LLM security for AI coding agents -- protect applications from prompt injection, PII leakage, and data exfiltration