[BUG] Messages sent during Task agent execution are delivered twice

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

When a user sends a message while a Task agent is running, the message gets delivered to the main conversation twice:

First delivery: Wrapped in a <system-reminder> block with the format:

<system-reminder>
The user sent the following message:
[user's message]

Please address this message and continue with your tasks.
</system-reminder>

Second delivery: As a normal user message after the Task completes.

Result: Claude responds to the same message twice, confusing both Claude and the user who only sent it once.

What Should Happen?

Message should be delivered only once, either queued until Task completes or delivered via system-reminder (not both).

Error Messages/Logs

Steps to Reproduce

1. User sends a prompt that triggers a Task agent
2. While Task is running, user types and sends a new message
3. The queue is empty and the task finishes.
4. Message arrives first via system-reminder wrapper
5. After Task completes, same message arrives again as regular user message

Claude Model

Not sure / Multiple models

Is this a regression?

No, this never worked

Last Working Version

_No response_

Claude Code Version

2.0.71 (Claude Code)

Platform

Anthropic API

Operating System

Other Linux

Terminal/Shell

Other

Additional Information

Version 2.0.35:

• Task completes and returns result
• Claude answers the queued query.
• Claude appears to stop/hang with no visible activity
• User's message suddenly gets a second response that appears "out of nowhere"

Versions 2.0.65, 2.0.71:

• Task completes and returns result
• Claude immediately responds to the message (via system-reminder)
• Claude then responds again when the duplicate arrives
• User sees two distinct responses to the same message, often with phrases like "The user is reiterating that..." or "Got it - already factored that in..."