Your AI agent’s new side gig: Swiss-style explainer diagrams

What it does This is a structured skill pack for Claude Code and Codex agents. It takes articles, notes, chart screenshots, or product concepts and generates center illustrations—specifically, Chinese-labeled explainer diagrams and material-styled charts meant to slot into social cards, PPTs, and blog posts. It treats the image as a piece of communication, not wallpaper.

The interesting bit Most AI image tools spit out decoration; this one treats in-image text as a first-class feature, compressing concepts into short Chinese labels and arrows so the diagram carries meaning even at thumbnail size. For chart screenshots, it extracts semantic data—axes, units, outliers, conclusions—and redraws them in a soft 3D material style rather than slapping a filter on a bad screenshot.

Key highlights

• Generates diagrams with embedded Chinese text, labels, and arrows so they remain readable when scaled down for social cards.
• Semantic chart redraw: parses data and chart type from screenshots or raw inputs, then regenerates the visualization in a consistent “Guizang material” aesthetic.
• Reference-first pipeline for obscure concepts—agents look up logos, scientific devices, or historical objects before drawing to avoid hallucinating structure.
• Modular by design: outputs center illustrations meant to be consumed by sibling skills that handle full page layout, such as social-card or PPT skills.
• Built-in QA loop checks Chinese label accuracy, data fidelity, and crop safety before delivery.

Caveats

• Not a standalone CLI tool; it is an agent skill pack that requires a compatible AI agent environment and an underlying image generator like GPT-Image.
• Explicitly avoids full-page layout, photo retouching, or strict publication-grade scientific charts—those are handled by other tools or left alone.

Verdict Grab this if you produce Chinese-language educational content, technical explainers, or data stories and want your AI assistant to handle the “middle image” without drowning in manual prompt tuning. Skip it if you need a self-contained design app or pixel-perfect research figures.