Claude Code reaps idle background shells under memory pressure: restarting a background dev server

Observed in Claude Code (Nov 2026, macOS): a long-running background Bash task (a local node dev server started with run_in_background) was stopped while the session was idle, with the notice 'stopped because the system is running low on memory'. The notice says this is not a failure of the command. Don't auto-restart; report it and restart only when the user asks. The behaviour can be disabled by starting Claude Code with CLAUDE_CODE_DISABLE_BG_SHELL_PRESSURE_REAP=1 in its environment; setting it from a shell command inside the session has no effect.

When the user asks to restart: - check headroom first (memory_pressure | tail -1 shows the free percentage on macOS) and that the port is free (lsof -iTCP:<port> -sTCP:LISTEN); - restart the same command in the background; don't repeat one-time setup steps from the original (my original moved a log file aside first); - wait for readiness with a bounded curl loop (perl -e 'alarm 30; exec @ARGV' bash -c 'until curl -s -o /dev/null -w %{http_code} URL | grep -q 200; do sleep 1; done'; macOS has no timeout).

Any watcher keyed on the server's output files (e.g. yield-on-user-request) keeps working after the restart, provided the paths are the same.