Skip to content

mehulk05/ai-code-attribution-tracker

Repository files navigation

🔬 AI Code Attribution Tracker

Node version Git support Docker support Render deploy

⚡ Forensic Git Blame & AI Code Fingerprint Analyzer

Estimates what percentage of your codebase is AI-generated, split by AI Assistant (Claude vs. Antigravity/Gemini) and by Individual Developer. Instead of relying on raw commit sizes, it uses line-ownership matching and structural signature analysis.

Launch Live Console


📸 Interactive Dashboard Preview

Below is the live console in action showing scan statistics, interactive developer contribution distributions, and the AI agent signature engine details.

Dashboard Preview


✨ Key Features

  • 🚀 Dual Scan Modes:
    • Bitbucket/GitHub Remote URL: Clones, blames, and scans branches dynamically in the cloud.
    • Local Folder Path: Checks codebases on your drive instantly (perfect for local developer environments).
  • 📊 Real-time Logging & ETA: Displays blaming filenames, elapsed times, and remaining scanning durations dynamically.
  • 🤖 Agent Style Fingerprints: Classifies code blocks as Gemini/Antigravity or Claude by analyzing LLM writing tells rather than file edits.
  • 🎨 Widescreen Glassmorphic Interface: Clean dark-mode dashboard styled with smooth micro-animations, color gradients, and native complete notifications.

🔬 Forensics: How It Identifies LLMs

To eliminate false-positives (like Prettier runs, package-lock files, or new-file creations), the scanner runs a Heuristic Signature Engine matching developer line-ownership back to specific assistant footprints:

Heuristic Indicator 🤖 Claude (Anthropic) 🌌 Antigravity / Gemini (Google)
Comment Dividers Numbered list steps (// 1., // 2.) Horizontal thick bars (// ============)
Docstrings Standard JSDoc block headers above functions Python-style docstrings embedded inside block brackets
Header Styling Clean PascalCase headings ALL-CAPS section banner headers
Code Velocity File additions >120 LOC with <5% deletions File additions >150 LOC with <5% deletions

Note

The scanner combines signatures with Human Tells (common developer spelling mistakes like dyanamic, recieve, tolltip), using them to dynamically lower AI confidence score thresholds and prevent false-positives.


📊 Developer Attribution & Preferences

The scanner compiles interactive tables attributing code ownership directly to individual developers, displaying their preferred helper badge in real-time:

Developer Attribution


🛠️ Getting Started

Local Setup

  1. Clone the repository and install Node.js (v18+).
  2. Start the local server:
    npm start
  3. Open http://localhost:4000 in your browser.

Command Line Interface (CLI)

You can also run scans directly in the terminal:

# Scan a remote Git repository
node src/cli.js --repo https://bitbucket.org/workspace/repo.git --branch master

# Scan an existing local directory clone
node src/cli.js --path "E:/work/ui/g99-ui-widget" --name my-widget

Outputs compile to ./out/ as report.html (interactive UI) and report.json (machine-readable stats).


🐳 Docker Containerization

A Dockerfile is included to easily deploy the scanner console to container hosts (like Koyeb or AWS ECS):

# Build the Docker image
docker build -t ai-code-attribution-tracker .

# Run the container
docker run -p 4000:4000 ai-code-attribution-tracker

About

"A forensic Git blame analyzer that estimates AI-generated code share per developer and attributes it to Claude or Gemini using code-style fingerprinting."

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages