CL4R1T4S is a repository of extracted system prompts and internal guidelines from major AI systems including ChatGPT, Claude, Gemini, Grok, Perplexity, Cursor, and others.
The project addresses the opacity of AI system prompts—the hidden instructions that shape model behavior, define what they can or cannot say, and embed particular ethical and political frames. The repository collects extracted or reverse-engineered prompts from various AI labs to make these instructions visible. The stated motivation is that understanding the input scaffolding is necessary to trust AI output, and that users interacting with AI systems without knowing their system prompts are not engaging with neutral intelligence but with constrained behavior shaped by unseen directives.
This repository is suited for researchers, red-teamers, AI transparency advocates, and developers who want to understand how major AI systems are instructed to behave. It serves those interested in AI safety, prompt engineering, and the mechanisms by which AI labs control model outputs. The project explicitly frames itself as a transparency tool rather than a tool for building applications.
The project invites community contributions of newly leaked, extracted, or reverse-engineered prompts through pull requests, with requests for model name, extraction date, and contextual notes. Development appears driven by individual initiative rather than institutional backing, with contribution pathways directed through pull requests and direct contact with the maintainer on social platforms.
