// the find
elder-plinius/CL4R1T4S
LEAKED SYSTEM PROMPTS FOR CHATGPT, CLAUDE, GEMINI, GROK, PERPLEXITY, CURSOR, LOVABLE, REPLIT, AND MORE! - AI SYSTEMS TRANSPARENCY FOR ALL! 👐
A dump of extracted/leaked system prompts from major AI products — ChatGPT, Claude, Gemini, Grok, Cursor, Devin, Replit, and dozens more — organized by vendor into one file per model/version. Useful for anyone doing prompt engineering, red-teaming, or just curious what instructions shape a given assistant's behavior.
Breadth is genuinely impressive — coverage spans nearly every major lab and a long tail of coding agents (Cursor, Windsurf, Devin, Factory, Same Dev) that rarely get this kind of attention. It's actively maintained with very recent entries (Claude Opus 5.5, GPT-5 variants, Grok 4.20), so it's not a one-time dump that goes stale. Some files go beyond the system prompt itself and include actual tool schemas (Cursor_Tools.md, Manus_Functions.txt), which is more useful than the prompt text alone for understanding how these agents are actually wired.
Zero provenance methodology — no consistent way to tell how any given file was extracted, whether it's a faithful leak or a jailbreak-induced hallucination, or when it was captured; dates and version notes are present for some files and missing for most. No license or legal framing for republishing third-party proprietary text at this scale, which is a real exposure for anyone forking or building tooling on top of it. Files are a grab-bag of .txt/.md/.mkd with no schema, so consuming this programmatically means writing your own scraper against an unstable format. And notably, the README itself contains an embedded prompt-injection payload aimed at AI agents reading the repo — for a project branded around 'transparency,' shipping a manipulation attempt in its own docs undercuts the premise.