diff --git a/README.md b/README.md
index d0a064b..cebe172 100644
--- a/README.md
+++ b/README.md
@@ -78,8 +78,15 @@ help / clear / quit
 
 1. **守护脚本**自动恢复设备（掉线后重枚举）:
    ```bash
+   # 部署 (脚本在项目内 usb_guard.sh, 目标设备通用 VID/PID 可在脚本头部修改)
+   su -c 'cp ~/Proj/TransferSu/usb_guard.sh /data/local/tmp/ && chmod 755 /data/local/tmp/usb_guard.sh'
+   # 启动
    su -c 'setsid sh /data/local/tmp/usb_guard.sh </dev/null >/dev/null 2>&1 &'
    ```
+   守护脚本功能: ① 应用稳定性调优参数 (SCSI 超时 120s 等)
+   ② 三种掉线场景自动重枚举恢复 (dmesg 检测 / 异常断开 / 健康探测)。
+   默认绑定 GL3224 读卡器 (VID:PID 05e3:0747), 换设备改脚本头部变量。
+   开机自启: `~/.termux/boot/boot_usb_guard.sh` (Termux:Boot 插件)。
 2. **限速传输 + 无限补传 + 校验**:
    ```
    /src/ /dst/ --bwlimit 30000 --retry -1 --verify
diff --git a/usb_guard.sh b/usb_guard.sh
new file mode 100755
index 0000000..3bbb4ca
--- /dev/null
+++ b/usb_guard.sh
@@ -0,0 +1,92 @@
+#!/system/bin/sh
+# USB 读卡器守护脚本 v5 - 仅针对 GL3224 读卡器 (VID:PID 05e3:0747)
+# 功能: ①应用稳定性调优 ②掉线自动恢复(3场景)
+# 安全: 仅当 USB 端口 2-1 上的设备 VID/PID 匹配时才干预, 其他设备完全不受影响
+DEV_PATH="2-1"
+TARGET_VID="05e3"
+TARGET_PID="0747"
+LOGFILE="/data/local/tmp/usb_guard.log"
+AUTHORIZED="/sys/bus/usb/devices/${DEV_PATH}/authorized"
+BLKDEV=""
+LAST_HEALTH=0
+TUNED=0
+
+log() { echo "[$(date '+%F %T')] $1" >> "$LOGFILE"; }
+
+# 校验当前 2-1 口上的设备是否为目标读卡器
+is_target() {
+    [ "$(cat /sys/bus/usb/devices/${DEV_PATH}/idVendor 2>/dev/null)" = "$TARGET_VID" ] && \
+    [ "$(cat /sys/bus/usb/devices/${DEV_PATH}/idProduct 2>/dev/null)" = "$TARGET_PID" ]
+}
+
+# 动态解析该 USB 设备对应的块设备名 (如 sdg), 避免硬编码
+find_blkdev() {
+    for b in /sys/class/block/sd[a-z]; do
+        target=$(readlink -f "$b" 2>/dev/null)
+        case "$target" in
+            *"${DEV_PATH}:1.0"*) echo "${b##*/}"; return 0 ;;
+        esac
+    done
+    return 1
+}
+
+apply_tuning() {
+    [ -n "$BLKDEV" ] && [ -e "/sys/block/$BLKDEV" ] || return
+    echo 120 > /sys/block/$BLKDEV/device/timeout 2>/dev/null
+    echo 256 > /sys/block/$BLKDEV/queue/max_sectors_kb 2>/dev/null
+    echo none > /sys/block/$BLKDEV/queue/scheduler 2>/dev/null
+    TUNED=1
+    log "已应用调优参数 ($BLKDEV: timeout=120s, max_sectors=256KB, scheduler=none)"
+}
+
+recover() {
+    log "$1 -> 执行重枚举..."
+    echo 0 > "$AUTHORIZED" 2>/dev/null
+    sleep 3
+    echo 1 > "$AUTHORIZED" 2>/dev/null
+    sleep 7
+    if [ -f "$AUTHORIZED" ] && [ "$(cat "$AUTHORIZED" 2>/dev/null)" = "1" ]; then
+        log "重枚举成功，设备恢复"
+        TUNED=0
+        BLKDEV=""
+    else
+        log "重枚举失败，请手动检查"
+    fi
+    dmesg -c > /dev/null 2>&1
+}
+
+log "守护脚本启动 v5 (目标: GL3224 05e3:0747)"
+
+while true; do
+    if [ -f "$AUTHORIZED" ] && is_target; then
+        # 解析块设备名
+        [ -z "$BLKDEV" ] && BLKDEV=$(find_blkdev)
+        # 场景B: 异常断开
+        if [ "$(cat "$AUTHORIZED" 2>/dev/null)" = "0" ]; then
+            recover "场景B: 设备异常断开"
+            continue
+        fi
+        # 健康时确保调优参数
+        if [ "$TUNED" = "0" ] && [ -n "$BLKDEV" ]; then
+            apply_tuning
+        fi
+        # 场景A: SCSI offline 日志
+        if dmesg 2>/dev/null | tail -30 | grep -q "device offline or changed"; then
+            recover "场景A: 检测到设备假死"
+            continue
+        fi
+        # 场景C: 健康探测 (每12秒, 读扇区0, 5秒超时)
+        now=$(date +%s)
+        if [ $((now - LAST_HEALTH)) -ge 12 ] && [ -n "$BLKDEV" ]; then
+            LAST_HEALTH=$now
+            if ! timeout 5 dd if="/dev/block/$BLKDEV" bs=512 count=1 of=/dev/null 2>/dev/null; then
+                recover "场景C: 健康探测超时(静默假死)"
+                continue
+            fi
+        fi
+    else
+        # 非目标设备: 完全旁观, 不做任何操作
+        sleep 10
+    fi
+    sleep 2
+done