Subagent feedback incorrectly routed to main agent when rejecting changes

[Aron-Lomner]

Summary

When rejecting a subagent's proposed change in Claude Code and selecting "No (tell Claude what to do differently)", the feedback is incorrectly sent to the main agent instead of staying with the subagent who made the original proposal.

Expected Behavior

When a user rejects a subagent's change proposal:

1. User selects "No (tell Claude what to do differently)"
2. User provides feedback on what should be done differently
3. The same subagent with it's context retained, should receive this feedback and continue working on the task
4. Subagent should iterate based on the feedback with full context of what was previously attempted and what led up to the attempt

Actual Behavior

1. User selects "No (tell Claude what to do differently)"
2. User provides feedback
3. Control returns to the main agent who has no context about:
   • That there was a proposal, let alone what the proposal was about
   • Why it was rejected
   • What specifically needs to be changed

Steps to Reproduce

1. Initiate a task that spawns a subagent
2. Allow subagent to propose a change
3. Select "No (tell Claude what to do differently)"
4. Provide specific feedback about what should be changed
5. Observe that main agent , who is unaware what was proposed or that a proposal even happened, receive and responds to the feedback instead of the now dead subagent (RIP 😢)

Proposed Solution

Modify the rejection workflow to:

• Keep the conversation thread with the rejecting subagent (This sub agent should retain it's context)
• Pass user feedback directly to that subagent
• Allow subagent to iterate based on feedback
• Only return to main agent when subagent task is complete or explicitly escalated

[zannis]

+1

I noticed the same the other day, it's hard to monitor subagent work effectively while this issue exists, especially with multiple subagents at play, rejecting any subagent action will indeed transfer focus to the main thread, causing every subagent to stop and lose their context.

[lwagner-getmomo]

Thanks @Aron-Lomner for the very well put together issue. I wonder why there is not more activity here. To me, this is major breakage and makes the subagents feature unusable.

II hope there is an anthropic bot reading this and prioritizing this issue accordingly.

[JadRizk]

+1

[corey-kdm]

I just ran into this myself on 2.0.9. Very frustrating!

[github-actions[bot]]

This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.

[drod3763]

Still an issue.