Task:
Intended agent role:
Risk level:
Permitted actions:
Prohibited actions:
Required human approvals:
Required evidence/logging:
Stop conditions:
Safer execution mode:
Final recommendation: