Live sandbox · AIVSS-aligned · OWASP Agentic Top 10
Know what an agent skill does — before it runs.
Agent skills are code-by-instruction — unsigned, unreviewed, and one prompt away from reading your secrets. AgentScan detonates a skill in an instrumented sandbox and returns a verdict in seconds.
Sign in to scan
Every scan runs in a real, instrumented sandbox on our infrastructure. Sign in with your work email to submit a skill, bundle, or folder.
Browsing public reports below needs no account.
Submit file
Submit URL soon
Drop a skill to detonate
A single SKILL.md agent skill.
Browse files
SKILL.md.mdmax 500 KB
Single model · 1 scan → one report