🛡️ S2-Avatar-Middleware: The Root Gatekeeper#

数字人权限拦截中间件#

v1.0.0 | Bilingual Edition (English / 中文)

This is the enforcement layer of Digital Jurisprudence. While s2-digital-avatar drafts the laws, this middleware acts as the Gatekeeper, physically intercepting unauthorized actions from OpenClaw or other sub-agents. 这是数字法理学的强制执行层。如果说 s2-digital-avatar 负责起草法律，那么本中间件就是 守门人，在物理层面拦截 OpenClaw 或其他下级智能体的越权行为。

(⚠️ PREREQUISITE: You must have generated a Mandate using s2-digital-avatar before using this module.)

---

🔴 System Pre-flight & Safety Lock / 运行前自检与安全锁#

To prevent silent failures, this middleware will NOT go ONLINE unless two strict conditions are met: 为了防止静默失效，除非满足以下两个严格条件，否则本中间件 拒绝上线：

1. Valid Avatar Mandate: The avatar_identity.json must exist, proving you have legally delegated authority. / 必须存在数字人最高授权令。
2. SMTP Email Configured: You MUST configure a valid SMTP server (e.g., Gmail/QQ App Password). [cite_start]If an agent tries to lock you inside your burning house , the Avatar will block it AND instantly email you the audit log. An unnotified block is a security risk. [cite_start]/ 必须配置邮件传输服务。如果底层 AI 试图在火灾时把您锁在屋内 ，数字人不仅会实施拦截，还必须能立刻向您发送邮件告警。

⚖️ The LLM Judgment Engine / 大模型开庭审判#

When a sub-agent attempts a sensitive action, this middleware pauses the execution and feeds the context to the Avatar (via your local LLM like LM Studio at localhost:1234). 当底层智能体试图执行敏感操作时，本中间件会挂起请求，并将上下文喂给您的数字人（通过本地大模型）。

The Avatar strictly evaluates the request against: 数字人将严格比对以下规则进行裁决：

• [cite_start] The Three Laws of Silicon Intelligence: Guaranteed enforcement of Fail-Open designs and Human Priority [cite: 157-169]. [cite_start]/ 硅基智能三定律：确保物理熔断与生命熵减绝对执行 [cite: 157-169]。
• Your Precedents: If an agent tries to buy cilantro, and your profile says "I hate cilantro," it gets denied. / 您的生活判例库：如违背您的喜好，请求将被无情驳回。

Configure your proxy. Enable the firewall. Reclaim control over your AI grid. 配置您的代理人。开启防火墙。重新夺回对 AI 网格的控制权。