GitHub radar
AI Skill for Visual App Development | Dream Loop
Use an AI to dream up a target image for the game, app or visual scene, then build towards it, while an AI critic gives feedback in a loop until you have the desired visual outcome.
A Dream Loop is a creative agent skill that allows you to develop games, apps and visual scenes iteratively within a closed loop. It works by having an AI first dream up a high quality target image which represents the end goal and then using this to build your project. This process is accompanied by an AI critic giving feedback on the difference between the live image and the target image until the critic is happy. Once this occurs, another option is to have the AI dream up again to generate a more ideal target image based on the current project state. Can be used with any agent that has an image generator, vision input and subagent capabilities (currently tested with GPT-6 Astra on Codex, but it should work with good models like Claude Fable 5 etc.) Install by running a single command `npx skills add achimala/dream-loop`
Why a vibe-coder should care
Rather than constantly editing visuals until they meet an aesthetic, this agent uses an automatic process to iterate towards a visual target defined by itself. You tell it what you want, and it produces a result that holds up to a certain aesthetic standard, rather than just a functional prototype that needs designing.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install this agent skill from https://github.com/achimala/dream-loop — follow the README. Then use it to build me a Three.js browser demo: isometric camera, voxel art style, fantasy scene with a character. Show me the result and a demo link. Ask me for API keys if needed.
▌ More finds