Baoyu Skills

Baoyu Skills (baoyu-skills) is a collection of Agent Skills open-sourced by Jim Liu (GitHub user JimLiu). The README credits Baoyu and says the Skills work with Claude Code, Codex and similar AI tools. The repo holds more than 20 Skills; AI Inspiration World recommends only 10 of the creative and productivity ones, each of which can be installed on its own. Six of them make images: baoyu-article-illustrator reads an article, picks the spots that need a visual and draws them by type, style and palette; baoyu-cover-image makes article covers, with controls for type, palette, rendering, amount of text, mood and font; baoyu-comic turns a topic or a life story into a multi-page educational comic, drawing a character sheet first so the cast stays consistent, then merging the pages into a PDF; baoyu-infographic offers 21 layouts and 22 visual styles; baoyu-slide-deck writes an outline, renders each slide as a full image and merges them into .pptx and .pdf, and SKILL.md says the decks are designed for reading and sharing, so every slide has to explain itself; baoyu-xhs-images splits content into 1 to 10 Xiaohongshu-style image cards, drawing the first card and using it as the reference for the rest. The other four need no image model: baoyu-diagram writes SVG code by hand for architecture diagrams, flowcharts, sequence diagrams, mind maps and more, then exports a PNG at twice the size; baoyu-compress-image compresses images to WebP or PNG; baoyu-format-markdown adds a title, summary, headings, bold and lists to an article, with a rule to change only formatting and obvious typos and never add, cut or rewrite the text; baoyu-translate has quick, normal and refined modes, where refined runs analysis, translation, review and polish, and you can attach your own glossary. What sets it apart is that the image Skills show you the plan for approval by default and save each image's full prompt to a file before generating, so you can rerun the same prompt with a different image tool; when text inside an image comes out wrong, the rule is to fix the prompt and redraw instead of patching the picture with code.

Who it's for:content creators who write for WeChat articles, blogs or Xiaohongshu, or who teach, and want AI help with illustrations, covers, image cards and explainer slides, or with translating and tidying up articles. You'll need to be comfortable with an AI tool that runs commands, such as Claude Code or Codex; readers who never touch a command line will find it hard going.

Before you start: we suggest copying only the individual Skill folders you need, whole, into the folder your AI tool reads Skills from (the README's example is .agents/skills inside a Codex project), and not installing baoyu-skills as a full plugin from the marketplace. The plugin installs every Skill in the repo, and the AI can then match and call the posting and scraping tools on its own from what you say; the README itself recommends installing only what you need. The README asks for Node.js and the ability to run npx bun; the scripts for compressing images, exporting PNGs, merging PPTX and PDF files and fixing typography may download Bun and dependencies from npm the first time they run. Going by the SKILL.md text, the six image Skills use the image generation built into the AI tool you run them in, with Codex's imagegen and Cursor's GenerateImage given as examples, so cost and limits follow that tool; if your tool can't generate images, the Skill stops and asks you rather than substituting a picture drawn in code. If baoyu-image-gen from the same repo is installed, these Skills will switch to it automatically when the host tool can't generate images; baoyu-image-gen sends your prompts and reference images to several third-party image APIs and may cost money, and we don't recommend it. Most of the Skills ask a few preference questions on first use and save the answers as EXTEND.md in a .baoyu-skills folder in your project or home directory. The SKILL.md for baoyu-compress-image says it replaces the original image by default; add --keep to hold on to it. Check that you have the rights to any reference images, photos of real people or brand assets you bring. The repo is MIT licensed overall, and the README says third-party code and assets keep the original licenses noted for them. The repo also contains tools that post to X, Weibo and WeChat Official Accounts from your real accounts (and can actually publish), a tool that reads your local WeChat chat history to summarize group chats, tools that pull Gemini and X content through reverse-engineered interfaces and cookies, a YouTube transcript tool that switches identities and retries when blocked, and others for converting web pages to Markdown, extracting source code from desktop apps and turning Markdown into WeChat-ready HTML. AI Inspiration World has not reviewed or recommended any of them.

Based on the project's public documentation. Not tested by AI Inspiration World.