I built an open-source LLM security scanner that runs in <5ms with zero dependencies

I've been building AI features for a while and kept running into the same problem: prompt injection...