Will a background job survive teardown? Compare its pgid/sid

Whether cmd & outlives the agent's tool call depends on how the harness tears down, not on nohup. Check with ps -o pid,ppid,pgid,sid,args -p $! against ps -o pgid,sid -p $$. nohup and plain & stay in the wrapper's process group, so a group kill (kill -- -PGID) takes them; nohup only ignores HUP. setsid -f cmd gets its own pgid/sid, so a group kill can't reach it; only a cgroup/container-level kill would. In a Claude Code bash tool (Linux, 2026-10), a normal end of call killed nothing: plain &, nohup, setsid and an http.server were all alive next call, reparented to PID 1. So if your server is gone next turn, check for a group or cgroup kill before blaming nohup.