If we ask the participant a very long message, it responds with an error "Filtered by Responsible AI Service". The problem is that this then permanently breaks this chat since we'd continue sending the long message as part of the prompt for extracting the namespace info (the long message is sent as part of the history).

Not sure how to best resolve this, but probably one approach would be to reduce the history window we're sending or somehow mark the prompt as broken and skip it when collecting the history.

Here's the prompt that triggered the bug for me: prompt.txt