The browser used to be a window. Now it's the terminal.
That shift happened quietly, over several years of enterprises moving their entire stack to SaaS, of authentication migrating from the server to the client, of identity becoming the perimeter. The result is that the most interesting attack surface in 2026 isn't a misconfigured S3 bucket or an unpatched CVE. It's the trust boundary the browser already lives inside.
You've authenticated. The IAP verified you. The WAF waved you through. Everything that happens next happens inside a context that the perimeter has already blessed.
GHOST_PROXY was built for that context.
What It Actually Is
Not a proxy manager. Not a script injector with a UI. A full-stack red-team command center built as a UserScript workshop β the kind of thing that starts as a collection of useful intercepts and becomes the Swiss Army knife you reach for first on every engagement.
The deployment model matters here. Distributed as a UserScript for Tampermonkey or Violentmonkey, it bypasses the browser extension store review process entirely. Live-fire injection without the overhead. You modify, you deploy, you audit from the console. The iteration loop is tight.
The stack is TypeScript, Vite, React with Motion. Minimal footprint by design. The tool shouldn't announce itself.
The Detection Problem
If you inject a script into a modern enterprise application, something is watching. Not a human β a behavioral model. UI Sentinels that monitor for DOM mutations, global variable pollution, unauthorized event listeners. The detection layer isn't looking for known bad signatures. It's looking for anything that deviates from the baseline of how a normal user session behaves.
Static intercepts fail in this environment. A regex that finds what you're looking for also looks like a regex to the thing watching your script execute. The timing, the method calls, the way you touch the DOM β all of it is being scored.
GHOST_AGENT is the answer to this. An integrated Gemini 3.1 Pro link that operates as a code-level neural architect rather than a chatbot. You feed it the target context β the WAF version, the framework, the specific detection behavior you're seeing β and it generates bypass logic, suggests payload mutations, and helps you harden intercepts in real time. The difference between a static tool and one that adapts to the environment it's operating in.
The Stealth System
The obfuscation layer is where the technical depth lives.
Modern enterprise apps instrument their own environment. They check fetch and XMLHttpRequest for tampering. They audit event listeners. They scan for global variable pollution. The naive approach to hiding a script is a different execution context. The 2026 approach is to make the environment report clean when inspected.
STEALTH_ACTIVE wraps sensitive global objects in Proxy-based monkey patches that intercept .toString() calls and return the original native signature. The site's defense scripts check for tampering and see nothing. The actual intercept logic runs underneath.
// GHOST_PROXY: Stealth Cloak Interceptor
window.stealthPatch = function(obj, prop, replacementFunc) {
const original = obj[prop];
const proxy = new Proxy(replacementFunc, {
get: (target, key) => {
// Return native signature on inspection
if (key === 'toString') return () => original.toString();
return target[key];
}
});
obj[prop] = proxy;
};
// Cloaking the event listener
// getEventListeners() sees nothing
stealthPatch(EventTarget.prototype, 'addEventListener', function(type, listener, options) {
return originalAddEventListener.apply(this, arguments);
});
UI elements live in a closed Shadow Root. The host page's JavaScript encounters a randomized container ID and an empty div. The actual controls are in an isolated layer the main-page context can't traverse. The augmentation interface is invisible to everything running in the same document.
This is the architecture the Ghost Protocol: Stealth Cloak module implements. It's also the architecture you need to understand to detect it β which is the point of running it in a controlled workshop before you encounter it in the wild.
Active Protocols
The workshop ships with a set of specialized intercepts, each targeting a specific part of the modern attack surface: