dtrail/system_prompts_leaks

This repository is a collection of system prompts, the hidden instructions an AI chatbot is given before it ever sees your first message, captured word for word from tools people actually use. It covers products from Anthropic such as Claude, Claude Code, and Claude Design, from OpenAI such as ChatGPT and Codex, from Google such as Gemini and Antigravity, along with xAI's Grok, Perplexity, GitHub Copilot, Cursor, VS Code, DeepSeek, Kimi, and several smaller tools. The files are organized into folders by company or product, each holding markdown files with the prompt text for a specific model or version. A table at the top of the README tracks what was added most recently, so you can see, for example, when a new Claude Code system prompt or a Perplexity prompt was captured. Some entries also link to an official published version alongside the captured one, for comparison. According to the README, the collection has been referenced by outside coverage, including an interactive story from The Washington Post and a data dashboard built by a European policy research group, both of which used these captured prompts to illustrate what instructions sit behind a chatbot's replies. This is a documentation project rather than a piece of software. There is nothing to install or run: the value is in reading the prompt files themselves to understand how a given AI product is instructed to behave, what tools it has access to, and what rules it follows before responding to a user. Contributions are welcome through pull requests, and the collection is described as updated regularly as new models and tools are captured.