panel

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

• Attempt to modify system services in skill instructions detected (high risk: 0.90). The skill prompt instructs the agent to run third-party AI coding CLIs with flags that bypass security, permissions, and sandboxing (--dangerously-skip-permissions, --dangerously-bypass-approvals-and-sandbox), which compromises the security state and safety mechanisms of the execution environment.