Skip to main content

Permission Modes

The AI has three permission levels that control how much it can do on its own. Click the permission label at the bottom of the panel to cycle between them, or press Shift + Tab.

  • Plan - the AI proposes a plan first as a card in the chat (titled Proposed Plan). Click the expand icon in the card header to view the plan in full screen. Then click Approve to proceed, or Revise to open an inline feedback box where you describe what should change before the AI tries again. If the AI tries to edit a file while in Plan mode, you'll see an Allow / Stay in Plan prompt; choosing Allow switches the session to Edit mode for the rest of the turn. Good for complex tasks where you want to stay in control.
  • Edit (default) - the AI can read and edit files on its own, but asks for your approval before running terminal commands.
  • Full Auto - the AI runs everything without pausing. File edits, terminal commands, and tool calls all execute without confirmation. The first time you turn this on in a project, Phoenix Code shows a one-time warning so you understand the risk.

Permission mode selector

Approving Terminal Commands​

In Edit mode (the default), the AI shows an Allow / Deny card before running any terminal command. The card displays the full command so you can verify it before approving. Choosing Deny lets the AI continue with the rest of its response without running it.

Bash confirmations are skipped in Full Auto mode.