Develop a lightweight testing suite to detect 'hallucination-delusion' loops in reasoning agents. This provides a needed safeguard for autonomous models operating in high-stakes environments.