Full Paper Audit Skill

SkillDev tools

Audit the entire paper -- cross-section consistency, all citations, all style issues

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Full Paper Audit Skill skill

What this skill tells your AI

The instructions your AI receives, as published by alexander-m-dickerson/ai-asset-pricing in .claude/skills/full-paper-audit/SKILL.md and read by ahel’s review.

Comprehensive audit of the entire paper for style compliance, factual consistency, citation correctness, and cross-section coherence.

Examples

  • /full-paper-audit -- run complete audit
  • /full-paper-audit --focus style -- style-only pass
  • /full-paper-audit --focus citations -- citation-only pass

Workflow

Step 1: Load All Context

  1. Read .claude/rules/academic-writing.md
  2. Read .claude/rules/banned-words.md
  3. Read guidance/paper-context.md (if it exists)
  4. Read .claude/rules/latex-citations.md (if it exists)
  5. Read main.tex in full

Step 2: Section-by-Section Audit

Discover sections by scanning for %% BEGIN: / %% END: markers in main.tex. Run /audit-section on each section sequentially.

Step 2b: Math Audit

For sections containing formal environments (any section with \begin{proposition}, \begin{theorem}, \begin{lemma}, or \begin{proof}), run /audit-math. Include SEVERITY SUMMARY and TOP PRIORITY FIXES in the master report.

Step 2c: Editorial Artifact Scan

Before proceeding to cross-section checks, grep the full manuscript for submission-blocking editorial artifacts in active prose (not LaTeX % comments):

  • [HUMAN EDIT, TODO, FIXME, XXX, [TBD], [PLACEHOLDER], [INSERT
  • Parenthetical editing notes: (change to, (should be, (need to, (fix this), (update this)
  • Missing-reference markers: [??], [?], [cite], [ref] Any hit is Critical and goes to the top of the priority fixes list.

Step 3: Cross-Section Consistency

Check that the SAME numbers are used consistently everywhere:

  • Do quantitative claims in the introduction match those in the results sections?
  • Does the conclusion match the findings reported in the body?
  • Are terminology choices consistent across all sections?
  • Are counts (sample size, number of variables, etc.) consistent everywhere they appear?

If guidance/paper-context.md exists, cross-reference all claims against its canonical values.

Step 3b: Caption Consistency

Run /audit-captions to check caption-level consistency. Include CRITICAL and IMPORTANT findings in the master report.

Step 4: Cross-Reference Audit

  • Check all \ref{} and \eqref{} resolve to valid labels
  • Check all tables and figures are referenced in text
  • Check no orphaned labels exist

Step 5: Citation Completeness

  • Check all \cite{} keys exist in .bib
  • Check all .bib entries are actually cited (flag unused entries)
  • Verify key citations via Perplexity (batch mode, high-priority entries first)

Step 6: Compile Master Report

Step 6b: Aggregate AI-Tell Statistics

After section-by-section audit, run a paper-wide pass for patterns that only emerge at scale:

  • AI-marker word frequency: Count total occurrences of all Kobak/Gray/Liang markers across the full paper. Report density per 1000 words. Flag if >2 per 1000 words.
  • Transition diversity: List all paragraph-opening words/phrases. Flag if any single opener appears 3+ times.
  • "By contrast" / "In contrast" density: Flag if >3 uses paper-wide.
  • Soft-ban accumulation: Sum all soft-ban word uses across sections. Flag if total exceeds 10.
  • Sentence length distribution: Sample 20 paragraphs. Report coefficient of variation in sentence length. Flag if CV < 0.25 (too uniform).
  • Intensive reflexive count: Count "itself"/"themselves" paper-wide. Flag if >4.

Output

FULL PAPER AUDIT
=================

OVERVIEW:
- Total issues: N
- Critical: M
- Suggestions: K
- Sections audited: [N body + M appendices]

CROSS-SECTION CONSISTENCY:
- [list of inconsistencies]

STYLE SUMMARY BY SECTION:
| Section | Banned Words | Passive | Vague Claims | Terminology | Total |
|---------|-------------|---------|--------------|-------------|-------|
| [name]  | ...         | ...     | ...          | ...         | ...   |
[etc.]

CITATION AUDIT:
- Total citations: N
- Verified: X
- Flagged: Y

CROSS-REFERENCES:
- Resolved: X
- Broken: Y

AI-TELL STATISTICS:
- AI-marker words: N (density: X per 1000 words)
- Unique paragraph openers: N out of M paragraphs
- Soft-ban total: N uses
- Sentence length CV: X (target: >0.30)

TOP 10 PRIORITY FIXES:
1. [most important issue]
[etc.]

Signals

GitHub stars
59
Forks
10
Last commit
Apr 2026
Advanced
Catalog kind
skill
Gateway key
full-paper-audit
Source
github.com/alexander-m-dickerson/ai-asset-pricing