Agent Security Scanner

•
124
•

Scans code for vulnerabilities, detects AI-hallucinated packages, and blocks prompt injection for AI coding agents and autonomous assistants.

The Agent Security Scanner is a comprehensive security tool designed for AI coding agents and autonomous assistants. It provides multi-layered protection by scanning code for over 1700 vulnerabilities across 12 languages using advanced AST and taint analysis, offering automated fixes for identified issues. It also proactively detects AI-hallucinated packages by checking against a registry of over 4.3 million known packages and acts as a prompt injection firewall with 56 rules to block malicious instructions and data exfiltration attempts. Integrated with popular AI clients like Claude Code and Cursor via MCP, it enhances the security posture of AI-driven development workflows.

주요 기능

01Code vulnerability scanning (1700+ rules, 12 languages) with AST & taint analysis
02Automated vulnerability fixing with 120 fix templates
03AI package hallucination detection across 4.3M+ packages in 7 ecosystems
04Prompt injection firewall (56 rules) for malicious instructions
05OpenClaw integration targeting autonomous AI threats like data exfiltration and credential theft
0610 GitHub stars

사용 사례

01Scan and auto-fix code vulnerabilities after writing or editing any code file
02Verify new dependencies for AI-hallucinated package names before installation or committing
03Detect prompt injection and malicious instructions from untrusted input before execution