Ironcurtain：运行时为自主智能体

A secure* 运行时为自主智能体 AI agents. Policy 从 plain-English constitutions. (*https://ironcurtain.dev)GitHub 523 ⭐ / 70 Fork，主要使用 TypeScript 开发。标签：agent,mcp,model-context-protocol,policy,sandbox。访问项目：https://github.com/provos/ironcurtain

项目介绍

A secure 运行时为自主智能体 AI agents. Policy 从 plain-English constitutions. (https://ironcurtain.dev)

核心功能

- The agent 是 untrusted. IronCurtain assumes 大语言模型 maycompromised 由 prompt injection、drift. Securitynot depend 在 the model "being good." - English in, enforcement out. You write intent ("no destructive Git operations without approval");系统 compiles it 进入 deterministic rules 是 enforced without further 大语言模型 involvement 在运行时. - Semantic interposition. Instead 的 givingagent raw 系统 access, all interactions go through [MCP](https://modelcontextprotocol.io/) servers (filesystem, Git, etc.). Every 工具 call passes throughpolicy 引擎 can allow, deny,、escalate到 the user 为 approval. - Defense 在 depth. Agent 代码 runs 在 a V8 isolate 与 no direct access 到 the 主机. The only way out 是 through semantically meaningful MCP 工具 calls、every one 是 checked against policy.

使用指南

- Docker Agent Mode — An external agent (Claude Code, Goose, etc.) runs inside a Docker container with no network access. IronCurtain mediates the external effects: LLM API calls pass through a TLS-terminating MITM proxy (host allowlist, fake-to-real key swap), MCP tool calls pass through the same policy engine, and package installations (npm/PyPI) go through a validating registry proxy. In both modes, the agent is untrusted. Security does not depend on the model following instructions — it is enforced at the boundary.

项目信息

项目地址

GitHub 仓库：<https://github.com/provos/ironcurtain>

> 完整 URL：[https://github.com/provos/ironcurtain](https://github.com/provos/ironcurtain)

如需查看完整源码、安装命令、使用示例、API 文档，请直接访问上方项目地址。所有信息均以 GitHub 仓库为准。