style-check

name	style-check
description	Analyze LaTeX text for violations of academic writing standards

Style Check Skill

When this skill is invoked, analyze the provided text for violations of the academic writing standards in .claude/rules/academic-writing.md.

Examples

• /style-check -- check the current .tex file
• /style-check introduction -- check only the introduction section
• /style-check projects/my_paper/latex/main.tex -- check a specific file

Input

The user provides a .tex file path or text block to analyze.

Workflow

1. Read .claude/rules/academic-writing.md for the full rule set
2. Read the project's CLAUDE.md for project-specific terminology
3. Read the target text (from file path or section markers)
4. Run all 13 analysis categories below
5. Check citations against the project's .bib file
6. Produce the structured report

Analysis Categories

1. Banned Word Scan

Check for all terms in academic-writing.md Section 1 and flag each occurrence:

• delve, crucial, comprehensive, multifaceted, utilize, leverage, facilitate
• endeavor, paramount, myriad, plethora, noteworthy
• "it is important to note", "it should be noted", "in this regard"
• "the fact that", "in order to", "due to the fact that"
• "a number of", "the vast majority of", "at the present time"
• "in the context of", "with respect to", "in terms of", "as such"
• All AI-marker words: underscores, showcasing, pivotal, intricate, meticulous, illuminate, unveil, bolster, realm, landscape, foster, encompass, aligns with, shed light on, profound, grappling, etc.

2. Throat-Clearing Detection

Flag openings that don't state a concrete finding:

• "The literature has long..."
• "Financial economists have wondered..."
• "An important question..."
• "Recent years have witnessed..."
• "It is well known that..."
• "There has been growing interest in..."
• "A growing body of literature..."

3. Passive Voice Check

Flag passive constructions:

• "it is shown that" -> "we show that"
• "it was found that" -> "we find that"
• "can be seen" -> "shows"
• "has been documented" -> "[Author] documents"

4. Superlative Stacking

Flag sequences of emphasis words:

• "crucial new insights into this important phenomenon"
• "significant novel contribution to this vital area"
• "robust striking results that highlight key findings"

5. Vague Quantitative Claims

Flag imprecise language when numbers should be given:

• "substantial premium" -> give the number
• "large bias" -> give the magnitude
• "many factors" -> give the count
• "significant at conventional levels" -> give the t-stat
• "economically large" -> give the percentage

6. Terminology Violations (project-configurable)

Check for incorrect terminology per the project's CLAUDE.md. Each project defines its own preferred/banned terminology. If no project-specific terminology is defined, skip this category.

7. Self-Praise Detection

Flag adjectives that praise the work:

• "striking results"
• "important contribution"
• "novel approach"
• "comprehensive analysis"
• "rigorous methodology"
• "impressive performance"

8. Domain-Specific Consistency (project-configurable)

Check for consistency in how results are presented, per the project's conventions. For example, if the project requires decomposing costs into components, flag any place where costs are lumped together.

9. Citation Check

For each \cite{} or \citep{} in the text:

• Extract the BibTeX key
• Verify it exists in the project's .bib file
• Flag any missing keys
• For new citations: note they should be verified via Perplexity (see .claude/rules/latex-citations.md)

10. Structural AI Tells

Check for patterns flagged by Kobak et al. (2025) and Liang et al. (2024):

• Em-dashes (---) in prose: rewrite with commas, semicolons, colons, or parentheses
• En-dashes (--) as parenthetical asides: only valid for ranges and compound modifiers
• "Together, these results..." as paragraph opener (max 1 per paper)
• "This finding" as sentence opener (max 1 per paper)
• Naked "this" without a noun ("This implies..." -> "This result implies...")
• "Importantly," / "Notably," / "Specifically," as sentence-opening adverbs
• "In this section, we..." throat-clearing
• "Overall," as paragraph opener
• Soft-ban overuse: "highlights" (max 2/paper), "insights" (max 1/paper)
• Uniform paragraph length: flag if 5+ consecutive paragraphs have similar length
• Consecutive same-structure paragraph openers: 2+ consecutive paragraphs with same grammatical pattern
• Content-free meta-announcements: "We now turn to...", "We next examine..."
• Closing-summary paragraphs: "In summary," within non-conclusion sections
• Gerund-opener density: 2+ gerund-phrase openers in same paragraph
• Sentence length uniformity: 5+ consecutive sentences of similar word count
• "First...Second...Third" overuse: more than one enumeration per subsection
• Intensive reflexive overuse: "itself", "themselves" as emphasis (max 2/section)

11. Hedge Words & Previewing (Cochrane, Nikolov)

Flag filler hedge words:

• "somewhat", "rather" (as hedge), "quite", "very" (as intensifier), "arguably", "perhaps"
• Replace with magnitude: "somewhat larger" -> "12 basis points larger"

Flag previewing/recalling:

• "as we show below", "we will show", "Recall from Section X"

Flag nominalizations:

• "conduct an analysis" -> "analyze"
• "provide evidence that" -> "show that"

12. Finance-Specific Formulaic Patterns

Flag cliche constructions:

• "Our findings contribute to the growing literature on..."
• "speaks to the broader debate"
• "a battery of robustness checks"
• "the economic mechanism is as follows"
• "We exploit variation in..."
• "is not statistically significantly different from zero"

13. Editorial Artifact Detection

Scan for leftover editing comments and placeholders in active prose:

• [HUMAN EDIT REQUIRED -- draft placeholders
• (change to... / (should be... / (fix this) -- editing notes
• TODO / FIXME / XXX -- developer-style markers
• [TBD] / [PLACEHOLDER] / [INSERT -- slot markers
• [??] / [?] / [cite] / [ref] -- missing-reference markers

Severity: Always Critical. These are submission-blocking artifacts.

Output Format

STYLE CHECK REPORT
==================

File: [filename]
Lines analyzed: [count]

BANNED WORDS FOUND:
- Line X: "delve" -> suggest "examine" or "investigate"         [Critical]
- Line Y: "utilize" -> suggest "use"                            [Critical]

THROAT-CLEARING:
- Line W: "The literature has long..." -> Start with concrete claim  [Critical]

PASSIVE VOICE:
- Line V: "it is shown that" -> "we show that"                 [Suggestion]

[... all categories ...]

SUMMARY:
- Total issues: N
- Critical (must fix): M
- Suggestions: K

Severity Levels

• Critical: Banned words, missing citation keys, structural AI tells (em-dashes, formulaic adverb openers), editorial artifacts
• Suggestion: Passive voice, vague claims, self-praise, naked "this", hedge words, nominalizations

Usage Notes

1. Run on any .tex file, text block, or full document
2. Prioritize issues by severity (Critical first)
3. Provide specific line numbers where possible
4. Always suggest concrete replacements
5. Focus on actionable feedback