rag-parse
Process/ingest the documents in /work/data/input into the knowledge stores (Obsidian vault + vector RAG index). Incremental: only new or changed files are processed, and files removed from input are pruned from the stores. Use when the operator wants to load, process, parse, ingest, index, or update their data. Triggers: parse, parse the files, process the input, ingest, index the documents, update the store, re-parse, load my data, refresh the index.
- License 许可
- MIT
- 最近更新
- 2026-11-11
- Stars
- ★ 1
- Forks
- 0
01 前言 Front Matter
- name
- rag-parse
- description
- Process/ingest the documents in /work/data/input into the knowledge stores (Obsidian vault + vector RAG index). Incremental: only new or changed files are processed, and files removed from input are pruned from the stores. Use when the operator wants to load, process, parse, ingest, index, or update their data. Triggers: parse, parse the files, process the input, ingest, index the documents, update the store, re-parse, load my data, refresh the index.
03 SKILL.md 正文
Parse input into the knowledge stores
Turn everything in /work/data/input into both knowledge stores with one
incremental, deterministic command. It extracts text + source locators, (re)builds
the vector index, and writes Obsidian source-pages — skipping unchanged files and
pruning ones the operator removed. It reads PDFs (OCR'ing scanned ones), text and
Markdown files, and images (OCR'd with tesseract).
kstore sync
Then:
- Report the summary line it prints (new / changed / unchanged / removed, OCR'd, vault pages).
- If it lists files as needs OCR (scanned PDFs, and
ocrmypdfwas not available), OCR each into the cache and re-run, e.g.:ocrmypdf --skip-text "/work/data/input/FILE.pdf" /work/data/store/cache/FILE-ocr.pdfthenkstore sync. (In this sandboxocrmypdfis installed, sosyncnormally OCRs scanned PDFs automatically.) - Never write extracted text, OCR output, or page images to
/work/data/output. That directory is for final deliverables only. All derived and transient artifacts belong under/work/data/store(the tools handle this).
Optional: richer Obsidian synthesis
sync populates the vault with one source-page per document (fast, mechanical).
For a cross-linked knowledge graph with claim ledgers and Maps of Content, the
operator can additionally ask for synthesis — run /skill:wiki-ingest on
specific documents. That is a heavier, LLM-driven layer on top of the source
pages; only do it when asked, and a few documents at a time (it consumes context).
04 同仓库更多 Skills
other
glove probe skill the session does not select.
greet
glove probe skill; greets through the plugin's script.
rag-query
Answer a question from the operator's parsed documents (the knowledge stores) and return sources for verification. Use this for ANY question about the operator's data, files, or documents — retrieve, do not answer from general knowledge. Retrieves from the Obsidian store, the vector index, or both (configurable). Triggers: what does, find, search the documents, according to the files, look up, where does it say, cite, question about the data, summarize the documents, is there anything about.