integration-tests

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill sets UV_DEFAULT_INDEX to https://pypi.org/simple and explicitly downloads/installs dependencies from PyPI at runtime (fresh isolated environments download optional dependencies), which causes fetching and execution of remote package code from that URL.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly instructs the agent to "request elevated command execution (sandbox_permissions=require_escalated)" to bypass execution-environment restrictions, which pushes the agent to obtain elevated privileges and potentially bypass security mechanisms.