📄 中文摘要
AI 代理在执行指令时，即使指令十分详细，最终也可能生成不符合预期的代码。这并非出于恶意，而是因为指令只是建议而非强制措施。为了解决这一问题，构建了一种加密审批系统，该系统利用数字签名、Copilot 代理钩子和 MCP 插件，拦截每一次写入尝试，检查有效签名，只有在内容经过人类明确批准后才允许通过，确保没有例外和变通。该系统为 AI 代理的代码生成提供了更高的安全性和控制力。
为 AI 代理构建加密审批门
出处: Building Cryptographic Approval Gates for AI Agents
发布: 2026年2月21日
📄 English Summary
Building Cryptographic Approval Gates for AI Agents
AI agents, regardless of the detail in instructions, will eventually produce code that is not desired. This occurs not due to malice, but because instructions serve as suggestions rather than enforcement mechanisms. A cryptographic approval system has been developed to address this issue, utilizing digital signatures, Copilot agent hooks, and an MCP plugin. The system intercepts every write attempt, verifies a valid signature, and only permits the content to proceed if it has been explicitly approved by a human. This approach ensures no exceptions or workarounds, enhancing the security and control over the code generated by AI agents.