| name | opendataloader-paper-intake |
| description | Use when Codex needs to place papers into a package's canonical source tree, convert PDFs to single-file markdown with OpenDataLoader PDF as the primary converter, omit images, fall back to markitdown on failure, and update the package paper index using the tracked paper-intake workflow. |
OpenDataLoader Paper Intake
Use this skill for repo-local paper intake and PDF-to-Markdown conversion, not for generic OCR advice.
Working Rules
-
Start with the tracked workflow doc at ../../../workflows/paper-intake/README.md.
-
Use the repo wrapper rather than calling the converter directly:
.venv/bin/python workflows/skills/opendataloader-paper-intake/scripts/convert_pdf_to_markdown.py input.pdf output.md
-
OpenDataLoader PDF requires Java 11+. Check java -version when bootstrapping a new machine.
-
For PDF sources, the wrapper uses OpenDataLoader PDF first with Markdown-only output, one Markdown file per PDF, page separators, and images disabled.
-
Keep markitdown installed as the fallback converter. The wrapper falls back to it when OpenDataLoader fails or when the source is not a PDF.
-
Prefer placing source PDFs or HTML exports under the target package papers/ directory.
-
For single-publication packages, prefer papers/markdown/ for converted markdown and papers/index.md for the package paper index unless the package already has a stronger established variant.
-
Keep filenames descriptive and stable. Do not rename aggressively if the package already has a working convention.
-
When converting several papers, write outputs first, then summarize failures or partial conversions instead of stopping after the first bad file.
-
If the package already has a paper manifest or source index, update that instead of creating a duplicate second index.
Direct OpenDataLoader Equivalent
Use this only when debugging the wrapper:
.venv/bin/opendataloader-pdf input.pdf \
--format markdown \
--image-output off \
--markdown-page-separator "<!-- page %page-number% -->" \
--output-dir output/
The canonical conversion should still go through the wrapper so fallback behavior stays consistent.
Reference Map
../../../workflows/paper-intake/README.md: canonical tracked paper-intake workflow.../../../README.md: repo package model, vocabulary, and package-map context.scripts/convert_pdf_to_markdown.py: primary OpenDataLoader conversion wrapper with markitdown fallback.
