reference reaper homebase socket sweep bug 20260715T212817

The orphan reaper (scripts/forge_reap_orphan_claudes.py, sweep_stale_sockets()) scanned every socket in /tmp/tmux-1000/ and unlinked any that didn't answer a tmux ls probe within 5s, with no exclusion for the homebase session. The idle Opus TUI often doesn't answer fast enough, so the reaper deleted its live socket every hour at :14/:15. Unlinking the socket doesn't kill the tmux server/Claude process, so this also caused a slow process leak (session count crept 17→21). The dedicated forge_tmux_socket_sweep.sh already protects homebase (two-strike probe + self-heal) but the reaper never got the same guard, so the sweep would see "no socket → down" at :20 and restart it, producing ~2 notifications/hour (~48/day) since 2026-07-14.

Why: Root-caused during a session investigating notification spam; fix was a one-line guard (PROTECTED_SOCKET_NAMES = ("homebase",)) added and verified (0 stale sockets removed on manual run), committed as ecf9bd2 and pushed to forge main.

How to apply: If home base flaps again or reaper/socket-sweep logic changes, check this exclusion still exists before assuming a new bug. forge-tmux-anchor.service was separately still failed (since 2026-07-14 06:52) as of this fix, cosmetic since the sweep keeps home base alive without it, but worth revisiting if that surfaces again.

[auto-memory session e889b9fd-c867-426f-af9a-9c04caa65c9c, confidence 0.75, mode staged]