`claude agents` bg-spare workers crash on init in 2.1.143 (regression from 2.1.142)

Summary

In claude 2.1.143 on macOS arm64, every worker spawned via the
--bg-spare mode crashes during init within ~60–300 ms. This breaks
claude agents (the FleetView agents view), whose daemon keeps a pool
of prewarmed bg-spare workers ready and immediately marks each newly
spawned worker as (crashed). Downgrading to 2.1.142 resolves the
issue. Main entrypoints (claude --version, claude --help,
interactive claude sessions) work fine on 2.1.143 — only the
--bg-spare code path is broken.

Environment

• claude version: 2.1.143 (Claude Code), Bun v1.3.14
• Platform: macOS 15 (Darwin 25.4.0), Apple Silicon (arm64)
• Binary: ~/.local/bin/claude → ~/.local/share/claude/versions/2.1.143

Reproduction

1. Have 2.1.143 installed.
2. Run claude agents to start the agents view.
3. Tail ~/.claude/daemon.log.

You will see a tight loop of:

\\\
[bg] bg spare spawned host pid=NNNNN
[bg] bg claimed-spare <id> (spare|fleet)
[bg] bg settled <id> (crashed)
\\\

with each spare crashing 60–300 ms after spawn. The view itself never
gets a working worker.

When the spare process error surfaces, the stack trace is the minified
Bun-bundled form:

\\\
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)
\\\

Workaround

Point the launcher symlink at 2.1.142:

\\\bash
ln -sfn ~/.local/share/claude/versions/2.1.142 ~/.local/bin/claude
\\\

After killing the running \claude agents\ + \claude daemon run\ processes
and relaunching, the daemon settles cleanly with no crashed spares.

Impact

\claude agents\ is unusable on 2.1.143 — the view stays empty because
no spare ever survives long enough to take work. Other claude
invocations are unaffected, so this regression can slip past install-
time smoke tests.