Be able to mark targets as "dangerous" and require to pass `--auto-approve`, or be prompted interactively