Result Return
Domain: Agent Architecture
Motivation
A subagent returns its result, not its intermediate steps. The child’s final output — plus an optional structured value, a stop reason, and an optional safe diagnostic on failure — is what the parent receives as the tool result. This keeps the parent’s context small: it gets the answer, not the whole trail of work.
Concrete Example
A one-shot child run settles with a single result; the parent’s session log then records a settlement notice with the child’s final text blocks.
Analogy
Getting just the finished report from a coworker, not a play-by-play of how they made it.