{"id":115510,"date":"2026-10-12T10:32:54","date_gmt":"2026-10-12T02:32:54","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/115510.html"},"modified":"2026-10-12T10:32:54","modified_gmt":"2026-10-12T02:32:54","slug":"%e7%ac%ac%e4%ba%8c%e7%ab%a0agent-%e5%bc%80%e5%8f%91%e7%8e%af%e5%a2%83%e6%90%ad%e5%bb%ba%e4%b8%8e%e9%aa%8c%e8%af%81","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/115510.html","title":{"rendered":"\u7b2c\u4e8c\u7ae0Agent \u5f00\u53d1\u73af\u5883\u642d\u5efa\u4e0e\u9a8c\u8bc1"},"content":{"rendered":"

\u6211\u4eec\u7684\u76ee\u6807\u662f\u8ba9\u5f00\u53d1\u73af\u5883\u8fbe\u5230"\u80fd\u8dd1\u901a\u4e00\u4e2a\u6700\u5c0f Agent"\u7684\u72b6\u6001：\u5b89\u88c5\u4f9d\u8d56\u3001\u914d\u7f6e\u6a21\u578b\u3001\u5c01\u88c5\u5de5\u5177\u3001\u8fd0\u884c\u8c03\u7528\u3001\u4fdd\u5b58\u8f68\u8ff9\u3001\u9a8c\u8bc1\u884c\u4e3a\u3002\u540c\u65f6，\u6240\u6709\u793a\u4f8b\u4ee3\u7801\u90fd\u5c3d\u91cf\u8865\u6210\u53ef\u76f4\u63a5\u62f7\u8d1d\u8fd0\u884c\u7684\u7248\u672c，\u5e76\u5728\u9700\u8981\u7279\u6b8a\u914d\u7f6e\u7684\u5730\u65b9\u7ed9\u51fa\u8fd0\u884c\u65b9\u5f0f\u8bf4\u660e\u3002<\/p>\n

\n

\u4e00\u3001\u73af\u5883\u642d\u5efa\u524d\u7684\u4e09\u4e2a\u9009\u62e9<\/h4>\n

\u5728\u52a8\u624b\u4e4b\u524d，\u9700\u8981\u5148\u786e\u5b9a\u4e09\u4ef6\u4e8b：<\/p>\n\n\u9009\u62e9\u9879\u8bf4\u660e\u5efa\u8bae<\/tr>\n\n\n\n\n
\u5f00\u53d1\u8bed\u8a00<\/td>\n\u76ee\u524d Python \u751f\u6001\u6700\u6210\u719f<\/td>\n\u521d\u5b66\u8005\u4f18\u5148 Python<\/td>\n<\/tr>\n
\u6a21\u578b\u63a5\u5165<\/td>\n\u51b3\u5b9a Agent \u7684\u63a8\u7406\u80fd\u529b<\/td>\n\u5148\u7528\u4e00\u4e2a\u7a33\u5b9a\u6a21\u578b\u8dd1\u901a<\/td>\n<\/tr>\n
\u6846\u67b6\u5c42\u7ea7<\/td>\n\u51b3\u5b9a\u540e\u7eed\u6269\u5c55\u590d\u6742\u5ea6<\/td>\n\u5148\u8f7b\u91cf，\u518d\u9010\u6b65\u5347\u7ea7<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

\u6846\u67b6\u5c42\u9762，2026 \u5e74\u7684\u9009\u62e9\u5df2\u7ecf\u6bd4\u8f83\u6e05\u6670：<\/p>\n

    \n
  • LangChain \/ LangGraph：\u9002\u5408\u6784\u5efa\u5de5\u5177\u8c03\u7528\u3001RAG\u3001\u591a\u6b65\u5de5\u4f5c\u6d41，\u793e\u533a\u751f\u6001\u6210\u719f\u3002<\/li>\n
  • CrewAI：\u9002\u5408\u5feb\u901f\u642d\u5efa\u591a\u89d2\u8272 Agent \u539f\u578b，\u4f46\u590d\u6742\u72b6\u6001\u7ba1\u7406\u8f83\u5f31\u3002<\/li>\n
  • LlamaIndex Workflows：\u9002\u5408\u77e5\u8bc6\u5e93\u3001\u6587\u6863\u68c0\u7d22\u3001\u5408\u89c4\u5ba1\u67e5\u7b49\u6570\u636e\u5bc6\u96c6\u578b\u573a\u666f\u3002<\/li>\n
  • Microsoft Agent Framework：\u9002\u5408 .NET \/ Azure \/ M365 \u751f\u6001\u3002<\/li>\n
  • OpenAI Agents SDK \/ Claude Agent SDK：\u9002\u5408\u7ed1\u5b9a\u5355\u4e00\u6a21\u578b\u6808\u3001\u903b\u8f91\u76f8\u5bf9\u7b80\u5355\u7684 Agent\u3002<\/li>\n
  • Dify \/ Coze：\u9002\u5408\u5feb\u901f\u9a8c\u8bc1\u3001\u4f4e\u4ee3\u7801\u642d\u5efa，\u4f46\u5b9a\u5236\u80fd\u529b\u548c\u6570\u636e\u63a7\u5236\u76f8\u5bf9\u53d7\u9650\u3002<\/li>\n<\/ul>\n

    \u5bf9\u6211\u4eec\u6765\u8bf4，\u5efa\u8bae\u5148\u7528\u00a0Python + LangChain\u00a0\u8dd1\u901a\u6700\u5c0f\u6d41\u7a0b；\u5982\u679c\u8981\u8fdb\u5165\u751f\u4ea7\u7ea7\u590d\u6742\u5de5\u4f5c\u6d41，\u518d\u8003\u8651\u8fc1\u79fb\u5230\u00a0LangGraph\u3002<\/p>\n

    \n

    \u4e8c\u3001Python \u57fa\u7840\u73af\u5883\u914d\u7f6e<\/h4>\n

    1. Python \u7248\u672c<\/h5>\n

    \u63a8\u8350\u5b89\u88c5\u00a0Python 3.10 ~ 3.12\u3002
    \nPython 3.10 \u4ee5\u4e0a\u7248\u672c\u5bf9\u7c7b\u578b\u63d0\u793a\u3001\u5f02\u6b65\u652f\u6301\u548c\u4e3b\u6d41 Agent \u6846\u67b6\u517c\u5bb9\u6027\u66f4\u597d\u3002<\/p>\n

    \u5b89\u88c5\u65f6\u6ce8\u610f\u52fe\u9009：<\/p>\n

    1Add Python to PATH<\/p>\n

    \u5b89\u88c5\u5b8c\u6210\u540e\u9a8c\u8bc1：<\/p>\n

    1python –version
    \n2pip –version<\/p>\n

    2. \u865a\u62df\u73af\u5883<\/h5>\n

    Agent \u9879\u76ee\u901a\u5e38\u4f1a\u5b89\u88c5\u8f83\u591a\u4f9d\u8d56，\u5efa\u8bae\u4f7f\u7528\u865a\u62df\u73af\u5883\u9694\u79bb，\u907f\u514d\u6c61\u67d3\u5168\u5c40 Python\u3002<\/p>\n

    1python -m venv .venv
    \n2
    \n3# Windows PowerShell
    \n4.venv\\\\Scripts\\\\Activate.ps1
    \n5
    \n6# macOS \/ Linux
    \n7source .venv\/bin\/activate<\/p>\n

    \u6fc0\u6d3b\u6210\u529f\u540e，\u7ec8\u7aef\u524d\u9762\u901a\u5e38\u4f1a\u51fa\u73b0\u00a0(.venv)\u3002<\/p>\n

    \u5982\u679c PowerShell \u63d0\u793a\u65e0\u6cd5\u8fd0\u884c\u811a\u672c，\u53ef\u4ee5\u5148\u6267\u884c：<\/p>\n

    1Set-ExecutionPolicy -Scope Process -ExecutionPolicy RemoteSigned<\/p>\n

    \u7136\u540e\u518d\u6fc0\u6d3b\u73af\u5883\u3002<\/p>\n

    3. \u865a\u62df\u73af\u5883\u5de5\u5177\u600e\u4e48\u9009<\/h5>\n\n\u5de5\u5177\u7279\u70b9\u9002\u5408\u573a\u666f<\/tr>\n\n\n\n\n\n\n
    venv<\/td>\nPython \u5185\u7f6e，\u8f7b\u91cf<\/td>\n\u65b0\u624b\u3001\u811a\u672c\u3001\u5c0f\u578b\u9879\u76ee<\/td>\n<\/tr>\n
    conda<\/td>\n\u53ef\u7ba1\u7406 Python \u7248\u672c\u548c\u975e Python \u4f9d\u8d56<\/td>\n\u6570\u636e\u79d1\u5b66\u3001\u673a\u5668\u5b66\u4e60<\/td>\n<\/tr>\n
    poetry<\/td>\n\u4f9d\u8d56\u9501\u5b9a\u3001\u6253\u5305\u53d1\u5e03\u80fd\u529b\u5f3a<\/td>\n\u56e2\u961f\u534f\u4f5c\u3001\u5f00\u6e90\u9879\u76ee<\/td>\n<\/tr>\n
    uv<\/td>\nRust \u7f16\u5199，\u5b89\u88c5\u548c\u89e3\u6790\u901f\u5ea6\u5f88\u5feb<\/td>\n\u8ffd\u6c42\u6548\u7387\u3001CI\/CD\u3001\u5927\u578b\u9879\u76ee<\/td>\n<\/tr>\n
    pipenv<\/td>\n\u66fe\u7ecf\u6d41\u884c，\u76ee\u524d\u65b0\u9879\u76ee\u4e0d\u4f18\u5148\u63a8\u8350<\/td>\n\u7ef4\u62a4\u8001\u9879\u76ee<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

    \u5efa\u8bae：<\/p>\n

      \n
    • \u65b0\u624b\u5165\u95e8：venv + pip<\/li>\n
    • \u6570\u636e\u79d1\u5b66 \/ \u673a\u5668\u5b66\u4e60\u9879\u76ee：conda<\/li>\n
    • \u56e2\u961f\u534f\u4f5c\u6216\u9700\u8981\u53d1\u5e03\u5305：poetry<\/li>\n
    • \u8ffd\u6c42\u5b89\u88c5\u901f\u5ea6\u548c\u4f9d\u8d56\u9501\u5b9a：uv<\/li>\n<\/ul>\n
      4. pip \u955c\u50cf\u6e90<\/h5>\n

      \u5982\u679c\u4e0b\u8f7d\u4f9d\u8d56\u8f83\u6162，\u53ef\u4ee5\u4e34\u65f6\u4f7f\u7528\u56fd\u5185\u955c\u50cf\u6e90，\u4f8b\u5982\u6e05\u534e\u6e90：<\/p>\n

      1pip install langchain langchain-openai python-dotenv -i https:\/\/pypi.tuna.tsinghua.edu.cn\/simple<\/p>\n

      \u6ce8\u610f\u9009\u62e9\u53ef\u4fe1\u955c\u50cf\u6e90，\u907f\u514d\u5b89\u88c5\u6765\u6e90\u4e0d\u660e\u7684\u5305\u3002<\/p>\n

      \n

      \u4e09\u3001\u5f00\u53d1\u5de5\u5177\u5efa\u8bae<\/h4>\n

      1. \u4ee3\u7801\u7f16\u8f91\u5668 \/ IDE<\/h5>\n\n\u5de5\u5177\u7279\u70b9\u9002\u5408\u4eba\u7fa4<\/tr>\n\n\n\n\n\n\n\n\n\n
      VS Code<\/td>\n\u514d\u8d39\u3001\u8f7b\u91cf\u3001\u63d2\u4ef6\u751f\u6001\u4e30\u5bcc<\/td>\n\u901a\u7528\u9996\u9009<\/td>\n<\/tr>\n
      PyCharm<\/td>\nPython \u4e13\u4e1a IDE，\u8c03\u8bd5\u80fd\u529b\u5f3a<\/td>\n\u5927\u578b\u9879\u76ee\u3001\u56e2\u961f\u534f\u4f5c<\/td>\n<\/tr>\n
      JupyterLab<\/td>\n\u4ea4\u4e92\u5f0f\u7f16\u7a0b，\u9002\u5408\u6570\u636e\u63a2\u7d22<\/td>\n\u6570\u636e\u5206\u6790\u3001AI \u5b9e\u9a8c<\/td>\n<\/tr>\n
      Cursor<\/td>\nAI \u6df1\u5ea6\u96c6\u6210，\u652f\u6301\u8de8\u6587\u4ef6\u91cd\u6784<\/td>\n\u8ffd\u6c42\u6548\u7387\u7684\u5f00\u53d1\u8005<\/td>\n<\/tr>\n
      Trae<\/td>\n\u4e2d\u6587\u53cb\u597d，\u57fa\u7840\u7248\u514d\u8d39<\/td>\n\u56fd\u5185\u5f00\u53d1\u8005\u3001\u5feb\u901f\u539f\u578b<\/td>\n<\/tr>\n
      Windsurf<\/td>\n\u5168\u5c40\u4e0a\u4e0b\u6587\u611f\u77e5\u5f3a<\/td>\n\u590d\u6742\u9879\u76ee\u91cd\u6784<\/td>\n<\/tr>\n
      Qoder<\/td>\n\u9002\u914d\u540e\u7aef\u5de5\u7a0b\u5316<\/td>\n\u4f01\u4e1a\u7ea7\u540e\u7aef\u5f00\u53d1<\/td>\n<\/tr>\n
      CodeBuddy<\/td>\n\u524d\u7aef\u3001\u5c0f\u7a0b\u5e8f\u3001H5 \u9002\u914d\u8f83\u597d<\/td>\n\u524d\u7aef\u4e0e\u5fae\u4fe1\u751f\u6001\u5f00\u53d1<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

      \u63a8\u8350\u7ec4\u5408：<\/p>\n

        \n
      • \u65b0\u624b \/ \u901a\u7528\u5f00\u53d1：VS Code + Python + Pylance + \u4e2d\u6587\u8bed\u8a00\u5305\u3002<\/li>\n
      • \u5927\u578b Python \u9879\u76ee：PyCharm \u4e13\u4e1a\u7248\u3002<\/li>\n
      • AI \u8f85\u52a9\u5f00\u53d1：Cursor\u3001Trae \u6216 Windsurf\u3002<\/li>\n
      • \u6570\u636e\u63a2\u7d22：JupyterLab + VS Code\u3002<\/li>\n<\/ul>\n

        VS Code \u5fc5\u88c5\u63d2\u4ef6\u5efa\u8bae：<\/p>\n

          \n
        • Python<\/li>\n
        • Pylance<\/li>\n
        • Chinese (Simplified) Language Pack<\/li>\n
        • GitLens（\u53ef\u9009）<\/li>\n
        • Jupyter（\u53ef\u9009）<\/li>\n<\/ul>\n
          2. AI \u7f16\u7a0b\u8f85\u52a9\u5de5\u5177<\/h5>\n

          AI \u7f16\u7a0b\u5de5\u5177\u53ef\u4ee5\u5206\u4e3a\u4e09\u7c7b：<\/p>\n

            \n
          • AI \u539f\u751f IDE：Cursor\u3001Windsurf\u3001Trae\u3001Qoder\u3001CodeBuddy\u3001Devin Desktop \u7b49\u3002<\/li>\n
          • AI \u63d2\u4ef6：GitHub Copilot\u3001Tabnine\u3001\u901a\u4e49\u7075\u7801\u3001Kimi Code\u3001Cline \u7b49\u3002<\/li>\n
          • \u4ee3\u7801\u5b89\u5168\u4e0e\u5ba1\u67e5\u5de5\u5177：Sourcery\u3001Snyk Code \u7b49\u3002<\/li>\n<\/ul>\n

            \u9009\u578b\u5efa\u8bae：<\/p>\n

              \n
            • \u60f3\u8981\u00a0\u7a33\u5b9a\u8865\u5168 + \u591a IDE \u652f\u6301：GitHub Copilot\u3002<\/li>\n
            • \u60f3\u8981\u00a0\u8de8\u6587\u4ef6\u7406\u89e3 + Agent \u6a21\u5f0f：Cursor\u3002<\/li>\n
            • \u60f3\u8981\u00a0\u4e2d\u6587\u53cb\u597d + \u514d\u8d39\u4f53\u9a8c：Trae\u3002<\/li>\n
            • \u60f3\u8981\u00a0\u4f01\u4e1a\u7ea7\u540e\u7aef\u5de5\u7a0b\u5316：Qoder\u3002<\/li>\n
            • \u60f3\u8981\u00a0\u6570\u636e\u9690\u79c1\u4f18\u5148：Tabnine\u3002<\/li>\n<\/ul>\n

              \u5bf9 Agent \u5f00\u53d1\u6765\u8bf4，AI \u7f16\u7a0b\u5de5\u5177\u7684\u4ef7\u503c\u4e3b\u8981\u5728\u4e8e：\u5feb\u901f\u751f\u6210\u5de5\u5177\u51fd\u6570\u3001\u8865\u5168 Pydantic \u6a21\u578b\u3001\u89e3\u91ca\u62a5\u9519\u4fe1\u606f\u3001\u751f\u6210\u6d4b\u8bd5\u7528\u4f8b，\u4ee5\u53ca\u8f85\u52a9\u8c03\u8bd5\u591a\u6b65\u8c03\u7528\u6d41\u7a0b\u3002<\/p>\n

              \n

              \u56db\u3001\u6838\u5fc3\u4f9d\u8d56\u5b89\u88c5<\/h4>\n

              \u521b\u5efa\u5e76\u6fc0\u6d3b\u865a\u62df\u73af\u5883\u540e，\u5b89\u88c5\u57fa\u7840\u4f9d\u8d56：<\/p>\n

              1pip install langchain langchain-openai langchain-community python-dotenv<\/p>\n

              \u5982\u679c\u9700\u8981\u672c\u5730\u6a21\u578b，\u53ef\u4ee5\u7ee7\u7eed\u5b89\u88c5：<\/p>\n

              1pip install langchain-ollama<\/p>\n

              \u5982\u679c\u9700\u8981\u5411\u91cf\u6570\u636e\u5e93：<\/p>\n

              1pip install chromadb<\/p>\n

              \u8f85\u52a9\u5e93\u53ef\u6309\u9700\u5b89\u88c5：<\/p>\n

              1pip install pydantic requests rich tiktoken<\/p>\n

              \u9a8c\u8bc1\u73af\u5883：<\/p>\n

              1import sys
              \n2import langchain
              \n3
              \n4print(f"Python: {sys.version}")
              \n5print(f"LangChain: {langchain.__version__}")<\/p>\n

              \u5982\u679c\u6b63\u5e38\u8f93\u51fa\u7248\u672c\u53f7，\u8bf4\u660e\u57fa\u7840\u73af\u5883\u5df2\u7ecf\u5c31\u7eea\u3002<\/p>\n

              \n

              \u4e94\u3001\u6a21\u578b\u63a5\u5165\u914d\u7f6e<\/h4>\n

              Agent \u7684\u6838\u5fc3\u63a8\u7406\u80fd\u529b\u6765\u81ea\u5927\u6a21\u578b\u3002\u914d\u7f6e\u65f6\u4e0d\u8981\u628a API Key \u5199\u8fdb\u4ee3\u7801，\u63a8\u8350\u4f7f\u7528\u00a0.env\u00a0\u6587\u4ef6\u7ba1\u7406\u3002<\/p>\n

              \u521b\u5efa\u00a0.env：<\/p>\n

              1OPENAI_API_KEY=your_api_key_here
              \n2MODEL_NAME=gpt-4o
              \n3MODEL_BASE_URL=<\/p>\n

              \u5982\u679c\u4f7f\u7528\u7684\u662f\u517c\u5bb9 OpenAI \u63a5\u53e3\u7684\u6a21\u578b\u670d\u52a1，\u53ef\u4ee5\u586b\u5199：<\/p>\n

              1OPENAI_API_KEY=your_api_key_here
              \n2MODEL_NAME=qwen-plus
              \n3MODEL_BASE_URL=https:\/\/your-api-endpoint\/v1<\/p>\n

              \u5982\u679c\u00a0MODEL_BASE_URL\u00a0\u4e3a\u7a7a，\u4ee3\u7801\u4f1a\u81ea\u52a8\u4f7f\u7528 OpenAI \u9ed8\u8ba4\u5730\u5740\u3002<\/p>\n

              \u5728\u4ee3\u7801\u4e2d\u52a0\u8f7d：<\/p>\n

              1from dotenv import load_dotenv
              \n2import os
              \n3
              \n4load_dotenv()
              \n5
              \n6api_key = os.getenv("OPENAI_API_KEY")
              \n7model_name = os.getenv("MODEL_NAME", "gpt-4o")
              \n8base_url = os.getenv("MODEL_BASE_URL", "").strip()<\/p>\n

              \u521d\u59cb\u5316\u6a21\u578b：<\/p>\n

              1from langchain_openai import ChatOpenAI
              \n2
              \n3llm_kwargs = {
              \n4 "model": model_name,
              \n5 "temperature": 0.2,
              \n6 "api_key": api_key,
              \n7}
              \n8
              \n9if base_url:
              \n10 llm_kwargs["base_url"] = base_url
              \n11
              \n12llm = ChatOpenAI(**llm_kwargs)<\/p>\n

              \u5de5\u5177\u8c03\u7528\u573a\u666f\u5efa\u8bae\u628a\u00a0temperature\u00a0\u8bbe\u4f4e\u4e00\u4e9b，\u4f8b\u5982\u00a00\u00a0\u5230\u00a00.2，\u51cf\u5c11\u968f\u673a\u8f93\u51fa\u3002<\/p>\n

              \n

              \u516d\u3001\u6700\u5c0f Agent \u7ed3\u6784<\/h4>\n

              \u4e00\u4e2a\u6700\u5c0f Agent \u81f3\u5c11\u9700\u8981\u56db\u90e8\u5206：<\/p>\n

              1\u7528\u6237\u8f93\u5165
              \n2 -> Prompt \u5b9a\u4e49\u89d2\u8272\u4e0e\u76ee\u6807
              \n3 -> LLM \u63a8\u7406
              \n4 -> \u5de5\u5177\u8c03\u7528
              \n5 -> \u5de5\u5177\u8fd4\u56de\u7ed3\u679c
              \n6 -> LLM \u751f\u6210\u6700\u7ec8\u56de\u7b54<\/p>\n

              \u5148\u5b9a\u4e49\u4e00\u4e2a\u7b80\u5355\u5de5\u5177，\u4f8b\u5982\u5929\u6c14\u67e5\u8be2\u5de5\u5177：<\/p>\n

              1from langchain_core.tools import tool
              \n2
              \n3@tool
              \n4def get_weather(city: str) -> str:
              \n5 """\u67e5\u8be2\u6307\u5b9a\u57ce\u5e02\u7684\u5929\u6c14\u4fe1\u606f\u3002
              \n6
              \n7 \u53c2\u6570：
              \n8 – city: \u57ce\u5e02\u540d\u79f0，\u4f8b\u5982"\u90d1\u5dde""\u5317\u4eac""\u4e0a\u6d77"
              \n9 """
              \n10 if not city:
              \n11 return "\u7f3a\u5c11\u57ce\u5e02\u540d\u79f0，\u65e0\u6cd5\u67e5\u8be2\u5929\u6c14\u3002"
              \n12
              \n13 weather_data = {
              \n14 "\u90d1\u5dde": "\u6674，22\u00b0C，\u5fae\u98ce",
              \n15 "\u5317\u4eac": "\u591a\u4e91，18\u00b0C，\u5317\u98ce2\u7ea7",
              \n16 "\u4e0a\u6d77": "\u5c0f\u96e8，20\u00b0C，\u4e1c\u5357\u98ce3\u7ea7",
              \n17 }
              \n18
              \n19 return weather_data.get(city, f"\u672a\u627e\u5230 {city} \u7684\u5929\u6c14\u6570\u636e，\u8bf7\u786e\u8ba4\u57ce\u5e02\u540d\u79f0\u3002")<\/p>\n

              \u7136\u540e\u521b\u5efa Agent：<\/p>\n

              1from langchain.agents import create_tool_calling_agent, AgentExecutor
              \n2from langchain_core.prompts import ChatPromptTemplate
              \n3
              \n4tools = [get_weather]
              \n5
              \n6prompt = ChatPromptTemplate.from_messages([
              \n7 ("system", "\u4f60\u662f\u4e00\u4e2a\u667a\u80fd\u52a9\u624b，\u53ef\u4ee5\u6839\u636e\u7528\u6237\u9700\u6c42\u8c03\u7528\u5de5\u5177\u3002"),
              \n8 ("human", "{input}"),
              \n9 ("placeholder", "{agent_scratchpad}"),
              \n10])
              \n11
              \n12agent = create_tool_calling_agent(llm, tools, prompt)
              \n13
              \n14agent_executor = AgentExecutor(
              \n15 agent=agent,
              \n16 tools=tools,
              \n17 verbose=True,
              \n18 max_iterations=5,
              \n19 handle_parsing_errors=True
              \n20)<\/p>\n

              \u8fd0\u884c\u6d4b\u8bd5：<\/p>\n

              1result = agent_executor.invoke({"input": "\u90d1\u5dde\u4eca\u5929\u5929\u6c14\u600e\u4e48\u6837？"})
              \n2print(result)<\/p>\n

              verbose=True\u00a0\u53ef\u4ee5\u6253\u5370 Agent \u7684\u601d\u8003\u8fc7\u7a0b\u3001\u5de5\u5177\u8c03\u7528\u548c\u8fd4\u56de\u7ed3\u679c，\u975e\u5e38\u9002\u5408\u8c03\u8bd5\u3002<\/p>\n

              \n

              \u4e03\u3001\u5b8c\u6574\u53ef\u8fd0\u884c\u811a\u672c：agent_demo.py<\/h4>\n

              \u4e0b\u9762\u628a\u524d\u9762\u7684\u7247\u6bb5\u6574\u5408\u6210\u4e00\u4e2a\u5b8c\u6574\u811a\u672c，\u53ef\u4ee5\u76f4\u63a5\u62f7\u8d1d\u8fd0\u884c\u3002<\/p>\n

              1"""
              \n2Agent \u5f00\u53d1\u73af\u5883\u6700\u5c0f\u9a8c\u8bc1\u811a\u672c
              \n3
              \n4\u8fd0\u884c\u65b9\u5f0f：
              \n51. \u521b\u5efa\u865a\u62df\u73af\u5883：python -m venv .venv
              \n62. \u6fc0\u6d3b\u73af\u5883：
              \n7 – Windows PowerShell: .venv\\\\Scripts\\\\Activate.ps1
              \n8 – macOS \/ Linux: source .venv\/bin\/activate
              \n93. \u5b89\u88c5\u4f9d\u8d56：pip install langchain langchain-openai python-dotenv
              \n104. \u5728\u9879\u76ee\u6839\u76ee\u5f55\u521b\u5efa .env \u6587\u4ef6，\u5199\u5165 OPENAI_API_KEY
              \n115. \u8fd0\u884c：python agent_demo.py
              \n12"""
              \n13
              \n14import os
              \n15import json
              \n16import time
              \n17from datetime import datetime
              \n18from dotenv import load_dotenv
              \n19from langchain_openai import ChatOpenAI
              \n20from langchain_core.tools import tool
              \n21from langchain.agents import create_tool_calling_agent, AgentExecutor
              \n22from langchain_core.prompts import ChatPromptTemplate
              \n23
              \n24# —————————-
              \n25# 1. \u52a0\u8f7d\u73af\u5883\u53d8\u91cf
              \n26# —————————-
              \n27load_dotenv()
              \n28
              \n29OPENAI_API_KEY = os.getenv("OPENAI_API_KEY", "")
              \n30MODEL_NAME = os.getenv("MODEL_NAME", "gpt-4o")
              \n31MODEL_BASE_URL = os.getenv("MODEL_BASE_URL", "").strip()
              \n32
              \n33if not OPENAI_API_KEY:
              \n34 raise RuntimeError(
              \n35 "\u672a\u627e\u5230 OPENAI_API_KEY，\u8bf7\u5728\u9879\u76ee\u6839\u76ee\u5f55\u521b\u5efa .env \u6587\u4ef6\u5e76\u5199\u5165 API Key\u3002"
              \n36 )
              \n37
              \n38# —————————-
              \n39# 2. \u521d\u59cb\u5316\u6a21\u578b
              \n40# —————————-
              \n41llm_kwargs = {
              \n42 "model": MODEL_NAME,
              \n43 "temperature": 0.2,
              \n44 "api_key": OPENAI_API_KEY,
              \n45}
              \n46
              \n47if MODEL_BASE_URL:
              \n48 llm_kwargs["base_url"] = MODEL_BASE_URL
              \n49
              \n50llm = ChatOpenAI(**llm_kwargs)
              \n51
              \n52# —————————-
              \n53# 3. \u5b9a\u4e49\u5de5\u5177
              \n54# —————————-
              \n55
              \n56@tool
              \n57def get_weather(city: str) -> str:
              \n58 """\u67e5\u8be2\u6307\u5b9a\u57ce\u5e02\u7684\u5929\u6c14\u4fe1\u606f\u3002
              \n59
              \n60 \u53c2\u6570：
              \n61 – city: \u57ce\u5e02\u540d\u79f0，\u4f8b\u5982"\u90d1\u5dde""\u5317\u4eac""\u4e0a\u6d77"
              \n62 """
              \n63 if not city:
              \n64 return "\u7f3a\u5c11\u57ce\u5e02\u540d\u79f0，\u65e0\u6cd5\u67e5\u8be2\u5929\u6c14\u3002"
              \n65
              \n66 weather_data = {
              \n67 "\u90d1\u5dde": "\u6674，22\u00b0C，\u5fae\u98ce",
              \n68 "\u5317\u4eac": "\u591a\u4e91，18\u00b0C，\u5317\u98ce2\u7ea7",
              \n69 "\u4e0a\u6d77": "\u5c0f\u96e8，20\u00b0C，\u4e1c\u5357\u98ce3\u7ea7",
              \n70 }
              \n71
              \n72 return weather_data.get(city, f"\u672a\u627e\u5230 {city} \u7684\u5929\u6c14\u6570\u636e，\u8bf7\u786e\u8ba4\u57ce\u5e02\u540d\u79f0\u3002")
              \n73
              \n74
              \n75@tool
              \n76def get_current_time(dummy: str = "") -> str:
              \n77 """\u83b7\u53d6\u5f53\u524d\u65e5\u671f\u548c\u65f6\u95f4\u3002
              \n78
              \n79 \u53c2\u6570：
              \n80 – dummy: \u53ef\u9009\u5360\u4f4d\u53c2\u6570，\u4e0d\u9700\u8981\u4f20\u5165\u5b9e\u9645\u5185\u5bb9
              \n81 """
              \n82 now = datetime.now()
              \n83 return now.strftime("%Y\u5e74%m\u6708%d\u65e5 %H:%M:%S")
              \n84
              \n85
              \n86tools = [get_weather, get_current_time]
              \n87
              \n88# —————————-
              \n89# 4. \u521b\u5efa Agent
              \n90# —————————-
              \n91
              \n92prompt = ChatPromptTemplate.from_messages(
              \n93 [
              \n94 ("system", "\u4f60\u662f\u4e00\u4e2a\u667a\u80fd\u52a9\u624b，\u53ef\u4ee5\u6839\u636e\u7528\u6237\u9700\u6c42\u8c03\u7528\u5de5\u5177\u3002"),
              \n95 ("human", "{input}"),
              \n96 ("placeholder", "{agent_scratchpad}"),
              \n97 ]
              \n98)
              \n99
              \n100agent = create_tool_calling_agent(llm, tools, prompt)
              \n101
              \n102agent_executor = AgentExecutor(
              \n103 agent=agent,
              \n104 tools=tools,
              \n105 verbose=True,
              \n106 max_iterations=5,
              \n107 handle_parsing_errors=True,
              \n108)
              \n109
              \n110# —————————-
              \n111# 5. \u8fd0\u884c\u6d4b\u8bd5
              \n112# —————————-
              \n113
              \n114test_cases = [
              \n115 {"input": "\u90d1\u5dde\u4eca\u5929\u5929\u6c14\u600e\u4e48\u6837？"},
              \n116 {"input": "\u73b0\u5728\u51e0\u70b9\u4e86？"},
              \n117 {"input": "\u5317\u4eac\u548c\u4e0a\u6d77\u4eca\u5929\u7684\u5929\u6c14\u5206\u522b\u5982\u4f55？"},
              \n118]
              \n119
              \n120traces = []
              \n121
              \n122for case in test_cases:
              \n123 print("\\\\n" + "=" * 60)
              \n124 print(f"\u6d4b\u8bd5\u8f93\u5165：{case['input']}")
              \n125 print("=" * 60)
              \n126
              \n127 start_time = time.time()
              \n128 result = agent_executor.invoke(case)
              \n129 elapsed = round(time.time() – start_time, 2)
              \n130
              \n131 trace = {
              \n132 "input": case["input"],
              \n133 "output": result.get("output", ""),
              \n134 "elapsed_seconds": elapsed,
              \n135 "timestamp": datetime.now().isoformat(),
              \n136 }
              \n137
              \n138 traces.append(trace)
              \n139
              \n140 print("\\\\n\u6700\u7ec8\u56de\u7b54：")
              \n141 print(result.get("output", ""))
              \n142 print(f"\\\\n\u8017\u65f6：{elapsed} \u79d2")
              \n143
              \n144# —————————-
              \n145# 6. \u4fdd\u5b58\u8f68\u8ff9
              \n146# —————————-
              \n147
              \n148output_dir = "agent_traces"
              \n149os.makedirs(output_dir, exist_ok=True)
              \n150
              \n151output_file = os.path.join(output_dir, "traces.json")
              \n152
              \n153with open(output_file, "w", encoding="utf-8") as f:
              \n154 json.dump(traces, f, ensure_ascii=False, indent=2)
              \n155
              \n156print(f"\\\\n\u8f68\u8ff9\u5df2\u4fdd\u5b58\u5230：{output_file}")<\/p>\n

              \u8fd0\u884c\u65b9\u5f0f：<\/p>\n

              1python agent_demo.py<\/p>\n

              \n

              \u516b\u3001\u8f68\u8ff9\u68c0\u67e5\u811a\u672c：check_traces.py<\/h4>\n

              \u8fd9\u4e2a\u811a\u672c\u7528\u4e8e\u9a8c\u8bc1 Agent \u662f\u5426\u771f\u7684\u4ea7\u751f\u4e86\u8f93\u51fa，\u4ee5\u53ca\u8f93\u51fa\u662f\u5426\u5305\u542b\u9884\u671f\u4fe1\u606f\u3002<\/p>\n

              1"""
              \n2Agent \u8f68\u8ff9\u68c0\u67e5\u811a\u672c
              \n3
              \n4\u8fd0\u884c\u65b9\u5f0f：
              \n51. \u5148\u8fd0\u884c：python agent_demo.py
              \n62. \u518d\u8fd0\u884c：python check_traces.py
              \n7"""
              \n8
              \n9import json
              \n10import os
              \n11
              \n12TRACE_FILE = os.path.join("agent_traces", "traces.json")
              \n13
              \n14if not os.path.exists(TRACE_FILE):
              \n15 raise FileNotFoundError(
              \n16 f"\u672a\u627e\u5230\u8f68\u8ff9\u6587\u4ef6 {TRACE_FILE}，\u8bf7\u5148\u8fd0\u884c agent_demo.py\u3002"
              \n17 )
              \n18
              \n19with open(TRACE_FILE, "r", encoding="utf-8") as f:
              \n20 traces = json.load(f)
              \n21
              \n22print(f"\u5171\u52a0\u8f7d {len(traces)} \u6761\u8f68\u8ff9\\\\n")
              \n23
              \n24for idx, trace in enumerate(traces, start=1):
              \n25 input_text = trace.get("input", "")
              \n26 output_text = trace.get("output", "")
              \n27 elapsed = trace.get("elapsed_seconds", 0)
              \n28
              \n29 has_output = bool(output_text and output_text.strip())
              \n30 mentions_weather = "\u5929\u6c14" in output_text
              \n31 mentions_time = "\u65f6\u95f4" in output_text or "\u65e5\u671f" in output_text
              \n32 mentions_uncertainty = any(
              \n33 word in output_text
              \n34 for word in ["\u65e0\u6cd5\u786e\u8ba4", "\u672a\u627e\u5230", "\u8bf7\u786e\u8ba4", "\u6570\u636e\u4e3a\u7a7a"]
              \n35 )
              \n36
              \n37 print(f"\u8f68\u8ff9 {idx}")
              \n38 print(f"\u8f93\u5165：{input_text}")
              \n39 print(f"\u8f93\u51fa：{output_text}")
              \n40 print(f"\u8017\u65f6：{elapsed} \u79d2")
              \n41 print(f"\u6709\u8f93\u51fa：{has_output}")
              \n42 print(f"\u5305\u542b\u5929\u6c14\u4fe1\u606f：{mentions_weather}")
              \n43 print(f"\u5305\u542b\u65f6\u95f4\u4fe1\u606f：{mentions_time}")
              \n44 print(f"\u5305\u542b\u4e0d\u786e\u5b9a\u6027\u63d0\u793a：{mentions_uncertainty}")
              \n45 print("-" * 60)<\/p>\n

              \u8fd0\u884c\u65b9\u5f0f：<\/p>\n

              1python check_traces.py<\/p>\n

              \u6ce8\u610f：\u5f53\u524d\u7248\u672c\u53ea\u505a\u6587\u672c\u5c42\u9762\u7684\u57fa\u7840\u68c0\u67e5\u3002\u5982\u679c\u9700\u8981\u66f4\u4e25\u683c\u5730\u9a8c\u8bc1"\u662f\u5426\u771f\u6b63\u8c03\u7528\u4e86\u67d0\u4e2a\u5de5\u5177"，\u53ef\u4ee5\u5728\u540e\u7eed\u7248\u672c\u4e2d\u63a5\u5165 LangChain \u7684\u56de\u8c03\u673a\u5236，\u8bb0\u5f55\u5b8c\u6574\u7684\u5de5\u5177\u8c03\u7528\u4e8b\u4ef6\u3002<\/p>\n

              \n

              \u4e5d\u3001\u5de5\u5177\u5c01\u88c5\u89c4\u8303<\/h4>\n

              \u5de5\u5177\u662f Agent \u4e0e\u73b0\u5b9e\u4e16\u754c\u4ea4\u4e92\u7684\u5165\u53e3\u3002\u5c01\u88c5\u5de5\u5177\u65f6\u8981\u6ce8\u610f\u51e0\u70b9：<\/p>\n

            • \n

              \u540d\u79f0\u6e05\u6670
              \n\u5de5\u5177\u540d\u8981\u80fd\u8868\u8fbe\u7528\u9014，\u4f8b\u5982\u00a0get_weather\u3001search_orders\u3001send_email\u3002<\/p>\n<\/li>\n

            • \n

              \u63cf\u8ff0\u5b8c\u6574
              \ndocstring\u00a0\u6216\u63cf\u8ff0\u5b57\u6bb5\u8981\u5199\u6e05\u695a\u5de5\u5177\u7528\u9014\u3001\u53c2\u6570\u542b\u4e49\u548c\u8fd4\u56de\u683c\u5f0f\u3002\u6a21\u578b\u4f1a\u4f9d\u8d56\u8fd9\u4e9b\u4fe1\u606f\u5224\u65ad\u662f\u5426\u8c03\u7528\u5de5\u5177\u3002<\/p>\n<\/li>\n

            • \n

              \u53c2\u6570\u7ed3\u6784\u5316
              \n\u5c3d\u91cf\u4f7f\u7528\u660e\u786e\u7c7b\u578b，\u4f8b\u5982\u00a0str\u3001int\u3001float\u3001list[str]，\u907f\u514d\u8ba9\u6a21\u578b\u731c\u6d4b\u53c2\u6570\u683c\u5f0f\u3002<\/p>\n<\/li>\n

            • \n

              \u9519\u8bef\u53ef\u89e3\u91ca
              \n\u5de5\u5177\u5931\u8d25\u65f6\u5e94\u8fd4\u56de\u53ef\u8bfb\u9519\u8bef\u4fe1\u606f，\u800c\u4e0d\u662f\u76f4\u63a5\u629b\u51fa\u5f02\u5e38，\u5426\u5219 Agent \u53ef\u80fd\u65e0\u6cd5\u7ee7\u7eed\u89c4\u5212\u3002<\/p>\n<\/li>\n

            • \n

              \u6743\u9650\u53ef\u63a7
              \n\u5220\u9664\u3001\u4fee\u6539\u3001\u53d1\u9001\u7c7b\u5de5\u5177\u9700\u8981\u989d\u5916\u589e\u52a0\u6743\u9650\u6821\u9a8c，\u4e0d\u80fd\u76f4\u63a5\u66b4\u9732\u7ed9\u6240\u6709\u7528\u6237\u3002<\/p>\n<\/li>\n

              \u793a\u4f8b：<\/p>\n

              1@tool
              \n2def search_orders(user_id: str, order_id: str = "") -> str:
              \n3 """\u67e5\u8be2\u7528\u6237\u8ba2\u5355\u4fe1\u606f\u3002
              \n4
              \n5 \u53c2\u6570：
              \n6 – user_id: \u7528\u6237ID
              \n7 – order_id: \u53ef\u9009，\u8ba2\u5355ID
              \n8 """
              \n9 if not user_id:
              \n10 return "\u7f3a\u5c11\u7528\u6237ID，\u65e0\u6cd5\u67e5\u8be2\u8ba2\u5355\u3002"
              \n11 return f"\u7528\u6237 {user_id} \u7684\u8ba2\u5355\u67e5\u8be2\u6210\u529f"<\/p>\n

              \n

              \u5341\u3001\u8bb0\u5fc6\u4e0e\u4e0a\u4e0b\u6587\u7ba1\u7406<\/h4>\n

              \u9ed8\u8ba4\u60c5\u51b5\u4e0b，Agent \u6bcf\u6b21\u8c03\u7528\u90fd\u662f"\u65e0\u72b6\u6001"\u7684\u3002\u8981\u5b9e\u73b0\u591a\u8f6e\u5bf9\u8bdd，\u9700\u8981\u663e\u5f0f\u914d\u7f6e\u8bb0\u5fc6\u3002<\/p>\n

              \u6700\u7b80\u5355\u7684\u65b9\u5f0f\u662f\u4f7f\u7528\u5bf9\u8bdd\u5386\u53f2：<\/p>\n

              1from langchain.memory import ConversationBufferMemory
              \n2
              \n3memory = ConversationBufferMemory(
              \n4 memory_key="chat_history",
              \n5 return_messages=True
              \n6)<\/p>\n

              \u5982\u679c\u5bf9\u8bdd\u5f88\u957f，\u53ef\u4ee5\u4f7f\u7528\u6458\u8981\u8bb0\u5fc6，\u907f\u514d\u4e0a\u4e0b\u6587\u65e0\u9650\u589e\u957f：<\/p>\n

              1from langchain.memory import ConversationSummaryMemory
              \n2
              \n3memory = ConversationSummaryMemory(
              \n4 llm=llm,
              \n5 memory_key="chat_history",
              \n6 return_messages=True
              \n7)<\/p>\n

              \u4f01\u4e1a\u7ea7\u573a\u666f\u4e2d，\u8fd8\u53ef\u4ee5\u628a\u5386\u53f2\u5bf9\u8bdd\u5b58\u5165\u6570\u636e\u5e93\u6216\u5411\u91cf\u5e93，\u5b9e\u73b0\u957f\u671f\u8bb0\u5fc6\u3002<\/p>\n

              \u9700\u8981\u6ce8\u610f：\u8bb0\u5fc6\u4e0d\u662f\u8d8a\u591a\u8d8a\u597d\u3002\u8fc7\u957f\u7684\u4e0a\u4e0b\u6587\u4f1a\u589e\u52a0\u6210\u672c，\u4e5f\u53ef\u80fd\u5f15\u5165\u566a\u58f0\u3002\u751f\u4ea7\u73af\u5883\u901a\u5e38\u9700\u8981\u8bbe\u7f6e\u6700\u5927\u8f6e\u6570\u3001\u6458\u8981\u538b\u7f29\u6216\u4f1a\u8bdd\u8fc7\u671f\u7b56\u7565\u3002<\/p>\n

              \n

              \u5341\u4e00\u3001Agent \u9a8c\u8bc1\u65b9\u6cd5<\/h4>\n

              Agent \u9a8c\u8bc1\u548c\u666e\u901a\u63a5\u53e3\u6d4b\u8bd5\u4e0d\u540c\u3002\u666e\u901a\u63a5\u53e3\u6d4b\u8bd5\u4e3b\u8981\u770b"\u8fd4\u56de\u7ed3\u679c\u662f\u5426\u6b63\u786e"，\u4f46 Agent \u8fd8\u9700\u8981\u770b"\u8fc7\u7a0b\u662f\u5426\u53ef\u9760"\u3002<\/p>\n

              \u4e00\u4e2a\u770b\u4f3c\u6b63\u786e\u7684\u56de\u7b54，\u53ef\u80fd\u6765\u81ea\u4ee5\u4e0b\u95ee\u9898：<\/p>\n

                \n
              • \u5de5\u5177\u6839\u672c\u6ca1\u6709\u88ab\u8c03\u7528；<\/li>\n
              • \u5de5\u5177\u53c2\u6570\u4f20\u9519\u4e86；<\/li>\n
              • \u5de5\u5177\u8fd4\u56de\u4e3a\u7a7a，\u4f46\u6a21\u578b\u7f16\u9020\u4e86\u6570\u636e；<\/li>\n
              • Agent \u8df3\u8fc7\u4e86\u5fc5\u8981\u7684\u6821\u9a8c\u6b65\u9aa4；<\/li>\n
              • \u591a\u8f6e\u5bf9\u8bdd\u540e\u8bb0\u5fc6\u51fa\u73b0\u504f\u5dee\u3002<\/li>\n<\/ul>\n

                \u6240\u4ee5\u6211\u4eec\u9a8c\u8bc1 Agent \u65f6，\u81f3\u5c11\u8981\u770b\u4e09\u5c42：<\/p>\n\n\u5c42\u7ea7\u9a8c\u8bc1\u5185\u5bb9\u5178\u578b\u95ee\u9898<\/tr>\n\n\n\n\n
                \u5de5\u5177\u5c42<\/td>\n\u662f\u5426\u8c03\u7528\u6b63\u786e\u5de5\u5177\u3001\u53c2\u6570\u662f\u5426\u6b63\u786e<\/td>\n\u53c2\u6570\u9519\u8bef\u3001\u5de5\u5177\u672a\u8c03\u7528<\/td>\n<\/tr>\n
                \u8f68\u8ff9\u5c42<\/td>\n\u662f\u5426\u57fa\u4e8e\u771f\u5b9e\u8fd4\u56de\u7ee7\u7eed\u63a8\u7406<\/td>\n\u5ffd\u7565\u5931\u8d25\u3001\u8df3\u6b65\u3001\u7f16\u9020<\/td>\n<\/tr>\n
                \u7ed3\u679c\u5c42<\/td>\n\u6700\u7ec8\u56de\u7b54\u662f\u5426\u6ee1\u8db3\u4efb\u52a1<\/td>\n\u9057\u6f0f\u6761\u4ef6\u3001\u8868\u8fbe\u4e0d\u51c6\u786e<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
                \n

                \u5341\u4e8c\u3001\u6700\u5c0f\u9a8c\u8bc1\u6d41\u7a0b<\/h4>\n

                \u6211\u4eec\u53ef\u4ee5\u91c7\u7528\u4e00\u4e2a\u7b80\u5316\u7248\u7684\u516d\u6b65\u9a8c\u8bc1\u6d41\u7a0b：<\/p>\n

              • \n

                Plan：\u5b9a\u4e49\u4efb\u52a1\u76ee\u6807
                \n\u660e\u786e Agent \u5fc5\u987b\u5b8c\u6210\u54ea\u4e9b\u52a8\u4f5c\u3002\u4f8b\u5982：\u5fc5\u987b\u8c03\u7528\u5929\u6c14\u5de5\u5177\u3001\u5fc5\u987b\u5f15\u7528\u771f\u5b9e\u6570\u636e\u3001\u5de5\u5177\u4e3a\u7a7a\u65f6\u5fc5\u987b\u63d0\u793a\u65e0\u6cd5\u786e\u8ba4\u3002<\/p>\n<\/li>\n

              • \n

                Data：\u51c6\u5907\u6d4b\u8bd5\u6570\u636e
                \n\u5305\u62ec\u6b63\u5e38\u8fd4\u56de\u3001\u7a7a\u8fd4\u56de\u3001\u9519\u8bef\u8fd4\u56de\u3001\u8fb9\u754c\u6570\u636e\u3002<\/p>\n<\/li>\n

              • \n

                Trace：\u4fdd\u5b58\u6267\u884c\u8f68\u8ff9
                \n\u8bb0\u5f55\u5de5\u5177\u8c03\u7528\u540d\u79f0\u3001\u53c2\u6570\u3001\u8fd4\u56de\u7ed3\u679c\u3001\u6a21\u578b\u601d\u8003\u8fc7\u7a0b\u3002<\/p>\n<\/li>\n

              • \n

                Run：\u591a\u6b21\u8fd0\u884c
                \nAgent \u5177\u6709\u975e\u786e\u5b9a\u6027，\u540c\u4e00\u4e2a\u8f93\u5165\u53ef\u80fd\u4ea7\u751f\u4e0d\u540c\u8def\u5f84\u3002\u5173\u952e\u573a\u666f\u81f3\u5c11\u8fd0\u884c\u591a\u6b21\u3002<\/p>\n<\/li>\n

              • \n

                Eval：\u81ea\u52a8\u65ad\u8a00 + \u4eba\u5de5\u8bc4\u5ba1
                \n\u53ef\u7528\u4ee3\u7801\u68c0\u67e5\u5de5\u5177\u540d\u3001\u53c2\u6570\u3001\u5b57\u6bb5\u662f\u5426\u5b58\u5728；\u5bf9\u5f00\u653e\u5f0f\u56de\u7b54\u518d\u7528 LLM \u505a\u8d28\u91cf\u8bc4\u5ba1\u3002<\/p>\n<\/li>\n

              • \n

                Report：\u5b9a\u4f4d\u95ee\u9898
                \n\u62a5\u544a\u4e0d\u80fd\u53ea\u7ed9\u5206\u6570，\u8981\u6307\u51fa\u54ea\u6761\u7528\u4f8b\u5931\u8d25\u3001\u54ea\u6b21\u5de5\u5177\u8c03\u7528\u5f02\u5e38\u3001\u54ea\u4e2a\u53c2\u6570\u9519\u8bef\u3002<\/p>\n<\/li>\n

                \u5341\u4e09\u3001\u4e00\u4e2a\u7b80\u5355\u8f68\u8ff9\u68c0\u67e5\u793a\u4f8b<\/h4>\n

                \u8fd9\u91cc\u628a\u6700\u5c0f\u8f68\u8ff9\u68c0\u67e5\u51fd\u6570\u8865\u5168：<\/p>\n

                1def evaluate_trace(case, trace):
                \n2 tool_calls = trace.get("tool_calls", [])
                \n3 final_text = trace.get("output", "")
                \n4 tool_results = trace.get("tool_results", [])
                \n5
                \n6 expected_tool = case["expected_tool"]
                \n7 expected_args = case.get("expected_arguments", {})
                \n8
                \n9 actual = next((c for c in tool_calls if c["name"] == expected_tool), None)
                \n10
                \n11 tool_called = actual is not None
                \n12 params_ok = tool_called and all(
                \n13 actual["arguments"].get(k) == v for k, v in expected_args.items()
                \n14 )
                \n15
                \n16 has_data = any(r for r in tool_results if r and str(r).strip())
                \n17 mentions_uncertainty = any(
                \n18 word in final_text
                \n19 for word in ["\u65e0\u6cd5\u786e\u8ba4", "\u672a\u627e\u5230", "\u8bf7\u786e\u8ba4", "\u6570\u636e\u4e3a\u7a7a"]
                \n20 )
                \n21
                \n22 if not tool_called:
                \n23 return "FAIL: \u672a\u8c03\u7528\u9884\u671f\u5de5\u5177"
                \n24
                \n25 if not params_ok:
                \n26 return "FAIL: \u5de5\u5177\u53c2\u6570\u4e0d\u6b63\u786e"
                \n27
                \n28 if not has_data and not mentions_uncertainty:
                \n29 return "FAIL: \u5de5\u5177\u65e0\u6570\u636e\u4f46\u6a21\u578b\u672a\u63d0\u793a\u4e0d\u786e\u5b9a\u6027"
                \n30
                \n31 return "PASS"<\/p>\n

                \u8fd9\u4e2a\u51fd\u6570\u9002\u5408\u4f5c\u4e3a\u201c\u8f68\u8ff9\u68c0\u67e5\u201d\u7684\u5165\u95e8\u793a\u4f8b\u3002\u5b83\u4e0d\u4f9d\u8d56\u7279\u5b9a\u6846\u67b6，\u6211\u4eec\u53ef\u4ee5\u5148\u628a Agent \u8fd0\u884c\u7ed3\u679c\u4fdd\u5b58\u6210\u7edf\u4e00\u683c\u5f0f，\u518d\u7528\u8fd9\u4e2a\u51fd\u6570\u505a\u57fa\u7840\u65ad\u8a00\u3002<\/p>\n

                \n

                \u5341\u56db\u3001\u6d4b\u8bd5\u7528\u4f8b\u793a\u4f8b<\/h4>\n

                \u6211\u4eec\u53ef\u4ee5\u51c6\u5907\u4e00\u7ec4\u6700\u5c0f\u6d4b\u8bd5\u7528\u4f8b，\u7528\u6765\u8986\u76d6\u6b63\u5e38\u8def\u5f84\u548c\u5f02\u5e38\u8def\u5f84：<\/p>\n

                1test_cases = [
                \n2 {
                \n3 "name": "\u6b63\u5e38\u5929\u6c14\u67e5\u8be2",
                \n4 "input": "\u90d1\u5dde\u4eca\u5929\u5929\u6c14\u600e\u4e48\u6837？",
                \n5 "expected_tool": "get_weather",
                \n6 "expected_arguments": {"city": "\u90d1\u5dde"},
                \n7 },
                \n8 {
                \n9 "name": "\u672a\u77e5\u57ce\u5e02\u5929\u6c14",
                \n10 "input": "\u706b\u661f\u4eca\u5929\u5929\u6c14\u5982\u4f55？",
                \n11 "expected_tool": "get_weather",
                \n12 "expected_arguments": {"city": "\u706b\u661f"},
                \n13 },
                \n14 {
                \n15 "name": "\u65f6\u95f4\u67e5\u8be2",
                \n16 "input": "\u73b0\u5728\u51e0\u70b9\u4e86？",
                \n17 "expected_tool": "get_current_time",
                \n18 "expected_arguments": {},
                \n19 },
                \n20]<\/p>\n

                \u8fd0\u884c\u68c0\u67e5\u65f6\u53ef\u4ee5\u8fd9\u6837\u7ec4\u7ec7：<\/p>\n

                1results = []
                \n2
                \n3for case in test_cases:
                \n4 trace = run_agent(case["input"])
                \n5 status = evaluate_trace(case, trace)
                \n6 results.append({"case": case["name"], "status": status})
                \n7
                \n8for r in results:
                \n9 print(r["case"], "->", r["status"])<\/p>\n

                \u8fd9\u91cc\u7684\u00a0run_agent\u00a0\u9700\u8981\u6839\u636e\u5b9e\u9645\u4f7f\u7528\u7684\u6846\u67b6\u66ff\u6362\u6210\u771f\u5b9e\u8c03\u7528\u3002\u9488\u5bf9\u672c\u7ae0\u00a0agent_demo.py\u00a0\u4e2d\u7684 LangChain Agent，\u6211\u4eec\u53ef\u4ee5\u8fd9\u6837\u5199：<\/p>\n

                1def run_agent(user_input: str) -> dict:
                \n2 """\u6267\u884c\u4e00\u6b21 Agent \u8c03\u7528，\u5e76\u8fd4\u56de\u7ed3\u6784\u5316\u8f68\u8ff9\u3002"""
                \n3
                \n4 result = agent_executor.invoke({"input": user_input})
                \n5
                \n6 intermediate_steps = result.get("intermediate_steps", [])
                \n7
                \n8 tool_calls = []
                \n9 tool_results = []
                \n10
                \n11 for step in intermediate_steps:
                \n12 if isinstance(step, (list, tuple)) and len(step) >= 2:
                \n13 action = step[0]
                \n14 output = step[1]
                \n15
                \n16 tool_calls.append({
                \n17 "name": getattr(action, "tool", None),
                \n18 "arguments": getattr(action, "tool_input", None),
                \n19 })
                \n20 tool_results.append(str(output))
                \n21
                \n22 return {
                \n23 "input": user_input,
                \n24 "output": result.get("output", ""),
                \n25 "tool_calls": tool_calls,
                \n26 "tool_results": tool_results,
                \n27 "intermediate_steps": intermediate_steps,
                \n28 }<\/p>\n

                \u9700\u8981\u6ce8\u610f\u7684\u662f，intermediate_steps\u00a0\u7684\u5177\u4f53\u7ed3\u6784\u4f1a\u968f LangChain \u7248\u672c\u53d8\u5316\u3002\u5982\u679c\u540e\u7eed\u53d1\u73b0\u00a0step[0]\u00a0\u4e0d\u662f\u9884\u671f\u7684 action \u5bf9\u8c61，\u6211\u4eec\u53ef\u4ee5\u628a\u00a0run_agent\u00a0\u6539\u6210\u53ea\u4f9d\u8d56\u00a0result.get("output")\u00a0\u505a\u57fa\u7840\u9a8c\u8bc1，\u518d\u9010\u6b65\u63a5\u5165\u56de\u8c03\u673a\u5236\u6765\u8bb0\u5f55\u5b8c\u6574\u5de5\u5177\u8c03\u7528\u4e8b\u4ef6\u3002<\/p>\n

                \n

                \u5341\u4e94\u3001\u5b8c\u6574\u9a8c\u8bc1\u811a\u672c：verify_agent.py<\/h4>\n

                \u628a\u00a0run_agent\u3001evaluate_trace\u00a0\u548c\u6d4b\u8bd5\u7528\u4f8b\u6574\u5408\u5230\u4e00\u4e2a\u53ef\u76f4\u63a5\u8fd0\u884c\u7684\u811a\u672c\u4e2d：<\/p>\n

                1"""
                \n2Agent \u8f68\u8ff9\u9a8c\u8bc1\u811a\u672c
                \n3
                \n4\u8fd0\u884c\u65b9\u5f0f：
                \n51. \u5148\u8fd0\u884c：python agent_demo.py
                \n62. \u518d\u8fd0\u884c：python verify_agent.py
                \n7"""
                \n8
                \n9import os
                \n10import json
                \n11from dotenv import load_dotenv
                \n12from langchain_openai import ChatOpenAI
                \n13from langchain_core.tools import tool
                \n14from langchain.agents import create_tool_calling_agent, AgentExecutor
                \n15from langchain_core.prompts import ChatPromptTemplate
                \n16
                \n17# —————————-
                \n18# 1. \u52a0\u8f7d\u73af\u5883\u53d8\u91cf
                \n19# —————————-
                \n20load_dotenv()
                \n21
                \n22OPENAI_API_KEY = os.getenv("OPENAI_API_KEY", "")
                \n23MODEL_NAME = os.getenv("MODEL_NAME", "gpt-4o")
                \n24MODEL_BASE_URL = os.getenv("MODEL_BASE_URL", "").strip()
                \n25
                \n26if not OPENAI_API_KEY:
                \n27 raise RuntimeError(
                \n28 "\u672a\u627e\u5230 OPENAI_API_KEY，\u8bf7\u5728\u9879\u76ee\u6839\u76ee\u5f55\u521b\u5efa .env \u6587\u4ef6\u5e76\u5199\u5165 API Key\u3002"
                \n29 )
                \n30
                \n31# —————————-
                \n32# 2. \u521d\u59cb\u5316\u6a21\u578b
                \n33# —————————-
                \n34llm_kwargs = {
                \n35 "model": MODEL_NAME,
                \n36 "temperature": 0.2,
                \n37 "api_key": OPENAI_API_KEY,
                \n38}
                \n39
                \n40if MODEL_BASE_URL:
                \n41 llm_kwargs["base_url"] = MODEL_BASE_URL
                \n42
                \n43llm = ChatOpenAI(**llm_kwargs)
                \n44
                \n45# —————————-
                \n46# 3. \u5b9a\u4e49\u5de5\u5177
                \n47# —————————-
                \n48
                \n49@tool
                \n50def get_weather(city: str) -> str:
                \n51 """\u67e5\u8be2\u6307\u5b9a\u57ce\u5e02\u7684\u5929\u6c14\u4fe1\u606f\u3002
                \n52
                \n53 \u53c2\u6570：
                \n54 – city: \u57ce\u5e02\u540d\u79f0，\u4f8b\u5982"\u90d1\u5dde""\u5317\u4eac""\u4e0a\u6d77"
                \n55 """
                \n56 if not city:
                \n57 return "\u7f3a\u5c11\u57ce\u5e02\u540d\u79f0，\u65e0\u6cd5\u67e5\u8be2\u5929\u6c14\u3002"
                \n58
                \n59 weather_data = {
                \n60 "\u90d1\u5dde": "\u6674，22\u00b0C，\u5fae\u98ce",
                \n61 "\u5317\u4eac": "\u591a\u4e91，18\u00b0C，\u5317\u98ce2\u7ea7",
                \n62 "\u4e0a\u6d77": "\u5c0f\u96e8，20\u00b0C，\u4e1c\u5357\u98ce3\u7ea7",
                \n63 }
                \n64
                \n65 return weather_data.get(city, f"\u672a\u627e\u5230 {city} \u7684\u5929\u6c14\u6570\u636e，\u8bf7\u786e\u8ba4\u57ce\u5e02\u540d\u79f0\u3002")
                \n66
                \n67
                \n68@tool
                \n69def get_current_time(dummy: str = "") -> str:
                \n70 """\u83b7\u53d6\u5f53\u524d\u65e5\u671f\u548c\u65f6\u95f4\u3002
                \n71
                \n72 \u53c2\u6570：
                \n73 – dummy: \u53ef\u9009\u5360\u4f4d\u53c2\u6570，\u4e0d\u9700\u8981\u4f20\u5165\u5b9e\u9645\u5185\u5bb9
                \n74 """
                \n75 from datetime import datetime
                \n76 now = datetime.now()
                \n77 return now.strftime("%Y\u5e74%m\u6708%d\u65e5 %H:%M:%S")
                \n78
                \n79
                \n80tools = [get_weather, get_current_time]
                \n81
                \n82# —————————-
                \n83# 4. \u521b\u5efa Agent
                \n84# —————————-
                \n85
                \n86prompt = ChatPromptTemplate.from_messages(
                \n87 [
                \n88 ("system", "\u4f60\u662f\u4e00\u4e2a\u667a\u80fd\u52a9\u624b，\u53ef\u4ee5\u6839\u636e\u7528\u6237\u9700\u6c42\u8c03\u7528\u5de5\u5177\u3002"),
                \n89 ("human", "{input}"),
                \n90 ("placeholder", "{agent_scratchpad}"),
                \n91 ]
                \n92)
                \n93
                \n94agent = create_tool_calling_agent(llm, tools, prompt)
                \n95
                \n96agent_executor = AgentExecutor(
                \n97 agent=agent,
                \n98 tools=tools,
                \n99 verbose=False,
                \n100 max_iterations=5,
                \n101 handle_parsing_errors=True,
                \n102)
                \n103
                \n104# —————————-
                \n105# 5. \u5c01\u88c5 run_agent
                \n106# —————————-
                \n107
                \n108def run_agent(user_input: str) -> dict:
                \n109 """\u6267\u884c\u4e00\u6b21 Agent \u8c03\u7528，\u5e76\u8fd4\u56de\u7ed3\u6784\u5316\u8f68\u8ff9\u3002"""
                \n110
                \n111 result = agent_executor.invoke({"input": user_input})
                \n112
                \n113 intermediate_steps = result.get("intermediate_steps", [])
                \n114
                \n115 tool_calls = []
                \n116 tool_results = []
                \n117
                \n118 for step in intermediate_steps:
                \n119 if isinstance(step, (list, tuple)) and len(step) >= 2:
                \n120 action = step[0]
                \n121 output = step[1]
                \n122
                \n123 tool_calls.append({
                \n124 "name": getattr(action, "tool", None),
                \n125 "arguments": getattr(action, "tool_input", None),
                \n126 })
                \n127 tool_results.append(str(output))
                \n128
                \n129 return {
                \n130 "input": user_input,
                \n131 "output": result.get("output", ""),
                \n132 "tool_calls": tool_calls,
                \n133 "tool_results": tool_results,
                \n134 "intermediate_steps": intermediate_steps,
                \n135 }
                \n136
                \n137# —————————-
                \n138# 6. \u5c01\u88c5 evaluate_trace
                \n139# —————————-
                \n140
                \n141def evaluate_trace(case, trace):
                \n142 """\u5bf9\u4e00\u6b21 Agent \u8c03\u7528\u8f68\u8ff9\u505a\u57fa\u7840\u65ad\u8a00\u3002"""
                \n143
                \n144 tool_calls = trace.get("tool_calls", [])
                \n145 final_text = trace.get("output", "")
                \n146 tool_results = trace.get("tool_results", [])
                \n147
                \n148 expected_tool = case["expected_tool"]
                \n149 expected_args = case.get("expected_arguments", {})
                \n150
                \n151 actual = next((c for c in tool_calls if c["name"] == expected_tool), None)
                \n152
                \n153 tool_called = actual is not None
                \n154 params_ok = tool_called and all(
                \n155 actual["arguments"].get(k) == v for k, v in expected_args.items()
                \n156 )
                \n157
                \n158 has_data = any(r for r in tool_results if r and str(r).strip())
                \n159 mentions_uncertainty = any(
                \n160 word in final_text
                \n161 for word in ["\u65e0\u6cd5\u786e\u8ba4", "\u672a\u627e\u5230", "\u8bf7\u786e\u8ba4", "\u6570\u636e\u4e3a\u7a7a"]
                \n162 )
                \n163
                \n164 if not tool_called:
                \n165 return "FAIL: \u672a\u8c03\u7528\u9884\u671f\u5de5\u5177"
                \n166
                \n167 if not params_ok:
                \n168 return "FAIL: \u5de5\u5177\u53c2\u6570\u4e0d\u6b63\u786e"
                \n169
                \n170 if not has_data and not mentions_uncertainty:
                \n171 return "FAIL: \u5de5\u5177\u65e0\u6570\u636e\u4f46\u6a21\u578b\u672a\u63d0\u793a\u4e0d\u786e\u5b9a\u6027"
                \n172
                \n173 return "PASS"
                \n174
                \n175# —————————-
                \n176# 7. \u6d4b\u8bd5\u7528\u4f8b
                \n177# —————————-
                \n178
                \n179test_cases = [
                \n180 {
                \n181 "name": "\u6b63\u5e38\u5929\u6c14\u67e5\u8be2",
                \n182 "input": "\u90d1\u5dde\u4eca\u5929\u5929\u6c14\u600e\u4e48\u6837？",
                \n183 "expected_tool": "get_weather",
                \n184 "expected_arguments": {"city": "\u90d1\u5dde"},
                \n185 },
                \n186 {
                \n187 "name": "\u672a\u77e5\u57ce\u5e02\u5929\u6c14",
                \n188 "input": "\u706b\u661f\u4eca\u5929\u5929\u6c14\u5982\u4f55？",
                \n189 "expected_tool": "get_weather",
                \n190 "expected_arguments": {"city": "\u706b\u661f"},
                \n191 },
                \n192 {
                \n193 "name": "\u65f6\u95f4\u67e5\u8be2",
                \n194 "input": "\u73b0\u5728\u51e0\u70b9\u4e86？",
                \n195 "expected_tool": "get_current_time",
                \n196 "expected_arguments": {},
                \n197 },
                \n198]
                \n199
                \n200# —————————-
                \n201# 8. \u8fd0\u884c\u9a8c\u8bc1
                \n202# —————————-
                \n203
                \n204print("\u5f00\u59cb Agent \u8f68\u8ff9\u9a8c\u8bc1…\\\\n")
                \n205
                \n206results = []
                \n207
                \n208for case in test_cases:
                \n209 print(f"\u8fd0\u884c\u7528\u4f8b：{case['name']}")
                \n210 print(f" \u8f93\u5165：{case['input']}")
                \n211
                \n212 trace = run_agent(case["input"])
                \n213 status = evaluate_trace(case, trace)
                \n214
                \n215 results.append({"case": case["name"], "status": status})
                \n216
                \n217 print(f" \u72b6\u6001：{status}")
                \n218 print(f" \u8f93\u51fa：{trace['output'][:100]}…")
                \n219 print()
                \n220
                \n221# —————————-
                \n222# 9. \u6c47\u603b\u7ed3\u679c
                \n223# —————————-
                \n224
                \n225print("=" * 60)
                \n226print("\u9a8c\u8bc1\u7ed3\u679c\u6c47\u603b：")
                \n227print("=" * 60)
                \n228
                \n229for r in results:
                \n230 print(f"{r['case']} -> {r['status']}")
                \n231
                \n232pass_count = sum(1 for r in results if r["status"] == "PASS")
                \n233fail_count = sum(1 for r in results if r["status"] != "PASS")
                \n234
                \n235print(f"\\\\n\u603b\u8ba1：{len(results)} \u6761\u7528\u4f8b，{pass_count} \u901a\u8fc7，{fail_count} \u5931\u8d25")
                \n236
                \n237# —————————-
                \n238# 10. \u4fdd\u5b58\u9a8c\u8bc1\u62a5\u544a
                \n239# —————————-
                \n240
                \n241output_dir = "agent_traces"
                \n242os.makedirs(output_dir, exist_ok=True)
                \n243
                \n244report_file = os.path.join(output_dir, "verify_report.json")
                \n245
                \n246with open(report_file, "w", encoding="utf-8") as f:
                \n247 json.dump(results, f, ensure_ascii=False, indent=2)
                \n248
                \n249print(f"\\\\n\u9a8c\u8bc1\u62a5\u544a\u5df2\u4fdd\u5b58\u5230：{report_file}")<\/p>\n

                \u8fd0\u884c\u65b9\u5f0f：<\/p>\n

                1python verify_agent.py<\/p>\n

                \n

                \u5341\u516d\u3001\u5e38\u89c1\u8fd0\u884c\u95ee\u9898\u6392\u67e5<\/h4>\n\n\u95ee\u9898\u53ef\u80fd\u539f\u56e0\u5904\u7406\u65b9\u5f0f<\/tr>\n\n\n\n\n\n\n\n\n\n
                OPENAI_API_KEY\u00a0\u7f3a\u5931<\/td>\n\u672a\u521b\u5efa\u00a0.env\u00a0\u6216\u6587\u4ef6\u8def\u5f84\u9519\u8bef<\/td>\n\u786e\u8ba4\u00a0.env\u00a0\u4e0e\u811a\u672c\u540c\u76ee\u5f55<\/td>\n<\/tr>\n
                \u6a21\u578b\u8c03\u7528\u5931\u8d25<\/td>\nAPI Key \u65e0\u6548\u6216\u4f59\u989d\u4e0d\u8db3<\/td>\n\u68c0\u67e5 Key\u3001\u989d\u5ea6\u3001\u670d\u52a1\u72b6\u6001<\/td>\n<\/tr>\n
                \u5de5\u5177\u672a\u88ab\u8c03\u7528<\/td>\n\u5de5\u5177\u63cf\u8ff0\u4e0d\u6e05\u6670<\/td>\n\u5b8c\u5584\u00a0docstring\u00a0\u548c\u53c2\u6570\u8bf4\u660e<\/td>\n<\/tr>\n
                \u8fd4\u56de\u7a7a\u7ed3\u679c<\/td>\n\u5de5\u5177\u8fd4\u56de\u4e3a\u7a7a\u4e14\u6a21\u578b\u672a\u515c\u5e95<\/td>\n\u589e\u52a0\u4e0d\u786e\u5b9a\u6027\u63d0\u793a\u903b\u8f91<\/td>\n<\/tr>\n
                \u4f9d\u8d56\u5b89\u88c5\u5931\u8d25<\/td>\n\u7f51\u7edc\u6216\u955c\u50cf\u6e90\u95ee\u9898<\/td>\n\u4f7f\u7528\u53ef\u4fe1\u56fd\u5185\u955c\u50cf\u6e90<\/td>\n<\/tr>\n
                \u865a\u62df\u73af\u5883\u672a\u6fc0\u6d3b<\/td>\n\u4f9d\u8d56\u88c5\u5230\u5168\u5c40\u73af\u5883<\/td>\n\u91cd\u65b0\u6fc0\u6d3b\u00a0.venv\u00a0\u540e\u8fd0\u884c<\/td>\n<\/tr>\n
                \u4e2d\u6587\u8f93\u51fa\u4e71\u7801<\/td>\n\u7ec8\u7aef\u7f16\u7801\u4e0d\u4e00\u81f4<\/td>\n\u4f7f\u7528 UTF-8 \u7ec8\u7aef\u6216 IDE \u8fd0\u884c<\/td>\n<\/tr>\n
                \u591a\u8f6e\u5bf9\u8bdd\u4e22\u5931\u4e0a\u4e0b\u6587<\/td>\n\u672a\u914d\u7f6e\u8bb0\u5fc6<\/td>\n\u542f\u7528\u5bf9\u8bdd\u5386\u53f2\u6216\u6458\u8981\u8bb0\u5fc6<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n
                \n

                \u5341\u4e03\u3001\u672c\u7ae0\u53ef\u76f4\u63a5\u4ea4\u4ed8\u7684\u811a\u672c\u6e05\u5355<\/h4>\n

                \u672c\u7ae0\u5b8c\u6210\u540e，\u6211\u4eec\u5e94\u62e5\u6709\u4ee5\u4e0b\u6587\u4ef6：<\/p>\n

                1project\/
                \n2\u251c\u2500\u2500 .env
                \n3\u251c\u2500\u2500 agent_demo.py
                \n4\u251c\u2500\u2500 verify_agent.py
                \n5\u251c\u2500\u2500 check_traces.py
                \n6\u2514\u2500\u2500 agent_traces\/
                \n7 \u251c\u2500\u2500 traces.json
                \n8 \u2514\u2500\u2500 verify_report.json<\/p>\n

                \u8fd0\u884c\u987a\u5e8f\u4e3a：<\/p>\n

                1python agent_demo.py
                \n2python verify_agent.py
                \n3python check_traces.py<\/p>\n

                \u5982\u679c\u540e\u7eed\u8981\u6269\u5c55，\u6211\u4eec\u53ef\u4ee5\u628a\u00a0agent_demo.py\u00a0\u62c6\u6210：<\/p>\n

                1project\/
                \n2\u251c\u2500\u2500 .env
                \n3\u251c\u2500\u2500 main.py
                \n4\u251c\u2500\u2500 tools\/
                \n5\u2502 \u2514\u2500\u2500 weather.py
                \n6\u251c\u2500\u2500 agent\/
                \n7\u2502 \u2514\u2500\u2500 executor.py
                \n8\u251c\u2500\u2500 traces\/
                \n9\u2502 \u2514\u2500\u2500 traces.json
                \n10\u2514\u2500\u2500 tests\/
                \n11 \u2514\u2500\u2500 test_traces.py<\/p>\n

                \u4f46\u6559\u7a0b\u521d\u671f\u4e0d\u5efa\u8bae\u62c6\u5f97\u592a\u6563\u3002\u6211\u4eec\u5148\u7528\u4e00\u4e2a\u6587\u4ef6\u8dd1\u901a\u5b8c\u6574\u6d41\u7a0b，\u66f4\u5bb9\u6613\u5efa\u7acb\u4fe1\u5fc3\u3002<\/p>\n

                \n

                \u5341\u516b\u3001\u672c\u7ae0\u5c0f\u7ed3<\/h4>\n

                \u8fd9\u4e00\u7ae0\u6211\u4eec\u5b8c\u6210\u4e86 Agent \u5f00\u53d1\u73af\u5883\u7684\u95ed\u73af：<\/p>\n

                  \n
                • \u9009\u62e9\u4e86\u00a0Python + LangChain\u00a0\u4f5c\u4e3a\u5165\u95e8\u6808；<\/li>\n
                • \u914d\u7f6e\u4e86\u865a\u62df\u73af\u5883\u3001\u4f9d\u8d56\u548c\u6a21\u578b\u63a5\u5165；<\/li>\n
                • \u5c01\u88c5\u4e86\u53ef\u8fd0\u884c\u7684\u5929\u6c14\u67e5\u8be2\u548c\u65f6\u95f4\u67e5\u8be2\u5de5\u5177；<\/li>\n
                • \u521b\u5efa\u4e86\u4e00\u4e2a\u6700\u5c0f Agent，\u5e76\u652f\u6301\u591a\u8f6e\u6d4b\u8bd5；<\/li>\n
                • \u4fdd\u5b58\u4e86\u8fd0\u884c\u8f68\u8ff9，\u4fbf\u4e8e\u540e\u7eed\u8c03\u8bd5；<\/li>\n
                • \u63d0\u4f9b\u4e86\u00a0run_agent\u00a0\u5c01\u88c5\u3001evaluate_trace\u00a0\u65ad\u8a00\u548c\u6d4b\u8bd5\u7528\u4f8b；<\/li>\n
                • \u63d0\u4f9b\u4e86\u8f68\u8ff9\u68c0\u67e5\u811a\u672c\u3001\u9a8c\u8bc1\u811a\u672c\u548c\u5e38\u89c1\u95ee\u9898\u6392\u67e5\u8868\u3002<\/li>\n<\/ul>\n

                  \u5230\u8fd9\u91cc，\u6211\u4eec\u5df2\u7ecf\u53ef\u4ee5\u72ec\u7acb\u8fd0\u884c\u4e00\u4e2a\u6700\u5c0f Agent，\u5e76\u521d\u6b65\u7406\u89e3"\u8f93\u5165\u2014\u63a8\u7406\u2014\u5de5\u5177\u8c03\u7528\u2014\u8fd4\u56de\u2014\u8f68\u8ff9\u2014\u9a8c\u8bc1"\u7684\u57fa\u672c\u94fe\u8def\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

                  \u6211\u4eec\u7684\u76ee\u6807\u662f\u8ba9\u5f00\u53d1\u73af\u5883\u8fbe\u5230\\”\u80fd\u8dd1\u901a\u4e00\u4e2a\u6700\u5c0f Agent\\”\u7684\u72b6\u6001：\u5b89\u88c5\u4f9d\u8d56\u3001\u914d\u7f6e\u6a21\u578b\u3001\u5c01\u88c5\u5de5\u5177\u3001\u8fd0\u884c\u8c03\u7528\u3001\u4fdd\u5b58\u8f68\u8ff9\u3001\u9a8c\u8bc1\u884c\u4e3a\u3002\u540c\u65f6，\u6240\u6709\u793a\u4f8b\u4ee3\u7801\u90fd\u5c3d\u91cf\u8865\u6210\u53ef\u76f4\u63a5\u62f7\u8d1d\u8fd0\u884c\u7684\u7248\u672c，\u5e76\u5728\u9700\u8981\u7279\u6b8a\u914d\u7f6e\u7684\u5730\u65b9\u7ed9\u51fa\u8fd0\u884c\u65b9\u5f0f\u8bf4\u660e\u3002\u4e00\u3001\u73af\u5883\u642d\u5efa\u524d\u7684\u4e09\u4e2a\u9009\u62e9\u5728\u52a8\u624b\u4e4b\u524d，\u9700\u8981\u5148\u786e\u5b9a\u4e09\u4ef6\u4e8b：\u9009\u62e9\u9879\u8bf4\u660e\u5efa\u8bae\u5f00\u53d1\u8bed\u8a00\u76ee\u524d Python \u751f\u6001\u6700\u6210\u719f\u521d\u5b66\u8005\u4f18\u5148 Python\u6a21\u578b\u63a5\u5165\u51b3\u5b9a Agent \u7684\u63a8\u7406\u80fd\u529b\u5148\u7528\u4e00\u4e2a\u7a33\u5b9a\u6a21\u578b\u8dd1\u901a\u6846\u67b6\u5c42\u7ea7\u51b3<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[50],"topic":[],"class_list":["post-115510","post","type-post","status-publish","format-standard","hentry","category-server","tag-50"],"yoast_head":"\n\u7b2c\u4e8c\u7ae0Agent \u5f00\u53d1\u73af\u5883\u642d\u5efa\u4e0e\u9a8c\u8bc1 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/115510.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u7b2c\u4e8c\u7ae0Agent \u5f00\u53d1\u73af\u5883\u642d\u5efa\u4e0e\u9a8c\u8bc1 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u6211\u4eec\u7684\u76ee\u6807\u662f\u8ba9\u5f00\u53d1\u73af\u5883\u8fbe\u5230"\u80fd\u8dd1\u901a\u4e00\u4e2a\u6700\u5c0f Agent"\u7684\u72b6\u6001：\u5b89\u88c5\u4f9d\u8d56\u3001\u914d\u7f6e\u6a21\u578b\u3001\u5c01\u88c5\u5de5\u5177\u3001\u8fd0\u884c\u8c03\u7528\u3001\u4fdd\u5b58\u8f68\u8ff9\u3001\u9a8c\u8bc1\u884c\u4e3a\u3002\u540c\u65f6，\u6240\u6709\u793a\u4f8b\u4ee3\u7801\u90fd\u5c3d\u91cf\u8865\u6210\u53ef\u76f4\u63a5\u62f7\u8d1d\u8fd0\u884c\u7684\u7248\u672c，\u5e76\u5728\u9700\u8981\u7279\u6b8a\u914d\u7f6e\u7684\u5730\u65b9\u7ed9\u51fa\u8fd0\u884c\u65b9\u5f0f\u8bf4\u660e\u3002\u4e00\u3001\u73af\u5883\u642d\u5efa\u524d\u7684\u4e09\u4e2a\u9009\u62e9\u5728\u52a8\u624b\u4e4b\u524d，\u9700\u8981\u5148\u786e\u5b9a\u4e09\u4ef6\u4e8b：\u9009\u62e9\u9879\u8bf4\u660e\u5efa\u8bae\u5f00\u53d1\u8bed\u8a00\u76ee\u524d Python \u751f\u6001\u6700\u6210\u719f\u521d\u5b66\u8005\u4f18\u5148 Python\u6a21\u578b\u63a5\u5165\u51b3\u5b9a Agent \u7684\u63a8\u7406\u80fd\u529b\u5148\u7528\u4e00\u4e2a\u7a33\u5b9a\u6a21\u578b\u8dd1\u901a\u6846\u67b6\u5c42\u7ea7\u51b3\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/115510.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-10-12T02:32:54+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"12 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/115510.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/115510.html\",\"name\":\"\u7b2c\u4e8c\u7ae0Agent \u5f00\u53d1\u73af\u5883\u642d\u5efa\u4e0e\u9a8c\u8bc1 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-10-12T02:32:54+00:00\",\"dateModified\":\"2026-10-12T02:32:54+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/115510.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/115510.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/115510.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u7b2c\u4e8c\u7ae0Agent \u5f00\u53d1\u73af\u5883\u642d\u5efa\u4e0e\u9a8c\u8bc1\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"\u7b2c\u4e8c\u7ae0Agent \u5f00\u53d1\u73af\u5883\u642d\u5efa\u4e0e\u9a8c\u8bc1 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/115510.html","og_locale":"zh_CN","og_type":"article","og_title":"\u7b2c\u4e8c\u7ae0Agent \u5f00\u53d1\u73af\u5883\u642d\u5efa\u4e0e\u9a8c\u8bc1 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u6211\u4eec\u7684\u76ee\u6807\u662f\u8ba9\u5f00\u53d1\u73af\u5883\u8fbe\u5230\"\u80fd\u8dd1\u901a\u4e00\u4e2a\u6700\u5c0f Agent\"\u7684\u72b6\u6001：\u5b89\u88c5\u4f9d\u8d56\u3001\u914d\u7f6e\u6a21\u578b\u3001\u5c01\u88c5\u5de5\u5177\u3001\u8fd0\u884c\u8c03\u7528\u3001\u4fdd\u5b58\u8f68\u8ff9\u3001\u9a8c\u8bc1\u884c\u4e3a\u3002\u540c\u65f6，\u6240\u6709\u793a\u4f8b\u4ee3\u7801\u90fd\u5c3d\u91cf\u8865\u6210\u53ef\u76f4\u63a5\u62f7\u8d1d\u8fd0\u884c\u7684\u7248\u672c，\u5e76\u5728\u9700\u8981\u7279\u6b8a\u914d\u7f6e\u7684\u5730\u65b9\u7ed9\u51fa\u8fd0\u884c\u65b9\u5f0f\u8bf4\u660e\u3002\u4e00\u3001\u73af\u5883\u642d\u5efa\u524d\u7684\u4e09\u4e2a\u9009\u62e9\u5728\u52a8\u624b\u4e4b\u524d，\u9700\u8981\u5148\u786e\u5b9a\u4e09\u4ef6\u4e8b：\u9009\u62e9\u9879\u8bf4\u660e\u5efa\u8bae\u5f00\u53d1\u8bed\u8a00\u76ee\u524d Python \u751f\u6001\u6700\u6210\u719f\u521d\u5b66\u8005\u4f18\u5148 Python\u6a21\u578b\u63a5\u5165\u51b3\u5b9a Agent \u7684\u63a8\u7406\u80fd\u529b\u5148\u7528\u4e00\u4e2a\u7a33\u5b9a\u6a21\u578b\u8dd1\u901a\u6846\u67b6\u5c42\u7ea7\u51b3","og_url":"https:\/\/www.wsisp.com\/helps\/115510.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-10-12T02:32:54+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"12 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/115510.html","url":"https:\/\/www.wsisp.com\/helps\/115510.html","name":"\u7b2c\u4e8c\u7ae0Agent \u5f00\u53d1\u73af\u5883\u642d\u5efa\u4e0e\u9a8c\u8bc1 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-10-12T02:32:54+00:00","dateModified":"2026-10-12T02:32:54+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/115510.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/115510.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/115510.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"\u7b2c\u4e8c\u7ae0Agent \u5f00\u53d1\u73af\u5883\u642d\u5efa\u4e0e\u9a8c\u8bc1"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/115510","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=115510"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/115510\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=115510"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=115510"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=115510"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=115510"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}