we need an agent loop that doesn't stop to ask for permissions to continue even though the initial user prompt instruction is clear to do otherwise.