Automating Prompt Leakage Attacks on Large Language Models Using Agentic Approach