Permissions Configuration

ship.json.permissions defines the operational boundary for an agent.

Minimal example

{
  "permissions": {
    "read_repo": true,
    "write_repo": {
      "requiresApproval": false
    },
    "exec_command": {
      "deny": ["rm"],
      "requiresApproval": false,
      "denyRequiresApproval": true,
      "maxOutputChars": 12000,
      "maxOutputLines": 200
    }
  }
}

Field semantics

• read_repo: allow repository reads.
• write_repo.requiresApproval: require approval before repository writes.
• exec_command.deny: blocked command prefix list.
• exec_command.requiresApproval: require approval for general commands.
• exec_command.denyRequiresApproval: whether denied commands can be manually approved.
• exec_command.maxOutputChars/maxOutputLines: command output limits.

Recommended policy

1. Keep baseline deny list for production and add risk-specific rules.
2. On CI/shared hosts, enable requiresApproval.
3. If output is truncated, narrow command scope first instead of unlimited output.

Related docs

• Configuration Overview
• ship.json Reference