How achimala/dream-loop works
dream-loop is an agent skill for iteratively building visually polished games, apps, or scenes. Its principal workflow is to generate a visual target, implement toward it, capture the running result, and have a separate vision-capable critic recommend changes; the agent repeats implementation and critique, optionally generating a better target. The repository also documents Plus and Pro workflows and includes a local preview/capture server and a Fal queue helper for image-to-3D assets. Source excerpts are partial, so the graph omits uncertain internal wiring from the skill references.
Main entry points: fal-batch.mjs Fal batch helper · preview-server.py Preview server · SKILL.md Dream-loop skill · workflow.md Plus workflow
Components 17
Visual iteration
- Dream-loop skill
SKILL.md - Plus workflow
references/plus-mode/workflow.md - Pro workflow
references/pro-mode/workflow.md
Asset generation
- Plus 3D assets
references/plus-mode/assets-3d.md - Pro 3D assets
references/pro-mode/assets-3d.md - Fal batch helper
scripts/fal-batch.mjs - Job plan and state
- Input images
- Downloaded GLB models
Preview and capture
- Preview server
scripts/preview-server.py - Screenshot captures
Other components
- User
- AI agent
- Image generation
- Built app or scene
- Vision critic
- Fal queue API
Connections 20
- User → AI agent: requests build
- AI agent → Dream-loop skill: follows workflow
- Dream-loop skill → Image generation: generates target
- AI agent → Built app or scene: implements target
- AI agent → Preview server: starts preview
- Preview server → Built app or scene: serves workspace
- AI agent → Preview server: requests capture
- Preview server → Screenshot captures: writes PNG
- AI agent → Vision critic: requests critique
- Screenshot captures → Vision critic: provides screenshot
- Vision critic → AI agent: returns feedback
- AI agent → Plus workflow: may follow
- AI agent → Pro workflow: may follow
- Plus workflow → Plus 3D assets: documents assets
- Pro workflow → Pro 3D assets: documents assets
- AI agent → Fal batch helper: runs asset jobs
- Fal batch helper → Job plan and state: reads and updates
- Fal batch helper → Input images: reads inputs
- Fal batch helper → Fal queue API: submits and collects
- Fal batch helper → Downloaded GLB models: downloads GLB