[Bug] Background agents output files remain empty (0 bytes) - run_in_background=true broken

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?

Bug Report: Background agents output files remain empty (0 bytes)

Summary

When launching sub-agents with run_in_background=true, the output files are created but remain empty (0 bytes). The agent work is executed (confirmed by system notifications showing tool usage and token consumption), but results are never written to the output files and cannot be retrieved.

Environment

• OS: Windows 10/11 (MINGW64_NT-10.0-26200)
• Platform: win32
• Claude Code Version: Latest as of January 9, 2026
• Model: claude-opus-4-5-20251101

Steps to Reproduce

1. Launch a sub-agent with run_in_background=true:

Task tool call:
- description: "Test background agent"
- prompt: "List all Use Cases in src/application/use-cases/"
- subagent_type: "codebase-analyzer"
- run_in_background: true

1. Wait for agent completion (system notifications confirm progress: "25 new tools used, 50237 new tokens")

1. Receive task completion notification with output file path

1. Attempt to read the output file

Expected Behavior

The output file should contain the agent's results/findings.

Actual Behavior

• Output file is created at the specified path
• File size remains 0 bytes even after agent completion
• Read tool returns: "Warning: the file exists but is shorter than the provided offset"
• TaskOutput tool returns status: completed but no content
• cat command on the file produces no output

Diagnostic Evidence

File listing showing 0-byte files:

$ ls -la /c/Users/.../tasks/
-rw-r--r-- 1 User 197121 0 Jan  9 17:49 a4b6c47.output
-rw-r--r-- 1 User 197121 0 Jan  9 17:49 aa80314.output
-rw-r--r-- 1 User 197121 0 Jan  9 17:49 aecb4d8.output

TaskOutput returns no content:

<retrieval_status>success</retrieval_status>
<task_id>a4b6c47</task_id>
<task_type>local_agent</task_type>
<status>completed</status>
<!-- No actual output/results -->

System notifications confirm agents ARE working:

Agent aecb4d8 progress: 25 new tools used, 50237 new tokens
Agent a4b6c47 progress: 10 new tools used, 29172 new tokens
Agent aa80314 progress: 35 new tools used, 35686 new tokens

Workaround

Synchronous agents work correctly. When launching agents WITHOUT run_in_background=true, results are returned properly:

Task tool call (synchronous):
- description: "Test synchronous agent"
- prompt: "List TypeScript files in src/domain/"
- subagent_type: "Explore"
- run_in_background: false (or omitted)

Result: Successfully returns agent findings

Comparison Table

| Mode | File Created | File Size | Results Returned |
|------|--------------|-----------|------------------|
| run_in_background=true | Yes | 0 bytes | No |
| run_in_background=false | Yes | 0 bytes | Yes (inline) |

Impact

• Cannot run multiple research agents in parallel
• Cannot perform deep codebase analysis that benefits from background processing
• Forces sequential agent execution, reducing efficiency
• Breaks workflows that depend on background agent results

Timeline

• Working: Before January 2026
• Broken: Noticed around January 7-9, 2026

Additional Context

This bug affects all agent types tested:

• Explore
• codebase-locator
• codebase-analyzer

The issue appears to be in the mechanism that writes agent results to the temporary output files, not in the agents themselves (they execute successfully based on token/tool usage metrics).

What Should Happen?

When launching sub-agents with run_in_background=true:

1. The output file should contain the agent's complete results/findings after task completion
2. The TaskOutput tool should return the actual content produced by the agent
3. The Read tool should be able to retrieve the agent's output from the specified output file

Currently, synchronous agents (without run_in_background) return results correctly - the same should work for background agents.

Error Messages/Logs

Steps to Reproduce

1. Launch any sub-agent with run_in_background=true parameter:

• Example: Task tool with subagent_type="Explore" or "codebase-analyzer"
• prompt: any research task like "List files in src/"
• run_in_background: true

1. Wait for agent completion (system notifications show progress: "X new tools used, Y new tokens")

1. Receive task completion notification with output_file path

1. Try to read the output file using:

• Read tool → returns "file exists but is shorter than the provided offset"
• TaskOutput tool → returns status "completed" but no content
• Bash: ls -la shows file size = 0 bytes
• Bash: cat produces no output

1. Compare with synchronous agent (same task without run_in_background=true):

• Results are returned correctly inline

Claude Model

Not sure / Multiple models

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.2

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

_No response_