A publishing toolbox disguised as a skill collection
Baoyu Skills packages more than 20 repeatable workflows for AI agents. The collection covers Xiaohongshu cards, infographics, SVG diagrams, cover art, slide decks, comics, article illustrations, translations, Markdown cleanup, HTML conversion, image compression, transcripts, and social posting. Many entries include TypeScript scripts, reference files, configuration rules, and checks around generated output.
The practical appeal is consistency. A writer can give an article to the cover skill, ask the illustration skill to find useful visual breaks, convert the result into WeChat-ready HTML, then stage it for publishing. A developer could instead install only the diagram skill and have Claude write a self-contained SVG. These jobs usually require long instructions and repeated corrections. A named skill preserves the procedure.
The repository is aimed at Claude Code, Codex, Cursor, and other agents that can read files and run commands. It is especially relevant to Chinese social publishing, but the main README is English and links to Chinese documentation.
Selective installation is the right way to use it
The one-line installation is honest as far as registration goes: Node.js and the ability to run npx bun are the stated prerequisites, then npx skills add jimliu/baoyu-skills installs from the repository. Claude's marketplace flow and project-level Codex directories are documented too. ClawHub users can install skills individually.
The README gives a warning that matters more than the quick command. Loading every skill adds unnecessary context to the agent on every run. Project-level Codex users are shown how to copy or link full skill directories into .agents/skills, and the WeChat example recommends a minimal three-skill set. Follow that advice. Begin with one end-to-end job, not every available capability.
Local-only tasks are the easiest trial. The diagram skill writes SVG directly rather than calling an image model. Markdown formatting, HTML conversion, and image compression also have clear inputs and outputs. Generated visual work is more involved. The shared image skill supports numerous providers, each with its own key, model behavior, reference-image rules, size limits, and concurrency settings. Codex users can route image work through their subscription, but that backend is not automatically selected.
This is why setup earns a middle score despite the simple install. Building a dependable production path means selecting a provider, storing secrets outside version control, testing output, and understanding which parts belong to the agent or a bundled script.
The visual workflows are unusually specific
The best parts of Baoyu Skills remove ambiguous art direction. The infographic skill offers named information structures such as funnels, comparison tables, timelines, and priority quadrants, then pairs them with documented visual treatments. The slide-deck skill separates texture, mood, typography, and density. The cover generator uses type, palette, rendering, text, and mood as independent controls.
That vocabulary gives users useful handles when an automatic choice misses and makes results easier to repeat. The README includes preview grids and command examples, so people can choose a direction before paying for generation. The diagram skill asks Claude to calculate and write real SVG, including embedded styling and dark-mode behavior.
There are limits. An open request asks for light and monochrome diagram themes because the current dark look can be unsuitable for professional documents. A subtitle-translation request points out that translated lines can differ in spoken duration. The existing translation modes improve editorial quality, but do not solve timing-aware localization. Test the exact artifact you need rather than assuming the collection's width means every workflow is finished.
Publishing is where convenience meets platform fragility
The social skills make sensible safety choices. X and Weibo browser flows fill the editor, then leave final review and publishing to the user. WeChat supports API, browser, and remote API methods. The fast API route needs credentials and an allowlisted IP. The remote route can use an SSH SOCKS5 tunnel without writing files to the server, while browser mode preserves a QR-login session. Multi-account profiles and scoped credentials are documented.
None of this makes third-party editors stable automation targets. Issue 197 describes the new WeChat editor exposing hidden or stale nodes that can break menu, editor, or save-button selection. It also reports that saving before image uploads finish can produce errors or failed uploads on slow networks. The proposed fixes had local tests, but the issue remained open. Anyone publishing revenue-critical material should keep the manual checkpoint, inspect uploaded images, and confirm the draft exists.
Active maintenance, with a wide surface to maintain
The repository's last push was July 4, 2026. Its latest release, version 2.5.2, arrived June 18 and fixed a Codex image-generation path that could reuse an unrelated existing image instead of proving a new generation occurred. Issues and pull requests were still being updated in August, including fixes for Korean transcript spacing and CSS parsing. The GitHub open count combines eight issues and pull requests, so it is a work queue rather than eight confirmed defects.
Maintenance activity is encouraging for a young project with many integrations. The harder question is whether one maintainer can keep pace with browser editors, model providers, agent conventions, and social APIs that change independently. Recent releases show real fixes, but the WeChat report demonstrates the ongoing cost.
Baoyu Skills is best treated as a well-stocked workshop. Take the few tools that match a real publishing process, pin and test them, then add more when the value is clear. Used that way, the collection saves instruction-writing and provides better creative controls than a generic prompt. Installed wholesale and trusted without review, it becomes a credential-heavy automation surface whose failures can land in public.