一键导入
wiki-ingest
Use when adding a new source to a wiki — a paper, article, URL, file, transcript, or any document. One ingest may touch 10-15 wiki pages.
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。
菜单
Use when adding a new source to a wiki — a paper, article, URL, file, transcript, or any document. One ingest may touch 10-15 wiki pages.
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
为文章生成配图。分析文章结构,判断哪些位置需要什么类型的插图(infographic/flowchart/comparison/framework/timeline/scene),生成 prompt 并调用 image-gen 生成图片,最后插入文章。当用户说"配图"、"给文章加插图"、"生成文章插图"、"illustrate this article"时触发。即使用户只说"帮这篇文章配个图"也应该触发。
AI image generation with OpenAI, Google and DashScope APIs. Supports text-to-image, reference images, aspect ratios, and direct Cloudflare R2 upload. Sequential by default; parallel generation available on request. Use when user asks to generate, create, or draw images.
快速信息推文 — 用户丢过来一条推文链接、新闻、开源项目、产品动态等素材,快速消化内容,写一条信息推文,判断该引用转发还是原创发布,确认后通过 xurl 直接发送。当用户说"帮我转发这个"、"这个发一条推"、"帮我分享这个"、"这条推翻译一下发出去"、"写个推发出去"并附带了链接或素材时触发。也适用于用户贴了一段英文内容想快速中文转发的场景。
Use when fact-checking a single wiki page against its cited sources — verifies that every footnote actually supports its claim and surfaces uncited factual claims. Run after ingesting a high-stakes page or any time you want confidence in one page's accuracy.
Use when bootstrapping a new personal wiki for any knowledge domain — research, codebase documentation, reading notes, competitive analysis, or any long-term knowledge accumulation project.
Use when asking a question against a personal wiki built with wiki-init and wiki-ingest. Do not answer from general knowledge — always read the wiki pages first.
| name | wiki-ingest |
| description | Use when adding a new source to a wiki — a paper, article, URL, file, transcript, or any document. One ingest may touch 10-15 wiki pages. |
Add a source to the wiki. Read it, discuss with the user, write a summary page, update entity/concept pages, and maintain the index, overview, and log.
Search for SCHEMA.md starting from the current directory and upward, or in common wiki locations (~/wikis/). If not found, tell the user to run wiki-init first.
Read SCHEMA.md to learn: wiki root path, page frontmatter format, cross-reference convention, log entry format, index category taxonomy.
The source can be:
raw/<filename> if not already therebrowse skill to fetch it; save to raw/<slug>.<ext>Read all content. For long sources, read in sections. Do not skip.
Tell the user:
wiki/index.md and relevant pages to check)Ask: "Anything specific you want me to emphasize or de-emphasize?"
Wait for the user's response before proceeding.
Lowercase, hyphens, no special characters.
Example: "Attention Is All You Need" → attention-is-all-you-need
Write wiki/pages/<slug>.md:
---
title: <source title>
tags: [<relevant tags>]
sources: [<slug>]
updated: <today>
---
# <Source Title>
**Source:** <original URL or file path>
**Date ingested:** <today>
**Type:** <paper | article | transcript | code | other>
## Summary
<2-3 paragraph synthesis — your own words, not abstract copy-paste>
## Key Takeaways
- <bullet>
## Entities & Concepts
<list of entities/concepts as [[slug]] links>
## Relation to Other Wiki Pages
<how this connects to or updates existing knowledge>
While drafting the Summary, Key Takeaways, and any other prose section, every
non-common-knowledge factual claim must carry a footnote. Read the Citations
section in SCHEMA.md for the full convention.
Two citation kinds, three valid targets:
Quote: [^N]: <target> <locator> — "<verbatim quote>"
Synthesis: [^N]: <target> <locator> [synthesis] — <what supports the claim>
<target> is one of:
[[source-slug]] — a source wiki page (preferred for the source you're ingesting)
raw/<file> — a path, for drive-by citations to other local files
<URL> — a live URL or post
For the source being ingested, use [[<this-source-slug>]] — wiki-ingest
is creating that page now, so the target exists by the time the page is read.
If you cannot produce either citation kind for a claim, you do not have a citation. Find one, weaken the claim ("the paper suggests..."), or drop it.
Footnotes go at the bottom of the page, below all sections. Number them sequentially in order of first reference.
Re-read the draft once. Scan for unfootnoted factual claims — this is the most common failure mode in long ingest sessions. For each, add a footnote or revise the wording. Only then move on to entity pages.
For each entity/concept touched by this source:
sources list, update updated date---
title: <Entity or Concept Name>
tags: [entity | concept]
sources: [<this-source-slug>]
updated: <today>
---
# <Name>
## Description
<synthesis across all sources that discuss this>
## Appearances in Sources
- [[source-slug]] — <one-line note>
## Related Concepts
- [[related-slug]] — <relationship>
Scan ALL existing pages in wiki/pages/ for any that mention this source's entities/concepts but don't yet link to the new page. Add [[new-slug]] references where appropriate.
This is the step most commonly skipped. A compounding wiki's value comes from bidirectional links.
wiki/index.mdAdd an entry under the correct category:
- [[<slug>]] — <one-line summary> _(ingested <date>)_
For any new entity/concept pages created, add those too.
wiki/overview.mdRe-read the current overview. If this source:
Update the frontmatter updated date.
wiki/log.md## [<today>] ingest | <source title>
Pages written: <slug>
Pages updated: <comma-separated list>
## April 27 update: or **Update:** followed by new content. Update the relevant section in-place, bump the updated frontmatter date, and record what changed in log.md. The log is the append-only record; pages are the current truth.wiki/pages/<slug>.md