Tsinghua University This document defines a framework for enabling seamless cross-domain interoperability among AI agents operating across different networks, administrative domains, and heterogeneous platforms. The framework addresses the challenges of identity federation, trust establishment, policy harmonization, and secure communication that arise when AI agents from distinct administrative realms need to collaborate on shared tasks. It specifies mechanisms for agent discovery, capability negotiation, trust delegation, and federated policy enforcement, enabling scalable and secure multi-domain AI collaboration without requiring centralized control.