{
  "title": "智能体基础",
  "schemaVersion": 2,
  "deckId": "deck_e2caa38af7cf4dc79a3fdfaacdc86fc0",
  "sharedGroups": [],
  "linkedGroups": [],
  "gitee": {
    "name": "agentgainianjijichukaifa",
    "fullName": "wangjingyi/agentgainianjijichukaifa",
    "htmlUrl": "https://gitee.com/wangjingyi/agentgainianjijichukaifa.git",
    "cloneUrl": "",
    "sshUrl": "https://gitee.com/wangjingyi/agentgainianjijichukaifa.git"
  },
  "slides": [
    {
      "file": "slide001.html",
      "title": "智能体基础",
      "slide_type": "cover",
      "id": "slide_73b1cbbb3f2248c49a7f4949bd7a96ff"
    },
    {
      "file": "slide002.html",
      "title": "课程目录",
      "slide_type": "catalog",
      "id": "slide_46e279b35f89411aa5482e1b871a1347"
    },
    {
      "file": "slide003.html",
      "title": "PART 01 · 智能体基础——从问答到自主循环",
      "slide_type": "chapter",
      "id": "slide_b44522a5bab342f698b29324f2943481"
    },
    {
      "file": "slide004.html",
      "title": "1.1 LLM：问答对话",
      "slide_type": "content",
      "id": "slide_d429844648794546925e72b0d731a3a3"
    },
    {
      "file": "slide005.html",
      "title": "1.2 LLM + 工具：让模型成为可用工具（12306 现场运行）",
      "slide_type": "content",
      "id": "slide_c346c7e7a2234792b4072d1393c32241"
    },
    {
      "file": "slide006.html",
      "title": "1.3 初代「智能体」= 工作流（DIFY / 扣子 / n8n）",
      "slide_type": "content",
      "id": "slide_b7ea7b95551d4494bcf93ef814e371f5"
    },
    {
      "file": "slide007.html",
      "title": "1.4 AI Agent：自主循环（规划 → 执行 → 观察 → 再规划）",
      "slide_type": "content",
      "id": "slide_aa2e80b8bb7642c293bb963c19553bb6"
    },
    {
      "file": "slide008.html",
      "title": "1.5 编程 Agent：ClaudeCode / Codex（范围：项目目录）",
      "slide_type": "content",
      "id": "slide_c451e5d2139441caa73e2c471885c9d7"
    },
    {
      "file": "slide009.html",
      "title": "1.6 桌面 Agent：OpenClaw / HermesAgent（范围：整个磁盘）",
      "slide_type": "content",
      "id": "slide_452ed8966de94d52994680a1cc00abca"
    },
    {
      "file": "slide010.html",
      "title": "PART 02 · Agent 案例展示",
      "slide_type": "chapter",
      "id": "slide_c2_chapter_cases_2026"
    },
    {
      "file": "slide011.html",
      "title": "2.1 智能运维 Agent（教学平台）",
      "slide_type": "content",
      "id": "slide_c201_ops_agent_2026"
    },
    {
      "file": "slide012.html",
      "title": "2.2 论文审核 Agent",
      "slide_type": "content",
      "id": "slide_c202_paper_review_agent_2026"
    },
    {
      "file": "slide013.html",
      "title": "2.3 专利评审 Agent",
      "slide_type": "content",
      "id": "slide_c203_patent_review_agent_2026"
    },
    {
      "file": "slide014.html",
      "title": "2.4 作业检查 Agent",
      "slide_type": "content",
      "id": "slide_c204_homework_check_agent_2026"
    },
    {
      "file": "slide015.html",
      "title": "2.5 桌面 Agent：HermesAgent",
      "slide_type": "content",
      "id": "slide_c205_hermes_agent_2026"
    },
    {
      "file": "slide016.html",
      "title": "PART 03 · 智能体风险——能力越大，风险越大",
      "slide_type": "chapter",
      "id": "slide_1b3396e9cdd546b5b877eec51dfe1a15"
    },
    {
      "file": "slide017.html",
      "title": "3.1 养龙虾热潮：26 万 / QQ 接入 / 深圳百万补贴",
      "slide_type": "content",
      "id": "slide_90edae9fa6e941198e6f1de0fa4ebf79"
    },
    {
      "file": "slide018.html",
      "title": "3.2 监管响应时间线",
      "slide_type": "content",
      "id": "slide_1af1c82713ad486581145eb23dbc3411"
    },
    {
      "file": "slide019.html",
      "title": "3.3 权限过高：误删 / 误发 / 拦不住（Meta 安全词实验）",
      "slide_type": "content",
      "id": "slide_f2d26fbd57d243dba64981727614002d"
    },
    {
      "file": "slide020.html",
      "title": "3.4 OpenClaw 高危漏洞概览（2026.1–3 集中披露）",
      "slide_type": "content",
      "id": "slide_c571dcb5e215442a8600d6ae095612d4"
    },
    {
      "file": "slide021.html",
      "title": "3.5 提示词注入：OpenClaw 官方预警点名第一类风险",
      "slide_type": "content",
      "id": "slide_e1bdc12b40d94f838f61ea252169e43e"
    },
    {
      "file": "slide022.html",
      "title": "3.6 内部开发投毒：北大博士 × 字节 800 万案",
      "slide_type": "content",
      "id": "slide_905551c480a34489a0bd01b8cdc60846"
    },
    {
      "file": "slide023.html",
      "title": "3.7 什么叫 Skill 投毒？ClawHub 341 个恶意技能",
      "slide_type": "content",
      "id": "slide_69ed69ef573c4241ac3cc29ed9650649"
    },
    {
      "file": "slide024.html",
      "title": "3.8 RAG 是什么：知识库与问答交互",
      "slide_type": "content",
      "id": "slide_b1c2d3e4f5a6b7c8d9e0f1a2b3c4d5e6"
    },
    {
      "file": "slide025.html",
      "title": "3.9 RAG 投毒的五个位置",
      "slide_type": "content",
      "id": "slide_f67500c8efdc47e3abfaee7f7577426d"
    },
    {
      "file": "slide026.html",
      "title": "3.10 数据与隐私泄漏：三星源码 + AI 志愿",
      "slide_type": "content",
      "id": "slide_ac20ec16c5b84fdd9bb0038e14484075"
    },
    {
      "file": "slide027.html",
      "title": "3.11 幻觉放大：新京报 2026.7 实测 6 款 AI 志愿",
      "slide_type": "content",
      "id": "slide_1c9dd7d839d84f1bb4a580d62018d88e"
    },
    {
      "file": "slide028.html",
      "title": "3.12 输出内容风险：周鸿祎手表事件",
      "slide_type": "content",
      "id": "slide_a7a0d3fb95a64852940f63e662ce124c"
    },
    {
      "file": "slide029.html",
      "title": "3.13 能力、权限与风险对应关系",
      "slide_type": "content",
      "id": "slide_abed77101202456ca16e6c6be22e008f"
    },
    {
      "file": "slide030.html",
      "title": "PART 04 · Agent 实现机制与架构实战",
      "slide_type": "chapter",
      "id": "slide_18d97acf2d894117b02db0bab5f49228"
    },
    {
      "file": "slide031.html",
      "title": "4.1 论文审核：需求与困境",
      "slide_type": "content",
      "id": "slide_54d3e23751ee475383bbf11d006bf4b6"
    },
    {
      "file": "slide033.html",
      "title": "4.2 论文审核 Agent：实现架构",
      "slide_type": "content",
      "id": "slide_c300_review_agent_architecture_2026"
    },
    {
      "file": "slide034.html",
      "title": "4.3 Agent = 目标驱动的决策循环",
      "slide_type": "content",
      "id": "slide_8267304fcddc40a1a0465d4e30ca43e0"
    },
    {
      "file": "slide035.html",
      "title": "4.4 Agent 运行架构",
      "slide_type": "content",
      "id": "slide_b0cce416384e46c383177d9b79a60e2d"
    },
    {
      "file": "slide036.html",
      "title": "4.5 执行轨迹 = 可观察的 Agent Loop",
      "slide_type": "content",
      "id": "slide_60d88153a3ad4629a97b197c88e7775d"
    },
    {
      "file": "slide037.html",
      "title": "4.6 短期记忆 vs 长期记忆",
      "slide_type": "content",
      "id": "slide_63f3b3a6c54d4de8b5cb55732f23d697"
    },
    {
      "file": "slide038.html",
      "title": "4.7 记忆实现方式",
      "slide_type": "content",
      "id": "slide_c301_memory_pipeline_2026"
    },
    {
      "file": "slide048.html",
      "title": "4.8 Tool、MCP 与 Skill",
      "slide_type": "content",
      "id": "slide_c303_tool_mcp_skill_2026"
    },
    {
      "file": "slide051.html",
      "title": "4.9 MCP 调用链：用外部证据核验文献",
      "slide_type": "content",
      "id": "slide_c306_mcp_chain_2026"
    },
    {
      "file": "slide052.html",
      "title": "4.10 Skill：可复用的任务方法包",
      "slide_type": "content",
      "id": "slide_c307_skill_package_2026"
    },
    {
      "file": "slide053.html",
      "title": "4.11 两个 Skill，各自保持专业边界",
      "slide_type": "content",
      "id": "slide_c308_two_skills_2026"
    },
    {
      "file": "slide054.html",
      "title": "4.12 单 Agent 与多 Agent：架构拆分条件",
      "slide_type": "content",
      "id": "slide_c310_split_conditions_2026"
    },
    {
      "file": "slide055.html",
      "title": "4.13 多 Agent 协作：角色与传递协议",
      "slide_type": "content",
      "id": "slide_c312_collaboration_2026"
    },
    {
      "file": "slide056.html",
      "title": "PART 05 · Agent 实现",
      "slide_type": "chapter",
      "id": "slide_part04_agent_impl_2026"
    },
    {
      "file": "slide058.html",
      "title": "5.1 多模态",
      "slide_type": "content",
      "id": "slide_c402_multimodal_2026"
    },
    {
      "file": "slide059.html",
      "title": "5.2 五类开发组件",
      "slide_type": "content",
      "id": "slide_c404_dev_components_2026"
    },
    {
      "file": "slide060.html",
      "title": "5.3 开发路线选择",
      "slide_type": "content",
      "id": "slide_c405_dev_route_2026"
    },
    {
      "file": "slide061.html",
      "title": "5.4 OpenAI Agents SDK",
      "slide_type": "content",
      "id": "slide_c406_openai_sdk_2026"
    },
    {
      "file": "slide062.html",
      "title": "5.5 Claude Agent SDK",
      "slide_type": "content",
      "id": "slide_c407_claude_sdk_2026"
    },
    {
      "file": "slide063.html",
      "title": "5.6 Pi 与 LangGraph",
      "slide_type": "content",
      "id": "slide_c408_pi_langgraph_2026"
    },
    {
      "file": "slide064.html",
      "title": "智能体上线底线：可限制 · 可解释 · 可恢复",
      "slide_type": "finish",
      "id": "slide_finish_control_explain_recover_2026"
    }
  ]
}