Analyze Code Skill

Extract implementation details from code repositories to write accurate methodology and experiment sections. Acts as a Code Analyst that understands code structure and translates it to paper descriptions.

Overview

This skill helps you:

• Understand code structure and key components
• Extract algorithms and architectures
• Generate pseudo-code from implementations
• Map code to methodology descriptions
• Extract hyperparameters and configurations

Commands

/analyze-code [--section <section>]

Extract implementation details from code.

Parameters:

• --section <section> (optional): Target paper section (e.g., methodology, experiments)

Section Options:

• methodology: Extracts algorithms, architectures, training procedures
• experiments: Extracts evaluation setup, metrics, configurations

Analysis Capabilities

Structure Analysis

• File organization and key components
• Dependencies and imports
• Main entry points and workflows

Implementation Extraction

• Algorithms: Step-by-step algorithm descriptions
• Architectures: Model structures and components
• Data Pipelines: Data loading and preprocessing
• Training Procedures: Training loops and optimization

Pseudo-code Generation

• Converts code to LaTeX algorithm blocks
• Extracts key steps and logic
• Preserves algorithm flow

Hyperparameters and Configuration

• Extracts hyperparameter values
• Identifies configuration files
• Documents training settings

Code Elements Mapped to Paper Descriptions

Code Element	Paper Description
Class definitions	Component descriptions
Forward pass	Algorithm steps
Hyperparameters	Paper configs
Loss functions	Training objectives
Data loaders	Dataset processing
Evaluation code	Experimental setup

Examples

# Analyze for methodology section
/analyze-code --section methodology

# Analyze for experiments section
/analyze-code --section experiments

# General code analysis
/analyze-code

Workflow

1. Add Code: Link repository or directory (can be done via codebase context)
2. Analyze: Run analyze-code to extract details
3. Integrate: Use output to write Methodology section with write-section

Output Format

Returns:

• Natural language description of implementation
• LaTeX algorithm environment code (for methodology)
• Key hyperparameters and configuration
• Training/evaluation procedure description (for experiments)

Integration with Writing

After analysis, use the output with write-section:

# 1. Analyze code
/analyze-code --section methodology

# 2. Write methodology using code analysis
/write-section methodology --from code/ --from notes/implementation/

Supported Code Types

• Python (most common)
• Configuration files (YAML, JSON)
• Training scripts
• Model definitions
• Evaluation code

Platform Compatibility

Works on all platforms (Claude Code, Codex, OpenCode, Gemini, Cursor). Command syntax may vary by platform - see platform-specific documentation in docs/.