菜单
Use for any wiki-related task — clipping sources, ingesting knowledge, checking health, or querying. Routes automatically so you don't need to remember sub-skill names.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when a wiki concept/entity/source page or a mm chapter needs a visual TL;DR card. Produces a `.mdx` file that renders through `<OneShotCard>` in the Astro site. Design-first workflow — different cards should LOOK different.
End-to-end pipeline for writing mental-model chapters: brainstorming → research & wiki building → editorial drafting → bilingual writing → review/optimize/enrich → wrap-up. Use this skill whenever the user wants to write a new mental-model chapter, continue work on an in-progress chapter, or asks about the mental-model writing workflow. Triggers on: “写 ch-XX”, “start chapter”, “mental model”, “心智模型”, “new chapter”, “下一篇”, “进入正式撰写”.
Use when a source document needs to be digested into the wiki. Without this, raw sources sit unconnected — no summary, no concept extraction, no cross-references, no discoverability. Takes a local file path or URL.
Use when you have a URL and want to preserve it locally before it disappears or changes. Web content is ephemeral; local markdown is permanent.
Health-check the wiki for contradictions, broken links, orphan pages, and missing concepts
Use when asking questions about ingested knowledge, comparing concepts across sources, or checking what the wiki knows about a topic. Prevents relying on stale memory when curated knowledge exists.
| name | llm-wiki |
| description | Use for any wiki-related task — clipping sources, ingesting knowledge, checking health, or querying. Routes automatically so you don't need to remember sub-skill names. |
| argument-hint | <natural language instruction> |
| user-invocable | true |
Route $ARGUMENTS to the correct sub-skill by intent. Never execute ingest/lint/query logic yourself — always delegate via the Skill tool.
User wants to save a web page locally as markdown.
Signals: "clip", "抓下来", "保存这个网页", "存到 sources", a bare URL with save intent, "archive this", "grab this page", "把这个存一下"
→ Invoke /llm-wiki-clip with the URL as argument.
User wants to digest material into the wiki.
Signals: "消化", "处理", "读一下这个", "加到 wiki", "ingest", "learn this", "把这篇文章吃掉", mentions a new article/paper/source to absorb
If given a URL with digest intent: recommend clip first for better quality, then ingest. Or chain both sequentially.
→ Invoke /llm-wiki-ingest with the source path or URL as argument.
User wants a wiki health check.
Signals: "检查", "体检", "扫一下", "lint", "健康", "矛盾", "断链", "orphan", "any broken links?", "wiki 有没有问题"
→ Invoke /llm-wiki-lint with scope if user specified one.
User is asking a question and wants answers drawn from the wiki.
Signals: question syntax, "什么是", "怎么理解", "比较", "分析", "总结", "explain", "what is", "how does", "compare"
→ Invoke /llm-wiki-query with the question as argument.
User wants to see source processing status.
Signals: "状态", "status", "哪些还没处理", "进度", "what's pending"
→ Handle inline: read INTERNAL_REFERENCES.md, group by Status, output:
## Ingested (N)
- [Source Title] — ingested on YYYY-MM-DD, touched N wiki pages
## Pending (N)
- [Source Title] — Local: path | Remote: url
## Partial (N)
- [Source Title] — ...
If intent is unclear, list the five operations with one-line descriptions. Ask user to pick. Do not guess.
If the request contains multiple intents ("clip this then ingest it", "消化完扫一下"), execute sub-skills sequentially in natural order: clip → ingest → lint → query.
Sub-skills internally leverage ljg- skills for enhanced quality. You don't need to invoke ljg skills directly — the sub-skills handle this. For awareness:
/ljg-paper for arXiv papers (Feynman-style extraction), /ljg-learn for core concept deepening (8-dimension anatomy), /ljg-card for optional visual cards./ljg-plain for readability spot-checks (9 red-line criteria), /ljg-rank for structural analysis of concept space (irreducible generators)./ljg-roundtable for multi-perspective debates on contested topics, /ljg-card for answer visualization.All ljg product handling (org files in ~/Documents/notes/, PNGs in ~/Downloads/) is managed by the sub-skills — artifacts are harvested into the project structure and originals cleaned up.