What Does Full Access Allow the Agent to Do?
Confirm the execution environment and acceptable modification scope before reducing step-by-step approvals.
Commands and Edits Proceed Directly
Full access allows corresponding terminal commands and file modifications to proceed without waiting for your approval each time. It is suitable for tasks where you have clearly defined the scope, understand the environment, and accept automatic execution. This reduces interruptions but does not judge the correctness of each result for you.
Selecting this does not mean all tools are always available; model, environment, and account restrictions still apply.
Check the Environment Before Enabling
The same command may have completely different effects on the cloud versus a personal computer. Do not rely solely on the permission name; first confirm the actual running location.
If the task only requires researching documentation, do not let the agent start modifying the system just to reduce waiting. You can write this round as a clearly read-only goal and require reporting if modifications are needed; whether to select full access should align with the actual allowed work scope.
- Verify the computer or cloud option in the environment selector.
- Specify the files and task scope allowed for operation.
- Keep recoverable copies of important files.
- Start by checking output on small tasks before scheduling continuous work.
When You Need to Regain Control
You can adjust subsequent behavior using the permission options provided in the interface; completed operations will not be automatically undone. If you find the direction is wrong, stop or clarify the task first, then check files and running processes. The local environment still uses the permissions of the user connected via the client, with no container isolation; full access does not automatically grant permissions the system does not originally have.