dream-loop: Agent Skill for 3D Visuals via Iterative Critic Feedback

Repositoryachimala/dream-loopTool

dream-loop is an open-source agent skill by Anshu Sharma (@anshuc) designed to build high-quality 3D scenes, games, and web apps through a closed feedback loop. Image generation creates a reference target, after which an AI critic visually inspects live output and directs iterative refinements.

Key points

  1. Core mechanism: Image generation dreams a target screenshot, the agent codes the scene, and a dedicated AI critic compares live renders against the target to direct adjustments.
  2. Iterative loop: Once the critic is satisfied, the agent can optionally dream an even better target based on the current implementation state.
  3. Prerequisites: Requires an agent with vision capabilities, image generation access (Codex, Grok, or Gemini API), and preferably subagent support.
  4. Tooling and 3D: Blender via MCP or scripting interface is recommended for custom 3D assets, alongside web-based 3D libraries like Three.js.
  5. Setup: Installable via npx skills add achimala/dream-loop or by copying the repository into the agent's skills directory.

Why it matters

Coding agents frequently miss aesthetic and visual nuances because they lack a systematic way to compare rendered output with high-fidelity visual expectations. dream-loop provides a closed-loop validation pattern using vision subagents.

Analysis

Targeted at developers building generative 3D environments, the repository leverages JavaScript, visual generative models, and Blender/Three.js integrations. Currently at an experimental stage tested primarily on GPT-6 Astra in Codex, it offers a working browser demo (Vesper) and a structured framework for visual prompt-driven development.

Mentioned links

Found in

Summary generated by AI (Gemini 3.8 Flash, 27 September 2026). It may contain errors – the original source is authoritative.

Related entries

Similar in content, found by the AI search.

  • Repository:MengTo/Skills

    Agent Skills: Reusable Playbooks for AI Coding Agents

    The GitHub repository MengTo/Skills provides a curated collection of Markdown-based 'agent skills'. It enables designers and builders to guide AI coding agents like Codex, Claude, and Cursor through complex web design, interaction, and 3D rendering tasks.

    6.2kstarsHTML

    AI & LLMs· Resource

  • Repository:img2threejs/img2threejs

    img2threejs: Procedural Code-Only Image-to-3D for Three.js

    img2threejs turns reference images into procedural, animation-ready Three.js models generated purely as code, avoiding traditional mesh extraction or photogrammetry.

    15.1kstarsPython

    AI & LLMs· Tool

  • Link:Matt Shumer

    The Gauntlet Loop: The Prompting Method Behind Claude of Duty

    Matt Shumer explains the 'Gauntlet Loop', the prompting technique used to build a 55,000-line 3D game using Claude Opus 5 inside Claude Code from a single prompt. The method delegates decomposition to a lead agent and pairs independent builders with isolated critics that evaluate outputs against an ambitious reference standard.

    AI & LLMs· Tutorial

Back to the knowledge base

Let's talk about your project

Get in touch

Locations

  • Mattersburg
    Johann Nepomuk Bergerstraße 7/2/14
    7210 Mattersburg, Austria
  • Vienna
    Ungargasse 64-66/3/404
    1030 Wien, Austria

Parts of this content were created with the assistance of AI.