open-managed-agents details proxy-based credential isolation for agents

On 2026-11-06 at 15:59, Juejin published an article titled "Agent 基建：基于 Proxy 的凭证不入沙箱与网络管控". It is the second entry in the infrastructure series of open-managed-agents, an open source cloud hosted Agent project whose code lives at github.com/superduck-ai/open-managed-agents. The piece explains how open-managed-agents can reach MCP Server endpoints and other external services that require a Token without letting credentials enter the sandbox. The approach uses a Proxy that holds the credentials on its own and enforces egress traffic control in one place. In the article, the authors position it as a ready to borrow pattern for teams building their own Agent platforms that need credential isolation and a clean network boundary. The material does not describe any earlier step or any follow up coverage after this one, so the picture stops at this single design write-up. No endpoint names, traffic figures, or adoption numbers for deployments are provided in the article, and the material neither assesses the mechanism nor reports results measured after it went live.