99. 参考资料

下面这些资料是写作本套文档时重点参考的一手来源。 涉及到技术细节和 API 的地方，请以官方文档的最新版本为准——Agent 生态演进快，2-3 个月 API 就可能发生变化。

1. 官方文档

1.1 LangChain / LangGraph (TypeScript)

• LangChain TypeScript Overview https://docs.langchain.com/oss/javascript/langchain/overview 

• LangChain Agents (createAgent) https://docs.langchain.com/oss/javascript/langchain/agents 

• LangChain Middleware https://docs.langchain.com/oss/javascript/langchain/middleware 

• LangChain Context Engineering https://docs.langchain.com/oss/javascript/langchain/context-engineering 

• LangChain Retrieval https://docs.langchain.com/oss/javascript/langchain/retrieval 

• LangChain Short-term Memory https://docs.langchain.com/oss/javascript/langchain/short-term-memory 

• LangChain Long-term Memory https://docs.langchain.com/oss/javascript/langchain/long-term-memory 

• LangChain Structured Output https://docs.langchain.com/oss/javascript/langchain/structured-output 

• LangChain Human-in-the-loop https://docs.langchain.com/oss/javascript/langchain/human-in-the-loop 

• LangGraph Overview https://docs.langchain.com/oss/javascript/langgraph/overview 

• LangGraph Quickstart https://docs.langchain.com/oss/javascript/langgraph/quickstart 

• LangGraph Streaming https://docs.langchain.com/oss/javascript/langgraph/streaming 

• LangGraph Interrupts（interrupt() + Command(resume)） https://docs.langchain.com/oss/javascript/langgraph/interrupts 

• LangGraph Persistence（Checkpointer） https://docs.langchain.com/oss/javascript/langgraph/persistence 

• LangGraph Memory https://docs.langchain.com/oss/javascript/langgraph/memory 

• LangGraph Frontend Overview https://docs.langchain.com/oss/javascript/langgraph/frontend/overview 

• useStream / Agent Chat UI https://docs.langchain.com/oss/javascript/langgraph/ui 

• LangGraph Human-in-the-loop https://docs.langchain.com/oss/javascript/langgraph/human-in-the-loop 

• LangGraph Platform https://docs.langchain.com/langgraph-platform/ 

• Deep Agents https://docs.langchain.com/labs/deep-agents 

1.2 Anthropic Claude

• Prompt Engineering Guide https://docs.claude.com/en/docs/build-with-claude/prompt-engineering/overview 
• Prompt Caching https://docs.claude.com/en/docs/build-with-claude/prompt-caching 
• Tool Use https://docs.claude.com/en/docs/build-with-claude/tool-use/overview 
• Extended Thinking https://docs.claude.com/en/docs/build-with-claude/extended-thinking 
• Building Effective Agents (Anthropic engineering blog, 2024-12) https://www.anthropic.com/research/building-effective-agents 
• Contextual Retrieval (2024-09) https://www.anthropic.com/news/contextual-retrieval 
• Claude Computer Use https://docs.claude.com/en/docs/build-with-claude/computer-use 
• Batch API https://docs.claude.com/en/docs/build-with-claude/batch-processing 

1.3 OpenAI

• Prompting https://platform.openai.com/docs/guides/prompting 
• Prompt Engineering https://platform.openai.com/docs/guides/prompt-engineering 
• Structured Outputs https://platform.openai.com/docs/guides/structured-outputs 
• Retrieval https://platform.openai.com/docs/guides/retrieval 
• File Search https://platform.openai.com/docs/guides/tools-file-search 
• Safety in building agents https://platform.openai.com/docs/guides/agent-builder-safety 
• Prompt Caching https://platform.openai.com/docs/guides/prompt-caching 
• Batch API https://platform.openai.com/docs/guides/batch 

1.4 Model Context Protocol (MCP)

• MCP Specification（当前版本 2025-06-18） https://modelcontextprotocol.io/specification/2025-06-18 
• MCP Introduction https://modelcontextprotocol.io/introduction 
• TypeScript SDK https://github.com/modelcontextprotocol/typescript-sdk 
• MCP Servers 生态 https://github.com/modelcontextprotocol/servers 

MCP 当前核心原语：

• Server 侧：Resources、Prompts、Tools
• Client 侧：Sampling、Roots、Elicitation

1.5 AG-UI

• AG-UI Introduction https://docs.ag-ui.com/introduction 
• AG-UI Concepts（Agents / Events / State） https://docs.ag-ui.com/concepts/agents 
• AG-UI Events 列表 https://docs.ag-ui.com/concepts/events 

1.6 前端 / UI

• Vercel AI SDK https://ai-sdk.dev/docs 
• Vercel Resumable Streams https://vercel.com/docs/functions/streaming/resumable-streams 
• CopilotKit https://docs.copilotkit.ai/ 
• assistant-ui https://www.assistant-ui.com/docs 

1.7 Observability

• LangSmith Observability (LangChain OSS JS) https://docs.langchain.com/oss/javascript/langchain/observability 
• LangSmith Observability (LangGraph OSS JS) https://docs.langchain.com/oss/javascript/langgraph/observability 
• LangSmith Observability Concepts https://docs.langchain.com/langsmith/observability-concepts 
• Langfuse Docs https://langfuse.com/docs 
• Arize Phoenix Docs https://arize.com/docs/phoenix 
• Helicone Docs https://docs.helicone.ai/ 
• OpenTelemetry Documentation https://opentelemetry.io/docs/ 
• OpenTelemetry GenAI Semantic Conventions https://opentelemetry.io/docs/specs/semconv/gen-ai/ 

1.8 Evals

• RAGAS https://docs.ragas.io/ 
• Promptfoo https://www.promptfoo.dev/docs/intro 
• DeepEval https://deepeval.com/docs/getting-started 
• LangSmith Evaluation https://docs.smith.langchain.com/evaluation 
• Phoenix Evals https://arize.com/docs/phoenix/evaluation 

1.9 Durable Execution

• Temporal https://docs.temporal.io/ 
• Inngest https://www.inngest.com/docs 
• Restate https://docs.restate.dev/ 
• Trigger.dev https://trigger.dev/docs 

1.10 Vector DB / Retrieval

• pgvector https://github.com/pgvector/pgvector 
• Qdrant https://qdrant.tech/documentation/ 
• Weaviate https://weaviate.io/developers/weaviate 
• Pinecone https://docs.pinecone.io/ 
• Turbopuffer https://turbopuffer.com/docs 
• Cohere Rerank https://docs.cohere.com/docs/rerank 
• Voyage AI Rerank https://docs.voyageai.com/docs/reranker 

1.11 Model Gateway / Router

• LiteLLM https://docs.litellm.ai/ 
• Portkey https://portkey.ai/docs 
• OpenRouter https://openrouter.ai/docs 

1.12 Guardrails

• Meta Llama Guard / Prompt Guard https://www.llama.com/docs/model-cards-and-prompt-formats/llama-guard-3/ 
• NVIDIA NeMo Guardrails https://docs.nvidia.com/nemo/guardrails/latest/ 
• Microsoft Presidio（PII） https://microsoft.github.io/presidio/ 
• ProtectAI Rebuff https://github.com/protectai/rebuff 

1.13 Security / Policy

• OWASP Top 10 for LLM Applications (2025) https://genai.owasp.org/llm-top-10/ 
• Open Policy Agent (OPA) https://www.openpolicyagent.org/docs/ 
• AWS Cedar Policy Language https://docs.cedarpolicy.com/ 

2. 经典论文

2.1 Agent / Planning

• ReAct: Synergizing Reasoning and Acting in Language Models https://arxiv.org/abs/2210.03629 
• Toolformer: Language Models Can Teach Themselves to Use Tools https://arxiv.org/abs/2302.04761 
• Tree of Thoughts https://arxiv.org/abs/2305.10601 
• Reflexion: Language Agents with Verbal Reinforcement Learning https://arxiv.org/abs/2303.11366 
• Voyager https://arxiv.org/abs/2305.16291 
• Language Agent Tree Search (LATS) https://arxiv.org/abs/2310.04406 
• Generative Agents: Interactive Simulacra of Human Behavior https://arxiv.org/abs/2304.03442 

2.2 RAG

• Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks https://arxiv.org/abs/2005.11401 
• HyDE: Precise Zero-Shot Dense Retrieval without Relevance Labels https://arxiv.org/abs/2212.10496 
• Lost in the Middle: How Language Models Use Long Contexts https://arxiv.org/abs/2307.03172 
• Microsoft GraphRAG https://arxiv.org/abs/2404.16130 

2.3 Security

• Prompt Injection attack against LLM-integrated Applications https://arxiv.org/abs/2306.05499 
• Indirect Prompt Injection https://arxiv.org/abs/2302.12173 
• CaMeL: Defeating Prompt Injections by Design (Google DeepMind, 2025) https://arxiv.org/abs/2503.18813 

3. 重要工程博客与分享

3.1 Anthropic

• Building Effective Agents https://www.anthropic.com/research/building-effective-agents 
• Contextual Retrieval https://www.anthropic.com/news/contextual-retrieval 
• How We Built Claude Code https://www.anthropic.com/news/claude-code 
• Engineering Blog https://www.anthropic.com/engineering 

3.2 LangChain

• Making it easier to build human-in-the-loop agents with interrupt https://blog.langchain.com/making-it-easier-to-build-human-in-the-loop-agents-with-interrupt/ 
• Deep Agents https://blog.langchain.com/deep-agents/ 
• LangGraph Blog https://blog.langchain.com/tag/langgraph/ 

3.3 其他值得读

• Cognition: Don’t Build Multi-Agents https://cognition.ai/blog/dont-build-multi-agents 
• Chip Huyen: AI Agents https://huyenchip.com/2025/01/07/agents.html 
• Simon Willison on Prompt Injection（持续更新） https://simonwillison.net/tags/prompt-injection/ 

4. 如何使用这些参考资料

4.1 学习路径建议

1. 先读本资料夹里的中文文档，建立完整知识地图
2. 对你最关心的模块，回到官方文档精读对应章节
3. LangChain / LangGraph / MCP / AG-UI / OTel GenAI 五个协议/框架建议直接读官方（API 变化快）
4. 重要论文读摘要 + 结论，不必逐字精读；真正要做的再细读

4.2 工程落地时的查阅优先级

碰到具体问题时，按这个优先级查：

1. 官方文档（LangChain/LangGraph/Anthropic/OpenAI/MCP）——最权威但更新快
2. 官方 changelog / release notes——找最近的 breaking changes
3. GitHub examples 仓库——例如 @langchain/langgraph 的 examples/ 目录
4. 工程博客（Anthropic、LangChain、Cognition）——看真实生产的坑
5. 论文——想深入理解某个模式时

4.3 版本敏感性提醒

下列技术当前（2026-04）演进很快，实际使用前一定对照最新文档：

• LangChain JS createAgent / middleware 的精确签名
• LangGraph JS 的 StateSchema / MessagesValue / ReducedValue API
• MCP 规范（下一版本可能替换 2025-06-18）
• OpenTelemetry GenAI Semantic Conventions（当前 experimental，stable 版本发布后字段名可能调整）
• Anthropic / OpenAI 的模型版本号和定价（至少每季度更新）
• AG-UI 事件 schema
• 各家 rerank / embedding 模型的最佳版本

规则：涉及具体代码和配置，永远以当前的官方文档为准。