[FEATURE]: Better execution permissions and sandboxing

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

OpenCode's current execution controls could be improved in three areas:

1. Make permission mode easier to access

The current permission settings are relatively hidden.

It would be useful to show the active permission mode directly in the conversation UI and allow quick switching during a session.

2. Add sandboxed execution

OpenCode has permission controls such as allow, ask, and deny, but lacks a built-in sandbox comparable to a restricted execution environment.

A sandbox could limit filesystem, network, and process access even when an action is approved.

3. Add optional model-assisted approval

For ask mode, an optional reviewer model could evaluate requested actions and decide whether to:

• auto-approve
• ask the user
• deny

Ideally, the reviewer model could be configured separately from the main agent, so a small and inexpensive model can handle approval checks.

These three features would make OpenCode safer and more convenient for autonomous execution.