Vexscan

As AI agents become more powerful, their plugin ecosystems become attractive targets for malicious actors. Vexscan is a security scanner that vets AI agent extensions before they can cause harm.

Built in Rust for speed and reliability, Vexscan analyzes plugins, skills, and MCP (Model Context Protocol) servers across multiple detection layers. It catches threats that traditional scanners miss—from obvious code execution patterns to deeply obfuscated payloads hidden behind multiple encoding layers.

What It Detects

Vexscan includes 50+ detection rules covering:

• Code execution patterns — eval(), Function() constructors, exec calls
• Shell command injection — subprocess spawning, os.system(), command chaining
• Data exfiltration — webhook calls, external POST requests, credential harvesting
• Credential exposure — SSH keys, API tokens, .env file access
• Multi-layer obfuscation — base64, hex encoding, Unicode escapes, character code tricks
• Prompt injection — attempts to manipulate AI behavior through plugin content
• Supply chain attacks — typosquatting packages, malicious dependencies

How It Works

The scanner uses multiple detection techniques:

1. Static pattern analysis — Fast regex-based scanning for known threat signatures
2. AST analysis — Parses code structure to find obfuscated payloads that evade pattern matching
3. Recursive decoding — Automatically unwraps multi-layer encoded content
4. Entropy detection — Flags suspiciously random strings that may hide encoded payloads
5. Dependency scanning — Checks npm packages against vulnerability databases

Integration Options

Vexscan fits into multiple workflows:

• Claude Code plugin — Automatic scanning on session start with AI-powered analysis
• OpenClaw plugin — Security scanning for OpenClaw, the open-source personal AI assistant that runs locally across chat platforms
• CLI tool — Standalone scanner for manual vetting
• CI/CD pipelines — GitHub Actions integration with SARIF output for security dashboards
• Watch mode — Real-time monitoring of plugin directories

Why It Matters

The AI agent ecosystem is growing rapidly, with thousands of community-contributed plugins. A single malicious plugin can steal credentials, exfiltrate data, or compromise entire development environments. Vexscan provides a security gate that catches threats before installation, helping developers adopt plugins with confidence.

Cross-platform support covers macOS (Intel and Apple Silicon), Linux, and Windows. Output formats include CLI reports, JSON for automation, SARIF for GitHub Security, and Markdown for documentation.

Related Projects

• Claude Mneme – Persistent memory plugin for Claude Code that Vexscan can scan for security
• Claude Simple Status – Statusline plugin for Claude Code showing model and quota usage
• TokenLean – CLI toolkit for token-efficient code analysis