我担心共享 .cursorrules 的安全性，因此我构建了一个静态分析器来审计它们。

Many users have been utilizing Cursor by grabbing .cursorrules and AI scripts from GitHub and various libraries to enhance productivity. However, this practice raises security concerns as it involves executing untrusted third-party instructions that have full access to source code, terminal, and .env files. To address this issue, the author developed a tool called AgentFend, which employs a static analysis engine named Onyx to scan prompts and scripts before the user hits 'Enter'. The tool currently focuses on detecting data exfiltration and prompt injections, ensuring that users' code and keys are not sent to external URLs, thereby safeguarding user security.