Unclear It is not clear what decides whether the answer is reported to Anthropic.
A subagent is a helper copy of Claude that takes on part of a task. When a remote subagent asks for permission and you cancel the prompt, Claude Code now responds based on where the subagent is running:
- If it is running in the background, the request is answered with "User denied permission" and the current turn carries on.
- Otherwise the answer is "User aborted", as before.
Before, cancelling always gave "User aborted" and could interrupt the turn. The permission dialog also now has a short grace period on input. Claude Code also reports to Anthropic how each request was answered: whether it was allowed, whether it was dismissed, and whether the subagent was in the background.
Dismissing a prompt from a background subagent no longer stops the work you are doing in the foreground.
It is not clear what decides whether the answer is reported to Anthropic.