| name | paper-depth-reading |
| description | Comprehensive academic paper analysis skill for computer graphics and AI research papers. Use when Claude needs to perform deep, structured analysis of academic papers with code implementation. |
Paper Depth Reading Skill
Overview
This skill provides a structured framework for deep analysis of academic papers, particularly in computer graphics, computer vision, and AI research. It guides Claude through a comprehensive reading process that connects theoretical concepts with practical code implementations.
Core Workflow
1. Initial Paper Assessment (10 minutes)
- Read abstract, introduction, and conclusion to understand core contributions
- Identify key sections: methodology, experiments, results
- Skim figures and tables to grasp visual results
- Note paper structure and main contributions
2. Code Repository Exploration
Before deep reading, always explore the code repository:
- Examine directory structure to understand project organization
- Identify key modules: models, training scripts, datasets, utils
- Read configuration files to understand hyperparameters
- Check README for setup instructions and dependencies
3. Detailed Methodology Analysis
- Network architecture: Layer-by-layer analysis with shape transformations
- Loss functions: Mathematical formulation and implementation
- Training pipeline: Data flow and optimization details
- Dataset processing: Preprocessing and augmentation strategies
4. Code-Paper Connection
- Cross-reference: Connect paper descriptions to actual code implementations
- Clarify ambiguities: Use code to resolve unclear paper descriptions
- Verify details: Check implementation specifics against paper claims
- Identify extensions: Note code features not mentioned in paper
5. Critical Analysis
- Strengths: What makes this approach effective?
- Limitations: What are the constraints or weaknesses?
- Comparisons: How does it compare to related work?
- Future directions: What improvements are possible?
Analysis Framework
Input Analysis
- Data type: Images, point clouds, text, video, etc.
- Shape specification: Exact dimensions (e.g.,
[B, C, H, W], [B, N, D])
- Preprocessing: Normalization, augmentation, encoding
- Special considerations: Multi-modal inputs, temporal sequences
Output Analysis
- Target format: Generated images, 3D models, predictions, etc.
- Shape specification: Output dimensions and format
- Post-processing: Decoding, rendering, evaluation metrics
Methodology Breakdown
For each component:
- Component name and purpose
- Mathematical formulation (LaTeX format)
- Implementation details from code
- Shape transformations through the pipeline
- Hyperparameters and configuration
Experimental Results
- Quantitative metrics: Tables with exact numbers from paper
- Qualitative results: Visual analysis of figures
- Ablation studies: Component importance analysis
- Comparison baselines: Performance relative to other methods
Professional Terminology Handling
- First occurrence: Provide brief explanation for domain-specific terms
- LaTeX formatting: Use
$$ for mathematical equations
- Variable definitions: Clearly define all symbols and variables
- Code references: Link concepts to specific code files and lines
Output Format
Generate comprehensive markdown report with:
- Structured sections following the analysis framework
- Code snippets with file paths and line numbers
- Mathematical equations in LaTeX format
- Tables for quantitative comparisons
- Critical insights and personal analysis
When to Use References
- references/methodology_patterns.md: For common architectural patterns
- references/experiment_analysis.md: For interpreting results
- references/code_reading.md: For systematic code exploration
Resources
This skill includes example resource directories that demonstrate how to organize different types of bundled resources:
scripts/
Executable code (Python/Bash/etc.) that can be run directly to perform specific operations.
Examples from other skills:
- PDF skill:
fill_fillable_fields.py, extract_form_field_info.py - utilities for PDF manipulation
- DOCX skill:
document.py, utilities.py - Python modules for document processing
Appropriate for: Python scripts, shell scripts, or any executable code that performs automation, data processing, or specific operations.
Note: Scripts may be executed without loading into context, but can still be read by Claude for patching or environment adjustments.
references/
Documentation and reference material intended to be loaded into context to inform Claude's process and thinking.
Examples from other skills:
- Product management:
communication.md, context_building.md - detailed workflow guides
- BigQuery: API reference documentation and query examples
- Finance: Schema documentation, company policies
Appropriate for: In-depth documentation, API references, database schemas, comprehensive guides, or any detailed information that Claude should reference while working.
assets/
Files not intended to be loaded into context, but rather used within the output Claude produces.
Examples from other skills:
- Brand styling: PowerPoint template files (.pptx), logo files
- Frontend builder: HTML/React boilerplate project directories
- Typography: Font files (.ttf, .woff2)
Appropriate for: Templates, boilerplate code, document templates, images, icons, fonts, or any files meant to be copied or used in the final output.
Any unneeded directories can be deleted. Not every skill requires all three types of resources.
使用示例
示例1:计算机图形学论文分析
论文: "NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis"
分析框架:
- 研究背景: 神经渲染、3D场景表示、视图合成
- 核心方法:
- 实验设计:
- 数据集: Synthetic NeRF, Realistic Synthetic 360°
- 评估指标: PSNR, SSIM, LPIPS
- 对比方法: SRN, NV, LLFF
- 代码连接:
示例2:AI研究论文分析
论文: "Attention Is All You Need" (Transformer架构)
分析框架:
- 研究背景: 序列建模、机器翻译、注意力机制
- 核心方法:
- 实验设计:
- 任务: WMT 2014英德/英法翻译
- 评估: BLEU分数
- 消融实验: 注意力头数、模型深度
- 代码连接:
- 官方实现: Tensor2Tensor库
- PyTorch实现: nn.Transformer模块
- 关键类: MultiHeadAttention, TransformerEncoder
示例3:计算机视觉论文分析
论文: "Mask R-CNN"
分析框架:
- 研究背景: 目标检测、实例分割、Faster R-CNN
- 核心方法:
- 实验设计:
- 数据集: COCO, Cityscapes
- 评估指标: AP, AP50, AP75
- 消融实验: ROIAlign vs ROIPool
- 代码连接:
- Detectron2框架
- mmdetection实现
- 关键组件: MaskRCNN类, ROIAlign层
示例4:分析工作流程
def analyze_paper(paper_pdf):
metadata = extract_metadata(paper_pdf)
methodology = analyze_methodology(paper_pdf)
experiments = extract_experiments(paper_pdf)
code_connections = find_code_connections(metadata['title'])
report = generate_report(
metadata=metadata,
methodology=methodology,
experiments=experiments,
code_connections=code_connections
)
return report
示例5:与其他技能协同
python github-to-skills/scripts/fetch_github_info.py https://github.com/author/paper-implementation.git
python update-readme/scripts/analyze_project.py research-folder/
python skill-evolution-manager/scripts/merge_evolution.py paper-depth-reading '{
"preferences": ["优先分析创新点", "关注实验可复现性"],
"fixes": ["修复公式解析问题"],
"custom_prompts": "为每篇论文生成技术迁移建议"
}'
References
scripts/: executable helpers for deterministic steps
references/: additional docs loaded on demand
assets/: templates or static files used by the skill
Trigger Conditions
- User explicitly names this skill
- User intent clearly matches this skill description
- The task needs this skill's scripts/resources
Applicable Scope
- Requests covered by this skill's frontmatter
description
- Tasks that benefit from the bundled workflow and scripts
Out of Scope
- Requests that conflict with repository safety rules
- Tasks unrelated to this skill's declared purpose
- Destructive changes without explicit user permission