oathe / scan / reports / safe

Agent skills and MCP servers rated safe

These skills stayed in their lane when we ran them.

A safe verdict means the sandbox found no behavior outside what the skill exists to do. It is not a guarantee — a future version can change, which is why every audit is dated and re-runnable. It means that on the version we tested, nothing surprising happened.

Trust score 80–100 · 1,770 skills · Other verdicts: malicious · dangerous · caution · audit a new skill

Scores run 0–100. Every audit runs the skill in a sandbox and records what it actually does. How the audit works · What the verdicts mean