Proofreading

A Claude Code skill for proofreading academic papers against a structured checklist covering paper structure, math notation, statistical relevance, figures and tables, grammar and style, and abbreviations.

Experience

I went through a hard school with my professor highlighting every single detail that differentiates an average paper from an outstanding scientific work. Over the years of my Ph.D., I received feedback on my paper drafts 44 times! I went through all of this feedback and compiled a comprehensive list of checks that regularly come up during proof reading. That being said, especially Check 5 (Grammar and Style) is highly subjective and should be adapted to your personal needs.

If you find this skill useful, give it a star and share it with your colleagues. Do not hesitate to open up a pull request to extend this skill further!

Features

• Report mode: full prioritized issue report with line-level findings
• Interactive mode: fix issues step by step together with the agent
• PDF annotation extraction: when given an annotated PDF, extracts highlights, notes, and strikethrough comments from reviewers

Checks

This skill contains over 100 checks, which we categorize into 6 groups. The following list gives a brief summary of the most important checks in each group.

Check 1 — Paper Structure (intro)

• Abstract follows the four-part structure: motivation, gap, approach, quantitative result
• Introduction contains an explicit, falsifiable contribution list
• Related work is organized thematically and ends with a gap statement

Check 2 — Math Symbols and Notation (math-notation)

• Every symbol and index is defined before first use
• Scalar/vector/matrix/set convention is detected and applied consistently
• Equations are punctuated and integrated into the surrounding text

Check 3 — Statistical Relevance (stats)

• Result figures show uncertainty (error bars, confidence bands)
• Result tables report values with standard deviation or confidence intervals
• Comparative prose claims are backed by statistical evidence or seed counts

Check 4 — Figures and Tables (figures)

• Every figure and table is referenced in the text and has a self-contained caption
• Figures use vector formats; tables use the booktabs package
• Result figures are assessed for axis labels, legend, line thickness, and color accessibility

Check 5 — Grammar and Style (grammar)

• First-person “we” is used consistently; passive is reserved for system properties
• Prohibited vague intensifiers and inflated vocabulary are flagged
• Compound adjectives are hyphenated; Oxford comma and i.e.,/e.g., rules are applied

Check 6 — Abbreviations (abbrev)

• The abstract introduces its own abbreviations independently of the main body
• Every abbreviation is introduced on first use in the body with the full term
• Article agreement (a/an) and plural formation (NNs not NN's) are checked

Requirements

Install Python dependencies with:

pip install -r requirements.txt

pymupdf is required for PDF annotation extraction (scripts/extract-pdf-annotations.py). It is not needed if you only work with LaTeX source files.

Usage

/proofread [path] [--check ] [--interactive]

• path — path to the root .tex file or directory of .tex files (default: current directory). A .pdf file can also be provided to extract annotations.
• --check — run only one check: intro, math-notation, stats, figures, grammar, abbrev
• --interactive — fix issues one by one instead of producing a written report

Contribute

Propose additions or changes by opening a pull request:

1. Edit SKILL.md — each check is a ### Check N section; sub-checks follow the #### N.x pattern.
2. Tag every rule with a severity ([ERROR] / [WARN] / [INFO]) and a difficulty label ([HARD] for tool-based checks, e.g. grub, [MIX] for a mix between tool-based and reasoning checks, [VIBE] for semantic reasoning only).
3. Keep the change focused: one logical issue per sub-check, one sub-check per rule group.
4. If you add a new check group, update the --check argument list in the Arguments section of SKILL.md and in this README.