- 官方上下文与会话主轴：OpenAI
Conversation state/Text generation、AnthropicMessages API/Define tools/Context windows、GoogleText generation、MicrosoftWork with chat completion models/Compaction - 检索、排序与压缩主轴：
RAG、Dense Passage Retrieval、Retrieval meets Long Context LLMs、RankRAG、RECOMP、Searching for Best Practices in RAG - 长上下文与位置偏置主轴：
Lost in the Middle、Context Length Alone Hurts LLM Performance Despite Perfect Retrieval、LongLLMLingua - 文档解析与结构化输入主轴：
Trafilatura、DocLLM、GoogleParse and chunk documents、MicrosoftBuild advanced retrieval-augmented generation systems、AnthropicIntroducing Contextual Retrieval - 企业 observation pipeline 主轴：Slack
Enterprise search、SourcegraphHow Cody understands your codebase/Context Retrieval and Evaluation、AnthropicEffective context engineering for AI agents - 输入安全与污染主轴：
Indirect Prompt Injection、Formalizing and Benchmarking Prompt Injection Attacks and Defenses、AgentPoison、NCSCPrompt injection is not SQL injection
附录 G：分章资料清单