chengyi-ai/cy-carousel-skill

What it solves

This project provides a structured workflow and toolset for creating high-quality, 10-page visual carousels (1440x1920) based on a single concept, paper, or product. It automates the layout and rendering of educational content, ensuring a consistent professional aesthetic (dark or light paper backgrounds, calligraphy titles, and highlighted paper strips) while enforcing strict factual accuracy and licensing compliance.

How it works

The system uses a series of Python scripts to move from a topic to a final package:

1. Topic Selection & Verification: Identifies a topic and verifies facts against primary sources using PyMuPDF to locate original sentences.
2. Asset Gathering: Collects open-license assets and records their SHA256 hashes and permissions.
3. Manual Layout (Hand-arrangement): Uses scripts/手排.py to programmatically place images, paper strips, red circles, and text elements (headlines, body text, labels) on each page.
4. Rendering: scripts/render.py converts the layout scripts into final images, handling font switching for English/numbers in calligraphy titles.
5. Checking: scripts/check_all.py runs a technical check for text overflow, missing page numbers, and other "red line" violations.
6. Packaging: scripts/package.py bundles the final images and metadata into a delivery folder.

Who it’s for

Content creators, researchers, and AI educators who want to produce visually striking, factually grounded educational carousels for social media without needing manual graphic design software.

Highlights

• Strict Quality Control: Implements a "red line" system to ensure factual traceability and the use of only open-license materials.
• Professional Typography: Supports specific font combinations (e.g., Ma Shan Zheng calligraphy) and precise layout rules for line breaks and alignment.
• Agent-Ready: Designed as a "Skill" for Claude Code and Codex, allowing AI agents to generate the layout scripts and manage the process.
• Cross-Platform Support: Works on Windows and Linux, with macOS Vision integration for specific OCR-based word-finding in scanned pages.