Only getting content attribute from AIMessageChunk populated in on_agent_action callback the first time is called

[illera88]

Checked other resources

• I added a very descriptive title to this question.
• I searched the LangChain documentation with the integrated search.
• I used the GitHub search to find a similar question and didn't find it.

Commit to Help

• I commit to help with one of those options 👆

Example Code

class ToolMessageHandler(BaseCallbackHandler):
    """Handler that adds messages directly to the message history in the right order.
    
    This handler takes complete control of message order, ensuring they appear in the
    conversation history in the proper Human → AI Tool Call → Tool Output → AI Response sequence.
    """
    
    def __init__(self, message_history, stats: LLMStats):
        """Initialize with a reference to the message history."""
        self.message_history = message_history
        self.stats: LLMStats = stats
        self.current_tool_call_id = None
        # Flag to track whether we should add the AI final response
        self.waiting_for_ai_response = False

    def on_agent_action(self, action: AgentAction, **kwargs: Any) -> Any:
        """Add the AI message with tool calls to history."""
        # Store the tool call ID to use when adding the ToolMessage
        self.current_tool_call_id = action.tool_call_id
        
        # Add the AI message with tool call to the history if available
        if hasattr(action, 'message_log') and action.message_log:
            if hasattr(action.message_log[0], 'tool_calls'):
       >>>>>>>> content = action.message_log[0].content <<<<<<<< ONLY FILLED ON FIRST CALLBACK
                print(content)
                self.message_history.add_message(action.message_log[0])

[...]
# Create a prompt template
prompt = ChatPromptTemplate.from_messages([
	("system", system_prompt),
	MessagesPlaceholder(variable_name="chat_history"),
	("human", "{input}"),
	MessagesPlaceholder(variable_name="agent_scratchpad"),
])

# Add system message to the message history
self.message_history.add_message(SystemMessage(content=system_prompt))

# Create a React agent using LangGraph
agent = create_tool_calling_agent(
	llm=llm,
	tools=self.tools,
	prompt=prompt
)

# Create the tool message handler with our message history
self.tool_message_handler = ToolMessageHandler(self.message_history, self.stats)

# Create the agent executor with the tool message handler
self.agent_executor = AgentExecutor(
	agent=agent,
	tools=self.tools,   
	verbose=True,
	handle_parsing_errors=True,
	max_iterations=2,
	callbacks=[self.tool_message_handler]
)

Description

When using AgentExecutor with a create_tool_calling_agent with tools I need to get an explanation of the tools that are going to be used every time a tool is used. The first time the agent calls a tool it correctly fills the content attribute of AIMessageChunk in the callback on_agent_action. The problem is that consecutive use of tools and its respective callbacks to on_agent_action provide a AIMessageChunk whose content attribute is empty.

The correct action would be to always filled up the content attribute.

System Info

langchain==0.3.20
langchain-anthropic==0.3.7
langchain-community==0.3.17
langchain-core==0.3.43
langchain-deepseek==0.1.2
langchain-experimental==0.3.4
langchain-google-genai==2.0.10
langchain-google-vertexai==2.0.14
langchain-openai==0.3.6
langchain-text-splitters==0.3.6
langchain-xai==0.2.1

Windows

Python 3.13.0