体系 / ai-driven

rule.agent.real-content-risk-gates · canon

Agents prove design with real content, hostile states, and risk gates

Generated interfaces are not refined until realistic content, complete state coverage, accessibility evidence, and consequence-aware AI review exist.

実践ガイダンス

Require purpose, seed content, preview matrix, key-moment tuning, user review, accessibility review, and a do-not-build threshold.

推奨

  • Test long, large, multilingual, RTL, denied, offline, destructive, and restored states
  • Preserve edit, undo, uncertainty, and proportional confirmation

非推奨

  • Treat placeholder content, happy paths, or generated code as production evidence

検証契約

state matrix, accessibility evidence, user feedback, and AI risk assessment

根拠リファレンス

実装・検証成果物