An error occurred (ValidationException) when calling the Converse operation: The content field in the Message object at messag es.5 is empty. Add a ContentBlock object to the content field and try again.

[parth-patel-flentas]

❗️ValidationException: The content field in the Message object is empty

💥 Error

I'm encountering the following error when using ChatBedrockConverse with LangGraph and AWS Bedrock:

An error occurred (ValidationException) when calling the Converse operation: 
The content field in the Message object at messages.5 is empty. 
Add a ContentBlock object to the content field and try again.

---

🧩 Context

I'm building a multi-agent LangGraph app using ChatBedrockConverse (claude-3-sonnet) with a supervisor and multiple domain agents.

Before invoking the graph, I sanitize the conversation history using this logic:

def is_valid_message_content(content) -> bool:
    if not content:
        return False

    if isinstance(content, str):
        return bool(content.strip())

    if isinstance(content, list):
        for block in content:
            if isinstance(block, dict):
                if block.get('type') == 'text' and block.get('text', '').strip():
                    return True
            elif isinstance(block, str) and block.strip():
                return True
        return False

    return True

Despite cleaning messages before calling .invoke(), I'm still hitting the error.

---

📌 Suspected Cause

The error seems to happen after tool_calls return and a message is added back to the graph with an empty content field.

Examples:

• An AIMessage with tool_calls but no content
• A message with content=[] or content=""
• A routing transfer message like "Transferring to supervisor" with no real content

---

🧪 Steps to Reproduce

1. Setup a LangGraph with ChatBedrockConverse and multiple agents
2. Add tool calls
3. Return messages back to the supervisor
4. See failure in .invoke() when message history includes an empty message

---

🧠 What I’ve Tried

• Adding this patch for tool_call-only messages:

if isinstance(msg, AIMessage):
    if hasattr(msg, 'tool_calls') and msg.tool_calls:
        if not msg.content or (isinstance(msg.content, str) and not msg.content.strip()):
            msg = AIMessage(
                content="[Tool execution]",
                tool_calls=msg.tool_calls
            )

• Filtering out placeholder messages like "Transferring to supervisor"
• Ensuring all BaseMessage objects have valid content

---

❓Questions

1. Is there a preferred way to patch or transform tool-call messages before calling Bedrock's converse?
2. Does ChatBedrockConverse require a specific content format when using tool calls?
3. Should I manually transform all content=[] to [{"type": "text", "text": "[placeholder]"}]?

---

📎 Dependencies

• langgraph: latest (0.0.28+)
• langchain-aws: latest
• aws-sdk: properly configured for Bedrock
• Model: anthropic.claude-3-sonnet-20250219-v1:0 via Bedrock

---

Any guidance on how to sanitize or transform messages for Bedrock's converse() would be appreciated!

Thanks 🙏

[zoltan-fedor]

I am hitting the same issue with LLama 3 on Bedrock.
Funny thing is that this used to work a few weeks / months ago, so either one of the Langchain / Langgraph module's introduced this issue or Bedrock did.

[cygniv404]

with create_supervisor and getting empty array [] as a response after transferring back to supervisor.

Do we have an explanation on this behaviour and how to mitigate it?