The system prompt is the model's secret

What this is

A free, step-by-step template from Taha's reels — the exact tools, prompts and links, laid out so you can follow along and ship. Comment the keyword on the reel and it also lands in your DMs.

Hey! 👋 Here's what I promised you:

A big part of the difference between AI agents isn't only the model — it's the system prompt: the instructions that tell the model who it is, how to plan and how to check its own work.

Anthropic's official system prompts

Anthropic publishes the system prompts it uses for the Claude apps (claude.ai, iOS and Android), one page per model, for free. Read them and study how they are built.

🔗System prompts (official docs)https://platform.claude.com/docs/en/release-notes/system-prompts/overview

To be clear: this is a docs page, not a plugin. It doesn't install anything and it doesn't rewrite your prompts — it shows you how Anthropic writes its own. Anthropic also notes these prompts are for the Claude apps and don't apply to the Claude API.

How to use it as a recipe

Open the page, pick a model, and read its prompt from top to bottom.
Notice the structure: it's split into clear tagged sections — who the assistant is, what it knows, how it should behave, how to format answers.
Write your own system prompt for your agent or app with the same kind of structure (template below).
Put it in the system prompt of the model you use — including a cheaper open model like GLM, DeepSeek or Kimi — and compare the results on your own tasks.

A simple template to start from

<role>
You are [who the agent is] helping [who the user is] with [the job].
</role>

<context>
[What the agent must know: the project, the tools it can use, the rules.]
</context>

<planning>
Before doing anything, write a short step-by-step plan. Then follow it one step at a time.
</planning>

<self_check>
Before you answer, check your work against the plan and the request. If something is wrong or missing, fix it first. If you are not sure about a fact, say so instead of guessing.
</self_check>

<output_format>
[How the answer should look: length, sections, code blocks, language.]
</output_format>

A good system prompt makes a cheaper model much more useful, but it doesn't turn it into a different model — results still depend on the model and the task. Test it on your own work.

Want more? Go further in the Atlas Room Skool class

This page gives you the tools. The Skool class takes you step by step — full courses and guides to build and sell your own AI app or SaaS, in Darija, with a new lesson every week.

Join the Skool class →

New to vibe coding? See what you'll learn →

🧠GRAPH

FIREBASE🔥