[Bug] Background daemon supervisor respawns every ~52s on macOS Apple Silicon, causing agent disconnects and task failures

Open 💬 3 comments Opened Jun 29, 2026 by Haohaokankan123

Bug Description
Background-agent daemon respawns every ~52s on macOS (Apple Silicon) — constant "Reconnecting…" + agents randomly turning red

Claude Code 2.1.195 (also reproduced on 2.1.193), macOS Apple Silicon, native install.

When I run agents in the Agents View, the background daemon supervisor dies and
respawns roughly every 52 seconds, nonstop. Every respawn makes my Agents View
flash "Reconnecting…" (freezes input ~1s, occasionally drops me from a session),
and when the respawn is a hard one it kills background workers, so agents suddenly
turn red / "failed" mid-task.

What ~/.claude/daemon.log shows:

  • A brand-new daemon supervisor (new pid) every ~52s, rebinding the same control

socket, for hours.

  • Shutdown reason is "cause=signal" — e.g.

"[supervisor] shutting down (cause=signal, uptime=6s, leases=2, live_workers=5)".

  • It also respawns with leases=0 (zero active agents), so it's not just multi-agent

contention.

  • On the red crashes the log shows workers dying, e.g. "bg adopt: adopted=2 dead=5".
  • The live daemon's parent process is a claude process (not launchd/pid 1) — it's

not detached, so when the short-lived helper that spawned it exits, the daemon is
SIGTERM'd and a replacement respawns.

Onset: it started suddenly at 2026-06-28 ~18:40 UTC after weeks of normal behavior
(daemon starts went from ~1/hour to ~96/hour and stayed there). It began the moment
I started keeping agents running continuously in the Agents View.

Things that did NOT fix it: full daemon restart, downgrading to 2.1.193, disabling
the autoupdater, reducing agent count, and manually pinning a detached
claude daemon run (origin=foreground) — the Agents View re-seizes the daemon and
kills the pinned one within ~15s.

Impact: makes running multiple background agents unusable — constant reconnect
flicker, dropped input, and agents dying mid-task. (Work isn't lost — red jobs
resume — but it's very disruptive.)

Environment Info

  • Platform: darwin
  • Terminal: xterm-256color
  • Version: 2.1.195
  • Feedback ID: da61aab9-c73f-4265-aa78-5919d3e3b5c2

Errors

[]

View original on GitHub ↗

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