Who is system_prompts_leaks for?

Suits prompt writers, agent builders and curious students; no setup is needed, only some familiarity with what a system prompt is.

What can you build with system_prompts_leaks?

Read how a vendor tells its model to handle tools, safety and formatting. Compare the prompts of two model versions to see what changed. Borrow structure and wording ideas when writing your own system prompts. Study how coding agents like Claude Code and Codex are instructed. Cite real prompt text in an article or a class on prompt engineering.

asgeirtj /

system_prompts_leaks

A library of system prompts from popular AI chatbots and coding tools

70.1k stars
1.3k stars since 3 Oct
Python
Updated today
Beginner friendly
Learning & resource lists

Open on GitHub (opens in a new tab)

Every chatbot has a hidden set of instructions, called a system prompt, that is sent before your first message. It sets tone, rules and how the model uses tools. Vendors rarely publish them. This repo collects the full text as files you can read in your browser.

Files are sorted by company: Anthropic, OpenAI, xAI and others. The Anthropic section covers Claude on web and mobile, many Claude Code versions, Claude Design, Cowork, and the Office and Chrome integrations. The OpenAI section covers ChatGPT models and Codex. A dated table at the top lists the newest additions.

It is a reference library, not software you run. The prompts are extracted from live products rather than published by the vendors, so treat each file as the author's capture at that time. The author invites pull requests from anyone who got a different result.

What it does

Many vendors in one place
Covers Anthropic and OpenAI in detail, with xAI entries too. The repo description also mentions Google, Cursor and Kimi.
Claude Code in depth
Lists prompts for Claude Code across models and surfaces: terminal, desktop, web and headless. It also includes subagent prompts, skills and slash commands.
Version history
Keeps older model versions beside new ones, so you can see how a prompt changed between releases.
Recent additions table
A dated table at the top of the README links to the latest prompts, so you can see what changed lately.

What you could build

Read how a vendor tells its model to handle tools, safety and formatting.
Compare the prompts of two model versions to see what changed.
Borrow structure and wording ideas when writing your own system prompts.
Study how coding agents like Claude Code and Codex are instructed.
Cite real prompt text in an article or a class on prompt engineering.

Get started

The README gives no install steps. Open the repo on GitHub, pick a vendor folder such as Anthropic or OpenAI, and read the Markdown file for the model you care about.

Commands are copied from the project's README. Read it on GitHub for the full setup.