[BUG] Background subagents silently auto-deny permissions (Write tool)

Status Closed — not planned
Reported on v2.1.71
Maintainer reply None cached
Activity 12 comments · opened Mar 9, 2026 · closed Jun 19, 2026

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 running subagents in background mode, permission requests for tool use are silently auto-denied without user prompts. Subagents report successful completion but operations never execute.
This appears to be a broader issue affecting multiple tool types in background mode (related to #30693 which affects Bash permissions).

Current behavior:

  • Subagents silently auto-deny Write permissions across all scenarios
  • Return success status despite operations failing
  • No user notification or error messages
  • Main agents detects failed tool usage after it read all subagent outputs

What Should Happen?

  • Permissions should be inherited from parent session, OR
  • User should be prompted for permission (forwarded to foreground), OR
  • Background subagents should fail with clear error messages indicating permission denial

Error Messages/Logs

No error provided

Steps to Reproduce

  1. Create multiple subagents to generate and write files
  2. Run them in parallel using background task execution
  3. Monitor subagent execution - reports show success
  4. Verify files exist on disk - they don't

Occurs in multiple scenarios:

  • Subagents with no tool allows provided (general purpose)
  • Subagents with tool allows explicitly configured (custom subagents)
  • Subagents initially run in foreground, allowed tool use and then moved to background afterwards
  • Subagents created with run_in_background: true from start

Claude Model

Not sure / Multiple models

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

Claude Code ( 2.1.71)

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

Windows Terminal

Additional Information

_No response_

View original on GitHub ↗

11 Comments

github-actions[bot] · 5 months ago

Found 3 possible duplicate issues:

  1. https://github.com/anthropics/claude-code/issues/27203
  2. https://github.com/anthropics/claude-code/issues/26851
  3. https://github.com/anthropics/claude-code/issues/30693

This issue will be automatically closed as a duplicate in 3 days.

  • If your issue is a duplicate, please close it and 👍 the existing issue instead
  • To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

thoeltig · 5 months ago

My issue is about Cloude Code when the main agents invokes the subagents. I don't know if this is the same internal logic or if these are different topics but they all share the same problem for the user that permissions are auto-denied silently.

christian1koch · 5 months ago

same issue, it never prompted anything, it just autodenied it

<img width="449" height="210" alt="Image" src="https://github.com/user-attachments/assets/85e48d4f-4953-4c56-87f2-2630cfcd3473" />

houserooms · 5 months ago

Still broken as of 25 March 2026. Filed #38859 (agent permissions) and #38662 (parent session permissions) with fresh diagnostic detail.

thoeltig · 4 months ago

The problem still occurs. This would be fixed if the feature #47339 was implemented.

cmc4ew · 4 months ago

still happening to me as well

nikhilsitaram · 4 months ago

Still happening!

betovildoza · 3 months ago

Consolidating discussion of this cluster (15+ open issues sharing the same root cause in the background-subagent permissions resolver) at #50267 — please add 👍 there if affected, so triage can see the aggregate impact.

BrendanC23 · 3 months ago

This is still a problem. Not stale.

github-actions[bot] · 2 months ago

Closing for now — inactive for too long. Please open a new issue if this is still relevant.

craiggill-sd · 2 months ago

Crazy the AI company that has all these AI resources closes COMPLETELY VALID issues with 0 human or AI agent resolution other an 'Hurr durr reopen because stale'.

Showing cached comments. Read the full discussion on GitHub ↗