new

README.md

@@ -60,10 +60,20 @@ driver.quit()
 
 
 ## 安装依赖
-建议使用 python3.9，安装依赖：
+安装python的3.9+版本。
 
+安装包依赖管理工具 poetry，打开 PowerShell 或命令提示符，使用以下命令下载并安装 Poetry：
+```cmd
+curl -sSL https://install.python-poetry.org/ | python
+```
+配置 poetry，将虚拟环境配置到当前项目目录：
 ```commandline
-pip install -r requirements.txt
+poetry config virtualenvs.in-project true
+```
+创建虚拟环境并安装依赖：
+```commandline
+poetry env use python
+poetry install
 ```
 
 ## 运行

agent/plan_agent.py

@@ -1,15 +1,16 @@
 import logging
 from agent.agent_base import AgentBase
-from tools.tools_base import ToolsBase
+from tools.tool_base import ToolBase
 from tools.tools_factory import ToolsFactory
 from work_principle.okr_principle import OKR_Object
 
 class PlanAgent(AgentBase):
     def __init__(self):
         self.logger = logging.getLogger(__name__)
-        super().__init__('你是一名计划拆解者，挑选工具列表中合适的工具将object拆解成key_result然后自动填充参数，建议拆解成多个key_result，' + 
-                         '多个key_result完成这一个目标，返回格式如下：[{"tools_name":"web_browser","request_param":"请求参数"}]')
-
+        super().__init__('#Goals:\n你是一名计划拆解者，将object拆解成key_result，\n#Workflow:1.')
+        '''
+        '''
+    # '多个key_result完成这一个目标，返回格式如下：{"tools_name":"web_browser","request_param":"请求参数"}'
     # 根据object和已有的工具能力拆解成key result，key
     def aligning(self, okr_object: OKR_Object):
         raw = okr_object.raw_user_task

main.py

@@ -6,25 +6,25 @@ import logging
 # 设置日志
 logging.basicConfig(level=logging.INFO)
 
+
 class AutoMate:
     def __init__(self):
         pass
-    
-    
+
     def rule_define(self):
         # 与用户对齐任务
         while True:
-            o_kr = OKR_Object("因为想要增加编程效率，对比一下copilot和curson谁更好用，比较提示词数量、安装易用性，给出不少于100字的文章")
+            o_kr = OKR_Object(
+                "因为想要增加编程效率，对比一下copilot和curson谁更好用，比较提示词数量、安装易用性，给出不少于100字的文章")
             ManagerAgent().optimization_Object(o_kr)
             r = input(f"最终对齐的任务是：{o_kr.raw_user_task}，一切都OK对吧？y/n\n")
             if r == "y":
                 break
-        
+
         # 让计划拆解者拆解任务
         PlanAgent().aligning(o_kr)
 
 
-
 if __name__ == "__main__":
     automator = AutoMate()
     automator.rule_define()

pyproject.toml

@@ -1,19 +0,0 @@
-[tool.poetry]
-name = "automate"
-version = "0.1.0"
-description = ""
-authors = ["yuruo"]
-readme = "README.md"
-
-[tool.poetry.dependencies]
-python = "^3.10"
-PyYAML = "6.0.1"
-requests = "2.31.0"
-selenium = "4.16.0"
-webdriver-manager = "4.0.1"
-langchain = "0.1.7"
-langchain-openai = "0.0.6"
-
-[build-system]
-requires = ["poetry-core"]
-build-backend = "poetry.core.masonry.api"

tests/test_plan_agent.py

@@ -1,10 +1,9 @@
-import unittest
 
 from agent.plan_agent import PlanAgent
 from work_principle.okr_principle import OKR_Object
 
 
-class TestPlanAgent(unittest.TestCase):
+class TestPlanAgent:
     def test_aligning(self):
         # Create an instance of the PlanAgent class
         plan_agent = PlanAgent()
@@ -15,5 +14,3 @@ class TestPlanAgent(unittest.TestCase):
         plan_agent.aligning(okr_object)
 
 
-if __name__ == '__main__':
-    unittest.main()

tests/test_web_browser.py

@@ -0,0 +1,39 @@
+import logging
+import unittest
+
+from langchain.agents import create_openai_functions_agent, AgentExecutor
+from langchain_core.messages import HumanMessage
+from langchain_core.prompts import SystemMessagePromptTemplate, PromptTemplate, MessagesPlaceholder, \
+    HumanMessagePromptTemplate, ChatPromptTemplate
+from langchain_core.utils.function_calling import convert_to_openai_function
+from langchain_openai import ChatOpenAI
+
+from tools.web_browser_tool import WebBrowserTool
+from utils.llm_util import LLMUtil
+
+
+class TestWebBrowser:
+    def test_web_browser(self):
+        model = LLMUtil().llm()
+        tools = [WebBrowserTool()]
+        model_with_functions = model.bind_functions(tools)
+        s = model_with_functions.invoke([HumanMessage(
+            content="帮我查询一下这个网页的内容 https://mbd.baidu.com/newspage/data/landingsuper?context=%7B%22nid%22%3A%22news_9510051560337988929%22%7D&n_type=-1&p_from=-1")])
+        print(s.additional_kwargs["function_call"])
+
+    def test_agent(self):
+        prompt = ChatPromptTemplate.from_messages(
+            [
+                SystemMessagePromptTemplate(
+                    prompt=PromptTemplate(input_variables=[], template='你是一个工作助手')),
+                MessagesPlaceholder(variable_name='chat_history', optional=True),
+                HumanMessagePromptTemplate(prompt=PromptTemplate(input_variables=['input'], template='{input}')),
+                MessagesPlaceholder(variable_name='agent_scratchpad')
+            ]
+        )
+        model = LLMUtil().llm()
+        tools = [WebBrowserTool()]
+        agent = create_openai_functions_agent(model, tools, prompt)
+        agent_executor = AgentExecutor(agent=agent, tools=tools, verbose=True, return_intermediate_steps=True)
+        # agent_executor.invoke({"input": "你好!你是谁"})
+        agent_executor.invoke({"input": "帮我查询一下这个网页的内容 https://mbd.baidu.com/newspage/data/landingsuper?context=%7B%22nid%22%3A%22news_9510051560337988929%22%7D&n_type=-1&p_from=-1"})

tools/llm_tools.py

@@ -1,10 +1,10 @@
 from langchain_core.messages import SystemMessage, HumanMessage
 from langchain_openai import ChatOpenAI
 
-from tools.tools_base import ToolsBase
+from tools.tool_base import ToolBase
 
 
-class LLMTools(ToolsBase):
+class LLMTools(ToolBase):
     def __init__(self):
         super().__init__()
         self.name = "llm_tools"

tools/tool_base.py

@@ -0,0 +1,13 @@
+from abc import ABC
+from typing import Any
+
+from langchain_core.tools import BaseTool
+
+from utils.config import Config
+
+
+class ToolBase(BaseTool, ABC):
+    def get_info(self):
+        return {"name": self.name, "description": self.description, "param": self.request_param, "return_content": self.return_content}
+
+

tools/tools_base.py

@@ -1,16 +0,0 @@
-from utils.config import Config
-
-
-class ToolsBase:
-    def __init__(self):
-        self.name = ""
-        self.description = ""
-        self.request_param = ""
-        self.return_content = ""
-        self.config = Config()
-        
-    def get_info(self):
-        return {"name": self.name, "description": self.description, "param": self.request_param, "return_content": self.return_content}
-
-    def run(self, param=None):
-        pass

tools/web_browser_element_tools.py

@@ -9,9 +9,10 @@ from webdriver_manager.opera import OperaDriverManager
 from selenium.webdriver.edge.service import Service as EdgeService
 from selenium.webdriver.common.by import By
 
-from tools.tools_base import ToolsBase
+from tools.tool_base import ToolBase
 
-class WebBrowserElement(ToolsBase):
+
+class WebBrowserElement(ToolBase):
     def __init__(self):
         super().__init__()
         self.name = "web_element"
@@ -28,5 +29,3 @@ class WebBrowserElement(ToolsBase):
         elif action == "send_text":
             driver.find_element(By.XPATH, xpath).send_keys(param["text"])
         return True
-
-       

tools/web_browser_tool.py

@@ -0,0 +1,55 @@
+import logging
+import os
+from typing import Optional, Type, Any
+
+from langchain_core.tools import BaseTool
+from selenium.webdriver.chrome.service import Service as ChromeService
+from langchain_core.callbacks import CallbackManagerForToolRun, AsyncCallbackManagerForToolRun
+from langchain.pydantic_v1 import BaseModel, Field
+from selenium import webdriver
+from webdriver_manager.chrome import ChromeDriverManager
+from webdriver_manager.microsoft import EdgeChromiumDriverManager
+from selenium.webdriver.edge.service import Service as EdgeService
+
+from tools.tool_base import ToolBase
+from utils.config import Config
+
+
+class SearchInput(BaseModel):
+    url: str = Field(description="要查询的网址")
+
+
+class WebBrowserTool(ToolBase):
+    name = "web_browser"
+    description = "利用浏览器访问url，得到网页源码"
+    args_schema: Type[BaseModel] = SearchInput
+
+    def _run(self, url: str, run_manager: Optional[CallbackManagerForToolRun] = None) -> str:
+        """Use the tool."""
+        config = Config()
+        # Download webdriver based on browser type
+        browser_type = config.BROWSER.get("browser_type")
+        if browser_type == "chrome":
+            options = webdriver.ChromeOptions()
+            options.add_argument("--headless")  # Enable headless mode
+            webdriver_manager = ChromeDriverManager()
+            driver = webdriver.Chrome(service=ChromeService(webdriver_manager.install()), options=options)
+        elif browser_type == "edge":
+            options = webdriver.EdgeOptions()
+            options.add_argument("--headless")  # Enable headless mode
+            webdriver_manager = EdgeChromiumDriverManager()
+            driver = webdriver.Edge(service=EdgeService(webdriver_manager.install()), options=options)
+        else:
+            return ""
+        driver.implicitly_wait(10)
+        driver.get(url)
+        driver.quit()
+        print(f"res {driver.page_source}")
+
+        return driver.page_source
+
+    async def _arun(
+            self, url: str, run_manager: Optional[AsyncCallbackManagerForToolRun] = None
+    ) -> str:
+        """Use the tool asynchronously."""
+        raise NotImplementedError("custom_search does not support async")

tools/web_browser_url_tools.py

@@ -5,10 +5,10 @@ from webdriver_manager.microsoft import EdgeChromiumDriverManager
 from selenium.webdriver.edge.service import Service as EdgeService
 from selenium.webdriver.chrome.service import Service as ChromeService
 
-from tools.tools_base import ToolsBase
+from tools.tool_base import ToolBase
 
 
-class WebBrowserUrl(ToolsBase):
+class WebBrowserUrl(ToolBase):
     def __init__(self):
         super().__init__()
         self.name = "web_browser"

utils/llm_util.py

@@ -0,0 +1,15 @@
+from langchain_openai import ChatOpenAI
+
+from utils.config import Config
+
+
+class LLMUtil:
+    def __init__(self):
+        super().__init__()
+        self.config = Config()
+        self.api_key = self.config.OPEN_AI.get("api_key")
+        self.base_url = self.config.OPEN_AI.get("api_url")
+
+    def llm(self):
+        return ChatOpenAI(temperature=0.7, model_name="gpt-3.5-turbo", openai_api_key=self.api_key,
+                          openai_api_base=self.base_url)