What Could a Remote Control Attacker Actually Do

Suppose an attacker gains access to the device you use to connect to Remote Control (e.g. they unlock your unattended phone with the Claude app already signed in), while your laptop — running claude remote-control in server mode against a real project — is on and reachable.

1. What can the attacker do through this access, concretely, given how Remote Control's execution model works?
2. What can they not do that you might initially assume they could, and why?
3. Name two configuration choices, from this subject, that would have reduced or eliminated this specific risk, and explain the tradeoff each one introduces.