Check Consistency Skill

SkillDev tools

Fast cross-section consistency scan for numbers, terminology, and references

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 Check Consistency 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/check-consistency/SKILL.md and read by ahel’s review.

Fast, focused scan for cross-section inconsistencies in main.tex. Lighter than /full-paper-audit -- designed for iterative use during editing.

Examples

  • /check-consistency -- full scan of main.tex
  • /check-consistency numbers -- only check quantitative claims
  • /check-consistency terminology -- only check terminology consistency

Workflow

Step 1: Load Reference Values

Read guidance/paper-context.md to get canonical values (if it exists):

  • Key quantitative results and their canonical magnitudes
  • Sample description (date range, number of observations, variable counts)
  • Any other numbers that appear in multiple sections

If no paper-context file exists, the skill still works by cross-referencing sections against each other (without a canonical reference).

Step 2: Quantitative Consistency

Grep main.tex for all quantitative claims and cross-reference:

  • Percentages and basis points mentioning specific variables or factors
  • Sample period mentions (start date, end date, number of periods)
  • Counts (variables, observations, subsamples, etc.)
  • Any number that appears in more than one section

Flag: mismatches between text claims and canonical values (if available), or between sections.

Step 3: Terminology Consistency

If guidance/paper-context.md defines a terminology table, grep across all sections for violations.

Also check for within-paper drift regardless of paper-context:

  • Same concept called different names in different sections
  • Inconsistent abbreviation introduction (defined in one section, used without definition in another)
  • Check against .claude/rules/banned-words.md for hard-banned terms

Step 4: Cross-Reference Integrity

  1. Extract all \ref{...} and \eqref{...} targets
  2. Extract all \label{...} definitions
  3. Flag any \ref or \eqref that points to a non-existent label
  4. Flag any \ref used where \eqref should be (equation references)
  5. Check that all tables and figures are cited at least once in the text

Step 5: Section Cross-References

Check that claims about other sections are accurate:

  • "As shown in Section X" -- does Section X actually show this?
  • "Table Y reports" -- does Table Y match the claim?
  • "See Appendix Z" -- does the appendix contain the referenced content?

Step 5b: Caption Consistency

Run /audit-captions. Include CRITICAL and IMPORTANT findings in the output.

Step 6: Output

CONSISTENCY CHECK
=================

QUANTITATIVE MISMATCHES: [N found]
- Line X: claims "[value]" but canonical value is "[value]" for [variable]
- Line Y: says "[count]" but paper-context.md says "[count]"
(or: "No paper-context.md — cross-section comparison only")

TERMINOLOGY VIOLATIONS: [N found]
- Line X: "[deprecated term]" (should be "[correct term]")
- Line Y: "[term A]" in this section vs "[term B]" in [other section]

REFERENCE INTEGRITY: [N issues]
- Line X: \ref{eq:decomp} should be \eqref{eq:decomp}
- Line Y: \ref{fig:missing} -- label not found

CROSS-SECTION CONSISTENCY: [N issues]
- Line X claims "Table 3 shows..." but Table 3 actually shows...

SUMMARY:
- Critical issues: N
- Warnings: M
- All clear: [list of checks that passed]

Signals

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