[BUG] Summarizing Conversation Manager throwing ValidationException

[flowconic]

Checks

• I have updated to the lastest minor and patch version of Strands
• I have checked the documentation and this is not expected behavior
• I have searched ./issues and there are no duplicates of my issue

Strands Version

1.3.0

Python Version

3.13.5

Operating System

macOS 15.5 (24F74) (sequoia)

Installation Method

pip

Steps to Reproduce

1. Install requirements

# Database
psycopg2-binary==2.9.*
asyncpg==0.30.*

# AI and chat related
requests==2.32.4
strands-agents[otel]>=1.3.0
strands-agents-tools>=0.2.1

2. Create chat summarisation agent

#file chat_summarization_agent.py
from models.nova import NOVA_PRO_FOR_SUMMARISATION
from strands import Agent
from strands.agent.conversation_manager import SummarizingConversationManager

SUMMARISATION_SYSTEM_PROMPT = """
You are a conversation summarizer. Provide a concise summary of the conversation history.

Format Requirements:
- You MUST create a structured and concise summary in bullet-point format.
- You MUST capture decisions made, actions agreed or goals discussed.
- You MUST NOT respond conversationally.
- You MUST NOT address the user directly.

Task:
Your task is to create a structured summary document:
- It MUST contain bullet points with key topics and questions covered
- It MUST contain bullet points for all significant tools executed and their results
- It MUST contain bullet points for any code or technical information shared
- It MUST contain a section of key insights gained
- It MUST format the summary in the third person

Example format:

## Conversation Summary
* Topic 1: Key information
* Topic 2: Key information
*
## Tools Executed
* Tool X: Result Y
"""


def get_custom_summarization_manager_agent(
    chat_uuid: str, user_uuid: str
) -> SummarizingConversationManager:

    custom_summarization_agent = Agent(
        name="Custom Summarization Agent",
        agent_id="custom_summarization_agent",
        description="An agent specialized in summarizing chat conversations.",
        model=NOVA_PRO_FOR_SUMMARISATION,
        system_prompt=SUMMARISATION_SYSTEM_PROMPT,
        # tools=[save_chat_summary],
        callback_handler=None,
    )
    custom_summarization_agent.state.set("chat_uuid", chat_uuid)
    custom_summarization_agent.state.set("user_uuid", user_uuid)

    conversation_manager = SummarizingConversationManager(
        summary_ratio=0.1,  # ideally increase
        preserve_recent_messages=1,
        summarization_agent=custom_summarization_agent,
    )

    return conversation_manager

3. Setup a coordinator agent (or any agent using this summarization agent)

# filename coordinator.agent.py

async def get_coordinator_agent(event_object):
    COORDINATOR_PROMPT_LOADED = ""
    agent_id = "coordinator_agent1"
    session_manager = get_session(event_object.user_uuid, event_object.chat_uuid)

    if session_manager.session_repository.is_new_session is True:
        # Parallel tasks to fetch initial memories and user profile
        COORDINATOR_PROMPT_LOADED = .....

    conversation_manager = get_custom_summarization_manager_agent(
        chat_uuid=event_object.chat_uuid, user_uuid=event_object.user_uuid
    )
    coordinator_agent = Agent(
        name="Coordinator Agent",
        model=ANTHROPIC_CLAUDE_SONNET_4,
        system_prompt=COORDINATOR_PROMPT_LOADED or COORDINATOR_PROMPT,
        tools=[
            http_request,
            calculator,
        agent_id=agent_id,
        trace_attributes={
            "user_uuid": event_object.user_uuid,
            "tenant_uuid": event_object.tenant_uuid,
            "chat_uuid": event_object.chat_uuid,
        },
        session_manager=session_manager,
        conversation_manager=conversation_manager,
    )

    # Set initial state values for the coordinator agent
    coordinator_agent.state.set("user_uuid", event_object.user_uuid)
    coordinator_agent.state.set("tenant_uuid", event_object.tenant_uuid)
    coordinator_agent.state.set("chat_uuid", event_object.chat_uuid)

    return coordinator_agent

Expected Behavior

When the summarization loads, it adds the previous summarisation from state without errors.

Actual Behavior

When Running the below errors come up

ERROR | strands.agent.conversation_manager.summarizing_conversation_manager | Summarization failed: An error occurred (ValidationException) when calling the ConverseStream operation: A conversation must start with a user message. Try again with a conversation that starts with a user message. | chat/.venv/lib/python3.13/site-packages/strands/agent/conversation_manager/summarizing_conversation_manager.py:166

The captured and loaded chat state from session manager:

{
  "state": {
    "chat_uuid": "5de0daca-b6a7-4b04-9247-c3d02fbb7ddf",
    "user_uuid": "00000000-0000-0000-0000-000000000000",
    "tenant_uuid": "00000000-0000-0000-0000-000000000000"
  },
  "agent_id": "coordinator_agent1",
  "created_at": "2025-08-05T05:43:11.538716+00:00",
  "updated_at": "2025-08-05T05:43:11.538724+00:00",
  "system_prompt": "You are a coordinator who is responsible for answering users chat questions across multiple domains",
  "conversation_manager_state": {
    "__name__": "SummarizingConversationManager",
    "summary_message": {
      "role": "assistant",
      "content": [
        {
          "text": "## Conversation Summary\n* Topic: User inquiry about assistance\n  * User asked how the assistant can help\n  * Assistant explained its role as a conversation summarizer\n* Decisions/Actions:\n  * Assistant committed to summarizing the conversation\n* Goals:\n  * Create a structured and concise summary document\n\n## Tools Executed\n* save_chat_summary: Summary of the conversation saved\n\n## Key Insights\n* The assistant's primary function is to provide structured and concise summaries of conversations.\n* The summary includes key topics, tools executed, code or technical information shared, and key insights gained."
        }
      ]
    },
    "removed_message_count": 1
  }
}

Additional Context

No response

Possible Solution

I believe the issue may be related to
strands/agent/conversation_manager/summarizing_conversation_manager.py and def reduce_context from line 117. which I added below with added comments.

Could the issue be that when the last line of code agent.messages[:] = [self._summary_message] + remaining_messages runs, it then loads the summarization. But as the summarisation state has captured role: assistant, this causes it to fail?

def reduce_context(self, agent: "Agent", e: Optional[Exception] = None, **kwargs: Any) -> None:
        """Reduce context using summarization.

        Args:
            agent: The agent whose conversation history will be reduced.
                The agent's messages list is modified in-place.
            e: The exception that triggered the context reduction, if any.
            **kwargs: Additional keyword arguments for future extensibility.

        Raises:
            ContextWindowOverflowException: If the context cannot be summarized.
        """
        try:
            # Calculate how many messages to summarize
            messages_to_summarize_count = max(1, int(len(agent.messages) * self.summary_ratio))

            # Ensure we don't summarize recent messages
            messages_to_summarize_count = min(
                messages_to_summarize_count, len(agent.messages) - self.preserve_recent_messages
            )

            if messages_to_summarize_count <= 0:
                raise ContextWindowOverflowException("Cannot summarize: insufficient messages for summarization")

            # Adjust split point to avoid breaking ToolUse/ToolResult pairs
            messages_to_summarize_count = self._adjust_split_point_for_tool_pairs(
                agent.messages, messages_to_summarize_count
            )

            if messages_to_summarize_count <= 0:
                raise ContextWindowOverflowException("Cannot summarize: insufficient messages for summarization")

            # Extract messages to summarize
            messages_to_summarize = agent.messages[:messages_to_summarize_count]
            remaining_messages = agent.messages[messages_to_summarize_count:]

            # Keep track of the number of messages that have been summarized thus far.
            self.removed_message_count += len(messages_to_summarize)
            # If there is a summary message, don't count it in the removed_message_count.
            if self._summary_message:
                self.removed_message_count -= 1

            # Generate summary

            # ******** NOTE that this returns return result.message which here would include the assistant role

            self._summary_message = self._generate_summary(messages_to_summarize, agent)

            # Replace the summarized messages with the summary

            # ********** NOTE As this creates the agent.messages, with the first message being what generate_summary returned, which has the role:assistant format. 
            agent.messages[:] = [self._summary_message] + remaining_messages

Related Issues

No response

[pgrayy]

Thank you for raising this issue. I'll start working on a fix. I think we could solve this in two ways:

1. Convert the role of the summarization message to a user role.
2. Leave the summarization as an assistant message but add a leading user message with a question such as "Please summarize our conversation".

[pgrayy]

Update: This appears to be a model specific issue. Nova Pro for example rejects messages starting with an assistant message but Claude does not.

[flowconic]

Thank you for starting this one. Confirming I was using NOVA Pro for this myself, as feel Claude is too costly and less performant and don't need any potential extra accuracy. Speed and cost more important

[pgrayy]

Fix: #698