Extracts text from local PDFs into Markdown clippings with fidelity flags for an Obsidian wiki.
npx skills add https://github.com/ehartye/wiki-master --skill clip-pdf-ehartye
Please help me install this Agent Skill. Skill: clip-pdf Source: https://github.com/ehartye/wiki-master/tree/main/skills/clip-pdf Command: npx skills add https://github.com/ehartye/wiki-master --skill clip-pdf-ehartye
--mode overrides when the detector misfires. - OCR fallback for scanned PDFs: Rasterizes pages with pdftoppm and recognizes them with Tesseract when no text layer exists, tagging results with extraction: ocr. - Fidelity and provenance tracking: Stamps clippings with fidelity: tabular|degraded, source URL, content hash, and topic, and records declines for thin extractions so they are not retried blindly. - Use Case: During a literature review, clip a two-column academic paper into raw/clippings/, verify the extraction, then hand it to the ingestion workflow to summarize and cross-reference it in the wiki. ## Quick Start Ask the agent to clip a local PDF into the wiki, for example: clip the paper at ~/Downloads/attention.pdf with source URL and topic set to my current research run.High-intent search queries and answers about installing and using this skill.
Run the clip-pdf.mjs script with the PDF path, an optional source URL, quality tier, and topic. It extracts text with pdftotext and writes a Markdown clipping with frontmatter into raw/clippings/, ready for ingestion into the wiki.
The clipper automatically falls back to OCR: it rasterizes pages with pdftoppm and recognizes them with Tesseract. You can force this path with --ocr, and OCR'd clippings are tagged extraction: ocr in frontmatter.
Reading-order mode emits a table column-by-column, detaching keys from values. The clipper detects tabular layouts and uses Xpdf's -table mode to preserve rows; without Xpdf it stamps the clipping extraction: table-flattened and fidelity: degraded instead of silently mispairing rows.
Only when forced: pdftotext defaults to Latin-1 on some builds, which corrupts accents and symbols when decoded as UTF-8. The clipper always passes -enc UTF-8, fixing accented text at the source without needing OCR.
Math and symbol fonts extract lossily and tables reconstructed from layout may mispair cells, so clippings carry fidelity flags. Degraded clippings should be paraphrased rather than quoted verbatim, and tabular pairings verified against the original PDF.
BraveDistributionDownload full-text PDFs from IEEE Xplore, ScienceDirect, and DOI links.
akillnessBuild and maintain a persistent markdown wiki from immutable raw sources.
f4rkh4dEnforces citationable claims and validates URLs/files in AI-generated documents.
rickyanandaExtract text from PDFs, scans, and office documents into markdown.