Clarify First

Enforce risk-based alignment for AI Agents. Stop Claude/Cursor from guessing on vague or high-impact tasks. 让 AI 先对齐、再动手：为 Agent 引入风险分级与澄清机制，拒绝盲目猜测，确保开发安全可控。