Menu
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.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
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.