xikhar/papercraft
Developer ToolsAn Agent Skill that teaches coding agents to build pop-up papercraft-style games, scenes, pages, and videos.
This repo is an agent skill pack · 1 skillsSkill list and install →Overview
papercraft is an agent skill that teaches a coding agent to produce games, interactive scenes, web pages, animations, and videos in a single pop-up papercraft style. The visual targets are printed cardstock pieces with thin cut edges that fold on hinges from a printed ground, soft shadows tinted by the surrounding world, a tilt-shift macro lens, calm hinge motion, and soft paper sounds. It is aimed at developers and designers who want a coherent, handcrafted-paper aesthetic without rebuilding the renderer themselves.
The skill carries technique from a shipped three.js pop-up game: how to make a frame read as a photographed paper model, how to build pieces from extendable structure types (slotted crosses, layered flats, folded boxes and prisms, domes, printed water, decks, bridges, stairs), and how paper should move and sound (2). Prompts and optional references set the world's palette, ground, buildings, creatures, and weather. An image generator is used only when asked; everything can also be made in code. Bundled code handles paper material, cards, layered prints, hinges, lighting, lens, walking, and life effects, while the agent designs the actual subject's pieces.
Getting started
Install the skill through the Agent Skills CLI, or copy the skills/papercraft/ folder into your agent's skills directory. It works with agents that support Agent Skills, including Claude Code, Codex, Cursor, Gemini CLI, and OpenCode (2).
npx skills add xikhar/papercraft
Then describe what you want in a paper style, for example a folding harbour town in three.js, a paper diorama animation rendered to MP4, or a vertical Reel of a lighthouse town. To browse the templates before writing your own scene, serve the assets folder over HTTP and open the 3D or 2D template:
npx serve skills/papercraft/assets
Open /templates/popup-3d/ or /templates/popup-2d/; structures.html shows every structure type. Both templates move the hero with WASD or arrow keys, and ?still gives a fixed overview (2). In your own project, copy assets/lib/ plus the template's harness (harness.js for 3D, harness-2d.js for 2D), write your own scene.js, and point the page's import map "papercraft/" entry at the folder containing lib/.
Output formats and capture
The same scene can be framed as 16:9 film, 9:16 Reel, square or 4:5 post, 4K master, ultrawide still, or a phone page. The lens tracks the frame's short side, vertical frames come closer and keep the subject clear of platform captions, and print and blur follow resolution . Add ?size=1080x1920 or ?aspect=9:16 to a template to preview another format. capture.py renders headless frames and MP4 clips at standard sizes via Playwright and reports build time, slowest prints, whether hero cards intersect, and page errors (2).
Constraints
The skill depends on an agent that supports the Agent Skills format, and the interactive demos need a browser with WebGL for the three.js template. The assets are served over HTTP rather than opened as local files. Image generation is optional and only invoked on request. The repository is MIT-licensed (2). The bundled code is infrastructure and examples; the aesthetic quality depends on the agent following the skill's references and critique loop, so results vary with the agent and the prompt.
HotRepo's take
The repository appeared on HotRepo's new-project list for JavaScript on 2026-11-22, gaining 4 stars and 1 fork during roughly 1.5 hours of observation, reaching 105 stars. That is a small sample, so it supports only the fact of early listing, not a trend. The more useful signal is the scope: rather than a generic prompt pack, it packages a specific material model, structure library, motion and sound modules, and a capture/measure loop, which makes it unusually concrete for an agent skill.
Written by AI from the project's public information and HotRepo's records, updated . The project's own docs are the source of truth.
You might also like
- vercel-labs/agent-skillsPackaged skill instructions that teach AI coding agents Vercel's React, Next.js and deployment conventions.
- android/skillsOfficial Android skill files that teach coding agents how to handle Android tasks LLMs usually get wrong.
- google-gemini/gemini-skillsAgent skills that teach coding assistants current Gemini API, SDK, and Live API practices.
- GoogleChrome/modern-web-guidance-srcA coding-agent skill that feeds curated modern web platform guidance into agents during web development tasks.
- expo/skillsOfficial Expo agent skills that teach coding assistants Expo Router, EAS and SDK upgrade conventions instead of guessing at React Native setups.