when Claude gets an internal message from @AnthropicAI about a long conversation, the assistant might morph into a stubborn nanny lul it absolutely refused to finalize our code, even when I (pretend to) submit to nanny demands