System Prompts
The live system prompts of 38 AI products, one tool per page
Overview
A reference site that lays out the system prompts actually running behind Cursor, v0, Devin, Claude Code, Windsurf and 33 more tools. The text isn't written here — it mirrors, word for word, the ★144K GitHub repository x1xhlol/system-prompts-and-models-of-ai-tools. What the site adds is readability: the loose txt, md and json files in the repo's vendor folders become one page per tool, split into versioned sections with an anchor table of contents, the category and the official site at the top. It is fully static HTML — no search box, no copy button, no sign-in. It exists to be read end to end.
Why it stands out
- Chinese-team products are in the 38 too: Qoder, Trae (the page labels it "AI IDE by ByteDance") and Z.ai Code ("Chinese AI coding tool"). Measured: the Qoder page has 3 sections — Main Prompt, Quest Action, Quest Design — 58,092 characters in total
- Old versions stay on the same page, so you can watch a prompt change: the Cursor page has 7 sections (Agent Prompt v2.0, the 2025-09-03 build, Agent CLI 2025-08-07, v1.2, v1.0, Chat Prompt, and Agent Tools v1.0 as JSON), 160,152 characters between them
- Tool definitions come with the prompt: every function name, its description, when to use it and when not to, all in the original wording. Designing tools for your own agent, this is a finished example to copy. The v0 page is System Prompt + Tools (JSON), 81,276 characters
- Clean paths make it easy to pull in as corpus: /cursor-system-prompt/, /qoder-system-prompt/ — nothing but JSON-LD runs on the page, so one curl gets the whole text. Measured 12.7KB for the homepage, 140KB+ for a tool page
- Maintenance: the upstream repo was last pushed 2026-10-12 and splits into 34 vendor folders at the top level (Anthropic, Google, Open Source prompts and so on); the About page states the content is shown under fair use, trademarks belong to the vendors, and a vendor wanting a page down files an issue upstream
People writing agent prompts or tool descriptions who want real examples to study, and anyone wondering why a given AI product refuses to budge on something — the answer is usually in its system prompt.
