Skip to content

CapyCreator · tool no. 3


.

Interrogates intent and scales scaffolding per model capability tier. All local.

CapyCreator builds a prompt for Claude, GPT, Gemini, DeepSeek, Qwen and other models from a short questionnaire, adding the structure smaller models need, entirely in your browser.

Updated · by Unfolding Dimensions

01 · The Ask & Model Dialect

Target: Gemini · Tier 2Derivation: med

Flash / Small model scaffolding: Injects explicit role, numbered steps, hard constraints, and strict negative 'Do NOT' boundaries. Intent questions are required.

Task detected: design review

The guide

How to write a better prompt for Claude, GPT, Gemini and DeepSeek

  1. Step 1

    Describe what you want and pick the model you will send it to, from a small flash model to a frontier one.

  2. Step 2

    Answer the short questionnaire. It asks only what this kind of task needs, and asks more of you when the model is smaller.

  3. Step 3

    Copy the assembled prompt — or run an optional polish pass through your own LLM provider first.

The same prompt does not suit every model. A small, fast model needs a role, numbered steps, explicit constraints and an output format spelled out. A frontier reasoning model does better with a clear statement of intent and room to think. CapyCreator scales the prompt it builds to the model you pick.

It covers Claude, DeepSeek, Gemini, GLM, GPT, Hunyuan and Qwen, and works as a system prompt generator as well as for one-off asks.

Questions

Do I need an API key?

No. Building the prompt happens entirely in your browser. A key is only needed for the optional polish pass.

Where does my text go when I polish?

Straight from your browser to the provider you chose — OpenRouter, OpenCode-Go, Nous Portal, Command Code or any OpenAI-compatible endpoint. It never passes through this site's server.

Where is my API key stored?

Only in your browser's local storage, on this device. It is never sent to this site.

Why does the model choice change the prompt?

Smaller models follow structure and drift without it; larger models can be over-constrained by it. Matching the prompt to the model gets better answers from both.