gitmyhub

sutniprompt

★ 20 updated 20d ago

My version-controlled (and modular!) AI system prompt to force strict formatting and prevent AI overreach. Tested daily on major LLMs for users who want control, not fluff. More info on README!

A ready-to-paste system prompt for Claude, ChatGPT, and Gemini that removes AI hedging and filler, gates vague requests behind targeted clarifying questions, and adds a timestamp and Wikipedia link to every response.

setup: easycomplexity 1/5

SutniPrompt is a system prompt designed to change how AI assistants like Claude, Gemini, and ChatGPT respond. A system prompt is a block of instructions given to an AI before the conversation starts, shaping its behavior for the entire session. This one is built for people who find commercial AI assistants too cautious, too verbose, or too prone to hedging, and who want a more direct, structured interaction.

The prompt has four main modules. The first sets the tone: the model should not announce that it is an AI, should not use phrases like "As an AI...", and should not express simulated emotions. The second focuses on reasoning: logic and code errors should be corrected with minimal explanation, and the model should not summarize itself at the end of responses. The third defines a gating rule: if a request is too vague to answer usefully, the model must stop and ask two to three targeted clarifying questions rather than guessing or producing a generic answer. This gate is bypassed for everyday tasks like drafting emails or writing code. The fourth module adds two mandatory behaviors to every response: a timestamp at the start (formatted as year, month, day, hour, minute, second, and timezone) and one relevant Wikipedia link at the end for further reading.

Deployment varies by platform. On Claude, the full prompt is pasted into the system prompt field. On Gemini, the numbered modules work better when inserted one at a time because the interface can have trouble parsing long structured blocks. On ChatGPT, character limits in the custom instructions field often prevent inserting the full prompt, so the README recommends pasting it as the first message in a new chat instead. A compressed version for ChatGPT is listed as a planned future addition.

The project is version-controlled and updated based on daily testing across these platforms. It is licensed under CC BY 4.0.

Where it fits