arabking

3 ms·

1. 1.▲
   by arabking 5mo ago

   I built this after the ClawHavoc campaign (341 malicious skills in 3 days, Jan 2026) and after Snyk's ToxicSkills audit, which showed that 13.4% of skills contain critical security issues. There was no OSS scanner, so I built one. 12 r
2. 2.▲
   Show HN: SkillGuard – scan agent skills for prompt injection payloads(github.com)

   2 points by arabking 5mo ago|1 comments
3. 3.▲
   by arabking 5mo ago

   The timing is striking. Most current AI security work focuses on the skill/tool layer — prompt injection, exfiltration via agent skills, rug pulls — and assumes the base model is trustworthy. A leaked cybersecurity-focused model flips