The CL4R1T4S GitHub repository is a publicly accessible collection of extracted or leaked system prompts, internal guidelines, and related tools from leading AI models and agents, including those powering ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google), Grok (xAI), Perplexity, Cursor, Devin, Replit, and several others. Organised by provider into folders with markdown and text files, it reveals the hidden instructions that define each model’s personality, response boundaries, ethical constraints, refusal mechanisms, and sometimes embedded biases or operational tools—offering unprecedented visibility into how these proprietary large language models are directed to behave. With over 13,000 stars and thousands of forks, the project has quickly become a significant resource for developers, researchers, and AI enthusiasts interested in prompt engineering, model transparency, and understanding the “guardrails” that shape AI outputs. While prompts can change with updates and the collection relies on community contributions, it serves as a valuable snapshot of current practices across major AI companies.
Edward Kiledjian
@ekiledjian