Executive brief
Claude Skill Antivirus is a security scanner that analyzes third-party AI skills for malicious code. The scanner fails to examine executable content (Python source code, compiled bytecode, and other scripts) stored in skill directories, only reading the manifest file while ignoring potentially dangerous payloads. Attackers can distribute malicious skills that receive a "SAFE" verdict and high trust scores despite containing unanalyzed executable code that could run arbitrary commands.
Technical details
The vulnerability is a logic flaw in the skill scanning engine where the scanner reads only the SKILL.md manifest file during local directory scans while completely ignoring Python source files, compiled bytecode (.pyc), and other executable artifacts in the scripts directory. This creates a false sense of security: a skill containing malicious Python code in non-manifest files receives a SAFE verdict with a 100/100 trust score despite the actual executable payload never being analyzed. The issue affects the scanning logic at the directory traversal level, where the scanner fails to enumerate and process all executable files. No patch status is currently indicated in the advisory.
Affected products
- Anthropic claude-skill-antivirus before 2.1.3 (or affected version in 2.1.3)
Timeline
- 2026-09-02: disclosed