[BUG] When multiple read-only agents run in the background, their output to Claude Code is empty every time.

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 dispatching multiple background read-only agents (e.g. 5 agents researching different topics simultaneously), each agent appears to work correctly - making significant progress with multiple web searches and fetches, consuming 12,000–18,000+ tokens each, and using 20+ tools but when they complete, the final result message returned to the orchestrator is empty.

Agents show substantial token usage and tool calls while running, confirming they ARE doing work
When an agent completes, the result message is empty
The agent output files at C:\Users\[user]\AppData\Local\Temp\claude\...\tasks\[id].output are also empty
This affects all agents, not just one - every agent that completed returned empty output.

What Should Happen?

Expected Behavior:
Each agent should return its research findings/output to the orchestrating agent when it completes.

Error Messages/Logs

Steps to Reproduce

Steps to Reproduce:

Start a task that dispatches multiple background (read-only) agents in parallel (I used 5 agents, each researching a single topic in a different country)
Wait for agents to complete their work
Observe that completed agents return empty results

Claude Model

Opus

Is this a regression?

No, this never worked

Last Working Version

_No response_

Claude Code Version

2.1.34

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Windows Terminal

Additional Information

I'm using Command Prompt in Windows 10.
I've recreated the problem multiple times.
The latest time when Claude Code realized the problem, it froze and had to be force-shut down.