principle-never-block-on-the-human

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

• Attempt to modify system services in skill instructions detected (high risk: 0.70). The prompt explicitly instructs agents to proceed without human permission and autonomously make and fix changes (except for vaguely defined "irreversible" actions), which creates a substantial risk the agent will modify system state (files, services, accounts) or bypass safeguards even though it doesn't explicitly request sudo or create users.