What problem does it solve? Academic manuscripts need a bibliography where every field is traceable to a real source and every claim about a cited work can be audited. This Skill maintains a verified reference base — BibTeX entries transcribed only from records fetched during the run, per-paper reading notes with auditable citable facts, and a provenance index — so citations never come from memory or guesswork. ## Core Features & Use Cases - Verified intake: Resolves arXiv ids, DOIs, URLs, or titles against DBLP, Crossref, Semantic Scholar, and arXiv with a three-field match rule, caching every payload and recording a % src: provenance line per entry. - Discovery and seeding: discover searches by topic from the paper's own story and proposes ranked candidates for the author to approve; seed converts imported STAR reference trees while preserving upstream citekeys. - Maintenance modes: tidy deduplicates offline, position clusters the base for related work, verify re-fetches and diffs every entry, and score refreshes impact metrics. - Use Case: While drafting a paper's related work section, run discovery to find candidate papers from your story notes, approve a subset, and receive fetched BibTeX entries plus reading notes whose citable facts a citation auditor can later verify. ## Quick Start Ask the assistant to run stage-refs-curator with an arXiv id or DOI to add that paper to the bibliography and write its reading note.
