citation-audit

What it does

Verify every \cite{...} in a paper against three independent layers:

1. Existence — the cited paper actually exists at the claimed arXiv ID / DOI / venue.
2. Metadata correctness — author names, year, venue, and title match canonical sources (DBLP, arXiv, ACL Anthology, Nature, OpenReview, etc.).
3. Context appropriateness — the cited paper actually supports the claim it is being used to support in the manuscript.

This skill is designed to run before submission to ensure bibliography integrity and is gated behind user prompts for specific citation-related languages.

How it works

• Step 1: Discover bib file(s) and all files containing \cite{...} calls. Audit each bib entry that is cited.
• Step 2: Extract (cite-key, file, line, surrounding_sentence) tuples and map citations to bib entries.
• Step 3: For each cited bib entry, invoke a fresh cross-model reviewer (mcp__codex__codex) with model gpt-5.5 to verify:
  • Existence: YES / NO / UNCERTAIN with verifying URL.
  • Metadata: whether author names, year, venue, and title are correct; report as correct / wrong / typo with expected values.
  • Context: for each use, verdict as SUPPORTS / WEAK / WRONG with one-sentence reasoning.
  • Overall VERDICT: KEEP / FIX / REPLACE / REMOVE.
• Step 4: Save reviewer responses to .aris/traces/citation-audit/<date>_runNN/<key>.md.
• Step 5: Aggregate into CITATION_AUDIT.json using the provided schema, with top-level verdict and per-entry details.
• Step 6: Generate human-readable CITATION_AUDIT.md documenting counts and per-entry actions; include sections for uncited entries if --uncited is used.
• Step 7: If FIX/REPLACE/REMOVE are chosen, prompt for user confirmation to apply changes to the bibliography or citations, with an option to AUTO_APPLY.
• Step 7 also includes a final recompile/check to ensure no new citation warnings appear.

When to use it

Run before submission after paper-write has produced the LaTeX draft and bib file, after paper-claim-audit has verified numerical claims, and before final paper-compile for submission. Do not run on a half-written draft.

What this skill touches

• Uses mcp__codex__codex with model gpt-5.5 and web lookups (DBLP/arXiv/venue databases).
• Reads references.bib or equivalent bib files and scans \cite{...} usages in TeX files.
• Writes outputs to CITATION_AUDIT.json and CITATION_AUDIT.md and per-entry trace files under .aris/traces/citation-audit/.

Caveats

• Requires web access for lookups; results depend on live sources.
• Fresh reviewer thread per audit; cannot reuse prior context.
• If a bib file cannot be parsed, behavior is defined to fall back to blocked or partial reports without altering verdict structure.