Paper Review Helper

Output structure: <paper-folder>/<pdf_id>/<pdf_id>.tex with images/ subfolder for figures.

<paper-folder>/
├── <pdf_id>/                 # Mathpix output (if converted)
│   ├── <pdf_id>.tex          # Converted LaTeX
│   └── images/               # Extracted figures
├── artifact/
│   ├── review-log.md         # Conversation log
│   ├── issues-major.md       # Major issues
│   ├── issues-minor.md       # Minor issues
│   └── programs/             # Math verification scripts
└── original.pdf              # Source PDF (if applicable)

Phase 2: Section-by-Section Review

Parse LaTeX structure: \section, \subsection, \begin{abstract}. For each section:

1. Chunk appropriately:

   • Section ≤5 paragraphs: review whole
   • Section >5 paragraphs: split by \subsection or paragraph groups

2. Present section with:

   1. Section text - save the partial tex to markdown format.
   2. Inline markers for issues (see Grammar Check Markers) in the markdown file.
   3. and provide the user with a path link to the file in the workspace.
   4. Figure refs: [Figure X: images/<filename> - <caption>] with a path link to the file in the workspace.

3. Ask user via AskUserQuestion:

   • "Questions about this section?"
   • "Any concerns or unclear points?"
   • "Rate clarity: clear / somewhat unclear / confusing"

4. Respond to requests:

   • Citations: WebSearch for DOI, author names, paper titles
   • Background: WebSearch, WebFetch for concepts
   • Math check: write script to artifact/programs/, run with Python/SymPy
   • Figures: read from images/ folder, describe or use /vision skill

5. Log to artifact/review-log.md:

## [Section Name] - [Timestamp]
### User Questions
- Q: ...
- A: ...
### Issues Identified
- [MAJOR] ...
- [MINOR] ...
### Tools Used
- WebSearch: "query" → finding

6. Classify issues:
   • Major → issues-major.md: methodology flaws, unsupported claims, logical errors
   • Minor → issues-minor.md: grammar, typos, unclear wording

Skip Supplementary/Appendix unless user requests.

Phase 3: Review Generation

1. Gather context via AskUserQuestion:

   • "Overall impression? (accept / minor revision / major revision / reject)"
   • "Editor's specific questions?"
   • "Journal level? (top-tier / mid-tier / specialized)"

2. Generate review to artifact/REVIEW.md:

# Review of [Paper Title]

## Summary
[1-2 sentences]

## Overall Recommendation
[User's decision + justification]

## Major Issues
1. **Issue**: [description]
   - **Location**: Section X, paragraph Y / Equation N
   - **Impact**: [why this matters]
   - **Suggestion**: [fix or note if unfixable]

## Minor Issues
[Grouped by type: grammar, clarity, formatting]

## User Misunderstandings Analysis
[If user had confusion during review:]
- **Confusion**: [what]
- **Cause**: paper vagueness / reader knowledge gap
- **Recommendation**: [should paper clarify?]

## Constructive Feedback
[Positives + specific improvements]

## Editor Questions Response
[If provided]

Grammar Check Markers

Inline markers when presenting text:

• [G: ...] - grammar error
• [C: ...] - clarity issue
• [?] - ambiguous/unsupported claim
• [REF?] - missing or questionable citation
• [EQ?] - equation to verify