Add a full-text extraction gate for backbone articles before manuscript drafting

[ibadmursalov-crypto]

Hello,

First of all, thank you for developing and sharing the MedSci skills. I have been reviewing the write-paper workflow, and I find the backbone article concept very useful, especially for medical manuscripts, diagnostic accuracy studies, and AI/prediction model papers.

I would like to suggest a possible improvement to the current backbone article workflow.

Current behavior

As I understand it, the write-paper skill currently selects a backbone article during Phase 0 by scanning manuscript/_src/refs.bib. It ranks candidate articles based on factors such as:

• similarity of study design;
• similarity of modality/domain;
• target journal;
• recency;
• availability of a Zotero PDF attachment.

After the user confirms the selected article, the citekey is saved in project.yaml::backbone_article.

This is a strong idea. However, I noticed that the workflow does not appear to require a mandatory full-text extraction step before manuscript drafting begins.

Concern

A BibTeX entry usually contains only metadata:

• title;
• authors;
• journal;
• year;
• DOI;
• sometimes abstract;
• sometimes a local PDF path.

This is enough to identify a potentially relevant article, but it is not enough to reliably extract the actual structure of the paper, such as:

• Methods subsection hierarchy;
• cohort description pattern;
• index test / model description;
• reference standard description;
• statistical analysis structure;
• table and figure layout;
• Results sequence;
• Discussion and limitations structure.

Because of this, the selected backbone article may function mainly as a citation-level anchor rather than as a true structural backbone, unless the full text is parsed before drafting.

This may be especially important for medical AI, prediction model, and diagnostic accuracy manuscripts, where reporting structure is closely tied to methodological quality.

Suggested improvement

I suggest adding a mandatory intermediate step after backbone article selection and before manuscript drafting:

Phase 0b: Backbone Full-Text Extraction Gate

After a backbone article is selected, the skill could:

1. Resolve the full-text source:

   • Zotero / Better BibTeX local PDF attachment;
   • local project PDF path;
   • user-uploaded PDF;
   • markdown or text full-text version.

2. If full text is available:

   • parse the article;
   • extract a structured backbone_map.md;
   • save it, for example, to:

   manuscript/_src/backbone/backbone_map.md

3. If full text is not available:

   • stop and ask the user to provide the PDF/full text;
   • or explicitly downgrade the article to “citation-level comparator only”;
   • avoid claiming that the manuscript is being structurally modeled on the backbone article.

Proposed project.yaml fields

backbone_article: citekey
backbone_status: proposed | confirmed | fulltext_available | extracted | waived
backbone_fulltext_path: manuscript/_src/backbone/backbone.pdf
backbone_map_path: manuscript/_src/backbone/backbone_map.md
backbone_use_level: structural | citation_comparator_only | waived


Thank you!!!