What problem does it solve?
NSFC grant proposals in LaTeX often contain broken citations, missing BibTeX keys, malformed DOIs, or sentences that cite papers whose content does not actually support the claim. Manually auditing every \cite command across a multi-file proposal is slow and error-prone. This Skill performs a read-only audit of citations and references, producing a structured review report without touching your .tex or .bib files.
Core Features & Use Cases
- Deterministic integrity checks: Parses the full \input/\include dependency tree, extracts every citation command, and detects missing bibkeys, duplicate entries, missing required fields (title/author/year), and invalid DOI formats.
- Optional online DOI verification: Queries Crossref and OpenAlex to confirm DOI existence and flag title mismatches between your BibTeX entry and the registered metadata.
- AI-assisted semantic review: Generates a structured JSON input so the host AI can judge, per citation, whether the sentence's claim actually matches the cited paper, flagging P0/P1 risks like misattribution or overclaiming.
- Use Case: Before submitting an NSFC proposal, run the audit on projects/NSFC_General to receive a report listing all missing bibkeys, suspicious DOIs, and sentences whose citations need manual re-checking.
Quick Start
Ask the AI to use nsfc-ref-alignment to check citation reliability for your proposal at projects/NSFC_General in read-only mode, outputting only a review report to ./references.