SceneProof gives coding agents source-grounded visual feedback for React and Three.js
ReyJ94 · reddit · 2026-07-25
SceneProof gives coding agents source-grounded visual feedback for React and Three.js
A developer built SceneProof, an open-source CLI designed to help coding agents understand visual issues that a screenshot alone cannot explain.
What it does
- Navigates a compact semantic tree of DOM or Three.js targets
- Inspects structure such as bounds, styles, geometry, materials, uniforms, lights, cameras, and relationships
- Renders components, UI regions, 3D objects, or source-camera views
- Generates a small 3D “Scout” portfolio with context, close detail, and shape-focused views
- Samples deterministic interaction states from a scene lifecycle
Requirements and scope
It currently supports TypeScript/JavaScript, React DOM/CSS/Tailwind v4, and Three.js. It requires Bun and local Chrome/Chromium.
The core idea is to give an agent source-derived evidence for visual correctness instead of relying on code plausibility or a low-information screenshot.
More from coding & agent
- Codex Micro’s live voice mode could make keyboards a secondary input device — dkundel · 2026-07-26
- Open-source Bento builds animated presentations from a single HTML file — vista8 · 2026-07-25
- Claude Code videos won’t teach you to build; pick one manual process and fix it this week — tomcrawshaw01 · 2026-07-25
- A Codex micro controller feels premium, customizable, and much better with remapped keys — dkundel · 2026-07-25
- Red Hat and Hugging Face meetup covered PyTorch profiling, SGLang, and TRL RL — ariG23498 · 2026-07-25
- AI product teams may be entering a slower phase after stacking plugins, connectors, MCP, and skills — iamrobotbear · 2026-07-25