[BUG] Planning mode prompting for approval of plans from Task() sub-agents

Environment

• Platform (select one):
• [x] Anthropic API
• [ ] AWS Bedrock
• [ ] Google Vertex AI
• [ ] Other: <!-- specify -->
• Claude CLI version: 1.0.43
• Operating System: macOS 14.3
• Terminal: iTerm2 / zsh

Bug Description

There's a confusing thing going on when I use plan mode and Claude decides to use a sub-agent for a task... the sub-agent itself seems to be in 'plan' mode (which is fine, I suppose), but the CLI then presents the output of that sub-agent as a "plan" to _me_ to approve, rather than to the 'parent' Claude that spawned it. I don't have easy visibility of what the 'parent' Claude instructed the sub-agent to do, so it's really difficult to know whether I should approve or reject it.

Normally, the output is clearly some sort of analysis, but it is definitely not what I've requested the parent 'Claude' to do in my prompt.

It feels like the right behaviour would be for the 'plan approval' handling to not be run at all when it's a sub-agent coming back with an answer for the 'parent' Claude.

Steps to Reproduce

1. Switch to plan mode (shift-tab twice)
2. Ask for a suitably complex thing to be planned such that Claude decides to use a sub agent.
3. See it happen

Expected Behavior

For sub-agents, I'd not expect to be asked to approve anything, it should report back to the agent that spawned it and it should decide what to do.

Actual Behavior

I get a confusing plan to approve each time, often followed by another plan from the parent to approve shortly afterwards.

Additional Context

I have a Stop hook configured (but it doesn't seem to be running when sub-agents finish, so I think that is not relevant).