WhiteNightShadow/firefox-reverse

What it solves

Firefox-Reverse addresses the difficulty of reverse-engineering web-based security mechanisms—such as dynamic signatures, tokens, and device fingerprints—that are often obfuscated, protected by virtual machines (JSVMP), or compiled into WebAssembly (WASM). Traditionally, this requires manual debugging in browser DevTools, which is time-consuming and prone to errors. This project automates the process of capturing, analyzing, and recreating these algorithms so they can be run independently in Node.js or Python scripts.

How it works

It is a modified version of Firefox that integrates an AI Agent directly into the browser sidebar. This agent has access to 71 specialized tools that allow it to perform tasks like network packet capture, JSVMP instruction tracing, WASM disassembly, and browser environment simulation. The agent can work in two modes: "Fully Automatic" (where it handles the entire process) or "AI-Assisted" (where it pauses for human guidance at each stage). Additionally, it manages isolated browser environments, each with its own profile, process, and fingerprint, ensuring that security analysis is not interfered with by the main browser session.

Who it’s for

This tool is designed for security researchers, developers working on API integrations, and those involved in authorized testing or CTF (Capture The Flag) competitions who need to reverse-engineer complex web-based security parameters.

Highlights

• Integrated AI Agent: A sidebar-based agent that autonomously executes the full "capture-locate-verify-recreate" workflow.
• Isolated Fingerprint Environments: Manages independent browser profiles with unique fingerprints, processes, and Marionette ports to prevent cross-session contamination.
• Engine-Level Observability: Uses deep integration with the SpiderMonkey/Gecko C++ engine to trace signatures and WASM boundaries, bypassing standard page-level anti-debugging techniques.
• Dual Workflow Modes: Offers both fully autonomous execution and human-in-the-loop guidance to suit different complexity levels.
• Model Agnostic: Supports a wide range of LLMs (DeepSeek, Claude, OpenAI, etc.) via local API key configuration.