paper-bundle

Convert your manuscript into a .paperbundle.zip containing a paper.json manifest and an assets/ directory, then upload it via the Import paper button on the paper-builder page. The importer builds the section tree, uploads images, and imports references automatically.

The skill reads Markdown and LaTeX directly, converts .docx via pandoc with media extraction, and reads PDFs with page ranges using pdfimages for figure extraction. You can also describe a paper from scratch and have the tree drafted manually.

Place references in citation order in the references[] array using unified metadata (title, authors, year, DOI), then cite with numeric [N] markers in body text. On import, the engine creates library entries and rewrites [N] markers to [@citekey] automatically.

A figure imports blank when its attrs.storage_path has no matching assets[] entry or the image file is missing from the assets/ directory. The build_bundle.py script flags this mismatch before you upload.

Yes, for paper_type grant you route pre-made PDFs like biosketches and budgets through grant_docs[] with matching assets[] entries instead of parsing them into sections. The narrative sections such as specific_aims remain normal prose sections.

Only tree.sections is strictly required, but unknown section or block kinds are silently coerced to custom or paragraph, and malformed table cells are dropped. Images are limited to 15 MB each, and the bundle never writes to Firestore or Storage directly.