I ran into problems trying to deploy AI agents in the wild because people were trying to hijack and jailbreak them. As a result, I started researching and have begun working on a tool to scan LLM system prompts for vulnerabilities so that they can be fixed pre-deployment or at least anticipated. I'm currently taking baby steps by pulling together various tools from the open-source domain and trying to make them easy to use.
It's completely free for now. The report gets emailed to you within 24 hours.