Paper Review Skill

Systematic LaTeX paper review for correctness, clarity, and consistency. Builds a high-level overview from introduction/technical overview sections, then reviews each section in parallel with shared context.

Works with OpenCode, Claude Code, and OpenAI Codex.

Install

Global installation (available in all projects):

curl -fsSL https://raw.githubusercontent.com/guruvamsi-policharla/paper-review-skill/main/install.sh | bash

Local installation (current project only):

curl -fsSL https://raw.githubusercontent.com/guruvamsi-policharla/paper-review-skill/main/install.sh | bash -s -- --local

Install for a specific tool only:

# OpenCode only
curl -fsSL https://raw.githubusercontent.com/guruvamsi-policharla/paper-review-skill/main/install.sh | bash -s -- --opencode

# Claude Code only
curl -fsSL https://raw.githubusercontent.com/guruvamsi-policharla/paper-review-skill/main/install.sh | bash -s -- --claude

# OpenAI Codex only
curl -fsSL https://raw.githubusercontent.com/guruvamsi-policharla/paper-review-skill/main/install.sh | bash -s -- --codex

Usage

Important: Run this skill from the directory containing your LaTeX source files (.tex files).

cd /path/to/your/paper

Then use the /paper-review command (OpenCode/Claude Code) or $paper-review (Codex):

# OpenCode / Claude Code
/paper-review review this paper

# OpenAI Codex
$paper-review review this paper

Review specific sections:

/paper-review review only the Methods and Results sections

Prioritize certain checks:

/paper-review focus on proof correctness

Updating

To update to the latest version, re-run the install command.

Tips for Best Results

Paper Structure

• Use clear section names: "Methods" works better than "Section 3"
• Define notation explicitly: Papers with a notation/preliminaries section get better consistency checking
• Use \section{} commands: The skill parses LaTeX structure to find sections

Multi-file Projects

The skill automatically resolves \input{} and \include{} commands, so splitting your paper across multiple .tex files works fine. Just run from the directory containing your main file.

What to Review

• Full review: Just say "review this paper" for a complete analysis
• Targeted review: Specify sections if you only changed part of the paper
• Focused review: Ask to prioritize specific aspects (proofs, clarity, notation)

Iterating on Feedback

After making changes based on the review:

/paper-review re-review the Methods section

Large Papers

For very long papers, consider reviewing a few sections at a time to get more detailed feedback.

Structure

The installer adds skills to the appropriate locations:

# OpenCode
.opencode/skills/paper-review/    # or ~/.config/opencode/skills/paper-review/
.opencode/command/paper-review.md # or ~/.config/opencode/command/paper-review.md

# Claude Code
.claude/skills/paper-review/      # or ~/.claude/skills/paper-review/

# OpenAI Codex
.agents/skills/paper-review/      # or ~/.agents/skills/paper-review/

Skill contents:

paper-review/
├── SKILL.md                  # Main orchestrator
└── prompts/
    ├── overview-builder.md   # Extracts paper context
    └── section-reviewer.md   # Reviews individual sections

What Gets Reviewed

Check Type	Description
Technical Correctness	Math errors, proof gaps, unsupported claims
Notation Consistency	Symbols matching definitions, undefined notation
Writing Clarity	Ambiguous statements, undefined terms
Structure/Flow	Logical organization, transitions, completeness

Issue Severity

Severity	Description
Critical	May indicate incorrectness (math errors, proof gaps)
Major	Significant quality impact (missing details, notation conflicts)
Minor	Should be fixed (typos, minor clarity issues)
Suggestions	Optional improvements (style, examples)

Output

Generates report.md in your paper directory with:

• Executive summary
• Paper overview (goals, contributions, notation dictionary)
• Detailed section reviews with categorized issues
• Summary statistics
• Prioritized action items

How It Works

1. Discovery - Parses LaTeX structure, resolves \input{} and \include{}
2. Overview Building - Extracts context from Introduction + Technical Overview
3. Parallel Review - Reviews each section with shared overview context
4. Report Generation - Aggregates findings into structured report

License

MIT