[BUG] Background agents: every daemon-spawned worker crashes "exit 1 before init" (macOS arm64)

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?

Summary

Every background-agent worker spawned by the on-demand daemon crashes immediately
with exit 1 before init. The Agents view and claude --bg are completely
unusable — no background session can start, and existing ones can't be resumed.
Interactive claude is unaffected.

Ruled out

• Not auth. The daemon separately hit auth_required (expired OAuth, headless

daemon can't refresh). Re-login + daemon restart fixed auth — the fresh daemon
authenticates cleanly — but workers still crash identically.

• Not a version regression. First seen on 2.1.142; persists on 2.1.143

(reported latest by claude update).

• Not system Bun. Updating system Bun to latest had no effect — consistent

with the crash being in the Bun runtime embedded in the compiled binary.

• Not project config. All settings.json files valid; claude -p from the

same cwd runs fine (full startup, exit 0); interactive claude is unaffected.

### Secondary issue

The crash detail stored in state.json is truncated (leading …) — the actual
error message and earlier stack frames are cut off, so a user has no way to see
what actually threw. The daemon should persist the full worker stderr.

What Should Happen?

Expected vs actual

Expected: the worker starts and runs the session.
Actual: the worker exits 1 before init; the daemon logs bg settled <id> (crashed)
~60–500ms after bg spare spawned host.

Error Messages/Logs

### daemon.log excerpt

  
  [2026-05-16T06:00:32.913Z] [supervisor] ─── daemon start ─── version=2.1.143 pid=92076 origin=transient
  [2026-05-16T06:00:33.050Z] [bg] bg spare spawned host pid=92103
  [2026-05-16T06:00:33.161Z] [bg] bg claimed-spare 66c5da22 (spare)
  [2026-05-16T06:00:33.615Z] [bg] bg settled 66c5da22 (crashed)
  [2026-05-16T06:00:54.844Z] [bg] bg claimed-spare a433d7e5 (fleet)
  [2026-05-16T06:00:54.851Z] [bg] bg spare spawned host pid=92410
  [2026-05-16T06:00:54.915Z] [bg] bg settled a433d7e5 (crashed)

Steps to Reproduce

Reproduction

1. Run claude --bg "reply with the single word OK" (any prompt).
2. The job is created, then fails within ~100ms.
3. claude logs <id> → job not found — it may have already exited.
4. ~/.claude/jobs/<id>/state.json shows:

``
exit 1 before init — …at uSK (/$bunfs/root/src/entrypoints/cli.js:9092:233) at async SKO (/$bunfs/root/src/entrypoints/cli.js:19558:68) at async h4O (/$bunfs/root/src/entrypoints/cli.js:19595:313) Bun v1.3.14 (macOS arm64)
``

100% reproducible. Also affects pre-warmed "spare" workers (no session/job
attached) and resuming existing Agents-view chats.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.143 (Claude Code)

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

iTerm2

Additional Information

_No response_