diff --git a/packages/secubox-aggregator/debian/changelog b/packages/secubox-aggregator/debian/changelog
index e3bdbd97..0ba9cf43 100644
--- a/packages/secubox-aggregator/debian/changelog
+++ b/packages/secubox-aggregator/debian/changelog
@@ -1,3 +1,14 @@
+secubox-aggregator (0.2.3-1~bookworm1) bookworm; urgency=medium
+
+  * #727 auto-heal watchdog: ship secubox-aggregator-watchdog.{sh,service,timer}.
+    The in-process aggregator is the hub/auth/menu SPOF; under a host load spike
+    its event loop can wedge and the socket stops answering (board-wide 502/000:
+    sparse navbar, login errors). The timer probes aggregator.sock every 2 min
+    and restarts the service after 2 consecutive failures. Enabled in postinst
+    (respects operator masking). Packages the live fix from the 2026-06-24 incident.
+
+ -- Gerald KERMA <devel@cybermind.fr>  Wed, 24 Jun 2026 15:10:00 +0000
+
 secubox-aggregator (0.2.1-1~bookworm1) bookworm; urgency=medium
 
   * Phase 7 follow-up (#498) — relax hardening for module sudoers :
diff --git a/packages/secubox-aggregator/debian/postinst b/packages/secubox-aggregator/debian/postinst
index 3124c926..e8b8a22d 100755
--- a/packages/secubox-aggregator/debian/postinst
+++ b/packages/secubox-aggregator/debian/postinst
@@ -17,6 +17,12 @@ case "$1" in
     systemctl enable secubox-aggregator.service
     systemctl start secubox-aggregator.service || true
 
+    # Auto-heal watchdog (#727): restart the aggregator if its socket wedges
+    # under load (the hub/auth/menu SPOF). Respect operator masking.
+    if [ "$(systemctl is-enabled secubox-aggregator-watchdog.timer 2>/dev/null)" != "masked" ]; then
+        systemctl enable --now secubox-aggregator-watchdog.timer 2>/dev/null || true
+    fi
+
     echo "secubox-aggregator: to migrate all installed SecuBox modules into"
     echo "  the aggregator (replaces per-module uvicorn processes) run :"
     echo "    sudo /usr/sbin/secubox-aggregator-migrate"
diff --git a/packages/secubox-aggregator/debian/rules b/packages/secubox-aggregator/debian/rules
old mode 100644
new mode 100755
index a829047e..9be08274
--- a/packages/secubox-aggregator/debian/rules
+++ b/packages/secubox-aggregator/debian/rules
@@ -14,6 +14,12 @@ override_dh_auto_install:
 	install -d $(CURDIR)/debian/secubox-aggregator/lib/systemd/system
 	install -m 644 systemd/secubox-aggregator.service \
 	  $(CURDIR)/debian/secubox-aggregator/lib/systemd/system/
+	install -m 644 systemd/secubox-aggregator-watchdog.service \
+	  $(CURDIR)/debian/secubox-aggregator/lib/systemd/system/
+	install -m 644 systemd/secubox-aggregator-watchdog.timer \
+	  $(CURDIR)/debian/secubox-aggregator/lib/systemd/system/
 	install -d $(CURDIR)/debian/secubox-aggregator/usr/sbin
 	install -m 755 sbin/secubox-aggregator-migrate \
 	  $(CURDIR)/debian/secubox-aggregator/usr/sbin/
+	install -m 755 sbin/secubox-aggregator-watchdog.sh \
+	  $(CURDIR)/debian/secubox-aggregator/usr/sbin/
diff --git a/packages/secubox-aggregator/debian/secubox-aggregator.postinst.debhelper b/packages/secubox-aggregator/debian/secubox-aggregator.postinst.debhelper
new file mode 100644
index 00000000..0349e052
--- /dev/null
+++ b/packages/secubox-aggregator/debian/secubox-aggregator.postinst.debhelper
@@ -0,0 +1,10 @@
+
+# Automatically added by dh_python3
+if command -v py3compile >/dev/null 2>&1; then
+	py3compile -p secubox-aggregator 
+fi
+if command -v pypy3compile >/dev/null 2>&1; then
+	pypy3compile -p secubox-aggregator  || true
+fi
+
+# End automatically added section
diff --git a/packages/secubox-aggregator/debian/secubox-aggregator.prerm.debhelper b/packages/secubox-aggregator/debian/secubox-aggregator.prerm.debhelper
new file mode 100644
index 00000000..5d4b4907
--- /dev/null
+++ b/packages/secubox-aggregator/debian/secubox-aggregator.prerm.debhelper
@@ -0,0 +1,10 @@
+
+# Automatically added by dh_python3
+if command -v py3clean >/dev/null 2>&1; then
+	py3clean -p secubox-aggregator 
+else
+	dpkg -L secubox-aggregator | sed -En -e '/^(.*)\/(.+)\.py$/s,,rm "\1/__pycache__/\2".*,e'
+	find /usr/lib/python3/dist-packages/ -type d -name __pycache__ -empty -print0 | xargs --null --no-run-if-empty rmdir
+fi
+
+# End automatically added section
diff --git a/packages/secubox-aggregator/sbin/secubox-aggregator-watchdog.sh b/packages/secubox-aggregator/sbin/secubox-aggregator-watchdog.sh
new file mode 100755
index 00000000..87890a76
--- /dev/null
+++ b/packages/secubox-aggregator/sbin/secubox-aggregator-watchdog.sh
@@ -0,0 +1,48 @@
+#!/bin/bash
+# SPDX-License-Identifier: LicenseRef-CMSD-1.0
+# Copyright (c) 2026 CyberMind — Gérald Kerma <devel@cybermind.fr>
+# Source-Disclosed License — All rights reserved except as expressly granted.
+# See LICENCE-CMSD-1.0.md for terms.
+#
+# SecuBox-Deb :: secubox-aggregator-watchdog
+#
+# Auto-heal the in-process aggregator — the hub/auth/menu single point of
+# failure. Under a host load spike its shared event loop can wedge and its
+# socket stops answering, taking down the navbar, login and service status
+# board-wide (incident 2026-06-24). Probe the socket; if /api/v1/hub/public/menu
+# stops answering for N consecutive checks, restart the service. Idempotent,
+# safe to run on a timer.
+set -uo pipefail
+readonly MODULE="secubox-aggregator-watchdog"
+readonly VERSION="1.0"
+
+SOCK="/run/secubox/aggregator.sock"
+# State lives in /run (root-owned), NOT the shared sticky /run/secubox: that dir
+# is 1777 and a stale secubox-owned file there can't be overwritten by this
+# (CSPN-hardened, CAP_DAC_OVERRIDE-less) root — which would silently freeze the
+# streak counter and stop the watchdog ever triggering.
+STATE="/run/secubox-aggregator-watchdog.fails"
+FAIL_THRESHOLD="${SECUBOX_AGG_WD_THRESHOLD:-2}"
+TIMEOUT="${SECUBOX_AGG_WD_TIMEOUT:-12}"
+
+# No socket yet (service still starting / not migrated) → nothing to heal.
+[ -S "$SOCK" ] || exit 0
+
+code=$(curl -s -o /dev/null -w "%{http_code}" --max-time "$TIMEOUT" \
+        --unix-socket "$SOCK" http://localhost/api/v1/hub/public/menu 2>/dev/null || echo 000)
+
+if [ "$code" = "200" ]; then
+    echo 0 > "$STATE" 2>/dev/null || true
+    exit 0
+fi
+
+n=$(( $(cat "$STATE" 2>/dev/null || echo 0) + 1 ))
+echo "$n" > "$STATE" 2>/dev/null || true
+logger -t "$MODULE" "aggregator probe failed (code=$code, streak=$n/$FAIL_THRESHOLD)"
+
+if [ "$n" -ge "$FAIL_THRESHOLD" ]; then
+    logger -t "$MODULE" "restarting secubox-aggregator (auto-heal)"
+    systemctl restart secubox-aggregator.service
+    echo 0 > "$STATE" 2>/dev/null || true
+fi
+exit 0
diff --git a/packages/secubox-aggregator/systemd/secubox-aggregator-watchdog.service b/packages/secubox-aggregator/systemd/secubox-aggregator-watchdog.service
new file mode 100644
index 00000000..4ee690db
--- /dev/null
+++ b/packages/secubox-aggregator/systemd/secubox-aggregator-watchdog.service
@@ -0,0 +1,9 @@
+[Unit]
+Description=SecuBox aggregator auto-heal watchdog
+Documentation=https://github.com/CyberMind-FR/secubox-deb/issues/727
+After=secubox-aggregator.service
+
+[Service]
+Type=oneshot
+ExecStart=/usr/sbin/secubox-aggregator-watchdog.sh
+Nice=10
diff --git a/packages/secubox-aggregator/systemd/secubox-aggregator-watchdog.timer b/packages/secubox-aggregator/systemd/secubox-aggregator-watchdog.timer
new file mode 100644
index 00000000..782fd2ca
--- /dev/null
+++ b/packages/secubox-aggregator/systemd/secubox-aggregator-watchdog.timer
@@ -0,0 +1,10 @@
+[Unit]
+Description=Probe + auto-heal secubox-aggregator every 2 min
+
+[Timer]
+OnBootSec=2min
+OnUnitActiveSec=2min
+AccuracySec=20s
+
+[Install]
+WantedBy=timers.target