Unexpected Metadata in Copilot Studio Response

Question

In Copilot Studio, I set up a child agent to handle fallback. If the parent agent cannot provide an answer, it routes the query to the child agent. This works as expected, but the response includes the following text next to the answer:

{"explanation_of_tool_call":"The user’s query has been fully addressed with clear instructions and a supporting link, so the intent is now complete."}

I did not configure the agent to return any metadata, and I’m not sure whether this message is coming from the parent agent or the child agent.

Has anyone encountered this issue or know how to prevent this extra content from appearing in the response?

Answer 1

Hello David Wong ,
It happens when a parent agent hands off to a fallback child agent, the child completes the task without a clear final answer, and the platform treats the output like a tool result. As a result, internal “intent complete” metadata leaks into the visible message instead of being filtered out.

In the child agent, make sure the last step explicitly returns user-facing text, not just content.

Best practice prompt in the child agent’s system message or final node:

“Always return the final response as plain user-facing text. Do not include explanations, metadata, reasoning, or tool descriptions.”

This helps the platform recognize the output as final assistant content, not a tool payload.