microsoft/semantic-kernel · error · RuntimeError

Unknown participant selected: {response.content}.

Error message

Unknown participant selected: {response.content}.

What it means

Thrown by the custom group-chat manager after the manager LLM picks the next participant. The model's response (parsed into StringResult) yielded a name that is not a key in participant_descriptions, so the manager refuses to hand off to an unknown agent. This is a guard against the LLM hallucinating an agent name.

Source

Thrown at python/samples/getting_started_with_agents/multi_agent_orchestration/step3b_group_chat_with_chat_completion_manager.py:265

        )

        response = await self.service.get_chat_message_content(
            chat_history,
            settings=PromptExecutionSettings(response_format=StringResult),
        )

        participant_name_with_reason = StringResult.model_validate_json(response.content)

        print("*********************")
        print(
            f"Next participant: {participant_name_with_reason.result}\nReason: {participant_name_with_reason.reason}."
        )
        print("*********************")

        if participant_name_with_reason.result in participant_descriptions:
            return participant_name_with_reason

        raise RuntimeError(f"Unknown participant selected: {response.content}.")

    @override
    async def filter_results(
        self,
        chat_history: ChatHistory,
    ) -> MessageResult:
        """Provide concrete implementation for filtering the results of the discussion.

        The manager will filter the results of the conversation after the conversation is terminated.
        """
        if not chat_history.messages:
            raise RuntimeError("No messages in the chat history.")

        chat_history.messages.insert(
            0,
            ChatMessageContent(
                role=AuthorRole.SYSTEM,
                content=await self._render_prompt(

View on GitHub (pinned to c028a0c7dc)

Solutions

1. Ensure participant names in the manager prompt exactly match the keys of participant_descriptions (case-sensitive).
2. Tighten the manager prompt to constrain output to the exact registered names, and lower temperature.
3. Normalize the model output (strip quotes/whitespace, case-fold) before the membership check, or add fuzzy matching against known names.
4. Inspect response.content printed in the error to see what the model actually returned and adjust the schema/prompt accordingly.

Example fix

# before
if participant_name_with_reason.result in participant_descriptions:
    return participant_name_with_reason.result
# after (normalize + fuzzy fallback)
name = participant_name_with_reason.result.strip().strip('"\'')
match = next((k for k in participant_descriptions if k.lower() == name.lower()), None)
if match:
    return match
raise RuntimeError(f"Unknown participant selected: {response.content}.")

name = participant_name_with_reason.result.strip().strip('"\'')
known = set(participant_descriptions)
assert name in known or name.lower() in {k.lower() for k in known}, \
    f'Unknown participant: {name}; known: {sorted(known)}'

def is_known_participant(name: str, descriptions: dict) -> bool:
    n = name.strip().strip('"\'').lower()
    return n in {k.lower() for k in descriptions}

try:
    return await manager.select_next_participant(chat_history)
except RuntimeError as e:
    if 'Unknown participant' in str(e):
        # normalize/reprompt the manager with stricter instructions
        ...
    raise

When it happens

Trigger: The manager completion returns a participant name with different casing/spelling than registered; the model invents a name not in the group; the structured StringResult parsing captured extra text in result.

Common situations: Model returns the agent description instead of its name, includes surrounding quotes/whitespace, or picks a role rather than the configured name; prompt describing participants is ambiguous; temperature too high causing creative names.

Related errors

• No messages in the chat history.
• Failed to parse response: {responseText}
• {nameof(A2AAgent)} is not for use with {nameof(AgentChat)}.
• Missing agent description: {agent.Name ?? agent.Id}
• Entry agent is not defined.

AI-assisted analysis of microsoft/semantic-kernel@c028a0c7dc (2026-08-13). Data as JSON: /api/errors/e4e2a05246312ea7. Report an issue: GitHub.