Paper Summarize

Paper Summarize Skill

This skill provides academic-grade paper summarization with dynamic Standard Operating Procedure (SOP) selection based on paper topic classification.

Capabilities

• Dynamic SOP Selection: Automatically selects appropriate analysis template based on paper type (method, dataset, multimodal, etc.)
• Rigorous Analysis: Follows top-tier conference review criteria (NeurIPS/ICML/ICLR/ACL)
• Structured Output: Generates comprehensive summaries with methodology critique, experimental assessment, strengths/weaknesses
• Local File Storage: Saves summaries to organized directory structure with proper naming
• Prompt Tracking: Maintains record of actual prompts used for reproducibility
• Dataset Focus: Explicit attention to training/evaluation datasets used in experiments

Supported Paper Types

• method: Algorithm/architecture papers
• dataset: Dataset/benchmark papers
• multimodal: Cross-modal learning papers
• tech_report: System/model release papers
• application: Applied AI papers
• survey: Survey/review papers
• rl_alignment: RL/Alignment/Safety papers
• speech_audio: Speech/audio processing papers
• benchmark: Evaluation/benchmark papers
• analysis: Empirical analysis papers

Usage

Input Requirements

• Paper title, authors, abstract
• Topic classification (one of supported types)
• Research context (keywords, subtopics)

Output Format

• Local file: {paper_title}.md in research/{domain}/ai_summaries/
• Content structure:
• Paper information (title, authors, venue, links)
• Core contribution summary
• Methodology critique (2000+ words)
• Experimental assessment (1000+ words, with dataset focus)
• Strengths and weaknesses
• Critical questions for authors
• Impact assessment

Quality Standards

• Methodology Critique: 2000+ characters, deep technical analysis including pipeline, novelty, mathematical principles, assumptions, prior art comparison, computational cost, and failure modes
• Experimental Assessment: 1000+ characters, rigorous evaluation with explicit focus on datasets used for training and testing, protocol rigor, baseline fairness, ablation completeness, and statistical significance
• Overall Analysis: 3000+ characters, critical perspective
• Technical Precision: Correct terminology, specific method names, exact metrics

Workflow Integration

This skill integrates with the broader research workflow:

1. Paper Discovery: Works with arXiv search results 2. Quality Filtering: Processes papers that pass relevance screening 3. Batch Processing: Can be called repeatedly for multiple papers 4. Report Generation: Outputs feed into final research report

Configuration

SOP templates are defined in:

• src/lib/agents/topic-sops.ts (primary location)
• summarization_prompt.ts (backup/reference)

Both files contain identical SOP definitions with shared output format requirements.

Examples

# Summarize a method paper
paper_summarize --title "SongEcho: Cover Song Generation" --topic "method" --abstract "..." --authors "..."

# Summarize a dataset paper  
paper_summarize --title "MusicSem: Language-Audio Dataset" --topic "dataset" --abstract "..." --authors "..."

Files Created

• research/{domain}/ai_summaries/{paper_title}.md
• research/{domain}/prompts/{paper_title}_prompt.txt
• Directory structure automatically created if missing

Paper Summarize Skill

Academic paper summarization with dynamic SOP selection based on paper topic classification.

Directory Structure

paper_summarize/
├── SKILL.md          # Skill definition and usage
├── README.md         # This file
├── scripts/          # Executable scripts (if any)
└── templates/        # SOP templates and prompt files

Integration

This skill is designed to work within the broader music generation research workflow:

1. Input: Papers that have passed quality filtering 2. Processing: Applies appropriate SOP template based on topic classification 3. Output: Saves detailed summaries to organized local directory structure 4. Tracking: Maintains prompt records for reproducibility

Usage in Research Workflow

When processing papers from arXiv search results:

# Example workflow integration
for paper in filtered_papers:
    topic = classify_paper(paper)  # method, dataset, multimodal, etc.
    summary = paper_summarize(
        title=paper.title,
        authors=paper.authors, 
        abstract=paper.abstract,
        topic=topic,
        domain="music_generation"
    )
    save_to_file(summary, f"research/music_generation/ai_summaries/{paper.title}.md")

Quality Assurance

• All summaries follow rigorous academic standards
• Methodology critique receives extra emphasis (1500+ characters minimum)
• Critical perspective maintained throughout (not just paper restatement)
• Technical precision with correct terminology and specific references

Paper Summarize Skill - Usage Example

Step 1: Classify Paper Topic

// Example paper classification
const papers = [
  {
    title: "SongEcho: Cover Song Generation via Instance-Adaptive Element-wise Linear Modulation",
    topic: "method", // method, dataset, multimodal, etc.
    authors: "Sifei Li, Yang Li, Zizhou Wang, et al.",
    abstract: "Cover songs constitute a vital aspect of musical culture..."
  },
  {
    title: "MusicSem: A Semantically Rich Language--Audio Dataset",
    topic: "dataset",
    authors: "Rebecca Salganik, Teng Tu, Fei-Yueh Chen, et al.", 
    abstract: "We introduce MusicSem, a semantically rich language-audio dataset..."
  },
  {
    title: "Art2Mus: Artwork-to-Music Generation via Visual Conditioning",
    topic: "multimodal",
    authors: "Ivan Rinaldi, Matteo Mendula, Nicola Fanelli, et al.",
    abstract: "Art2Mus enables artwork-to-music generation through visual conditioning..."
  }
];

Step 2: Apply Appropriate SOP Template

// For each paper, select SOP based on topic
function getPaperSummary(paper) {
  const { title, authors, abstract, topic } = paper;
  
  // Load appropriate SOP template
  const sopTemplate = getSOPForTopic(topic);
  
  // Fill template variables
  const userPrompt = sopTemplate
    .replace('{{title}}', title)
    .replace('{{authors}}', authors)
    .replace('{{abstract}}', abstract)
    .replace('{{topic}}', topic);
    
  // Combine with system prompt
  const fullPrompt = SYSTEM_PROMPT + '\n\n' + userPrompt;
  
  return fullPrompt;
}

Step 3: Generate and Save Summary

// Generate summary using AI model
const summary = generateSummary(fullPrompt);

// Save to organized directory structure
const safeTitle = title.replace(/[^a-zA-Z0-9]/g, '_');
const filePath = `research/music_generation/ai_summaries/${safeTitle}.md`;
writeFile(filePath, summary);

// Save prompt for reproducibility
const promptPath = `research/music_generation/prompts/${safeTitle}_prompt.txt`;
writeFile(promptPath, fullPrompt);

Expected Output Structure

File: SongEcho_Towards_Cover_Song_Generation_via_Instance-Adaptive_Element-wise_Linear_Modulation.md

# SongEcho: Towards Cover Song Generation via Instance-Adaptive Element-wise Linear Modulation

## 论文信息
- **标题**: SongEcho: Towards Cover Song Generation via Instance-Adaptive Element-wise Linear Modulation
- **作者**: Sifei Li, Yang Li, Zizhou Wang, Yuxin Zhang, Fuzhang Wu, Oliver Deussen, Tong-Yee Lee, Weiming Dong
- **机构**: 中国科学院自动化研究所、中国科学院大学人工智能学院、中国科学院软件研究所、康斯坦茨大学、国立成功大学
- **发表**: ICLR 2026
- **arXiv**: [2602.19976](https://arxiv.org/abs/2602.19976)
- **代码**: [GitHub](https://github.com/lsfhuihuiff/SongEcho_ICLR2026)

## 核心贡献
[1500+字的方法论深度剖析]
[1000+字的实验评估]
[完整的分析正文，总计3000+字]

Quality Assurance Checklist

• [ ] Methodology critique ≥ 1500 characters
• [ ] Experimental assessment ≥ 1000 characters
• [ ] Total analysis ≥ 3000 characters
• [ ] Critical perspective (not just restatement)
• [ ] Technical precision with correct terminology
• [ ] Proper file naming and directory structure
• [ ] Prompt saved for reproducibility