Cataloging the hidden instructions behind custom GPTs

What it does

This repository collects system prompts from ChatGPT and custom GPTs into a searchable catalog. It positions itself as an educational resource for learning how system prompts are structured and how custom GPTs are built. The project also links to guides on extracting these instructions and protecting them from leakage.

The interesting bit

Rather than treating extracted prompts as mere curiosities, the project frames them as cautionary tales for prompt injection security. The maintainers note that GPT authors have tightened their defenses after seeing their instructions broken down here, suggesting the collection doubles as a public awareness exercise.

Key highlights

• Large collection of real-world system prompts (browsable via TOC.md)
• idxtool for locally indexing the catalog
• Linked guides for getting started, security hardening, and contributing
• Automated TOC generation via GitHub Actions
• Explicit focus on educational value and injection awareness

Caveats

• The README is mostly a landing page; actual prompts and methodologies reside in linked files we can’t preview here
• GitHub classifies the repo as “HTML,” which is misleading given the content is primarily text-based prompts

Verdict

A useful reference if you build custom GPTs and want to study real instruction patterns—or learn how they leak. If you need a programmatic framework rather than a reading library, look elsewhere.