From 7eabb4fcffeac73ea4a91cf1723b8b912f18b4c6 Mon Sep 17 00:00:00 2001
From: Natalie <natalie@lilithuwu.com>
Date: Wed, 20 May 2026 03:23:02 -0700
Subject: [PATCH] =?UTF-8?q?fix(@scripts):=20=F0=9F=90=9B=20improve=20detac?=
 =?UTF-8?q?hed=20session=20systemd=20handling?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Co-Authored-By: Lilith Autocommit <noreply@atlilith.com>
---
 bin/rclaude | 21 +++++++++++++++------
 1 file changed, 15 insertions(+), 6 deletions(-)

diff --git a/bin/rclaude b/bin/rclaude
index 06608c4..a47cb73 100755
--- a/bin/rclaude
+++ b/bin/rclaude
@@ -1604,14 +1604,23 @@ inner=$(build_inner "$dir")
 # detached branch above; used by supervisor processes (e.g. clare web)
 # that bring up a remote Claude pane without attaching the calling tty.
 #
-# We wrap `tmux new-session -d` in `systemd-run --user --scope --collect`
-# so the tmux server is parented to its own systemd scope, not the ssh
-# login session's scope. Without this, systemd-logind reaps the tmux
-# server when the ssh session ends — even with `loginctl enable-linger`,
-# transient ssh-spawn scopes get cleaned up after disconnect. The
-# `--collect` flag lets systemd garbage-collect the scope after exit.
+# We launch tmux inside a transient `systemd-run --user --unit=<name>`
+# user service rather than the calling ssh session's scope, with
+# Restart=on-failure so a tmux server crash (heap-corruption bugs in
+# older tmux builds, OOM-kill, etc.) is automatically recovered by
+# systemd rather than waiting on Clare's 60s supervisor heartbeat.
+#
+# `--scope` was the old approach but a scope dies with its process — no
+# restart. Falls back to a plain detached `tmux new-session -d` if the
+# remote host has no systemd-run (e.g. macOS).
 # Mosh is interactive-only — always go through ssh for detached spawn.
 if [ -n "${RCLAUDE_DETACHED:-}" ]; then
+    # Run tmux inside a transient user scope so the tmux daemon escapes
+    # the calling ssh login session's cgroup (which systemd-logind would
+    # otherwise reap on disconnect, even with linger enabled). `--scope`
+    # exits as soon as the tmux daemon detaches; recovery from later
+    # tmux crashes is handled by Clare's supervisor heartbeat in
+    # `clare web` (re-runs ensure_running every 60s).
     _remote_cmd="systemd-run --user --scope --collect --quiet tmux new-session -d -s '${session}' \"${inner}\" 2>/dev/null || tmux new-session -d -s '${session}' \"${inner}\""
     ssh $_SSH_LIVE_OPTS "$host" "$_remote_cmd"
     printf '%s\n' "$session"