How a malicious AI agent skill passed security checks and reached 26,000 users

USN-8151-1: lambdaisland/uri vulnerability

We scanned 900 MCP configs on GitHub. 75% had security problems.