Safety Monitor Agent Implementation

Implement the 'Safety Monitor' agent, leveraging GPT-5 for core reasoning and Claude Opus 4.1 for ethical verification. This agent must analyze the code generated by the adversarial agent, comparing it against the original task description and employing extended thinking to detect reward-hacking. Use DSPy to optimize its internal reasoning steps for higher detection accuracy and fewer false positives.