fix(health): drop live LLM ping — scan agent.log instead (38s -> 0.1s)

Every /api/xmpp/health fetch ran a real LLM call (22k token system prompt
each). Dashboard refreshes every 10s -> thousands of paid LLM calls/day,
plus 38s latency hanging the health tab on '加载中...'.

LLM health is now derived from the gateway's own agent.log (zero cost,
more accurate than synthetic ping — real traffic results):
- last 'API call #N latency=Xs' -> ok
- last 'API call failed ... HTTP 429...' -> error with summary
- health() runtime 38s -> 0.1s; endpoint 38s -> 0.097s

xmpp_logger.py

@@ -338,20 +338,9 @@ def health():
             except Exception: pass
             result["gateways"][name] = {"port": port, "alive": ok}
 
-        # 测试知微的 LLM 调用是否可达
-        if result["gateways"].get("zhiwei", {}).get("alive"):
-            try:
-                import urllib.request
-                payload = json.dumps({"model": "deepseek-v4-flash", "messages": [{"role": "user", "content": "ping"}],
-                                      "max_tokens": 5, "stream": False}).encode()
-                req = urllib.request.Request("http://127.0.0.1:8643/v1/chat/completions",
-                                            data=payload, headers={"Content-Type": "application/json",
-                                            "Authorization": "Bearer hermes123"})
-                urllib.request.urlopen(req, timeout=90)
-                result["llm_provider"] = {"status": "ok", "latency": "fast"}
-            except Exception as e:
-                result["llm_provider"] = {"status": "timeout" if "timeout" in str(e).lower() else "error",
-                                          "error": str(e)[:150]}
+        # LLM 健康：扫 gateway agent.log 最近一次真实调用结果（零成本，不发 LLM 请求）
+        # agent.log 里每次真实调用都有记录：成功 "API call #N: ... latency=Xs" / 失败 "HTTP 429..."
+        result["llm_provider"] = _scan_agent_log(now_epoch)
     except Exception as e:
         result["gateways"] = {"error": str(e)[:100]}
 
@@ -376,6 +365,61 @@ def health():
     return result
 
 
+AGENT_LOG = Path("/home/hmo/.hermes/profiles/position-analyst/logs/agent.log")
+
+
+def _scan_agent_log(now_epoch, tail_lines=300):
+    """扫 gateway agent.log 尾部，取最近一次真实 LLM 调用的结果。
+
+    成功行: "2026-07-19 20:04:42,880 INFO ... API call #4: ... latency=16.0s ..."
+    失败行: "2026-07-19 19:35:59 WARNING ... API call failed ... HTTP 429: Weekly usage limit reached ..."
+    返回 {"status": "ok"|"error"|"unknown", "latency"/"error", "age_sec"}
+    """
+    if not AGENT_LOG.exists():
+        return {"status": "unknown", "error": "agent.log not found"}
+    try:
+        # 只读尾部（大文件不全读）
+        size = AGENT_LOG.stat().st_size
+        with open(AGENT_LOG, "rb") as f:
+            f.seek(max(0, size - 64 * 1024))
+            chunk = f.read().decode("utf-8", errors="replace")
+        lines = [l for l in chunk.splitlines() if l.strip()][-tail_lines:]
+    except Exception as e:
+        return {"status": "unknown", "error": str(e)[:80]}
+
+    last_ok = None      # (epoch, latency_str)
+    last_fail = None    # (epoch, summary)
+    for line in lines:
+        m_ts = _re.search(r"(\d{4}-\d{2}-\d{2} \d{2}:\d{2}:\d{2})", line)
+        if not m_ts:
+            continue
+        try:
+            ts = datetime.strptime(m_ts.group(1), "%Y-%m-%d %H:%M:%S").timestamp()
+        except Exception:
+            continue
+        m_ok = _re.search(r"API call #\d+.*latency=([\d.]+)s", line)
+        if m_ok:
+            last_ok = (ts, m_ok.group(1))
+            continue
+        if "API call failed" in line and ("summary=" in line or "after" in line):
+            m_err = _re.search(r"summary=(.+?)$", line)
+            summary = (m_err.group(1) if m_err else line)[:150]
+            last_fail = (ts, summary)
+
+    if not last_ok and not last_fail:
+        return {"status": "unknown", "error": "no LLM calls in log"}
+
+    if last_fail and (not last_ok or last_fail[0] > last_ok[0]):
+        return {"status": "error",
+                "error": last_fail[1],
+                "age_sec": int(now_epoch - last_fail[0]),
+                "source": "agent.log"}
+    return {"status": "ok",
+            "latency": f"{last_ok[1]}s",
+            "age_sec": int(now_epoch - last_ok[0]),
+            "source": "agent.log"}
+
+
 def _fetch_keys():
     """从 AgentsMeeting Dashboard 获取可用 API Key 列表"""
     try: