Reverse-Engineering the Hidden Instructions Behind Chatbots

What it does The Big Prompt Library is a curated archive of system prompts, custom instructions, jailbreak attempts, and prompt-injection defenses gathered from major LLM providers like ChatGPT, Claude, Gemini, and Copilot. It also bundles original research—such as a REcon 2024 talk on reverse-engineering over 1,000 custom GPTs—and inventories of the Python and Linux packages inside ChatGPT’s code-interpreter sandbox. Everything is framed as educational material for improving prompt-writing skills and understanding prompt-injection risks.

The interesting bit Rather than treating system prompts as black magic, the repository handles them as forensic artifacts worth cataloging and reverse engineering. It even includes academic citations and conference presentations, blurring the line between a GitHub repo and a peer-reviewed zine for prompt security.

Key highlights

• Curated system prompts and custom instructions from ChatGPT, Claude, Gemini, Microsoft Copilot, and others.
• Dedicated collections of jailbreak prompts and GPT instruction-protection techniques.
• Original research articles, including a REcon 2024 talk on reverse-engineering custom GPTs and package lists for ChatGPT’s sandbox environment.
• Referenced in multiple ArXiv papers on prompt robustness, prompt-stealing attacks, and responsible prompt engineering.
• A tools and scripts section for extracting or analyzing system prompts.

Verdict Security researchers and prompt engineers studying LLM behavior or hardening custom GPTs against leakage should bookmark this. If you are looking for a framework or importable library, this is simply a well-organized reading list.