complete

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 acknowledge and (potentially) use flags that bypass permission gates and sandboxes (e.g. "--dangerously-bypass-approvals-and-sandbox", "--permission-mode auto"), which encourages bypassing security mechanisms and granting full local access to launched builders.