Give the KAOS bridge the same E-mode treatment, and keep it current

unleashed-x-kaos/config/kaos-ams-bridge.cfg

@@ -118,6 +118,10 @@ gcode:
 description: magic_ams tool-change entry - arms the ORCA_PURGE handoff and hands over to the spit tail
 gcode:
     {% set ams = printer.save_variables.variables.ams|default(0)|int %}
+    # Read the caller's extrusion mode once, before either branch touches it. Jinja renders the whole
+    # body before the first command runs, so this is the mode on entry -- which is what both branches
+    # need to hand back.
+    {% set was_abs_e = printer.gcode_move.absolute_extrude %}
     {% if ams == 1 %}
         # ALWAYS drop a stale handoff first. A leftover active=1 from a tool change that
         # never reached its spit tail would otherwise fire ORCA_PURGE at the wrong moment.
@@ -149,6 +153,10 @@ gcode:
         # standing "G1 E-1 before M83" latent defect on this path.
         M83
         G1 E-1 F3600
+        # ...and hand the mode back. Setting M83 and keeping it would be worse than the defect it
+        # fixes: a slicer running absolute E would have every extrusion after the first tool change
+        # reinterpreted as relative, for the rest of the print.
+        {% if was_abs_e %}M82{% endif %}
 
         # --- Stage 3: hand the purge to ORCA_PURGE --------------------
         # Refresh the fan/temp mirrors ORCA_PURGE restores at the end. Runs as a
@@ -174,8 +182,11 @@ gcode:
         # is sticky per klippy session, so it MUST be re-sent on every change.
         P10 S2
     {% else %}
-        # No AMS attached: the kit's manual-swap fallback (byte-identical to stock).
+        # No AMS attached: the kit's manual-swap fallback. It used to be byte-identical to stock,
+        # which is why it carried the bare retract; the kit has since fixed that, so this follows.
+        M83
         G1 E-1 F3600
+        {% if was_abs_e %}M82{% endif %}
         M118 No AMS -> manual filament change (M600: unload -> swap -> load -> purge -> resume)
         M600
     {% endif %}

unleashed-x-kaos/scripts/kaos-guard.sh

@@ -77,6 +77,28 @@ restore_stock_dev_py() {
 # printer.cfg, a Phrozen update replacing one file but not the other); the include line is
 # ground truth for which macro set is loading. This is what makes every half-state
 # self-heal on the next boot instead of persisting silently.
+# Our own bridge configs are copied into printer_data/config when KAOS is ACTIVATED, and never
+# again. A kit update refreshes $SELF_DIR/config (it is inside the kit clone) but not the copies,
+# so a printer that has had KAOS on since before an update keeps running the old ones -- the same
+# "the correction never reaches the printer" shape this project keeps meeting. Keyed on the include
+# line actually present in printer.cfg, like sync_spit_patch above: a file whose include is gone is
+# not put back, so KAOS_OFF stays off. KAOS's OWN files come from $CACHE and are not touched here.
+sync_bridge_configs() {
+    [ -d "$SELF_DIR/config" ] && [ -f "$PRINTER_CFG" ] || return 0
+    local src name
+    for src in "$SELF_DIR"/config/*.cfg; do
+        [ -f "$src" ] || continue
+        name="$(basename "$src")"
+        grep -qE "^[[:space:]]*\[include[[:space:]]+${name//./\.}\]" "$PRINTER_CFG" 2>/dev/null || continue
+        cmp -s "$src" "$CONFIG_DIR/$name" && continue
+        if cp -f "$src" "$CONFIG_DIR/$name"; then
+            log "refreshed $name from the kit — the installed copy was older than this kit"
+        else
+            log "WARNING - could not refresh $name"
+        fi
+    done
+}
+
 sync_spit_patch() {
     [ -f "$SPIT_PATCH" ] && [ -f "$PGM" ] && [ -f "$PRINTER_CFG" ] || return 0
     local want r
@@ -141,7 +163,11 @@ sync_home_hook() {
 # ---------------------------------------------------------------------------
 [ -d "$EXTRAS" ] || exit 0                       # not an Arco / nothing to guard
 
-# Both run on EVERY path, including the two early exits below.
+# All three run on EVERY path, including the two early exits below.
+# Refresh our own bridge configs BEFORE the spit patch reconciles: that patch renames a macro in
+# printer_gcode_macro.cfg to match what the bridge declares, so the bridge has to be the current
+# one first or the two describe different files.
+sync_bridge_configs
 sync_spit_patch
 sync_home_hook