LLMZ+: Contextual Prompt Whitelist Principles for Agentic LLMs