Identifying and Mitigating Vulnerabilities in LLM-Integrated Applications