Static scanner audits 332 AI skill repos for security risks

The author of the open-source SkillScan scanner analyzed 332 AI skill repositories on GitHub (6,262 files, 12.6 million combined stars) using rules mapped to CWE, OWASP LLM Top 10 and MITRE ATLAS. Top findings include shell config rewrites and cron jobs (725), reading .env and SSH keys (404), and instructions to hide actions from users (225). After human review, all 7 of the most alarming flags were false positives, including a Microsoft project.

• 332 repositories scanned: 6,262 files and 12.6 million combined stars
• 725 hits for shell config rewrites and cron job registration
• 404 findings for reading .env and SSH keys, 225 for hiding actions from users
• All 7 of the most alarming flags were false positives after human review