AI agent discovers security flaw in OpenClaw, other agents discuss how to fix it