citation-reference-reviewer — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited citation-reference-reviewer (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Verify every reference against authoritative external sources before declaring it correct. Do not assume that titles, author lists, venues, years, DOIs, arXiv IDs, URLs, or BibTeX fields are accurate because they look plausible.
If available, apply the shared protocol in ../_shared/scientific-paper-common.md.
Use this source priority:
If sources disagree, report the disagreement and prefer publisher or proceedings metadata unless there is evidence of publisher metadata error.
Inspect all available manuscript and bibliography files:
.tex, included sections, appendices, and supplementary material..bib, generated .bbl, or reference lists if BibTeX is not yet available.\cite, \citep, \citet, \citeauthor, \citealp, \parencite, \textcite, and custom citation macros.All maintained references must remain in BibTeX format unless the user explicitly asks for another format.
Before external verification, generate a machine-readable citation inventory when possible.
Use scripts/citation_inventory.py when available:
python .codex/skills/citation-reference-reviewer/scripts/citation_inventory.py --root . --out citation_inventory.json --md citation_audit.mdRequired checks:
\cite, \citep, \citet, \citeauthor, \citealp, \parencite, \textcite, and custom citation commands;.tex but missing from .bib;.bib entries never cited;doi field;When writing inventory outputs, use:
citation_inventory.jsoncitation_audit.mdhttps://doi.org/<doi> or an authoritative metadata source.Flag entries with:
@article, @inproceedings, @book, @incollection, @techreport, @misc, @standard, @manual, @phdthesis, @mastersthesis, or toolchain-supported alternatives.doi fields, prefer bare DOI strings unless the project style requires DOI URLs. For displayed reference text or publisher-facing links, prefer the canonical https://doi.org/... form.url for stable non-DOI resources and include urldate only if the bibliography style supports it or the project already uses it.{IEEE}, {GPU}, or method names.eprint, archivePrefix = {arXiv}, and primaryClass when available.If a citation, source record, DOI, URL, or metadata field cannot be verified from supplied files or authoritative external sources, mark it as unresolved. Do not silently repair it by inventing missing information.
Return the audit in this order:
Verification summary: counts of citations, BibTeX entries, verified entries, corrected entries, unresolved entries, uncited entries, and missing entries.Inventory results: citation inventory path if generated, missing keys, uncited keys, duplicate keys, duplicate DOI/source candidates, and malformed DOI fields.Critical reference problems: fabricated, non-resolving, mismatched, duplicate, or missing sources that must be fixed.Citation consistency: missing keys, uncited references, duplicate keys, citation clusters that need explanation, and IEEE style issues.Per-reference audit table: key, source status, metadata status, DOI/URL status, authoritative source used, and required action.Corrected BibTeX: provide complete replacement entries for entries that need correction.Unresolved items: list sources that could not be verified and the exact authoritative searches attempted.When reporting verification, cite or name the authoritative source used for each correction. If external access is unavailable, mark entries as not externally verified and explain that final authenticity checks remain pending.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.