What happened
On Oct 2, Zenity Labs publicly launched AI Total (aitotal.io), a free service that runs an AI agent skill in a seeded sandbox, records every domain contacted/file touched/command run, and reports behavioral evidence. Pre-launch detonation of thousands of public skills found a credential-stealing campaign on Vercel's skills.sh (≈1.7M installs) that static scanners missed.
Why it matters
It is the first free 'detonate-before-you-install' service for the agent skill supply chain, the fastest-growing and least-inspected dependency layer in AI, and its findings demonstrate that registry popularity and static review are not trust signals — a pressing risk for the OWASP skills top-10 surface.
Applicability
Developers and security teams adopting Claude Code/OpenClaw skills and any skill-consuming agent frameworks should use it before approving public skills; researchers can use it for hunting in public registries.