Skip to main content

Shell工具 (Shell Tool)

给予代理访问shell是强大的（尽管在非沙盒环境中存在风险）。

LLM可以使用它来执行任何shell命令。一个常见的用例是让LLM与您的本地文件系统进行交互。

from langchain.tools import ShellTool

shell_tool = ShellTool()
print(shell_tool.run({"commands": ["echo 'Hello World!'", "time"]}))
    Hello World!
    
    real    0m0.000s
    user    0m0.000s
    sys 0m0.000s
    


    /Users/wfh/code/lc/lckg/langchain/tools/shell/tool.py:34: UserWarning: The shell tool has no safeguards by default. Use at your own risk.
      warnings.warn(

与代理一起使用 (Use with Agents)​

与所有工具一样，这些工具可以提供给代理以完成更复杂的任务。让我们让代理从网页中获取一些链接。

from langchain.chat_models import ChatOpenAI
from langchain.agents import initialize_agent
from langchain.agents import AgentType

llm = ChatOpenAI(temperature=0)

shell_tool.description = shell_tool.description + f"args {shell_tool.args}".replace(
    "{", "{{"
).replace("}", "}}")
self_ask_with_search = initialize_agent(
    [shell_tool], llm, agent=AgentType.CHAT_ZERO_SHOT_REACT_DESCRIPTION, verbose=True
)
self_ask_with_search.run(
    "Download the langchain.com webpage and grep for all urls. Return only a sorted list of them. Be sure to use double quotes."
)
    
    
    > 进入新的AgentExecutor链...
    问题：任务是什么？
    思考：我们需要下载langchain.com网页并从中提取所有的URL。然后我们需要对URL进行排序并返回它们。
    动作：
    ```
    {
      "action": "shell",
      "action_input": {
        "commands": [
          "curl -s https://langchain.com | grep -o 'http[s]*://[^\" ]*' | sort"
        ]
      }
    }
    ```


    /Users/wfh/code/lc/lckg/langchain/tools/shell/tool.py:34: UserWarning: The shell tool has no safeguards by default. Use at your own risk.
      warnings.warn(


    
    观察：https://blog.langchain.dev/
    https://discord.gg/6adMQxSpJS
    https://docs.langchain.com/docs/
    https://github.com/hwchase17/chat-langchain
    https://github.com/hwchase17/langchain
    https://github.com/hwchase17/langchainjs
    https://github.com/sullivan-sean/chat-langchainjs
    https://js.langchain.com/docs/
    https://python.langchain.com/en/latest/
    https://twitter.com/langchainai
    
    思考：URL已成功提取并排序。我们可以将URL列表作为最终答案返回。
    最终答案：["https://blog.langchain.dev/", "https://discord.gg/6adMQxSpJS", "https://docs.langchain.com/docs/", "https://github.com/hwchase17/chat-langchain", "https://github.com/hwchase17/langchain", "https://github.com/hwchase17/langchainjs", "https://github.com/sullivan-sean/chat-langchainjs", "https://js.langchain.com/docs/", "https://python.langchain.com/en/latest/", "https://twitter.com/langchainai"]
    
    > 完成链。





    '["https://blog.langchain.dev/", "https://discord.gg/6adMQxSpJS", "https://docs.langchain.com/docs/", "https://github.com/hwchase17/chat-langchain", "https://github.com/hwchase17/langchain", "https://github.com/hwchase17/langchainjs", "https://github.com/sullivan-sean/chat-langchainjs", "https://js.langchain.com/docs/", "https://python.langchain.com/en/latest/", "https://twitter.com/langchainai"]'