| name | paper-citation-audit |
| description | Use when auditing manuscript claims, citations, references, DOI/BibTeX metadata, citation support strength, unsupported claims, fabricated references, or claim-evidence-boundary maps. |
Paper Citation Audit
Use this skill when a manuscript needs evidence-grounded citation checking. The goal is to prevent fabricated references, weak support, and unsupported scientific claims.
Inputs
- Manuscript text or selected claims.
- Existing references or BibTeX if available.
- Target venue and subject area.
- Allowed sources or search constraints if any.
Workflow
- Segment the passage into auditable claims.
- For each claim, identify phenomenon, entity, relationship, context, and boundary.
- Preserve existing citation anchors.
- Search or inspect candidate sources when available.
- Grade support strength.
- Flag unsupported, overclaimed, stale, or metadata-only citations.
- Suggest safer citation wording.
- Return a claim-evidence-boundary table.
Support Grades
- Strong: directly supports the same relationship in a comparable context.
- Partial: supports one component or a narrower setting.
- Background: establishes context but not the main claim.
- Limiting: contradicts or narrows the claim.
- Metadata-only: candidate found but abstract/full text not checked.
Citation Gate
- Do not invent DOI, volume, issue, page, title, author, or journal metadata.
- Do not cite a paper for a claim it does not support.
- Do not upgrade background evidence into direct support.
- Keep unsupported claims visible instead of hiding them with vague citations.
When To Read References
- For claim segmentation, read
references/claim-segmentation.md.
- For audit output format, use
../../templates/claim-evidence-boundary.md.
