21K stars for AI agent clip-art? The baoyu-skills phenomenon

What it does

baoyu-skills is a sprawling plugin pack of 20+ “skills” for AI agents — primarily Claude Code, Codex, and OpenClaw. The headline acts are visual content generators: Xiaohongshu-style image cards, professional infographics, SVG diagrams, slide decks, comics, and article illustrations. Each skill is a structured prompt-and-template system that the agent invokes via slash commands, producing publication-ready output from raw text or markdown files.

The interesting bit

The sheer combinatorial explosion. The infographic skill alone offers 20 layout types (funnel, fishbone, iceberg, Venn, etc.) crossed with 17 visual styles (claymation, IKEA-manual, corporate-Memphis, pixel-art, knolling). The Xiaohongshu generator adds a “Style × Layout × Palette” matrix. Rather than one-size-fits-all AI slop, the system forces the agent to analyze content and recommend a specific structure — then generates within tight aesthetic guardrails. It’s prompt engineering as product design.

Key highlights

• Agent-native workflow: Skills install as plugins; you type /baoyu-infographic article.md --layout pyramid --style technical-schematic and the agent handles the rest
• Self-contained SVG output: The diagram skill has Claude write actual SVG code directly, not generate images via DALL-E
• Granular publishing: Individual skills can be published to ClawHub under MIT-0; the repo warns against bulk-installing to avoid context bloat
• Non-interactive mode: --yes flag for scheduled/CI tasks
• Multi-language support: --lang flag for localized output

Caveats

• Requires Node.js and npx bun capability; not a standalone tool
• The README is overwhelmingly a catalog of options — actual implementation details of how the agent “writes SVG code” or handles errors are unclear
• 20+ skills in one repo risks the exact context-overhead problem the author warns about

Verdict

Worth a look if you’re already living in Claude Code or Codex and need to produce visual content at volume — social posts, documentation diagrams, pitch decks. If you don’t use agent-centric IDEs, this is just an elaborate prompt library you can’t easily access.