns a structured report with findings anchored to principles or smells, each with a concrete fix, ordered by importance.

It covers meaningful naming, small single-purpose functions, DRY, SOLID, error handling, TDD and Kent Beck's simple design rules, component cohesion and coupling, the Clean Architecture Dependency Rule, concurrency defense, and craftsmanship disciplines like estimation and saying no.

Yes, it is language-agnostic and prompt-only with no runtime dependencies. The principles apply to any codebase the agent can read; examples are illustrative and the agent is instructed to write in the user's language and idioms.

A dedicated right-sizing section treats needless complexity as a smell and matches structure to the problem. Full Clean Architecture layering is reserved for systems with diverging reasons to change; a short script gets one testable seam, not ports and adapters.

As an advisory knowledge skill, models rarely auto-invoke it as a first action. The README recommends preloading it via a subagent's skills field, a SessionStart hook, a CLAUDE.md instruction, or manual /clean-code invocation for reliable application.

Skip ports, adapters, and composition roots for small scripts or single-purpose tools without diverging reasons to change. The skill directs you to introduce a boundary only when a concrete pressure appears, such as a second caller, a swapped database, or an untestable seam.