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.
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 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/。