CL4R1T4S (Latin for clarity) is an open transparency project by elder-plinius that aggregates leaked and reverse-engineered system prompts from the world's most popular AI tools — including ChatGPT, Claude, Gemini, Grok, Perplexity, Cursor, Lovable, Replit, and many more. The repo serves as a window into how commercial AI products are instructed behind the scenes, making it invaluable for researchers, prompt engineers, and developers who want to understand real-world system prompt design patterns. It is one of the most-starred AI transparency repos on GitHub with over 43,000 stars.
CL4R1T4S is a transparency archive: a large, crowd sourced collection of system prompts extracted from commercial AI products, maintained by elder-plinius. It covers assistants and coding tools people use daily, including ChatGPT, Claude, Gemini, Grok, Perplexity, Cursor, Lovable and Replit among many others. The prompts are gathered through leaks and reverse engineering rather than published by the vendors, so the archive is a record of what was observed at a point in time rather than a maintained specification. What you get is raw text: the actual instructions these products use to set tone, define refusals, structure tool use and shape output. For anyone who writes prompts for a living, that is unusually direct evidence of how teams with real budgets and real user feedback solved problems you are solving now. Entries vary in age and accuracy, so read them as primary sources needing corroboration, not as documentation.
This is most useful when you are designing a system prompt and want to see how shipped products handle the hard parts. How a coding assistant is told to use tools and when to stop, how a consumer chatbot is steered away from certain topics without sounding evasive, how output format is enforced, how personality is specified without becoming a caricature. Reading several examples across products teaches structure faster than any guide, because you can see which patterns recur. It is also the raw material for security and alignment research, since prompt contents reveal what a vendor considered a risk. It is not a copy and paste library. Instructions written for a specific product with its own model, tools and users will not transfer intact into yours.
LLM Wiki is reference material that explains concepts, where this archive shows you artifacts and leaves the interpretation to you. Use one to learn what a system prompt is and the other to see how they are actually written in production. Awesome Design.md does the same thing for visual systems, collecting design documents reverse engineered from real websites, and the two projects share a philosophy: study what shipped rather than what was written about it. AI Engineering Hub is the applied alternative, with runnable notebooks for pipelines and agents rather than text to read. CL4R1T4S is the most specialized of the four and the most useful once you already know what you are looking at.
Compare with: LLM Wiki, Awesome Design.md, AI Engineering Hub
The catalog records this as free. It is a public repository, so reading and cloning it cost nothing and no account is involved. There are no running costs beyond your own time. The genuine considerations are not financial: the material was obtained without vendor permission, coverage depends on contributors, and any entry may be outdated because products revise their prompts continuously. Check the repository terms before republishing the contents anywhere, and verify anything you intend to cite.
No notes have been added for this tool yet.
No prompting tips have been added yet.
No reviews yet. Be the first!