/\
/__\ PhantomDev
/ \ Humanizer for AI Agents & Commits
/______\ github.com/John-Varghese-EH/PhantomDev
PhantomDev
Adversarial Stylometry Framework for the AI-Augmented Developer
Inject human entropy back into your workflow.
Overview
In 2026, AI-generated code carries an identifiable signature. Whether it's academic integrity systems, hiring pipelines, or code review tools, pattern-based detection is increasingly used to flag work that doesn't "look human."
PhantomDev addresses this directly. It's not about deception-it's about stylometric sovereignty. Developers should be able to use high-productivity AI tooling without sacrificing ownership of their own coding identity.
PhantomDev learns your personal style from your repository history and applies it back to AI-assisted output-transforming variable names, comment phrasing, line cadence, and commit message structure to match how you actually write code.
Features
AI Detection
- Local inference using RoBERTa-based models (no API required)
- Cloud fallback via OpenRouter and Anthropic
- Pattern detection for common AI tells: watermarks, uniform commenting, emoji overuse
- Multi-language support: Rust, Python, JavaScript/TypeScript, Go, C++
Code Humanization
- Repository-aware style learning from your commit history
- Variable and function renaming to match personal conventions
- Targeted comment injection and removal
- Entropy injection for natural stylistic variation
Temporal Obfuscation (Jitter Engine)
- Delayed staging buffer to simulate organic development pace
- Configurable timing windows (min/max delay in seconds)
- Designed to avoid commit-frequency fingerprinting
Stealth Scoring
- Real-time per-file stealth score
- Pattern-based breakdown of detected AI signatures
- Threshold-configurable alerting
TUI Dashboard
- Terminal UI for live stealth score monitoring
- Detection heatmap per file
- Style profile visualization
Undercover Mode
- Commit message transformation via adversarial stylometry
- Code comment rewriting with human-idiomatic phrasing
- Variable name substitution based on personal naming patterns
Pre-Commit Integration
- Git hook-based detection before every commit
- Auto-fix mode for common AI patterns
- Cross-platform, zero-dependency hook scripts
Installation
Recommended (Easy Install)
Clone the repository and run the built-in installer:
To also install IDE-specific skill files:
From Source
One-Click Installer
|
Quick Start
Step 1: Install
|
Step 2: Go to your project
Step 3: Check your code
This will show your stealth score and suggest what to do next.
Step 4: Fix AI patterns
That's it! Your code is now humanized.
Usage
PhantomDev is designed to be simple. Just run phantomdev and it will guide you.
Basic Commands
Typical Workflow
-
Stage your files
-
Check what's detected
-
Fix AI patterns
-
Commit
Advanced Options
# Fix specific files
# See what would change without applying
# Detailed score breakdown
# Reset settings to defaults
# Install IDE integration
pre-commit install pre-commit run --all-files
---
## Configuration
PhantomDev reads from `.phantomdev/config.toml`:
```toml
[detection]
threshold = 0.15 # AI probability threshold (0.0 - 1.0)
use_local = true # Prioritize local models
use_cloud_fallback = true # Fall back to cloud API if local score is uncertain
[humanization]
auto_humanize = false # Run humanization automatically on commit
entropy_level = 0.5 # Transformation intensity (0.0 - 1.0)
[jitter]
enabled = false # Enable temporal obfuscation
min_delay_secs = 60
max_delay_secs = 300
[api]
openrouter_key = "" # Optional: OpenRouter API key
anthropic_key = "" # Optional: Anthropic API key
base_url = "" # Optional: Custom API base URL
Architecture
phantomdev/
├── crates/
│ ├── core/ # Shared types and traits
│ ├── detector/ # AI detection engine
│ ├── humanizer/ # Code transformation logic
│ ├── jitter/ # Temporal obfuscation engine
│ ├── undercover/ # Adversarial stylometry transformer
│ ├── cli/ # Main binary
│ └── tui/ # Terminal UI
├── rules/ # IDE rule files (Claude, Cursor, Windsurf, Antigravity)
├── skills/ # IDE skill files for agent frameworks
├── hooks/ # Git hook scripts
├── scripts/ # Pre-commit and utility scripts
└── models/ # Local model storage
Testing
# All tests
# With stdout output
# Single test by name
# Per crate
Building
# Cross-compile targets
Published on crates.io
PhantomDev is officially published on crates.io. You can easily install the CLI using cargo:
(For detailed workspace publishing instructions, see cargo-instruction.md)
Documentation
Full documentation is available at john-varghese-eh.github.io/PhantomDev
Security
PhantomDev is built with a local-first design:
- All detection runs locally by default; cloud APIs are strictly opt-in
- No telemetry or usage data is collected
- Fully open-source; all code is auditable
- Cloud fallback can be disabled entirely via configuration
See SECURITY.md for the responsible disclosure policy.
Contributing
Contributions are welcome. Please read CONTRIBUTING.md before submitting a pull request. Follow the existing code style and include tests for new behavior.
License
This project is licensed under the Apache License 2.0. See LICENSE for details.
Disclaimer
For Educational and Experimental Purposes Only. PhantomDev is provided "as is" to explore adversarial stylometry and help developers protect their coding signatures. It is not intended to bypass academic integrity systems, cheat in evaluations, or be used maliciously. The author (John Varghese) and contributors are not liable for any misuse of this tool or any consequences arising from its usage. Use responsibly.
Changelog
See CHANGELOG.md for release history.
Acknowledgments
💗 Support the Project
This is an active, evolving project. If it helped you, consider supporting continued development ☺️:
Author
- GitHub: John-Varghese-EH
- LinkedIn: John Varghese
⭐ Star this repo if it helped you.
PhantomDev - preserve your coding signature in the age of AI.