Claude Code Background Subagents Now Surface Permission Prompts in Your Main Session
Claude CodeClaude Code v2.1.186 changes how background subagents handle permission requests. Previously they would auto-deny any tool requiring approval, causing silent failures. Now those prompts surface in the main session, labeled with which subagent is asking. Pressing Esc denies only that specific tool call, leaving the rest of the background session intact.
Sources & Mentions
5 external resources covering this update
Background subagents don't prompt for tool permissions (Issue #34095)
GitHub
Background subagents continue after stop (Issue #28570)
GitHub
Claude Code Subagents: How to Create, Use, and Debug Them
Builder.io
Claude Code Subagents: A 2026 Practical Guide
Tembo
I ignored Claude Code's subagents until I realized what I was missing
XDA Developers
Background subagents stop failing silently
Before v2.1.186, any tool call inside a background subagent that needed user approval was silently auto-denied. The background task would simply fail, often with no visible output, leaving users to guess why a parallel agent produced nothing.
Prompts now route to the main session
With this release, those permission prompts surface in your main session and are labeled with the identity of the subagent making the request. You approve or deny each call in context. Pressing Esc denies only that specific tool call, leaving all other background agents and the main session running.
Why it matters
Long-running parallel agent teams become far more reliable because failures are now visible rather than silent. The change also enforces Agent(type) deny rules and Agent(x,y) restrictions for named subagents, closing a prior gap, and it aligns background subagents with foreground subagents, which have always passed permission prompts through to the user.