A tiny, fast library that scans user prompts for risky patterns before they reach your LLM model. It flags likely prompt-injection attempts so you can block, review, or route them differently—without making a model call.