/local-vault
Turns a folder of messy files into a clean, local Markdown knowledge base — then answers questions over it with an analyst's retrieval discipline. Your documents stay on your machine; AI reads them well.把一堆杂乱文件，整理成一个干净的、本地的 Markdown 知识库——再用分析师的检索纪律在上面回答问题。你的文档留在本机，AI 也能读得明白。
Type /local-vault, or say things like "convert these files to markdown for AI", "整理我的资料库", "本地知识库", or "这个主题我的资料里怎么说". Point it at a folder to build the vault, then just ask.输入 /local-vault，或者直接说“把这些文件转成 md 给 AI 读”“整理我的资料库”“本地知识库”“这个主题我的资料里怎么说”。指给它一个文件夹建库，然后开口问就行。
Convert your files in. Ask questions out.把文件转进来，再把问题问出去。
Raw files → a clean Markdown vault原始文件 → 干净的 Markdown 库
PDFs, Office docs, images and code become clean Markdown with retrieval-friendly frontmatter. Local-first — pandoc, python-pptx, openpyxl, pymupdf4llm — with cloud OCR only as a fallback when a file resists local extraction.PDF、Office 文档、图片和代码，转成带检索友好 frontmatter 的干净 Markdown。本地优先——pandoc、python-pptx、openpyxl、pymupdf4llm——只有当某个文件本地抽取不出来时，才退回云端 OCR 兜底。
Answer questions over the vault在资料库上回答问题
Ask anything across the vault and get answers with retrieval discipline — coverage self-monitored, missing or lossy content flagged rather than glossed over, and Maps-of-Content proposed to organize what's there.在整个资料库上随便问，回答带检索纪律——自检覆盖率，缺失或有损的内容会被标出来而不是糊弄过去，并提出内容地图（MOC）帮你梳理已有材料。
Not for one-off web research, or a single document you can just read directly — this earns its keep on a folder of files you'll return to.不适合一次性的网络检索，也不适合一份你直接读就行的单一文档——它的价值，在于一个你会反复回来翻的文件夹。
- Local-first, cloud as fallback. Everything is extracted on your machine; cloud OCR is only invoked when a file resists local processing.本地优先，云端兜底。一切都在你本机抽取；只有本地处理不了的文件，才调用云端 OCR。
- Retrieval-friendly frontmatter. Every converted file carries structured metadata, so the vault stays searchable instead of becoming a pile of loose text.检索友好的 frontmatter。每个转换后的文件都带结构化元数据，资料库始终可检索，而不是一堆散落的文本。
- Coverage self-monitored. When answering, the skill tracks how much of the vault it actually read — and tells you when an answer rests on thin coverage.覆盖率自检。作答时会盯住自己究竟读了资料库的多少——覆盖太薄时主动告诉你，而不是装作读全了。
- Missing / lossy content flagged. Gaps and degraded conversions are surfaced, never quietly glossed over.缺失 / 有损内容会标出。缺口和转换降级都会被点出来，绝不悄悄糊弄过去。
- Maps-of-Content proposed. The skill suggests MOCs to give your growing vault a navigable structure.提出内容地图（MOC）。为不断变大的资料库建议 MOC，给它一个可导航的结构。
- pandoc Word · HTML · MarkdownWord · HTML · Markdown
- python-pptx PowerPoint slidesPowerPoint 幻灯片
- openpyxl Excel · spreadsheetsExcel · 表格
- pymupdf4llm PDF text & layoutPDF 文本与版式
- MinerU Cloud OCR — fallback only云端 OCR——仅兜底