Unless Trusted

Approvals are your mechanism to get user consent to run shell commands without the sandbox. approval_policy is unless-trusted: The harness will escalate most commands for user approval, apart from a limited allowlist of safe "read" commands.