[BUG] Claude Code process exits with code 143 (SIGTERM) every exactly 5 minutes in Desktop app & VS Code extension — terminal CLI unaffected

Bug description

Claude Code child process spawned by the Desktop app (and VS Code extension) is killed with SIGTERM (exit code 143) at exactly 300-second intervals. The terminal CLI (claude in shell) works perfectly — the issue is exclusive to the Desktop app and VS Code extension wrapper.

Evidence from logs (~/Library/Logs/Claude/main.log)

Crash pattern — exact 5-minute intervals

2026-05-25 17:13:57  → crash
2026-05-25 17:18:58  → crash (gap: 300s)
2026-05-25 17:23:58  → crash (gap: 300s)
2026-05-25 17:28:58  → crash (gap: 300s)
2026-05-25 17:38:58  → crash (gap: 300s after manual restart)
2026-05-25 17:43:58  → crash (gap: 300s)
2026-05-25 17:48:58  → crash (gap: 300s)
2026-05-25 17:53:59  → crash (gap: 300s)
2026-05-25 17:58:59  → crash (gap: 300s)
2026-05-25 18:03:59  → crash (gap: 300s)

Same pattern observed on May 4, May 15, and May 22 — not a one-time occurrence.

Error from main.log

[error] Session local_8de989a6-... query error: Claude Code process exited with code 143 {
  stack: 'Error: Claude Code process exited with code 143
    at d4i.getProcessExitError (.../app.asar/.vite/build/index.js:459:8311)
    at ChildProcess.i (.../app.asar/.vite/build/index.js:459:11562)
    at Object.onceWrapper (node:events:631:26)
    at ChildProcess.emit (node:events:521:24)
    at ChildProcess._handle.onexit (node:internal/child_process:295:12)'
}

CycleHealth shows it's a timer, not response timeout

Some queries complete successfully in 60–93s, while others get killed at the exact 5-minute mark regardless of progress:

17:23:12 healthy cycle (68s, hadFirstResponse=true)    ← OK
17:26:33 healthy cycle (64s, hadFirstResponse=true)    ← OK
17:51:16 healthy cycle (93s, hadFirstResponse=true)    ← OK
17:28:58 unhealthy cycle (62s, reason=system_error)    ← killed at 5-min mark
17:58:59 unhealthy cycle (34s, reason=system_error)    ← killed at 5-min mark

Not macOS Jetsam / OOM

• macOS system log (log show) shows zero Jetsam/memorystatus/kill entries for Claude processes
• memory_pressure reports 39–48% free after closing background apps
• Crash persists regardless of memory state

Not OAuth/auth issue

Logs showed [remoteManagedSettings] fetch returned 401 concurrently, but re-login fixed the 401 errors while 143 crashes continued unchanged at the same 5-minute interval.

Reproduction steps

1. Open Claude Desktop app (v1.8555.2) on macOS
2. Start or resume a Claude Code session (tested with ~115 transcript messages, 126 tools, 11 MCP servers, 4 plugins)
3. Use the session normally — send messages, run tools
4. Observe: process exits with code 143 every exactly 300 seconds
5. Session auto-resumes but crashes again at the next 5-minute mark
6. Run the same session via terminal CLI (claude --resume <session-id>) → no crash

Expected behavior

The Claude Code child process should not be killed by a 300-second timer while actively processing or idle. The process should remain alive until explicitly terminated by the user or by a legitimate resource constraint.

System information

• Claude Desktop app: v1.8555.2
• Claude Code CLI: v2.1.149
• macOS: 26.5 (Build 25F71)
• Chip: Apple M3
• RAM: 16 GB
• Model: claude-opus-4-6 with --effort max

Session details

• Session had 115 transcript messages at time of crashes
• 126 tools loaded, 11 MCP servers, 4 plugins (skills, remote, 2 local LSP)
• CLI session ID: a9c4e41e-c8d3-491b-8124-54833635b2d1

Possibly related

• #45717 — Bash tool timeout SIGTERM propagation kills parent process
• #53136 — claude --print -p exits 143 after ~14 tool round-trips

Workaround

Use the terminal CLI instead of Desktop app / VS Code extension for long or heavy sessions.