[SYS-032] Ultrareview crash kills Remote Control session — cascading failure, no isolation

Summary

When crashes mid-run (in this case due to server rate limiting), it takes down the active Remote Control session along with it. Two independent subsystems — a cloud code review skill and a mobile RC connection — should never share failure modes.

Reproduction

1. Connect Claude mobile app via Remote Control
2. Run from the desktop CLI
3. Ultrareview hits server rate limit mid-run and crashes
4. Remote Control session disconnects simultaneously

What should happen

Ultrareview crashes → ultrareview session ends → RC stays connected. The user's mobile session should be completely unaffected by what happens in a spawned cloud skill.

Impact

User loses both the review output AND their mobile session in a single event. Must reconnect RC manually. In this case it was the second consecutive failed ultrareview of the day — both sessions consumed credits, neither delivered output, and the second one also killed the RC connection.

Context

This is the third distinct ultrareview bug filed on launch day (2026-04-16):

• #49465: Credit consumed on failed launch
• #49466: Not available over Remote Control
• This issue: RC killed by ultrareview crash

The pattern suggests ultrareview's remote agent session shares connection/process state with the parent CLI session rather than running as a fully isolated subprocess.

Environment

• Claude Code v2.1.111
• Windows 10, Claude iOS mobile app via Remote Control
• Ultrareview crash cause: 'Server is temporarily limiting requests (not your usage limit) · Rate limited'