Every AI tool you've ever used has a hidden set of instructions written by its creators. Think of it like a staff handbook — the rules and personality the AI follows before you've even typed a word. These are called system prompts, and they're usually kept private.
Someone found a way to extract them from over 25 major AI tools — Cursor, Claude Code, Lovable, Devin — and published the whole thing openly. Over 20,000 lines. It went viral almost immediately: 118,000 people have saved it.
You don't need to read all 20,000 lines. But there's something quietly important here.
The gap between an AI that feels sharp and useful versus one that feels scattered and unreliable? It often comes down to how well someone wrote those hidden instructions. These aren't magic — they're craft. Careful, deliberate writing about how the AI should think, prioritise, and behave.
If you're working with an agency or freelancer to build something AI-powered, this is good context to have. The quality of those instructions shapes everything the user experiences.
You can browse the repo here: https://github.com/x1xhlol/system-prompts-and-models-of-ai-tools
System prompt — The hidden instructions given to an AI before any conversation begins. Like a briefing you give a new hire before their first client call.
AI agent — An AI that doesn't just answer questions but takes a sequence of actions to complete a task, like a junior assistant who works through a to-do list.
Scaffold — The underlying structure or template that tells an AI how to organise its work. Think of it as the skeleton before the meat.
Something worth sitting with: if you use AI tools in your business, do you know what instructions are quietly shaping their behaviour?