diff --git a/scripts/cron_health_monitor.py b/scripts/cron_health_monitor.py
index 32701435..1402e8ab 100644
--- a/scripts/cron_health_monitor.py
+++ b/scripts/cron_health_monitor.py
@@ -172,7 +172,33 @@ def main():
             alerts.append(f"🔴 XMPP bridge检测失败: {e}")
             state[bridge_key] = {"last_alerted": datetime.now().timestamp()}
 
-    # ── 检查4：price_monitor数据新鲜度（直接查DB，不依赖job记录）─
+    # ── 检查4：gateway LLM推理是否正常（发真实请求测响应时间） ─
+    gateway_key = "gateway_down"
+    try:
+        import urllib.request, json, time
+        payload = json.dumps({
+            "model": "position-analyst",
+            "messages": [{"role": "user", "content": "ok"}],
+            "max_tokens": 5
+        }).encode()
+        req = urllib.request.Request("http://127.0.0.1:8643/v1/chat/completions",
+            data=payload,
+            headers={"Authorization": "Bearer hermes123", "Content-Type": "application/json"})
+        t0 = time.time()
+        resp = urllib.request.urlopen(req, timeout=30)
+        elapsed = time.time() - t0
+        data = json.loads(resp.read())
+        usage = data.get("usage", {})
+        tok = usage.get("total_tokens", 0)
+        if elapsed > 25 and is_new_alert(gateway_key, state):
+            alerts.append(f"🔴 gateway响应慢: {elapsed:.0f}s ({tok}tok)")
+            state[gateway_key] = {"last_alerted": datetime.now().timestamp()}
+    except Exception as e:
+        if is_new_alert(gateway_key, state):
+            alerts.append(f"🔴 gateway推理失败: {e}")
+            state[gateway_key] = {"last_alerted": datetime.now().timestamp()}
+
+    # ── 检查5：price_monitor数据新鲜度（直接查DB，不依赖job记录）─
     price_key = "price_stale"
     try:
         import sqlite3