- 官方交接与 ownership 主轴：OpenAI Agents SDK
Sessions/Results/Agent orchestration/Handoffs/Guardrails/Tracing/Human-in-the-loop、AnthropicComputer use tool、MicrosoftHandoff/Checkpoints/Using Workflows as Agents/Concurrent/Hand off to a live agent - 线程连续性与恢复主轴：LangGraph
Persistence/Interrupts/Durable execution - 最终控制权与人工接管主轴：OpenAI
ChatGPT Agent System Card/Operator System Card、EU AI ActArticle 14 - 责任链与制度边界主轴：EU AI Act
Article 25、NIST AI RMF Playbook - 交接面风险主轴：
Red-Teaming LLM Multi-Agent Systems via Communication Attacks - 工程案例主轴：Anthropic
How we built our multi-agent research system
附录 G：分章资料清单