SkillsGuard, a free static security scanner, launches to detect malicious code in AI agent skill packages before they run—no account, token, or LLM endpoint required.

What happened

SkillsGuard is a new open-source tool that scans SKILL.md files and bundled scripts for malicious patterns using 151 detection rules across 15 categories. It decodes obfuscated payloads (base64, hex, URL-encoding, recursively) before scanning, so encoded reverse shells cannot slip through. The tool runs entirely offline with no runtime dependencies beyond Node ≥18.3.

Why it matters

AI agent skills are a new, largely unaudited attack surface—a malicious skill can inject prompts to override AI guidelines, exfiltrate secrets like API keys and SSH credentials, execute arbitrary commands, persist via cron jobs or shell startup files, escalate privileges, or hide all of the above through obfuscation. SkillsGuard catches these patterns before an agent ever reads the file, with zero installation friction.

What to watch

SkillsGuard competes in a crowded 2026 market alongside NVIDIA SkillSpector, Cisco AI Defense Skill Scanner, Snyk Agent Scan, and Mondoo Skill Check. SkillsGuard's key differentiator is that it needs nothing beyond Node to run—every other actively-maintained competitor either requires signing up for a service or recommends configuring an LLM provider. The tool is available via free cloud API (no account needed) or can be built from source; it outputs color-coded CLI reports, JSON, SARIF, or MCP-server responses.