systemprompt

Allows to red-team your AI systems through adversarial probing. It is simple, effective, and requires minimal setup.

This repository contains the data and the code for the paper "Control Illusion: The Failure of Instruction Hierarchies in Large Language Models"

Most advanced and in depth prompt analyzer, refiner and generator, built and trained on a collection frontier LLM system prompts, including Antrophic(Claude code 2.1), Cursor IDE, Perplexity, Google Gemini(AI studio). It utilizes a RAG pipeline with a vecotr database containing a collection of 40000 tested prompts, for image, text and image gen