SuperAPI AI Directory

citation-management

Writing & research

View repository

Updated 10-15 20:06

About this skill

Citation Management

Use this skill when a task asks you to validate, repair, deduplicate, or format academic citations. Treat citation work as evidence work: never invent metadata, venues, DOIs, authors, or publication years.

Workflow

  1. Parse every citation entry first. Record the key, title, authors, venue, year, DOI/URL/PMID/arXiv ID, and any suspicious fields.
  2. Verify identifiers before prose. Prefer DOI resolution, Crossref, PubMed, arXiv, ACL Anthology, DBLP, publisher pages, or other primary metadata sources over search-result snippets.
  3. Check whether title, authors, venue, year, and identifier all refer to the same work. A real DOI attached to the wrong title is still a bad citation.
  4. Flag hallucinated or fake citations when identifiers do not resolve, the claimed venue appears nonexistent, or no credible source supports the title.
  5. Keep uncertain cases separate from confirmed fake cases unless the requested schema has no uncertainty bucket.
  6. When producing BibTeX, normalize braces and escaping, preserve meaningful capitalization, and include only fields supported by evidence.
  7. Return exactly the requested output format. If JSON is requested, return valid JSON only.

Evidence Standards

  • A valid DOI is strong evidence only if it resolves to the same title and authors in the citation.
  • Missing DOI alone does not make a citation fake; books, older papers, and proceedings may lack DOIs.
  • Generic titles, generic author names, unverified venues, impossible DOI prefixes, and absent source coverage are warning signs, not final proof by themselves.
  • If the task provides a source-of-truth excerpt, use it as evidence and state any mismatch against it.