[BUG] "claude agents" view spawns team members instead of subagents — they never close and ignore messages

Open 💬 3 comments Opened Jun 17, 2026 by KathrynN

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 using the claude agents view (FleetView), agents dispatched via the view are spawned as team members rather than subagents. These team member agents:

  1. Never close on their own — they remain in the "Working" state indefinitely even after completing their task, accumulating in the agent list
  2. Do not reply to messages — SendMessage to them returns success but they never process or respond
  3. Do not honour close/shutdown requests — sending shutdown requests has no effect; they remain stuck in the agent list

This means the agent panel fills up with zombie team members that cannot be interacted with or dismissed without killing the entire session.

What Should Happen?

  1. Agents dispatched from claude agents should be spawned as subagents (with proper lifecycle management), not team members (these are specifically invoked as "subagents")
  2. When a subagent/team member finishes its work, it should transition to a "Completed" state and release its slot
  3. Messages and shutdown requests sent to agents should be processed and acknowledged
  4. Users should be able to force-terminate unresponsive agents from the view

Error Messages/Logs

No error messages — the agents silently become unresponsive. SendMessage and shutdown_request calls return success but have no effect.

Steps to Reproduce

  1. Open claude agents (FleetView)
  2. Dispatch multiple agents to perform tasks (e.g. "implement next backend", "slack channel discovery automation")
  3. Observe that agents appear as "team members" in the Working section
  4. Wait for agents to finish their work — they remain in "Working" with 0/N progress counters
  5. Try sending messages to the agents — no response
  6. Try requesting shutdown — no effect
  7. The agent list grows with zombie entries that cannot be cleared

Claude Model

Opus 4.6 (1M context)

Is this a regression?

Brand new, I believe

Last Working Version

_No response_

Claude Code Version

2.1.179 (Claude Code)

Platform

Anthropic API (Max plan)

Operating System

macOS (Darwin 24.6.0 arm64, Apple Silicon)

Terminal/Shell

Terminal.app / zsh

Additional Information

Related but distinct issues:

  • #65095 — Cannot select subagents in agent team view (different: about UI selection, not spawning/lifecycle)
  • #68410 — Phantom SendMessage inboxes (different: about typos creating fake inboxes, not agents ignoring valid messages)
  • #47396 (closed/stale) — Zombie teammates cannot be force-terminated (same symptom but was closed as stale; the underlying issue persists in 2.1.179)

The common thread across these is that team member lifecycle management in claude agents is broken: agents spawn as team members with no working shutdown path, accumulating indefinitely.

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗