Background commands can now wake the subagent that started them

What

A subagent is a helper Claude starts to handle part of a task. A background command is a shell command left running while Claude carries on. A subagent can now pause until its background command ends and be woken by it:

• The Bash tool result has a new optional field, backgroundWakesOwner, passed on as wakesItsOwner. It is true when the subagent that owns the backgrounded command waits for it to end and is woken by it.
• The message Claude gets about a backgrounded command has a new variant for owners woken by completion. Before, the only variant was the "After your reply you may not hear of it" text, keyed on backgroundShellsKeptElsewhere.
• The stop notice now says Claude will be notified only under a condition, for example: "You will be notified when it completes, even after your final response, unless this session goes idle first, so if you need its result, wait for it before you reply."
• The related MCP tool-call registration now passes backgroundedToDeliverMessage.
• Local agent tasks also record launchModel.

The stock Bash tool description captured under this release is unchanged.

Why

A subagent can wait on a long command instead of finishing before the result arrives, and Claude is told more exactly whether it will hear back about a background command, so it is less likely to reply before a result it needs.