This repository provides a tool for exploring the client-side codebase of the GitHub Copilot VSCode extension, specifically version 1.57.7193. It aims to demystify how Copilot processes code and generates suggestions by visualizing module dependencies, displaying code, and offering insights into module functionality, targeting developers interested in understanding AI-assisted coding tools.

How It Works

The tool analyzes the Copilot extension's JavaScript bundle, reverse-engineering module structures and dependencies. It employs automated techniques, including Codex-based naming and categorization of modules, alongside manual annotations for clarity. The output is presented in a web interface with three panels: a dependency graph, prettified code, and module metadata, facilitating navigation and comprehension of the extension's internal workings.

Quick Start & Requirements

• Install dependencies: npm install
• Run the analysis: node index.js (optional, if modifying extension code)
• Install manifest-ml: pip install manifest-ml
• Create codeviz/.openai-api-key with your OpenAI API key (optional).
• Start the web app: python3 codeviz/app.py
• Access at localhost:5000.
• Requires Node.js, Python 3, and an OpenAI API key for certain features.

Highlighted Details

• Detailed breakdown of "Prompting" modules, including prompt-extractor.js and get-prompt-actual.js.
• Analysis of "Completion" workflows: Inline-Completions (e.g., 3197) and Panel Completion (e.g., 2388).
• Insights into "Telemetry" data collection, including suggestion effectiveness measurement.
• Identification of network modules like OpenAIFetcher (4419).

Maintenance & Community

This is a personal project with no explicit mention of maintainers, community channels, or a roadmap. Contributions via issues or PRs are welcomed.

Licensing & Compatibility

The repository does not specify a license.

Limitations & Caveats

The project is described as "hacky" and "work in progress," with potential for inaccuracies in module naming, categorization, and analysis. Some features, like snippet description, may not function optimally. Annotations are stored in local storage, requiring manual export.