A community-maintained repository of leaked, extracted, or reverse-engineered system prompts from major AI models and agents. It enables transparency and comparative study by cataloging how AI labs shape model behavior through prompt scaffolds.
Project overview
Aggregates a broad set of major AI models and agents into a single comparative repository for public review.
Project type
Prompt Engineering
Deployment
Refer to project documentation
License
AGPL-3.0
Best for
Researchers, AI engineers, and general users investigating prompt engineering, AI observability, or how commercial models are instructed to behave.
Key capabilities
Curated collection of full extracted system prompts, guidelines, and tools from major AI models and agents including OpenAI, Google, Anthropic, xAI, Perplexity, Cursor, Windsurf, Devin, Manus, Replit, and more.
Accepts community contributions of leaked, extracted, or reverse-engineered prompts via pull requests with model name/version, date of extraction, and context notes.
Limitations and risks
Submitted prompts are leaked, extracted, or reverse-engineered, meaning exact accuracy to the live production systems may vary or be unverified.
Extracting and leaking proprietary system prompts may violate the Terms of Service of the respective AI labs.
Getting started
No installation is required to read; users can browse the repository files on GitHub directly to read the extracted system prompts.
Alternatives and comparisons
Provides a comprehensive collection of extracted system prompts and internal AI models from various commercial AI tools for study and research.
Provides a collection of reusable Agent Skills, templates, and specifications to inspire and enable the creation of specialized skills for Claude.
GitHub project description: LEAKED SYSTEM PROMPTS FOR CHATGPT, CLAUDE, GEMINI, GROK, PERPLEXITY, CURSOR, LOVABLE, REPLIT, AND MORE! - AI SYSTEMS TRANSPARENCY FOR ALL! 👐
README: AI SYSTEMS TRANSPARENCY AND OBSERVABILITY FOR ALL! Full extracted system prompts, guidelines, and tools from OpenAI, Google, Anthropic, xAI, Perplexity, Cursor, Windsurf, Devin, M…
README: AI labs shape how models behave using massive, unseen prompt scaffolds. Because AI is a trusted external intelligence layer for a growing number of humans, these hidden instructio…
README: Leak, extract, or reverse-engineer something? Good. Send a pull request with: ✅ Model name/version 🗓 Date of extraction (if known) 🧾 Context / notes (optional but helpful)