📄 中文摘要
对八个框架进行了调查，以映射代理团队中谁可以调用谁，包括向上、横向和递归的通信模式。研究重点在于当领导代理将工作委派给团队成员时，该成员是否可以向上升级、与同级交流或招募自己的子团队。这些问题决定了多代理系统是层级结构、网状结构还是单向管道的树形结构。调查的框架包括Claude Code、OpenClaw、CrewAI、AutoGen、LangGraph、OpenAI Agents SDK、Semantic Kernel和Google ADK，结果显示大多数系统的限制远超预期。
代理如何调用代理
发布: 2026年3月11日
📄 English Summary
How agents call agents
A survey of eight frameworks was conducted to map the communication patterns in agent teams, including upward, lateral, and recursive interactions. Key questions include whether a member can escalate back to a leader, communicate with peers, or recruit a sub-team after being delegated work. These factors determine whether a multi-agent system is hierarchical, mesh-like, or a tree of one-way pipes. The frameworks surveyed include Claude Code, OpenClaw, CrewAI, AutoGen, LangGraph, OpenAI Agents SDK, Semantic Kernel, and Google ADK, revealing that most systems are far more restricted than their marketing suggests.