feat: Yahoo Finance SSH proxy + Xiaomi MiMo LLM support

- Add SSH proxy via foreign server for Yahoo Finance bypass
- Add fallback in load_ohlcv and get_YFin_data_online
- Add LLM JSON decode retry (3 attempts) in NormalizedChatOpenAI
- Add yahoo_proxy.py utility module
- Add test_smoke.py and patches/

tradingagents/dataflows/stockstats_utils.py

@@ -1,6 +1,8 @@
 import logging
 import os
 import time
+import json
+import subprocess
 from typing import Annotated
 
 import pandas as pd
@@ -19,6 +21,82 @@ logger = logging.getLogger(__name__)
 # enough to catch the year-old frames yfinance occasionally returns (#1021).
 MAX_OHLCV_STALE_DAYS = 10
 
+_ssh_config_cache = None
+
+def _get_ssh_config():
+    """Load SSH credentials for foreign server from vault (cached)."""
+    global _ssh_config_cache
+    if _ssh_config_cache is not None:
+        return _ssh_config_cache
+    result = subprocess.run(
+        ["bash", "-c",
+         'source <(bash /root/.openclaw/workspace/scripts/vault.sh decrypt '
+         '/root/.openclaw/workspace/secrets/credentials.env.enc) '
+         '&& echo "$FOREIGN_HOST $FOREIGN_PORT $FOREIGN_PASS"'],
+        capture_output=True, text=True, timeout=10
+    )
+    parts = result.stdout.strip().split()
+    if len(parts) != 3:
+        raise RuntimeError(f"Failed to load SSH config: {result.stderr}")
+    _ssh_config_cache = (parts[0], int(parts[1]), parts[2])
+    return _ssh_config_cache
+
+
+def _fetch_via_ssh(url: str, timeout: int = 30) -> str:
+    """Fetch a URL through the foreign server via SSH tunnel."""
+    host, port, password = _get_ssh_config()
+    cmd = [
+        "sshpass", "-p", password,
+        "ssh", "-o", "StrictHostKeyChecking=no",
+        "-o", "ConnectTimeout=15",
+        "-p", str(port), f"root@{host}",
+        f'curl -s "{url}" -H "User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36"'
+    ]
+    result = subprocess.run(cmd, capture_output=True, text=True, timeout=timeout)
+    if result.returncode != 0:
+        raise RuntimeError(f"SSH fetch failed: {result.stderr}")
+    return result.stdout
+
+
+def _yf_download_via_proxy(symbol: str, start: str, end: str) -> pd.DataFrame:
+    """Download OHLCV data via SSH proxy when local Yahoo Finance is rate-limited."""
+    import datetime
+    # Convert dates to timestamps
+    start_ts = int(pd.Timestamp(start).timestamp())
+    end_ts = int(pd.Timestamp(end).timestamp())
+    
+    url = (f"https://query2.finance.yahoo.com/v8/finance/chart/{symbol}"
+           f"?period1={start_ts}&period2={end_ts}&interval=1d")
+    
+    data_str = _fetch_via_ssh(url)
+    parsed = json.loads(data_str)
+    
+    result = parsed.get("chart", {}).get("result", [])
+    if not result:
+        return pd.DataFrame()
+    
+    ts = result[0].get("timestamp", [])
+    indicators = result[0].get("indicators", {}).get("quote", [{}])[0]
+    
+    if not ts:
+        return pd.DataFrame()
+    
+    df = pd.DataFrame({
+        "Date": pd.to_datetime(ts, unit="s"),
+        "Open": indicators.get("open", []),
+        "High": indicators.get("high", []),
+        "Low": indicators.get("low", []),
+        "Close": indicators.get("close", []),
+        "Volume": indicators.get("volume", []),
+    })
+    
+    # Convert None values to NaN
+    for col in ["Open", "High", "Low", "Close", "Volume"]:
+        df[col] = pd.to_numeric(df[col], errors="coerce")
+    
+    df = df.dropna(subset=["Close"])
+    return df
+
 
 def yf_retry(func, max_retries=3, base_delay=2.0):
     """Execute a yfinance call with exponential backoff on rate limits.
@@ -36,6 +114,7 @@ def yf_retry(func, max_retries=3, base_delay=2.0):
                 logger.warning(f"Yahoo Finance rate limited, retrying in {delay:.0f}s (attempt {attempt + 1}/{max_retries})")
                 time.sleep(delay)
             else:
+                logger.warning("Yahoo Finance local exhausted, will let caller handle fallback")
                 raise
 
 
@@ -122,6 +201,82 @@ def _assert_ohlcv_not_stale(
         )
 
 
+def _load_ohlcv_alpha_vantage(symbol: str, curr_date: str) -> pd.DataFrame:
+    """Fetch OHLCV data from Alpha Vantage, filtered to prevent look-ahead bias."""
+    from .alpha_vantage_common import _make_api_request
+    from io import StringIO
+
+    canonical = normalize_symbol(symbol)
+    curr_date_dt = pd.to_datetime(curr_date)
+    today_date = pd.Timestamp.today()
+    start_date = today_date - pd.DateOffset(years=5)
+    start_str = start_date.strftime("%Y-%m-%d")
+
+    # Determine outputsize based on how far back we need
+    days_from_today = (today_date - start_date).days
+    outputsize = "compact" if days_from_today < 100 else "full"
+
+    params = {
+        "symbol": canonical,
+        "outputsize": outputsize,
+        "datatype": "csv",
+    }
+
+    try:
+        csv_text = _make_api_request("TIME_SERIES_DAILY", params)
+    except Exception as e:
+        raise NoMarketDataError(symbol, canonical, f"Alpha Vantage error: {e}") from e
+
+    if not csv_text or not csv_text.strip():
+        raise NoMarketDataError(symbol, canonical, "Alpha Vantage returned empty data")
+
+    try:
+        df = pd.read_csv(StringIO(csv_text))
+    except Exception as e:
+        raise NoMarketDataError(symbol, canonical, f"Failed to parse Alpha Vantage CSV: {e}") from e
+
+    if df.empty:
+        raise NoMarketDataError(symbol, canonical, "Alpha Vantage returned no rows")
+
+    # Alpha Vantage CSV has 'timestamp' column; normalize to 'Date'
+    date_col = df.columns[0]  # usually 'timestamp'
+    df = df.rename(columns={date_col: "Date"})
+
+    # Keep only standard OHLCV columns
+    keep_cols = ["Date"]
+    col_map = {}
+    for av_col, std_col in [
+        ("open", "Open"), ("high", "High"), ("low", "Low"),
+        ("close", "Close"), ("adjusted_close", "Close"), ("volume", "Volume"),
+    ]:
+        for c in df.columns:
+            if c.lower().replace(" ", "_") == av_col:
+                if std_col not in keep_cols:
+                    keep_cols.append(std_col)
+                    col_map[c] = std_col
+                break
+    df = df.rename(columns=col_map)
+    # Prefer adjusted_close over close for "Close"
+    if "adjusted_close" in [c.lower().replace(" ", "_") for c in df.columns]:
+        for c in df.columns:
+            if c.lower().replace(" ", "_") == "adjusted_close":
+                df["Close"] = pd.to_numeric(df[c], errors="coerce")
+                break
+
+    available = [c for c in ["Date", "Open", "High", "Low", "Close", "Volume"] if c in df.columns]
+    df = df[available]
+
+    # Filter to curr_date
+    df["Date"] = pd.to_datetime(df["Date"], errors="coerce")
+    df = df.dropna(subset=["Date"])
+    df = df[df["Date"] <= curr_date_dt].sort_values("Date")
+
+    if df.empty:
+        raise NoMarketDataError(symbol, canonical, "No data on or before " + curr_date)
+
+    return df
+
+
 def load_ohlcv(symbol: str, curr_date: str) -> pd.DataFrame:
     """Fetch OHLCV data with caching, filtered to prevent look-ahead bias.
 
@@ -138,24 +293,24 @@ def load_ohlcv(symbol: str, curr_date: str) -> pd.DataFrame:
     config = get_config()
     curr_date_dt = pd.to_datetime(curr_date)
 
+    # Check which vendor to use for OHLCV
+    vendor_config = config.get("data_vendors", {}).get("core_stock_apis", "yfinance")
+    use_alpha_vantage = "alpha_vantage" in vendor_config
+
     # Cache uses a fixed window (5y to today) so one file per symbol.
     today_date = pd.Timestamp.today()
     start_date = today_date - pd.DateOffset(years=5)
     start_str = start_date.strftime("%Y-%m-%d")
-    # yfinance ``end`` is EXCLUSIVE; request tomorrow so today's row is included
-    # when curr_date is the current day (#986). Look-ahead is still prevented by
-    # the curr_date filter below.
     end_str = (today_date + pd.Timedelta(days=1)).strftime("%Y-%m-%d")
 
+    cache_prefix = "AV" if use_alpha_vantage else "YFin"
     os.makedirs(config["data_cache_dir"], exist_ok=True)
     data_file = os.path.join(
         config["data_cache_dir"],
-        f"{safe_symbol}-YFin-data-{start_str}-{end_str}.csv",
+        f"{safe_symbol}-{cache_prefix}-data-{start_str}-{end_str}.csv",
     )
 
-    # A cached file may be empty if a prior fetch failed (unknown symbol,
-    # transient rate limit). Treat an empty/columnless cache as a miss and
-    # re-fetch rather than serving the poisoned file forever.
+    # A cached file may be empty if a prior fetch failed. Treat empty cache as miss.
     data = None
     if os.path.exists(data_file):
         cached = pd.read_csv(data_file, on_bad_lines="skip", encoding="utf-8")
@@ -163,30 +318,53 @@ def load_ohlcv(symbol: str, curr_date: str) -> pd.DataFrame:
             data = cached
 
     if data is None:
-        downloaded = yf_retry(lambda: yf.download(
-            canonical,
-            start=start_str,
-            end=end_str,
-            multi_level_index=False,
-            progress=False,
-            auto_adjust=True,
-        ))
-        downloaded = _ensure_date_column(downloaded.reset_index())
-        # Only cache real data — never persist an empty frame.
-        if downloaded.empty or "Close" not in downloaded.columns:
-            raise NoMarketDataError(
-                symbol, canonical, "Yahoo Finance returned no rows"
-            )
+        if use_alpha_vantage:
+            downloaded = _load_ohlcv_alpha_vantage(symbol, curr_date)
+        else:
+            downloaded = None
+            try:
+                raw = yf_retry(lambda: yf.download(
+                    canonical,
+                    start=start_str,
+                    end=end_str,
+                    multi_level_index=False,
+                    progress=False,
+                    auto_adjust=True,
+                ))
+                downloaded = _ensure_date_column(raw.reset_index())
+            except Exception as yf_err:
+                pass  # Will try proxy below
+
+            # If local yfinance failed or returned empty, try SSH proxy
+            if downloaded is None or downloaded.empty or "Close" not in downloaded.columns:
+                if downloaded is None or downloaded.empty:
+                    logger.warning(f"Yahoo Finance local failed, trying SSH proxy...")
+                try:
+                    downloaded = _yf_download_via_proxy(canonical, start_str, end_str)
+                except Exception as proxy_err:
+                    raise NoMarketDataError(
+                        symbol, canonical, f"Yahoo Finance proxy also failed: {proxy_err}"
+                    ) from proxy_err
+                if downloaded.empty:
+                    raise NoMarketDataError(
+                        symbol, canonical, f"Yahoo Finance proxy also returned no data"
+                    )
+            if downloaded.empty or "Close" not in downloaded.columns:
+                raise NoMarketDataError(
+                    symbol, canonical, "Yahoo Finance returned no rows"
+                )
+            downloaded = _clean_dataframe(downloaded)
+
         downloaded.to_csv(data_file, index=False, encoding="utf-8")
         data = downloaded
 
-    data = _clean_dataframe(data)
+    if not use_alpha_vantage:
+        data = _clean_dataframe(data)
 
     # Filter to curr_date to prevent look-ahead bias in backtesting
     data = data[data["Date"] <= curr_date_dt]
 
-    # Reject a stale frame (latest row far older than curr_date) rather than
-    # feeding year-old prices into indicators (#1021).
+    # Reject a stale frame
     _assert_ohlcv_not_stale(data, curr_date, symbol, canonical)
 
     return data