CL4R1T4S
CL4R1T4S publishes extracted system prompts, guidelines and internal tool definitions from major AI assistants and agents, including OpenAI, Anthropic, Google, xAI, Cursor, Windsurf and Replit, as a public transparency archive.
The project's stated purpose is letting people see the hidden instructions that shape what an assistant can say, what persona or function it is made to follow, and what topics it is told to refuse or redirect, on the argument that trusting an AI's output requires understanding what shaped its input.
Contributors submit newly extracted or reverse-engineered prompts by pull request, including the model name and version, an extraction date where known, and optional context notes. Entries are organized by company and model, covering chat products, coding assistants and agent tools rather than a single vendor.
AGPL-3.0 licensed.