luffysolution-svg/omnischolar/paper-reading

Prepare a paper for local reading

This Skill owns attachment selection and MinerU publication, not interpretation. Its successful handoff is an exact publication.markdownPath, paper identity, and parser limitations.

Workflow

1. Resolve the bibliographic parent and attachments with zotero_item in aggregate mode. If more than one PDF is available, list attachment keys and filenames, ask the user to choose, and do not parse or synchronize until an explicit attachmentKey is confirmed.
2. Reuse an already known, readable publication.markdownPath unless the user requested a fresh parse or the source identity/version is inconsistent. Do not query or upload again merely because interpretation was requested.
3. When parsing is needed, call omnischolar_parse directly after the user has authorized the selected PDF's external upload. The parse call performs MinerU cache lookup or upload before publication safety checks; do not make omnischolar_sync action=plan a mandatory parse preflight.
4. Do not set force unless the user explicitly requests a fresh parse and understands that it may upload the PDF again. Never use it to bypass a conflict, exclusion, missing publication, or interrupted transaction.
5. From the result, verify publication.markdownPath, bibliographic frontmatter, metadata.json, and sibling assets. Parsed image filenames begin with the Zotero key. Report OCR, formula, table, image, or layout limitations that affect downstream reading.
6. Hand all interpretation, evidence location, and figure/table/equation reading to literature-reading. Do not write an interpretation file from this Skill.

Read references/sync.md only for an explicit synchronization, repair, exclusion, conflict, or recovery request. Those states govern publication, not whether MinerU may probe its cache or process an explicitly authorized PDF first.

If MinerU is unavailable, Zotero indexed text may answer a narrow question when coverage is sufficient and the limitation is explicit. Do not claim a complete paper interpretation from metadata, notes, annotations, or incomplete indexed text.