#232232nd of 259 in Skills, plugins and prompts
dream-loop
Agent skill that builds visuals by looping against an AI-generated target screenshot
- Stars
- 1.7k
- License
- MIT
- Last commit
- Sep 2026
- Language
- JavaScript
Missed the latest GitHub check; these facts may be older.
Overview
An agent skill that has the model generate a target screenshot with image generation, build a game or scene toward it, then have a separate critic agent compare the live screenshot to the target. The loop repeats until the critic is satisfied, and can optionally generate a new target. The README says it is tested only with GPT-6 Astra in Codex.
Who it is for: Developers using coding agents to generate polished games, apps or 3D scenes
Strengths
- Closed loop with a separate critic agent comparing screenshots to a target image
- Installs with `npx skills add` or by cloning into a skills directory
- Works with built-in or API image generation (e.g. a Gemini key)
- Live demo and example prompt show a Three.js isometric scene
Weaknesses
- Tested only with GPT-6 Astra in Codex; other models are untested
- Needs an agent with image generation and vision input
- Subagents strongly preferred, so results may degrade without them
- Blender needed for custom 3D modeling; no resource requirements documented
What it needs
- Needs Image generation (built-in or API, e.g. Gemini), Vision-capable agent, Blender (optional)
- Models: GPT-6 Astra (Codex)
Also in Skills, plugins and prompts
See all 259| Rank | Project | Score |
|---|---|---|
| 1 | impeccableDesign skill with 24 slash commands for AI coding agents' frontend output | 96 out of 100 |
| 2 | cavemanCuts token use in coding agents with terse replies and compressed inputs | 95 out of 100 |
| 3 | skillsAgent skills for planning, TDD, debugging and code review | 94 out of 100 |
| #4 | eccAgents, skills, hooks and rules bundle for Claude Code, Codex and other harnesses | 94 out of 100 |
| #5 | oh-my-claudecodeMulti-agent orchestration plugin and CLI for Claude Code | 94 out of 100 |