Permission modes
Permission modes control what an agent is allowed to do within a task. They define when the agent can edit files and run commands, and when it must ask for approval.
The available modes depend on the selected agent provider, but they represent the same intent: require approval, auto-approve edits, or grant full access. Claude Agent and Junie also support a dedicated Plan mode.
Select permission mode
Open the task in the Chat tool.
In the task toolbar, click the permission mode selector.
Select a mode that matches your intent:
Ask for approval – prompt for approval before editing files or running commands. Depending on the agent, this mode can be named Default or Chat.
Auto-approve edits – automatically approve file edits. Depending on the agent, this mode can be named Auto-Edit or Agent.
Full access – automatically approve tool calls and allow the broadest set of actions for the selected agent. Use this mode only if you trust the project and understand the risks.
Plan – analyze and propose a plan without editing files or running commands. This mode is available for Claude Agent and Junie. Learn more about the Plan mode.
If available, select the thinking level.
Some agents provide a Thinking Level selector. Thinking level controls the model's reasoning effort: by default, supported models adjust the amount of reasoning dynamically based on the complexity of your request.
Higher thinking levels typically produce more carefully reasoned results, but they take longer and use more tokens.
To change the thinking level, click the thinking icon in the task toolbar and choose a level.
