This skill designs and generates widescreen illustrations that sit inside Chinese articles, posts, blog entries and Notion or workflow documents. The aim is not commercial illustration, slide-style infographics or cute cartoons but a clean, quirky hand-drawn diagram that turns one key judgment, process, structure, state or metaphor from the text into a single picture. Images use a pure white background, black line art, plenty of empty space and a few handwritten Chinese annotations.
A recurring mascot called Xiaohei, a solid black figure with white dot eyes and thin legs, has to take part in the central action of every image. The workflow reads the article, picks the points that deserve a picture, such as core judgments, breakpoints, loops and before-and-after contrasts, and writes a shot list saying where each image goes and what it shows. It plans four to eight images by default and generates each one separately with the built-in image tool, without waiting for confirmation once you ask for generation. Each result is checked against a QA list, redone if it looks like a slide or has typos, and saved under an assets folder with numbered names.
The folder includes reference files for style, the mascot, composition patterns, a prompt template and the QA checklist, an agent configuration file for OpenAI, and a set of example images used only to calibrate the look, which the agent is told not to copy.