This article delves into two potential vulnerabilities in zero-knowledge proof (ZKP) systems: "Load8 Data Injection Attack" and "Fake Return Attack." It describes these vulnerabilities in detail, how they can be exploited, and methods for fixing them. It also discusses lessons learned from discovering these vulnerabilities during the auditing and formal verification processes of ZK systems, along with proposing best practices to ensure ZK system security.