| name | pdf-paper-source |
| description | Convert user-selected PDF literature into check-paper paper source directories. Trigger when the user asks to import, organize, convert, or prepare PDFs as paper author and paper-id article.md data sources for ppc scan, ingest, or sync. |
PDF Paper Source
Target Format
Create one directory per paper:
paper/<AUTHOR>/<PAPER_ID>/
article.md
fetch-result.json
source.pdf
article.md is required. Preserve the original PDF as source.pdf; write fetch-result.json when metadata is available.
Import Workflow
- Resolve PDF inputs from user-provided paths, attached files, or directories. Do not invent missing papers.
- Resolve the target author. If omitted, inspect:
target/debug/ppc authors
target/debug/ppc config --show
- Convert PDFs with the bundled helper:
python3 skills/pdf-paper-source/scripts/import_pdfs.py \
--project-root /Users/hanlife02/code/check-paper \
--author "AUTHOR" \
/path/to/paper.pdf
If the helper cannot extract text, manually create the same layout using pdftotext, pypdf, PyPDF2, or OCR. Keep extracted text faithful; do not summarize the PDF into article.md.
- Inspect generated
article.md frontmatter and first page text. Fix obvious title/year/DOI mistakes before ingest.
- Validate:
target/debug/ppc scan --author "AUTHOR"
target/debug/ppc ingest --author "AUTHOR"
Use target/debug/ppc sync --author "AUTHOR" only when the user wants ingest plus analysis queueing.
article.md Rules
Use YAML frontmatter:
---
title: "Paper title"
doi: "10.xxxx/xxxxx"
year: "2024"
source: "pdf_import"
has_fulltext: true
access_method: "user_selected_pdf"
text_chars: 12345
---
# Paper title
- DOI: `10.xxxx/xxxxx`
- Year: `2024`
- Source: User-selected PDF
## Article Body
Full extracted text here.
For scanned PDFs or very short extracted text, report that OCR is needed instead of creating low-quality source text silently.
Naming
Use deterministic ASCII slugs:
- DOI present:
<year>-<normalized-doi>
- DOI missing:
<year-or-undated>-<title-or-filename-slug>
Never overwrite an existing paper directory without explicit user approval. Add a numeric suffix if importing a distinct file with the same slug.
After Import
After successful ingest, the user can ask:
target/debug/ppc ask --author "AUTHOR" "这几篇论文的主要贡献是什么？"
For LLM analysis:
target/debug/ppc analyze --author "AUTHOR" --limit 5
