@forge-kit-dev/module-clean-code

Robert C. Martin's Clean Code principles, narrowed to the 15 rules we can actually enforce on a React/Next.js codebase without drowning developers in false positives. 10 of them are mechanical (block); 5 are advisory (Evaluator rubrics).

Mechanical rules (pre-commit block)

| Rule | Source | Owner | |---|---|---| | @forge-kit-dev/forge/component-max-lines | Clean Code ch. 3 "Small!" | custom ESLint rule (this repo) | | @forge-kit-dev/forge/no-boolean-flag-arg | Clean Code ch. 3 "flag arguments" | custom ESLint rule (this repo) | | max-params (limit 3) | Clean Code ch. 3 "few arguments" | ESLint core | | no-console (allow warn/error) | soft SRP discipline | ESLint core | | complexity (limit 12) | cognitive-complexity proxy | ESLint core |

v0.2 will add @typescript-eslint/naming-convention, react/boolean-prop-naming, and @typescript-eslint/no-explicit-any once the CLI wizard can install peer plugins automatically.

Advisory rubrics (Evaluator)

• r-clean-code-intent — are variables and functions named after what they mean, not what they are?
• r-clean-code-srp — does this component have exactly one reason to change?
• r-clean-code-boundary — is error handling explicit, or buried under nulls and silent catches?

Every criterion is scored 0 / 5 / 10. The Evaluator is forbidden from picking 7 — see RubricScoreSchema in @forge-kit-dev/schemas for why. In short: a forced binary-ish choice stops the Evaluator from praising mediocre code.

Skills

• clean-code-component-size — activated at Generator stage when the model is about to write a new React component.
• clean-code-naming — activated at Planner stage, before any file is created, so names start out intention-revealing.
• clean-code-srp — activated at Evaluator stage, after a sprint completes, so the reviewer can spot hidden responsibilities.

Why only 50 lines?

Clean Code's explicit advice is "functions should hardly ever be 20 lines long". We chose 50 as a compromise because modern React components include a lot of JSX that Martin's original advice didn't anticipate. If your components routinely run past 50, consider whether composition or hook extraction is missing — not whether the rule is wrong.