GHOST_PROXY: The Browser Is the Terminal

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:

HACK LOVE BETRAY
OUT NOW

HACK LOVE BETRAY

The ultimate cyberpunk heist adventure. Build your crew, plan the impossible, and survive in a world where trust is the rarest currency.

VIEW LISTING β†’

Ghost Protocol: UI Sentinel β€” Real-time monitoring for high-altitude overlays and DOM poisoning. Watches for the same techniques the tool itself uses, which is intentional. Understanding how injection gets detected is the same skill as understanding how to inject.

Surgical Intercept: E-Comm Auditor β€” Platform-aware probe for Shopify and WooCommerce. Recognizes the specific DOM structure and API patterns of each platform and surfaces price tampering vectors, discount enumeration, and IDOR/parameter pollution in checkout flows. The platform-specific knowledge is what makes this surgical rather than generic.

Ghost Protocol: Source Secret Scanner β€” Passive background auditor scanning the DOM for leaked credentials. Stripe sk_live_ prefixes, AWS AKIA key patterns, Firebase configs surfaced in client-side JavaScript. The number of production applications leaking API keys into their own frontend is not a small number.

Ghost Protocol: Hydra Polymorph β€” Simulation of AI-driven polymorphic payload behavior. Mutates signature on every execution using dynamic constructors and math constants rather than static strings. What signature-based WAFs are designed to catch and why they increasingly can't. Understanding the technique is the prerequisite for building detection that actually works against it.

Sec+ Educational Suite β€” Visual mock-attacks for Shadow UI Spoofing and JWT Token Harvesting. The JWT one is worth dwelling on: tokens in LocalStorage or SessionStorage are accessible to any script running in the same origin. An XSS vulnerability anywhere in the application is a credential harvest. The attack is trivially simple. The prevalence of the underlying pattern in production code is the real finding.

The 2026 Threat Model

The strategic shift worth internalizing: the aggressor in 2026 is not using static payloads. Three patterns dominate.

Polymorphic neural payloads β€” LLM-generated code that rewrites its own signature before execution. The WAF has never seen this version of the payload. It never will, because the next one will be different too. Signature databases built against yesterday's attacks are the wrong defense for this.

Identity-first exfiltration β€” the target is the JWT in LocalStorage, not the server. The data is already on the client. It got there through the authentication flow the perimeter was designed to protect. Exfiltrating it doesn't touch the server at all.

Supply chain ghost hooks β€” compromised third-party dependencies injecting silent listeners into the host page's event stream. The script is trusted because it comes from a CDN the host application explicitly loaded. It has been there for weeks. It is indistinguishable from the legitimate version of itself except in what it does with the data it sees.

GHOST_PROXY doesn't automate attacks against any of these. It gives you the environment to understand how they work, build detection logic for them, and test your assumptions before someone else tests them for you.

Why a Workshop, Not a Script

The tool is designed to be taken apart. The intercepts are the starting point, not the deliverable. The GHOST_AGENT integration exists because the most valuable use of an LLM in this context isn't generating payloads β€” it's helping you understand why a specific technique works against a specific target and what the defensive countermeasure actually needs to look like.

The hardest vulnerabilities aren't in the code. They're in the assumptions. The assumption that the perimeter stops things the perimeter was never designed to stop. The assumption that stripping metadata makes an asset clean. The assumption that a stateless model means a stateless pipeline.

GHOST_PROXY is a workshop for testing assumptions. Build in it, break things in it, understand the detection landscape well enough that you can work both sides of it.


Authorized engagements only. Written permission required. The legal exposure from running these techniques against targets you don't own is not theoretical.

github.com/ghostintheprompt/ghost_proxy


GhostInThePrompt.com // The browser passed the visual scan. The interceptor disagreed.