- 官方循环与编排主轴：OpenAI
A practical guide to building AI agents/Agents/New tools for building agents/ CookbookReasoning function calls/Build a coding agent with GPT 5.1、AnthropicBuilding effective agents/Claude Agent SDK/Effective context engineering for AI agents、GoogleChoose a design pattern for your agentic AI system/Choose your agentic AI architecture components、MicrosoftWorkflows - Checkpoints/Handoff - 研究范式主轴：
ReAct、Plan-and-Solve Prompting、Tree of Thoughts、Language Agent Tree Search、Reflexion、Self-Refine - 长任务与代码编排主轴：
Voyager、SWE-agent - 循环评测与可观测性主轴：Anthropic
Demystifying evals for AI agents、OpenAITrace grading、LangChainMulti-turn Evals、WebArena、ToolSandbox - 失控、停止与监督主轴：
Operator System Card
附录 G：分章资料清单