AI · GitHub repo

system_prompts_leaks

Curated collection of leaked system prompts from major AI assistants and models.

Visit github.com ↗

This repository compiles verbatim system prompts, agent instructions, and tool definitions extracted from major AI chatbots including Anthropic Claude, OpenAI ChatGPT, Google Gemini, and Cursor. It serves as an extensive public archive of real-world prompt engineering, tool use patterns, and personality guidelines used in commercial AI products.

What it's for

  • Study production prompt engineering and guardrail design techniques
  • Inspect tool calling and agent instructions used by tools like Claude Code and Codex
  • Compare persona, tone, and safety directives across different commercial LLMs
  • Reverse-engineer agentic workflows, subagent architectures, and memory structures

Features

  • Verbatim system prompts for Claude, ChatGPT, Gemini, Grok, Perplexity, and others
  • Dedicated prompt breakdowns for coding tools like Claude Code, Codex, and Cursor
  • Documentation of injected safety guidelines, reminders, and persona instructions
  • Subagent, command, and skill instructions for modern agentic products
  • Public domain (CC0 1.0) community-maintained documentation

Pricing

Open source. Free (CC0-1.0 license)

Details

Type
GitHub repo
Category
AI
Topic
Prompt Engineering
Open source
Yes
Platforms
Web
For
Prompt engineers, AI developers, security researchers, and LLM enthusiasts.
Tags
prompts, system prompts, prompt engineering, claude, chatgpt, gemini, ai agents

Why it was saved

  • @nick_saraev, 26 January 2026: Provides direct reference material to study how Anthropic structures system prompts, agentic workflows, memory systems, coding tools, and tone instructions.

Related

All 18 in Prompt Engineering →

Updated