Validate AI agent system prompts for identity clarity, capability limits, anti-hallucination instructions and jailbreak