Agent规划系统 - ReAct与Plan-Execute

前置知识：需要先掌握 AI Agent基础

本文重点：Agent规划框架与实现

---

一、规划策略概述

Agent规划策略：
1. ReAct (Reasoning + Acting)
   - 思考 → 行动 → 观察 → 循环
   - 适合交互式任务
2. Plan-and-Execute
   - 先制定完整计划
   - 再逐步执行
   - 适合复杂多步骤任务
3. Self-Reflection
   - 执行后反思
   - 持续改进

---

二、ReAct框架

2.1 ReAct循环

from langchain.agents import create_react_agent
from langchain_openai import ChatOpenAI
from langchain.tools import Tool
# 定义工具
tools = [
    Tool(name="search", func=search_func, description="搜索信息"),
    Tool(name="calculate", func=calc_func, description="计算")
]
# ReAct提示模板
react_prompt = """
你需要回答问题。可以使用以下工具：
{tools}
使用格式：
Question: 输入问题
Thought: 思考该做什么
Action: 工具名称
Action Input: 工具输入
Observation: 工具输出
... (重复 Thought/Action/Observation)
Thought: 我知道最终答案了
Final Answer: 最终答案
Question: {input}
Thought: {agent_scratchpad}
"""
# 创建Agent
llm = ChatOpenAI(model="gpt-4")
agent = create_react_agent(llm, tools, react_prompt)

2.2 自定义ReAct Agent

class ReActAgent:
    """自定义ReAct Agent"""
    
    def __init__(self, llm, tools, max_iterations=10):
        self.llm = llm
        self.tools = {tool.name: tool for tool in tools}
        self.max_iterations = max_iterations
    
    def run(self, question):
        scratchpad = ""
        
        for _ in range(self.max_iterations):
            thought = self._think(question, scratchpad)
            scratchpad += f"\nThought: {thought}"
            
            if "Final Answer:" in thought:
                return self._extract_answer(thought)
            
            action, action_input = self._parse_action(thought)
            
            if action in self.tools:
                observation = self.tools[action].func(action_input)
                scratchpad += f"\nAction: {action}\nAction Input: {action_input}\nObservation: {observation}"
        
        return "达到最大迭代次数"

---

三、Plan-and-Execute

from langchain_experimental.plan_and_execute import (
    PlanAndExecute,
    load_agent_executor,
    load_chat_planner
)
llm = ChatOpenAI(model="gpt-4")
# 创建规划器
planner = load_chat_planner(llm)
# 创建执行器
executor = load_agent_executor(llm, tools, verbose=True)
# 创建Plan-and-Execute Agent
agent = PlanAndExecute(
    planner=planner,
    executor=executor,
    verbose=True
)
# 执行复杂任务
result = agent.invoke("帮我研究Python异步编程的最佳实践并总结")

---

四、LangGraph工作流

from langgraph.graph import StateGraph, END
from typing import TypedDict
class AgentState(TypedDict):
    messages: list
    current_step: str
    result: str
# 定义节点
def analyze(state):
    state["current_step"] = "analyze"
    return state
def plan(state):
    state["current_step"] = "plan"
    return state
def execute(state):
    state["current_step"] = "execute"
    return state
# 构建图
workflow = StateGraph(AgentState)
workflow.add_node("analyze", analyze)
workflow.add_node("plan", plan)
workflow.add_node("execute", execute)
workflow.set_entry_point("analyze")
workflow.add_edge("analyze", "plan")
workflow.add_edge("plan", "execute")
workflow.add_edge("execute", END)
# 编译运行
app = workflow.compile()
result = app.invoke({"messages": ["任务"], "current_step": "", "result": ""})

---

参考资源

• ReAct论文
• LangGraph文档

---

返回：AI Agent智能体 最后更新: 2026年4月20日

访问 -- 赞 -- 踩 --

讨论与反馈