Understanding System Prompts

Note: This repository serves as a summary of mainstream GitHub repositories containing system prompts. It will be automatically updated.

What is a System Prompt?

A System Prompt (often called a "system message," "metaprompt," or "custom instructions") is the initial set of instructions given to an Artificial Intelligence model before it interacts with a user. It serves as the AI's "context" or "persona."

Think of it as the Job Description and Rulebook for the AI.

Key Components of a System Prompt

Most typically, effective system prompts include the following sections:

1. Identity & Role: Who is the AI? (e.g., "You are a helpful coding assistant," "You are a creative writer.")
2. Capabilities & Tools: What can the AI do? ("You have access to a browser," "You can run Python code," "You cannot access the internet.")
3. Tone & Style: How should the AI speak? ("Be concise," "Use a formal tone," "Explain like I'm 5.")
4. Constraints & Safety: What is forbidden? ("Do not reveal user data," "Do not generate harmful content," "Never mention you are an AI.")
5. Output Format: How should the instructions be delivered? ("Use Markdown," "Respond in JSON," "Always start with 'Here is your answer'.")

Refined System Prompts Collection

Large Language Models (LLMs)

• Google Gemini: Focused on React/Tailwind frontend engineering.
• Anthropic Claude: The Sonnet 4.5 prompt, featuring Artifacts and Safety protocols.
• OpenAI / ChatGPT: (GPT-4o) Features "Canvas" (canmore), Browsing, and Python Sandbox rules.

AI Integrated Development Environments (IDEs)

• Cursor: Features deep 'Codebase Search' and 'Context' tools.
• Windsurf: Features the 'Cascade' flow and 'Memory' systems.

Why Reading These Prompts is Useful

Analyzing these prompts helps you understand:

• Hidden Mechanisms: How the model is steered towards certain job roles.
• Better Self-prompting: How to write better system prompts.
• Feature Engineering: How "magic" features (like reading a file) are actually just instructions text.