Add Creator Skills

Add creator skills

Turn GitHub skill repos or articles about skills into catalog entries in this repository. Do not add this maintainer skill itself to catalog/skills.yaml.

When to use

User provides one or more links such as:

• https://github.com/owner/repo
• a path to SKILL.md / tree/ / blob/
• a WeChat article (mp.weixin.qq.com) or other post that introduces a skill

Workflow

1. Classify each URL: GitHub vs article.
2. Dry-run first. Print import / skip, then vendor only the import rows.
3. GitHub: scan every SKILL.md, keep creator-aligned skills, skip cataloged/rejected. Do not block on missing or unknown licenses.
4. Article: fetch body, extract GitHub / skills.sh links, then follow the GitHub path. If the page embeds a full skill and has no GitHub source, copy it as source.type: url.
5. Regenerate README. Refresh stars for new git sources.
6. Polish catalog summary into one Chinese sentence. Classify by reading the skill — follow catalog/classification-guide.md: set role (publishing or creation), then one primary content_types, necessary platforms, optional single format. Ingest leaves tags empty.
7. Tell the user what changed (imported / skipped, plus classification and summary edits).
8. Commit and push immediately. Do not wait for another confirmation.

After import: report, commit, push

Report in this shape, then git:

Imported: <id> from <repo or url> (role, platforms, content type)
Skipped: ...
Catalog polish: ...

Then add the touched catalog, skills/<id>/, README, and star cache. Commit with a message like Add <skill-id> from <source>. If you also changed this maintainer skill, include that in the same commit or a second one. Push HEAD to origin/main.

Follow this repository's existing git author and GitHub push method. Do not update git config. Do not force-push. Strip a Co-authored-by: Cursor trailer if a hook adds it.

Commands

npm run ingest -- --dry-run <url...>
npm run ingest -- <url...>
npm run ingest -- --from-dir <skill-dir> --url <article-url>

Single known GitHub skill (existing path):

npm run vendor -- --repo <url> --path skills/<id> --platforms ... --content-types ...

Non-GitHub copy:

npm run vendor -- --from-url <article-url> --from-dir <skill-dir>

After git imports, npm run ingest already calls refresh-stars and generate-readme.

Article fetch order

1. npm run ingest -- --dry-run <article> (plain fetch).
2. If output contains ARTICLE_FETCH_REQUIRED, fetch with WebFetch, browser, or URL-to-markdown.
3. Pull github.com / skills.sh links from the markdown and run ingest on those URLs.
4. If the article embeds YAML frontmatter + SKILL.md body and has no GitHub repo: write files under a temp dir and run --from-dir --url.

Creator filter

Keep a skill only if name, description, path, or body match creator platforms/types (WeChat, 小红书, YouTube, X, 微博, 文章/图片/图文/视频/幻灯片/漫画/信息图, etc.). Drop infra/CI/K8s-style skills unless they are clearly for content production.

Classification (agent, not script)

After vendor, read skills/<id>/SKILL.md and catalog/classification-guide.md.

First decide role:

• publishing — post/upload/draft to a platform (post-to-*, video-publisher, WeChat draft upload). README lists these only under 内容发布.
• creation — generate images, articles, videos, slides, etc. README lists under 按平台 / 按创作类型 / 按创作形式.

Then for role: creation, write one content_types, only the platforms the skill serves, and at most one formats. For role: publishing, still record platforms and the main published content_types, but the skill must not appear outside 内容发布 in README.

B站 / b站 / 哔哩哔哩 → bilibili.

Do not commit without role, platforms, and content_types.

License field

Record SPDX when the repo declares one; otherwise write Unknown. Never block import on a missing LICENSE file.

URL copies: record the article URL in SOURCE.md, sync: false, no star badge.

Multi-skill repos

Scan the whole tree (e.g. skills/*/SKILL.md plus a root SKILL.md). Import every creator match that is not already in the catalog. Skip the rest with reasons.

Output

Before writing, list:

• import — will vendor
• skip — already cataloged, rejected, or not creator-related
• article — fetch still needed, or GitHub links extracted

This skill is a maintainer tool. Never catalog it as a public creator skill.