メインコンテンツまでスキップ

Part I—Foundational Principles: Adapting Human-Centric Review for AI

Decomposing the Gold Standard​

AI as "Enhanced Linter" is insufficient. Core of what human review observes:

  • Design: Does change fit architecture?

  • Functionality: Does it work as intended? / User value

  • Complexity / Maintainability: Simplification room / Future understandability

  • Tests: Appropriate and inclusive?

  • Naming / Comments / Style / Docs: Consistent readability

Health of Review itself:

  • Small, Focused (~400 lines / review target)

  • Collaborative / Non-adversarial, Start with Questions not assumptions

  • Polite and Constructive feedback

Role as Architecture Guardian​

AI should become an Architecture Suitability Checker visualizing dependency graphs, boundaries, and layers. Embed Continuous Improvement into decision logic so valuable changes are not blocked by minor nits.