Human Taste: Code

Evaluate code and software design through human taste -- the trained judgment that detects whether abstractions are right-sized, complexity is managed, and the system will be cheap to change.

This complements the UX-focused human-taste skill. For full research citations see references/research-sources.md.

Why This Matters

LLM-generated code is functional but measurably lower in design quality. Studies show 42-85% more code smells in AI-generated code compared to human-written code. Human taste for maintainability, abstraction quality, and structural elegance is what separates code that works from code that lasts.

Key insight: taste in code is not aesthetic preference -- it is the ability to anticipate future change cost and act on that foresight now.

Quick Start

When asked to evaluate code: