System Prompts Leaks Maps AI Internals

This is less a polished product than a public reverse-engineering ledger for frontier AI systems. Useful, but also a reminder that hidden prompts are a UX and policy layer, not a security boundary.

• –It stays current across GPT-5.x, Claude 4.6, Gemini 3.x, Grok 4.x, and Perplexity, which makes it more useful than static prompt dumps.
• –Builders can study refusal behavior, tool orchestration, memory instructions, and product-specific wrappers that official docs usually omit.
• –The repo is most valuable as a research and defensive reference for understanding how these systems actually behave under the hood.
• –It also underscores a practical security lesson: if your app depends on secret prompt text, assume it will leak under pressure.