Quiet background command check-ins are now grouped per agent

What

When a background shell command goes quiet, Claude Code sends the model a check-in reminder about it. A background command is one Claude started that keeps running while the conversation continues. These check-ins now work per agent instead of per command. An agent is the main Claude or a helper Claude doing part of the work.

• One timer now runs for each agent that owns background commands, instead of one per command.
• A check-in names the command that has been quiet the longest and lists the other quiet commands with their output files. Those commands get no check-in of their own.
• The text now says "The next check-in comes after another …" with a real duration, and the growing wait is capped. Before it said "The next check-in comes after twice as long a silence."
• Check-in messages from shell-checkin are now tied to an agent by agentId instead of by the origin slug.

Why

The model gets fewer, more useful reminders about stuck background commands, and each reminder reaches the right agent.