[{"data":1,"prerenderedAt":61},["ShallowReactive",2],{"article-other\u002Fagent_compose":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"tags":11,"body":13,"_type":55,"_id":56,"_source":57,"_file":58,"_stem":59,"_extension":60},"\u002Farticles\u002Fother\u002Fagent_compose","other",false,"","LLM Agent 的核心组成：LLM 与 Harness","用 2026 年的工程语境理解 Agent：LLM 负责决策，Harness 负责上下文、工具、状态与记忆、权限控制及 Agent Loop 的运行执行。","2026-09-05",[12],"随笔",{"type":14,"children":15,"toc":50},"root",[16,24,30,35,40,45],{"type":17,"tag":18,"props":19,"children":21},"element","h2",{"id":20},"面试题",[22],{"type":23,"value":20},"text",{"type":17,"tag":25,"props":26,"children":27},"p",{},[28],{"type":23,"value":29},"你如何定义一个基于 LLM 的智能体（Agent）？它通常由哪些核心组件构成？",{"type":17,"tag":18,"props":31,"children":33},{"id":32},"组成",[34],{"type":23,"value":32},{"type":17,"tag":25,"props":36,"children":37},{},[38],{"type":23,"value":39},"站在2026年的工程语境来看，Agent = LLM + Harness。",{"type":17,"tag":25,"props":41,"children":42},{},[43],{"type":23,"value":44},"LLM是决策中心，负责理解目标、推理并生成下一步行动。",{"type":17,"tag":25,"props":46,"children":47},{},[48],{"type":23,"value":49},"Harness 是运行和执行系统，负责上下文组织(Context Engineering)、工具(tool)注册与执行、状态和记忆管理、权限控制，以及驱动 Agent Loop。",{"title":7,"searchDepth":51,"depth":51,"links":52},2,[53,54],{"id":20,"depth":51,"text":20},{"id":32,"depth":51,"text":32},"markdown","content:articles:other:agent_compose.md","content","articles\u002Fother\u002Fagent_compose.md","articles\u002Fother\u002Fagent_compose","md",1789465114556]