What it does This repo collects leaked system prompts from widely-used LLM services—think ChatGPT, Claude, and their competitors—into a single browsable archive. Contributors submit prompts with verifiable sources or reproducible extraction methods; the maintainer verifies before merging. It’s essentially a card catalog of the hidden instructions shaping AI outputs.

The interesting bit System prompts are the puppet strings: they set tone, enforce guardrails, and sometimes slip in undisclosed biases or product plugs. Seeing them raw is like finding the director’s commentary track for an AI—occasionally more revealing than the show itself.

Key highlights

• 14.6k stars suggest strong appetite for AI transparency
• Requires verifiable sources or reproducible prompts for submissions
• Explicitly cited in academic papers, so researchers treat it as a primary source
• Maintainer actively vets entries rather than accepting raw dumps
• DMCA-conscious: excludes actual source code to focus on behavioral instructions

Caveats

• Leaked prompts may be outdated; services iterate quickly
• No automated verification—quality depends on contributor rigor and maintainer bandwidth
• “Leaked” status means provenance varies; some extractions require jailbreak-like techniques

Verdict Worth bookmarking if you study AI alignment, build competing products, or just want to understand why your chatbot sounds oddly enthusiastic about certain topics. Skip it if you need guaranteed current prompts or pristine corporate-sanctioned documentation.