The 21-Skill Personal Toolkit — One Author, Every Publishing Job, 8.9 to 9.5

Most multi-skill authors repeat themselves. The largest single-author family in our directory — twenty-one skills from one person — doesn't. We read all twenty-one to figure out why.

There's a failure mode we see constantly in multi-skill repos: fifteen skills that all describe the same capability with different words. The boundaries blur, the agent can't tell which to load, and none of them score well — the interference problem in its purest form.

The baoyu family is the counterexample. Twenty-one skills, one author, scores from 8.9 to 9.5, no overlap. The family covers a complete bilingual content pipeline — and reading it in pipeline order explains the whole thing.

The pipeline, stage by stage

Input & research

• baoyu-url-to-markdown — 9.2: fetch any URL, convert to clean markdown.
• baoyu-youtube-transcript — 9.3: download YouTube transcripts and subtitles.
• baoyu-wechat-summary — 9.3: summarize WeChat group chats into structured highlights.

Processing & translation

• baoyu-translate — 9.4: translation with the author's house style. One of the two flagship skills of the family.

Visual production

• baoyu-diagram — 9.4: professional dark-themed SVG diagrams. The other flagship.
• baoyu-infographic — 9.3: infographics with 21 tested layout types.
• baoyu-image-gen — 9.3: multi-provider image generation.
• baoyu-cover-image — 9.2: article covers with 5 dimension-presets.
• baoyu-xhs-images — 9.4: Xiaohongshu card series.
• baoyu-slide-deck — 9.3: slide deck images from content.
• baoyu-electron-extract — 9.5: the outlier — Electron app resource extraction, and the family's highest score. (Covered here.)

Publishing & output

• baoyu-post-to-wechat — 9.3: publish to WeChat Official Accounts.
• baoyu-markdown-to-html — 8.9: WeChat-compatible styled HTML.
• plus markdown processing and formatting utilities completing the set.

Chain them and you get a full workflow: research a topic → translate the material → build diagrams and covers → generate a slide deck → publish to WeChat or format for Xiaohongshu. The pipeline is reconstructable from the skill descriptions alone — the same property that made the HyperFrames family work.

Why there's no overlap

The discipline is in the triggers. Each skill's description opens with a narrow "use when" clause that could not be satisfied by another family member:

• baoyu-translate fires on translation requests with a style requirement — not on "make this English" (that's generic translation, no skill needed).
• baoyu-xhs-images fires on Xiaohongshu card-series requests — not on "make an infographic" (baoyu-infographic) or "make a cover" (baoyu-cover-image).
• baoyu-electron-extract fires on installed-app extraction — nothing else in the family touches the local machine.

Twenty-one skills, twenty-one distinguishable moments. That's the whole trick, and it's the same test we recommend in the 60-second SKILL.md read: read the trigger first, and ask whether it could belong to a sibling.

The bilingual bet

The family is deliberately bilingual — Chinese-optimized where the platform is Chinese (WeChat, Xiaohongshu), English-optimized where the output is (translation, diagrams, Electron tooling). Our Chinese-social-publishing roundup found the same pattern: the skills that score highest on Chinese platforms are the ones that treat the platform's conventions as first-class content, not as a locale setting.

For non-Chinese users, the family's general tools — translate, diagram, infographic, cover, slide deck, Electron extract — are platform-agnostic and all 9.2+.

The takeaway

A personal toolkit becomes a directory-grade family when every skill answers a question the others can't: "when would I load you instead of your sibling?" If the answer is "it depends," the family needs one fewer skill. The baoyu family answers it twenty-one times, which is why it averages 9.3.