Integration
Safety Confirmation
To ensure that any Agent action with potential impact on your local system remains fully under user control, Z Code enables a manual confirmation step before execution by default.
This is part of the product’s core safety strategy. It gives you the final veto over Agent behavior and helps prevent accidental or unauthorized actions.
Safety tip: Review the command, impact range, and risk notice carefully before confirming.
Workflow
- Trigger confirmation: when the Agent attempts a potentially risky action such as running a script, accessing the network, or modifying important files, execution pauses automatically
- Show the request: the UI displays a permission confirmation panel including the current execution state, the command being prepared, and related risk reminders
- User decision: you choose how to handle the request and confirm before the Agent continues or stops
Decision Options
| Option | Description | Recommended Scenario |
|---|---|---|
| Allow | Authorize the Agent to perform this one action only | Temporary or one-off tasks that still require review each time |
| Always Allow | Skip future confirmations for the same kind of command | When you already trust this class of action and expect to repeat it frequently |
| Reject | Stop the current action from continuing | When the command, path, or risk explanation does not match expectations |
Note: Always Allow relaxes confirmation frequency for similar commands within the current session, so only use it when you fully trust the action.
Interaction tip: The permission panel supports
Tabor the arrow keys to switch between options, thenEnterto confirm.
Typical Scenarios
- Running third-party scripts: when the Agent attempts to run Python, Shell, or Node.js scripts inside your project
- Network requests: when the Agent needs to access external APIs through
curlor similar tools - System-level commands: when a command may change system configuration or delete files