Answer me with HTML
Your coding agent hands back a built page instead of a wall of prose
Emerged finds new products worth a look, daily — what’s new today
dream-loop
An agent skill that builds 3D scenes by chasing an AI-generated target image
It gives a coding agent a picture to aim at and a critic to answer to, which is what separates a striking 3D scene from the grey boxes agents usually produce.
dream-loop is a skill you add to a coding agent. The agent first generates a target screenshot with an image model, then builds towards it in Three.js, optionally modelling assets in Blender. A subagent acts as critic, compares each render with the target and sends back notes, and the loop repeats until the two match. The Vesper demo, a walkable drowned cathedral running above 60fps in the browser, was made this way. Install with npx skills add achimala/dream-loop. The agent needs image generation, vision input and subagents. MIT licensed.
More like this
Reef
Open-source infrastructure for agents that keep improving from use
Recto
A free desktop canvas where Claude Code, Codex and Cursor draw UI you can edit
DeepSeek Harness
DeepSeek's open-source agent harness where every capability is a plugin
Whenever
Hosted automations your coding agent writes in TypeScript
M3E Canvas
Sketch Material 3 Expressive screens and turn them into prompts for AI coding