review-17-citation-audit
Verify citations, reference formatting, DOI/PMID accuracy, source quality, publication year, and claim-citation alignment. Use before submission, major revision, or when adding/replacing references in
SKILL.mdskills/review-17-citation-audit/SKILL.md
GitHubReview 17 Citation Audit
Purpose
Prevent fabricated, mismatched, outdated, or unsupported citations.
References are a priority evidence layer, not decorative support. Treat every added, replaced, or retained citation as a claim-level evidence decision.
Required Inputs
- Full manuscript draft.
- Reference library export.
- Evidence matrix.
- Target citation style.
- User's journal/source exclusions, if any.
Output
Create 13_citation_audit/citation_audit_report.md and 13_citation_audit/claim_citation_check.csv with:
- Missing citation list.
- DOI/PMID checks.
- Citation-to-claim alignment.
- Reference style issues.
- Papers requiring replacement or stronger support.
- Source quality and publication-year risks.
Strict Reference Rules
- Do not invent references, DOI, PMID, author names, journal names, volume, issue, pages, article numbers, or publication years.
- Prefer references from 2021-2026. If the user explicitly allows a 2020 start year, keep 2020 papers only when they are high-impact foundational evidence, guidelines, consensus statements, or major reviews.
- Require DOI and PMID for core biomedical references unless the user explicitly approves an exception. Flag any missing DOI or PMID for replacement.
- Verify existence through PubMed and publisher/DOI sources first. Use Web of Science or Google Scholar only as additional support or fallback discovery, not as the sole final proof when PubMed/publisher data are available.
- Prefer high-impact, field-relevant journals, JCR Q1-Q2, CAS high-partition sources, guidelines, consensus statements, and major original studies or authoritative reviews.
- Avoid MDPI, Frontiers, Scientific Reports, Medicine, and similar user-excluded or quality-risk sources for this user's SCI review work unless no credible alternative exists and the user explicitly accepts the risk.
- Check whether each reference supports the exact claim, not merely the broad topic.
- Do not use correlation evidence to support causal wording.
- Do not use basic/mechanistic evidence as direct support for clinical efficacy unless the sentence is explicitly framed as mechanistic or hypothesis-generating.
- For paragraph-level citation insertion, ensure every substantive paragraph has at least one appropriate citation and key factual claims have direct support.
Completion Gate
Pass only when every key claim has an appropriate citation; no reference is knowingly fabricated, unverifiable, mismatched, outdated without justification, or inconsistent with the user's source-quality rules; and all core references have DOI/PMID or documented user-approved exceptions.
Stop Rules
Do not invent citations to fill gaps. Mark gaps as citation needed. If no suitable reference can be found under the required year, DOI/PMID, and journal-quality constraints, report the gap and propose a replacement search strategy instead of lowering the evidence standard silently.