Curating the hidden instructions behind popular chatbots

What it does

This repository gathers leaked system prompts from widely used LLM-based services into a single, structured collection. Contributors submit documents through pull requests or issues, and the maintainer verifies each leak against reproducible prompts or verifiable sources before merging. The result is a curated reference library of instructions that normally live behind an API.

The interesting bit

The collection is explicitly cited in academic papers, making it a de facto research dataset for studying commercial prompt engineering. The maintainer also walks a careful line around DMCA takedowns by refusing sensitive commercial source code, focusing only on the prompt text itself.

Key highlights

• Only accepts submissions with verifiable sources or reproducible prompts.
• Maintained through both PRs and Issues, with the owner doing manual vetting.
• Explicitly referenced in research papers studying LLM behavior.
• Avoids hosting sensitive source code to minimize legal risk.

Caveats

• The README does not list which specific services or models are represented.
• It is unclear what threshold makes a prompt “reproducible” versus simply claimed.

Verdict

Useful for researchers, prompt engineers, and the AI-curious who want to see how closed chatbots are instructed. Not a tool for developers building or deploying their own models.