system_prompts_leaks asgeirtj

Extracted system prompts from Anthropic - Claude Fable 5, Opus 5, Claude Design, Claude Code. OpenAI - ChatGPT GPT-5.6-Sol, Codex. Google - Gemini 3.5 Flash, 3.1 Pro, Antigravity. xAI - Grok, Cursor, Copilot, VS Code, Perplexity, and more. Updated regularly.

主要言語
JavaScript
Stars
62,010
Forks
10,124

タグ

  • 教程/资源
  • 提示词工程
  • 历史归档
  • 数据集
  • 安全

概要

该项目专门收集并整理各主流AI产品被泄露的系统提示词(System Prompts),涵盖Anthropic Claude、OpenAI ChatGPT/Codex、Google Gemini、xAI Grok、Cursor、Copilot、Perplexity等众多产品。这些系统提示词是AI在接收用户消息前所预设的隐藏指令与规则,决定了AI的行为方式和限制边界。项目内容逐字记录原始提示词,持续更新最新版本,已被《华盛顿邮报》等媒体引用,适合AI研究者、安全分析师及对大模型内部机制感兴趣的开发者参考。

README

> **The Washington Post** built its interactive story on prompts from this repo: [See the hidden rules behind AI. Then use them to rewrite this article.](https://wapo.st/49t4gSb) (May 11, 2026) > > **CEPS' AI World** built a live data dashboard from this repo's files: [System prompts and what they …

查看完整页面