- 生命周期与上下文主轴：OpenAI
Sessions、Context management - 控制权与治理主轴：OpenAI Agents SDK
Guardrails/Handoffs/Tracing、AnthropicHooks/Handle approvals and user input - 状态与恢复主轴：LangGraph
Persistence/Interrupts - 权限与部署边界主轴：Anthropic
Securely deploying AI agents、MCPAuthorization/Security Best Practices - 平台观测主轴：Microsoft
Trace and Observe AI Agents - 本章落点主轴：分析现有 Agent 系统时，应从生命周期、状态、动作、控制权和治理边界往里收，而不是从提示词往外散读
附录 G：分章资料清单