agentleFS
Boom5426Boom5426/Nature-Paper-Skills/skills/core/citation-verifier/SKILL.md

citation-verifier

Skill661 starsChanged 36 days agoMITView on GitHub

Use when checking manuscript citations, bibliography hygiene, DOI or PMID completeness, placeholder references, or BibTeX consistency before submission or revision.

7 other projects use a version of this, 1 of them edited.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

14 rules. 0 are common, 13 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 6 edited versions we read keep these.

  • 13

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use this skill when citation integrity is the bottleneck.

    line 10
  2. Very specific:

    Use this skill together with scientific-writing or submission-audit, not instead of them.

    line 12
  3. Distinctive:

    Use this skill when:

    line 25
  4. Distinctive:

    Do not use this skill for:

    line 32
  5. Distinctive:

    Start with a local citation scan before any online verification.

    line 39
  6. Distinctive:

    Use it to find:

    line 49
  7. Distinctive:

    Treat imported BibTeX and Google Scholar-style exports as draft metadata, not as verified truth.

    line 57
  8. Distinctive:

    Remove or flag all placeholder citations.

    line 62
  9. Distinctive:

    Check BibTeX-key uniqueness and obvious identifier gaps.

    line 63
  10. Distinctive:

    Only then finalize formatting for the venue.

    line 72
  11. Distinctive:

    When you report field-level mismatches, grade each one so authors fix the dangerous ones first:

    line 76
  12. Distinctive:

    Author-order reversal: verify the first author specifically, not just that all names are present.

    line 84
  13. Distinctive:

    Prefer this order when moving beyond local scans:

    line 90
  14. Distinctive:

    When using this skill, report:

    line 108

“Common” is measured: kept by at least 6 in 10 of the 6 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Citation Verifier

Overview

Use this skill when citation integrity is the bottleneck. It is narrower and more operational than general writing skills: the goal is to catch unverifiable references, missing identifiers, duplicated BibTeX keys, placeholder citations, and mismatches between local citation artifacts before they become submission problems.

Use this skill together with scientific-writing or submission-audit, not instead of them.

This is stage one of three. It shows the bibliography is well-formed; it cannot show that the cited papers are real, and it cannot show that a real paper supports the sentence citing it. Hand the surviving entries to reference-audit-guide (optional set, installed with --set all), which queries CrossRef, Semantic Scholar, arXiv, and PubMed to confirm existence and match metadata, then to claim-source-verification for claim-to-evidence support. When reference-audit-guide is not installed, say that existence was not machine-verified rather than letting a clean local scan stand in for it. A fabricated citation carrying a well-formed DOI passes this skill and fails the next one.

When To Use

Use this skill when:

  • a manuscript is close to submission or resubmission
  • the bibliography may contain placeholders, guessed entries, or stale metadata
  • a .bib file needs a hygiene pass
  • the draft mixes DOI, arXiv, PMID, or plain-text references inconsistently
  • you need to verify that local citation artifacts are ready before manual or online source verification

Do not use this skill for:

  • generic literature discovery from scratch
  • full manuscript restructuring
  • venue formatting that does not depend on citation integrity

Local First Pass

Start with a local citation scan before any online verification.

The helper script:

python ~/.agents/skills/citation-verifier/scripts/scan_citations.py path/to/file_or_dir [...]
# Claude Code (global install): replace ~/.agents/skills with ~/.claude/skills
# Claude Code (project-local install): replace ~/.agents/skills with .claude/skills

Use it to find:

  • [CITATION NEEDED] or similar placeholders
  • PLACEHOLDER_... citation keys
  • TODO: verify markers
  • DOI, arXiv, and PMID tokens
  • LaTeX \cite{...} keys
  • duplicate BibTeX keys inside .bib files

Treat imported BibTeX and Google Scholar-style exports as draft metadata, not as verified truth.

Verification Order

  1. Scan the local draft and bibliography artifacts.
  2. Remove or flag all placeholder citations.
  3. Check BibTeX-key uniqueness and obvious identifier gaps.
  4. Normalize metadata fields that are commonly wrong in copied BibTeX:
    • author list and spelling
    • year
    • venue or journal name
    • volume, issue, pages, or article number
    • DOI versus preprint identifier
  5. For every citation that matters to a scientific claim, verify the paper exists in trusted sources.
  6. If a claim depends on a specific result, confirm the cited paper actually supports that claim. For a systematic pass over many such claims, hand off to claim-source-verification: existence and metadata checks do not catch a real paper made to carry a conclusion it does not reach.
  7. Only then finalize formatting for the venue.

Severity Grading

When you report field-level mismatches, grade each one so authors fix the dangerous ones first:

  • Critical (must fix): wrong or fabricated first author, reversed author order, a DOI that resolves to a different paper, a title whose core terms do not match, or a page range off by five or more.
  • Warning (verify): volume-year inconsistent with the DOI-registration year, a missing or extra issue number, a missing middle initial, a page range off by one to four, or an ambiguous publisher-name variant.
  • Info (note only): title case versus sentence case, abbreviated versus full journal name, and punctuation or conjunction differences (and versus &).

Two checks are easy to miss and worth running explicitly:

  • Volume-year versus DOI-year: a citation whose stated volume-year differs from the year encoded in the DOI or the online-publication date is a common sign of a miscopied entry.
  • Author-order reversal: verify the first author specifically, not just that all names are present; a reversed or shifted author list still passes a set-membership check.

Roll each reference up to one confidence verdict: verified, check_suggested (has a Warning), needs_fix (has a Critical), or unverifiable (no source could confirm it).

Trusted Verification Sources

Prefer this order when moving beyond local scans:

  • publisher page or DOI resolver
  • PubMed for biomedical papers
  • arXiv for preprints
  • Crossref or Semantic Scholar for metadata cross-checking
  • Google Scholar only as a fallback discovery aid, not as the sole source of truth

Common Failure Modes

  • writing BibTeX from memory
  • leaving placeholder citation keys in the draft
  • assuming an arXiv preprint and published version are interchangeable
  • citing a paper correctly but using it to support a claim it does not make
  • letting duplicate BibTeX keys silently overwrite entries
  • blindly trusting copied Scholar or BibTeX metadata for venue names, pages, years, or DOI fields

Output Standard

When using this skill, report:

  • placeholder or unverifiable items
  • duplicate or suspicious BibTeX keys
  • missing or inconsistent DOI/arXiv/PMID information
  • suspicious metadata fields that need manual correction
  • claims that still need source verification
  • the minimum safe next step before submission

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.