如何解决LangChain中agent_scratchpad应为消息列表的ValueError？

问题场景 #

使用Python 3.9，搭配LangChain 0.3.x系列依赖包，运行结构化聊天Agent代码时，抛出错误：

ValueError: variable agent_scratchpad should be a list of base messages, got of type <class 'str'>

错误原因 #

create_structured_chat_agent在LangChain 0.3+版本中，要求agent_scratchpad必须是LangChain消息对象列表，但原代码中使用FakeMessagesListChatModel返回的是仅包含JSON字符串的AIMessage，无法被Agent正确解析为消息列表格式。

解决步骤 #

需要调整两处代码：

• 修改Fake模型的响应消息结构，加入ToolCall类型的消息，匹配结构化聊天Agent的预期格式
• 确保Agent执行时的消息流转符合新版本要求

修改后的完整代码 #

import asyncio
import json
from langchain.agents import AgentExecutor, create_structured_chat_agent, Tool
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.messages import AIMessage, ToolMessage, HumanMessage
from langchain_community.chat_models.fake import FakeMessagesListChatModel

# 1. 定义工具函数
def simple_tool_function(input: str) -> str:
    """A simple tool that returns a fixed string."""
    print(f"Tool called with input: '{input}'")
    return "The tool says hello back!"

tools = [
    Tool(
        name="simple_tool",
        func=simple_tool_function,
        description="A simple test tool.",
    )
]

# 2. 调整Fake模型的响应格式：包含ToolCall和对应的ToolMessage
responses = [
    # 第一步：Agent发起工具调用
    AIMessage(
        content="",
        tool_calls=[{
            "name": "simple_tool",
            "args": {"input": "hello"},
            "id": "tool_call_1"
        }]
    ),
    # 第二步：模拟工具返回结果的ToolMessage
    ToolMessage(
        content="The tool says hello back!",
        tool_call_id="tool_call_1"
    ),
    # 第三步：Agent返回最终答案
    AIMessage(
        content="The tool call was successful. The tool said: 'The tool says hello back!'"
    )
]

llm = FakeMessagesListChatModel(responses=responses)

# 3. 保持原有Prompt结构不变
prompt = ChatPromptTemplate.from_messages([
    ("system", """Respond to the human as helpfully and accurately as possible. You have access to the following tools:

{tools}

Use a json blob to specify a tool by providing an action key (tool name) and an action_input key (tool input).

Valid "action" values: "Final Answer" or {tool_names}

Provide only ONE action per $JSON_BLOB, as shown:

{{
"action": $TOOL_NAME,
"action_input": $INPUT
}}

Follow this format:

Question: input question to answer
Thought: consider previous and subsequent steps
Action:

$JSON_BLOB

Observation: action result
... (repeat Thought/Action/Observation as needed)
Thought: I know what to respond
Action:

{{
"action": "Final Answer",
"action_input": "Final response to human"
}}

Begin! Reminder to ALWAYS respond with a valid json blob of a single action. Use tools if necessary. Respond directly if appropriate. Format is Action:```$JSON_BLOB```then Observation"""),
    ("human", "{input}"),
    MessagesPlaceholder(variable_name="agent_scratchpad"),
])

# 4. 创建Agent和执行器
agent = create_structured_chat_agent(llm, tools, prompt)
agent_executor = AgentExecutor(
    agent=agent,
    tools=tools,
    verbose=True,
    handle_parsing_errors=True,
    max_iterations=3
)

# 5. 运行Agent
result = asyncio.run(agent_executor.ainvoke({"input": "call the tool"}))

关键修改点说明 #

1. 消息结构调整：将原有的纯JSON字符串AIMessage替换为带tool_calls参数的AIMessage，并添加ToolMessage模拟工具返回结果，让Agent能正确识别工具调用流程。
2. 移除冗余JSON包装：最终的答案消息直接返回文本内容，无需再用JSON包裹，符合新版本Agent的输出要求。

内容的提问来源于stack exchange，提问作者hitesh