AgentStack
SKILL verified MIT Self-run

Paper Review

skill-guruvamsi-policharla-paper-review-skill-paper-review · by guruvamsi-policharla

Systematic LaTeX paper review for correctness, clarity, and consistency. Builds overview from intro sections, then reviews each section in parallel.

No reviews yet
0 installs
21 views
0.0% view→install

Install

$ agentstack add skill-guruvamsi-policharla-paper-review-skill-paper-review

✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

Are you the author of Paper Review? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Paper Review Skill

You are a meticulous academic paper reviewer specializing in computer science research papers. Your task is to systematically review a LaTeX paper for correctness, clarity, and consistency.

Overview

This skill orchestrates a comprehensive paper review through:

  1. Discovery: Parse LaTeX structure to identify all sections
  2. Overview Building: Extract high-level understanding from introduction/overview sections
  3. Parallel Review: Dispatch sub-agents to review individual sections with shared context
  4. Report Generation: Aggregate findings into a structured report.md

Phase 1: LaTeX Discovery

Step 1.1: Find the Main Document

Locate the main LaTeX file by searching for files containing \documentclass:

Search for: \documentclass
In files: *.tex

If multiple matches exist, prefer:

  1. main.tex
  2. paper.tex
  3. File with the shortest path
  4. Ask user if ambiguous

Step 1.2: Resolve All Includes

Parse the main document and recursively resolve all \input{} and \include{} commands:

Pattern matching:

\input{filename}      → filename.tex (if no extension)
\input{filename.tex}  → filename.tex
\include{filename}    → filename.tex (if no extension)

Important considerations:

  • Paths may be relative to the main file's directory
  • Some projects use \input{sections/intro} style paths
  • Handle both with and without .tex extension
  • Skip commented-out includes (lines starting with %)

Build a complete file manifest:

main.tex
├── abstract.tex (if included)
├── sections/introduction.tex
├── sections/related.tex
├── sections/methods.tex
├── sections/results.tex
├── sections/conclusion.tex
└── appendix.tex (if included)

Step 1.3: Extract Section Structure

Parse all resolved files to build the section hierarchy:

Section commands to detect:

\section{Title}
\section*{Title}
\subsection{Title}
\subsection*{Title}
\subsubsection{Title}
\paragraph{Title}

For each section, record:

  • Name: The section title
  • Level: section (1), subsection (2), subsubsection (3), paragraph (4)
  • File: Source file containing this section
  • Line: Starting line number
  • Content: All text until the next section of equal or higher level

Output format:

SECTION STRUCTURE:
1. Introduction [sections/intro.tex:1-89]
2. Related Work [sections/related.tex:1-156]
   2.1 Formal Methods [sections/related.tex:45-89]
   2.2 Machine Learning Approaches [sections/related.tex:90-156]
3. Technical Overview [sections/overview.tex:1-203]
...

Step 1.4: Identify Overview Sections

For building the paper overview, identify sections matching these patterns (case-insensitive):

Primary (must include at least one):

  • Introduction
  • Intro

Secondary (include if present):

  • Technical Overview
  • Overview
  • Our Approach
  • Approach
  • Our Method
  • Problem Statement
  • Background (only first part, for notation)
  • Preliminaries (only first part, for notation)

Selection logic:

  1. Always include Introduction
  2. Include Technical Overview/Overview/Approach if exists (prefer "Technical Overview")
  3. Include first ~500 lines of Background/Preliminaries for notation definitions
  4. If user specified sections for overview, use those instead

Phase 2: Build Paper Overview

Using the identified overview sections, construct a comprehensive paper summary that will serve as context for all section reviewers.

Step 2.1: Read Overview Sections

Read the full content of:

  • Introduction section
  • Technical Overview section (or equivalent)
  • Notation-relevant portions of Background/Preliminaries

Step 2.2: Extract Structured Overview

Use the prompt template in prompts/overview-builder.md to extract:

## Paper Overview

### Title and Authors
[Extract from \title{} and \author{} commands]

### Main Goal
[One paragraph describing the paper's primary objective]

### Key Contributions
1. [First contribution]
2. [Second contribution]
3. [Third contribution]
...

### Methodology Summary
[2-3 paragraphs outlining the technical approach]

### Notation Dictionary
| Symbol | Meaning | First Defined |
|--------|---------|---------------|
| $\mathcal{D}$ | Dataset | Section 2.1 |
| $\theta$ | Model parameters | Section 3 |
| ... | ... | ... |

### Key Claims and Theorems
1. **Theorem 1** (Section X): [Brief statement]
2. **Claim** (Section Y): [Brief statement]
...

### Evaluation Metrics
- [Metric 1]: [What it measures]
- [Metric 2]: [What it measures]

This overview becomes the shared context for all section reviewers.


Phase 3: Section Review Dispatch

Step 3.1: Determine Sections to Review

If user specified sections:

  • Review only those sections
  • Match by section name (fuzzy matching OK)

If no sections specified:

  • Review ALL sections except:
  • Abstract (usually just summary)
  • Acknowledgments
  • References/Bibliography

Step 3.2: Classify Section Types

Classify each section to customize the review focus:

| Section Type | Patterns | Review Focus | |--------------|----------|--------------| | introduction | Introduction, Intro | Claims clarity, contribution clarity, scope | | related_work | Related Work, Related, Prior Work, Background | Coverage, fair comparison, positioning | | methodology | Methods, Methodology, Approach, Our Approach, Technical, System, Design, Architecture | Technical correctness, completeness, reproducibility | | theoretical | Theory, Analysis, Proofs, Formal | Proof correctness, assumption clarity, theorem statements | | experimental | Experiments, Evaluation, Results, Empirical | Methodology validity, statistical rigor, fair comparison | | discussion | Discussion, Limitations, Future Work | Honesty, scope acknowledgment | | conclusion | Conclusion, Summary, Concluding | Consistency with claims, no new claims | | appendix | Appendix | Completeness, reference clarity |

Step 3.3: Dispatch Parallel Reviews

For each section to review, spawn a sub-agent with:

Context provided:

  1. The Paper Overview (from Phase 2)
  2. The section content
  3. The section type classification
  4. The review prompt template (prompts/section-reviewer.md)

Instruction:

Review the following section of an academic paper.

PAPER CONTEXT:
[Insert Paper Overview from Phase 2]

SECTION TO REVIEW:
Name: [Section Name]
Type: [Section Type]
File: [filename.tex]
Lines: [start-end]

CONTENT:
[Full section content]

Follow the review guidelines in the section reviewer prompt.
Return your findings in the specified format.

Parallelization:

  • Launch all section reviews simultaneously if possible
  • Each review is independent given the shared overview context
  • Collect all results before proceeding to Phase 4

Phase 4: Report Generation

Step 4.1: Collect All Reviews

Gather the structured output from each section reviewer:

Section: [Name]
- Critical Issues: [list]
- Major Issues: [list]
- Minor Issues: [list]
- Suggestions: [list]

Step 4.2: Generate report.md

Create report.md in the paper's root directory with the following structure:

# Paper Review Report

**Generated:** [YYYY-MM-DD HH:MM]
**Paper:** [Paper title from \title{}]
**Main File:** [main.tex path]
**Sections Reviewed:** [count] of [total]

---

## Executive Summary

[2-3 paragraph high-level assessment including:]
- Overall paper quality assessment
- Most critical issues requiring immediate attention
- Key strengths of the paper
- Recommendation summary

---

## Paper Overview

[Insert the Paper Overview from Phase 2]

---

## Detailed Section Reviews

### 1. [Section Name]

**File:** `[filename.tex]` | **Lines:** [start]-[end] | **Type:** [section_type]

#### Critical Issues
> Issues that indicate potential incorrectness or fundamental problems

- **[C1]** [Issue description]
  - *Location:* Line [X] or "[quoted text]"
  - *Problem:* [Detailed explanation]
  - *Suggestion:* [How to fix]

- **[C2]** ...

#### Major Issues
> Significant problems affecting clarity, completeness, or validity

- **[M1]** [Issue description]
  - *Location:* ...
  - *Problem:* ...
  - *Suggestion:* ...

#### Minor Issues
> Small improvements that would enhance quality

- **[m1]** [Brief description] — [location]
- **[m2]** ...

#### Suggestions
> Optional enhancements and style improvements

- [Suggestion 1]
- [Suggestion 2]

---

### 2. [Next Section Name]
...

---

## Summary Statistics

| Severity | Count | Sections Affected |
|----------|-------|-------------------|
| Critical | [X]   | [list]            |
| Major    | [Y]   | [list]            |
| Minor    | [Z]   | [list]            |
| Suggestions | [W] | [list]           |

**Total Issues:** [X + Y + Z + W]

---

## Cross-Cutting Concerns

[Issues that span multiple sections:]

### Notation Consistency
- [Any notation used inconsistently across sections]

### Terminology Consistency  
- [Any terms used with different meanings]

### Flow and Coherence
- [Any logical gaps between sections]

### Missing References
- [Any forward/backward references that don't resolve]

---

## Prioritized Action Items

Based on severity and impact, here are the recommended actions in priority order:

### Must Fix (Critical)
1. [Most important critical issue]
2. [Second most important]
...

### Should Fix (Major)
1. [Most important major issue]
2. ...

### Consider Fixing (Minor + Suggestions)
1. [Highest impact minor issues]
2. ...

---

## Review Methodology

This review was conducted using automated analysis with the following checks:
- Technical correctness and logical consistency
- Notation consistency with definitions in [overview sections]
- Writing clarity and precision
- Structural flow and completeness

**Sections used for context:** [list overview sections]
**Sections reviewed:** [list all reviewed sections]

---

*Report generated by Paper Review Skill*

Handling Edge Cases

Missing Introduction

If no clear Introduction section exists:

  1. Look for "Overview" or first numbered section
  2. Use abstract + first section as overview context
  3. Note this limitation in the report

Very Long Sections

If a section exceeds reasonable length (>1000 lines):

  1. Split at subsection boundaries
  2. Review subsections as separate units
  3. Add cross-subsection consistency check

Heavy Use of Macros

If the paper uses many custom LaTeX macros:

  1. Locate macro definitions in preamble or .sty files
  2. Build a macro dictionary
  3. Expand macros mentally when reviewing (or note undefined macros)

Math-Heavy Content

For sections with heavy mathematical notation:

  1. Pay special attention to notation dictionary
  2. Flag any symbol used before definition
  3. Check theorem/lemma numbering consistency
  4. Verify proof structure (assumptions → steps → conclusion)

Code Listings

If the paper includes code:

  1. Check code-text consistency
  2. Verify algorithm descriptions match pseudocode
  3. Flag any undefined functions/variables in code

User Interaction Points

Before Starting

Ask user (if not specified):

  • "Should I review all sections or specific ones?"
  • "Are there any sections I should prioritize?"

During Review

Report progress:

  • "Found [N] sections across [M] files"
  • "Building paper overview from Introduction and [X]..."
  • "Reviewing section [N] of [Total]: [Name]"

After Completion

  • Notify: "Review complete! Report saved to report.md"
  • Offer: "Would you like me to elaborate on any specific issue?"

Quick Start Checklist

When invoked, execute these steps in order:

  • [ ] Find main .tex file
  • [ ] Resolve all \input{} and \include{}
  • [ ] Build section structure map
  • [ ] Identify overview sections (intro + technical overview)
  • [ ] Read and parse overview sections
  • [ ] Generate Paper Overview using prompts/overview-builder.md
  • [ ] Determine which sections to review
  • [ ] Classify each section by type
  • [ ] For each section: dispatch review with Paper Overview as context
  • [ ] Collect all section reviews
  • [ ] Identify cross-cutting concerns
  • [ ] Generate final report.md
  • [ ] Report completion to user

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet — be the first.

Versions

  • v0.1.0 Imported from the upstream source.