Menu
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.
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
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 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.
| name | llm-wiki-query |
| description | 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. |
| argument-hint | <question> |
| user-invocable | true |
从 wiki 中检索信息回答问题。每个论点带页面引用，只用 wiki 内容作答。
$ARGUMENTS 是用户的问题（自然语言）。
读取 wikis/index.md，浏览所有 entry 的标题和简介，判断哪些页面与问题相关。
读取所有可能相关的 wiki 页面。如果一个页面引用了其他页面，且那些页面也可能相关，一并读取。宁可多读几个页面，不要遗漏关键信息。
基于读取的 wiki 内容回答问题。
（见 [page-name](wikis/category/slug.md)）。没有引用支撑的论点不要写。如果回答综合了多个页面的分析、发现了页面之间未被记录的关联、或产生了新的概念梳理，在回答末尾提示用户：
这个回答可以回写为 wiki 页面
wikis/concepts/[suggested-slug].md。要回写吗？
如果用户同意：
wikis/index.md 添加 entry。wikis/log.md 末尾追加一条 query write-back 记录。如果问题涉及比较或争议（如"隐式循环 vs 显式图哪个好？"、"context compaction 和 offloading 的取舍？"），wiki 内容可能只有各方各自的陈述而缺乏碰撞。
此时建议用户调用 /ljg-roundtable 做一次虚拟圆桌辩论——从 wiki 中挑选代表性立场（如 Anthropic 观点 vs LangChain 观点），让它们碰撞，提取最深层分歧。
产物处理：ljg-roundtable 输出 org 文件到 ~/Documents/notes/。
wikis/concepts/implicit-vs-explicit-loop.md）。~/Documents/notes/ 中的 org 文件。如果回答涉及结构性对比、流程、层级关系等空间性强的内容，offer 用 /ljg-card -i 生成信息图。
产物处理：ljg-card 输出 PNG 到 ~/Downloads/。
assets/ 目录并嵌入页面。wikis/，不碰 docs/ 或 patterns/。