#!/usr/bin/env python3
"""
stale_push_wlin.py — 按5步逻辑推送自选股买入区提醒 + 自动触发重评

5步逻辑：
1. 筛选 is_watchlist=true 且价在买入区
2. RR<1.5/无止盈位/非买入signal → 标记 STRATEGY_STALE → 触发自动重评
3. 可推的：计算每手买入金额和现金占比
4. 发现 STRATEGY_STALE → 后台跑 per_stock_reassess.py 自动重评

no_agent模式：有推送→输出；无→静默
搭配 cron: no_agent=True, 交易日每30分跑一次
"""
import subprocess
import sys
import re
import json
import os
import threading
import time
from datetime import datetime, time

# 市场时段检查
_MARKET_HOURS = {
    'ashare': (time(9, 30), time(15, 0)),
    'hk':     (time(9, 30), time(16, 0)),
}

def is_ashare(code: str) -> bool:
    """判断是否A股代码"""
    return code.isdigit() and (code.startswith(('6', '5')) or len(code) in (6,))

def market_is_open(code: str, now: datetime = None) -> bool:
    """检查某股票对应市场是否在交易时段内"""
    if not code:
        return True
    now = now or datetime.now()
    t = now.time()
    code_str = str(code)
    if code_str.startswith(('0', '1')) and len(code_str) == 5:
        # 港股
        start, end = _MARKET_HOURS['hk']
    else:
        # A股（含ETF、科创板）
        start, end = _MARKET_HOURS['ashare']
    return start <= t <= end
try:
    from urllib.request import Request, urlopen
except ImportError:
    from urllib2 import Request, urlopen
# 6维评分系统
sys.path.insert(0, "/home/hmo/MoFin/scripts")
from stock_scorer import score_future_outlook, is_hk_stock, settlement_delay_note

# ── 趋势检查 ────────────────────────────────────────────────────
def fetch_trend_data(code):
    """取均线数据判断趋势状态。返回 (current_price, ma5, trend_label) 或 None"""
    try:
        prefix = "sh" if code.startswith(('60','68','51','56','50')) else "sz" if code.startswith(('00','30','15')) else "hk"
        url = f"http://qt.gtimg.cn/q={prefix}{code}"
        req = Request(url, headers={'User-Agent': 'Mozilla/5.0'})
        resp = urlopen(req, timeout=5).read().decode('gbk')
        fld = resp.split('=')[1].strip().strip('"').strip(';').split('~')
        current = float(fld[3]) if len(fld) > 3 else 0
    except:
        return None

    try:
        url = f"http://ifzq.gtimg.cn/appstock/app/fqkline/get?param={prefix}{code},day,,,30,qfq"
        req = Request(url, headers={'User-Agent': 'Mozilla/5.0'})
        resp = urlopen(req, timeout=5).read().decode('utf-8')
        data = json.loads(resp)
        day_key = 'qfqday' if prefix != 'hk' else 'day'
        bars = data.get('data', {}).get(f'{prefix}{code}', {}).get(day_key, [])
    except:
        return None

    if not bars or current <= 0:
        return None
    closes = [float(b[2]) for b in bars]
    if len(closes) < 5:
        return None

    def ma(n):
        return sum(closes[-n:]) / n
    ma5 = ma(5)
    ma10 = ma(10) if len(closes) >= 10 else None
    ma20 = ma(20) if len(closes) >= 20 else None

    # 趋势分析
    pct_above_ma5 = (current - ma5) / ma5 * 100
    uptrend = False

    if ma20 and ma10:
        if ma5 > ma10 > ma20:
            trend_label = "多头排列"
            uptrend = True
        elif current < ma5 and ma5 < ma10 and current < ma10:
            trend_label = "空头排列"
        elif current > ma5 and ma5 > ma10:
            trend_label = "短期转强"
            uptrend = True
        else:
            trend_label = "震荡"
            if current > ma5 > ma10:
                uptrend = True
    else:
        trend_label = "数据不足"

    return {
        'price': current,
        'ma5': round(ma5, 2),
        'ma10': round(ma10, 2) if ma10 else None,
        'ma20': round(ma20, 2) if ma20 else None,
        'pct_above_ma5': round(pct_above_ma5, 1),
        'trend': trend_label,
        'uptrend': uptrend,
    }

# ── XMPP
XMPP_BRIDGE = "http://127.0.0.1:5805/"
XMPP_USER = "hmo@yoin.fun"

STALENESS_REPORT = "/home/hmo/web-dashboard/data/strategy_staleness_report.json"
DETECTOR = "/home/hmo/.hermes/profiles/position-analyst/scripts/stale_detector.py"
PORTFOLIO_PATH = "/home/hmo/web-dashboard/data/portfolio.json"
REGEN_SCRIPT = "/home/hmo/.hermes/profiles/position-analyst/scripts/per_stock_reassess.py"
REGEN_LOCK = "/tmp/.stale_push_wlin_regen.lock"
MACRO_CTX = "/home/hmo/web-dashboard/data/macro_context.json"
MARKET_JSON = "/home/hmo/web-dashboard/data/market.json"
COOLDOWN_PATH = "/home/hmo/web-dashboard/data/push_cooldown.json"

NON_BUY_SIGNALS = ["观望", "弱势持有", "深套持有"]


def load_macro_line():
    """加载大盘和市场的简要描述"""
    parts = []
    try:
        # 优先 DB
        import sqlite3
        db = sqlite3.connect("/home/hmo/MoFin/data/mofin.db")
        row = db.execute(
            "SELECT structure FROM macro_context_log "
            "WHERE has_valid_data=1 ORDER BY created_at DESC LIMIT 1"
        ).fetchone()
        db.close()
        if row and row[0]:
            m = json.loads(row[0])
        else:
            raise ValueError("no db data")
        overall = m.get("overall", "neutral")
        desc = m.get("description", "")
        if "bearish" in overall:
            parts.append("大盘偏弱")
        elif overall == "bullish":
            parts.append("大盘偏强")
        elif desc:
            parts.append(f"大盘{desc}")
    except Exception:
        try:
            with open(MACRO_CTX) as f:
                m = json.load(f).get("structure", {})
            overall = m.get("overall", "neutral")
            desc = m.get("description", "")
            if "bearish" in overall:
                parts.append("大盘偏弱")
            elif overall == "bullish":
                parts.append("大盘偏强")
            elif desc:
                parts.append(f"大盘{desc}")
        except Exception:
            pass
    try:
        with open(MARKET_JSON) as f:
            mk = json.load(f)
        mood = mk.get("mood", "")
        if mood:
            parts.append(f"市场{mood}")
    except Exception:
        pass
    return " | ".join(parts) if parts else ""


def is_actionable(cur, timing_signal=""):
    """检查信号是否可操作。空文本/含非买入关键词 → 不可操作"""
    if not cur and not timing_signal:
        return False  # 空文本默认不安全
    for kw in NON_BUY_SIGNALS:
        if cur and kw.lower() in cur.lower():
            return False
        if timing_signal and kw.lower() in timing_signal.lower():
            return False
    return True


def trigger_regen_sync(stock_codes=None):
    """同步执行指定个股的重评（等重评完再发报告）"""
    if not stock_codes:
        return
    try:
        cmd = ["python3", REGEN_SCRIPT] + stock_codes
        subprocess.run(cmd, capture_output=True, text=True, timeout=60)
    except subprocess.TimeoutExpired:
        print("[REGEN] 重评超时（60s）", file=sys.stderr)
    except Exception as e:
        print(f"[REGEN] 重评失败: {e}", file=sys.stderr)


def load_cash():
    """从 portfolio.json 实时读可用现金（可用 ≈ 实时买力），不硬编码"""
    try:
        with open(PORTFOLIO_PATH) as f:
            data = json.load(f)
        if isinstance(data, dict):
            # 先读 cash_available（拆分了可用/冻结），fallback 到 cash
            return data.get("cash_available", data.get("cash", 0))
        if isinstance(data, list) and len(data) > 1 and isinstance(data[1], dict):
            return data[1].get("cash_available", data[1].get("cash", 0))
        return 0
    except Exception:
        return 0


_HK_LOT_CACHE = {}

def hk_lot_size(code):
    """从腾讯行情API获取港股实际每手股数（字段[60]），带缓存"""
    if code in _HK_LOT_CACHE:
        return _HK_LOT_CACHE[code]
    try:
        url = f"http://qt.gtimg.cn/q=hk{code}"
        req = Request(url, headers={"User-Agent": "curl/7.81"})
        with urlopen(req, timeout=5) as r:
            text = r.read().decode("gbk")
        raw = text.split("=", 1)[1].strip().strip('"').strip(";")
        fld = raw.split("~")
        lot = int(fld[60]) if len(fld) > 60 and fld[60] else 1000
        _HK_LOT_CACHE[code] = lot
        return lot
    except Exception:
        _HK_LOT_CACHE[code] = 1000
        return 1000


def lot_cost(code, price):
    if str(code).startswith("688"):
        return 200 * price
    elif len(str(code)) == 5:
        lot = hk_lot_size(code)
        try:
            sys.path.insert(0, '/home/hmo/MoFin')
            from hk_rate import hkd_to_cny
            rate = hkd_to_cny()
        except Exception:
            rate = 0.87
        return int(lot * price * rate)
    else:
        return 100 * price


def push_to_xmpp(text):
    """通过知微 HTTP bridge 推送到老爸私信"""
    if not text.strip():
        return
    try:
        payload = json.dumps({
            "to": XMPP_USER,
            "body": text.strip(),
            "type": "chat",
        }).encode("utf-8")
        req = Request(XMPP_BRIDGE, data=payload, headers={"Content-Type": "application/json"})
        urlopen(req, timeout=5)
    except Exception as e:
        print(f"[XMPP推送失败] {e}", file=sys.stderr)


def load_cooldown():
    try:
        with open(COOLDOWN_PATH) as f:
            return json.load(f)
    except Exception:
        return {}


def save_cooldown(cd):
    try:
        with open(COOLDOWN_PATH, "w") as f:
            json.dump(cd, f, indent=2)
    except Exception:
        pass


def in_cooldown(code, action_type, cooldown_dict, minutes=30):
    key = f"{code}_{action_type}"
    last = cooldown_dict.get(key, 0)
    elapsed = datetime.now().timestamp() - last
    return elapsed < minutes * 60, elapsed, key


def main():
    r = subprocess.run(
        ["python3", DETECTOR], capture_output=True, text=True, timeout=60
    )
    if r.returncode != 0 and r.stderr:
        print(f"[stderr] {r.stderr.strip()}", file=sys.stderr)

    wl_lines = [
        l for l in r.stdout.split("\n")
        if "[WL_IN]" in l and "[自选]" in l
    ]
    if not wl_lines:
        return 0

    # 读 stale report
    try:
        with open(STALENESS_REPORT) as f:
            report = json.load(f)
    except Exception:
        report = {"flagged": []}
    code_cur = {i["code"]: i.get("current", "") for i in report.get("flagged", [])}

    # 加载冷却状态
    cooldown = load_cooldown()
    now_ts = datetime.now().timestamp()

    # 读 decisions.json 获取完整策略数据
    code_data = {}
    try:
        with open("/home/hmo/web-dashboard/data/decisions.json") as f:
            dec = json.load(f)
        for e in dec.get("decisions", []):
            code_data[e["code"]] = e
    except Exception:
        pass

    cash = load_cash()
    stocks = []
    stale_list = []
    all_candidates = []  # 所有在买入区的自选（stale+non-stale）

    for l in wl_lines:
        m = re.match(r'\[WL_IN\](?:\s+\[\w+\])*\s+\[自选\]\s+(\S+)\((\d+)\)', l)
        if not m:
            continue
        name, code = m.group(1), m.group(2)
        pm = re.search(r'价(\d+\.\d{2})', l)
        if not pm:
            continue
        price = float(pm.group(1))
        zm = re.search(r'买入([\d.]+)~([\d.]+)', l)
        if not zm:
            continue
        buy_low, buy_high = float(zm.group(1)), float(zm.group(2))
        is_stale = "[STRATEGY_STALE]" in l
        cur = code_cur.get(code, "")

        all_candidates.append((name, code, price, buy_low, buy_high, cur, is_stale))

        if not is_actionable(cur, code_data.get(code, {}).get("timing_signal", "")) or is_stale:
            stale_list.append((name, code, price, buy_low, buy_high, cur))
            continue

        lot = lot_cost(code, price)
        ratio = lot / cash if cash > 0 else 999
        stocks.append((name, code, price, buy_low, buy_high, lot, ratio))

    if not stocks and not stale_list:
        return 0

    now = datetime.now().strftime("%H:%M")
    lines = []

    # 市场背景
    macro_line = load_macro_line()
    if macro_line:
        lines.append(f"【市场背景】{macro_line}")

    # [关键修复: 2026-06-25] 所有预推票先重评，再出报告
    # 不只是 stale 的重评，所有在买入区的自选都先刷新策略，确保推荐不滞后
    to_reassess = list(set(s[1] for s in stocks) | set(s[1] for s in stale_list))
    if to_reassess:
        trigger_regen_sync(to_reassess)
        # 重评完成，re-read decisions.json 获取最新策略
        code_data = {}
        try:
            with open("/home/hmo/web-dashboard/data/decisions.json") as f:
                dec = json.load(f)
            for e in dec.get("decisions", []):
                code_data[e["code"]] = e
        except Exception:
            pass

    # 重新过滤：重评后可能有策略变化（止盈/止损/信号变动）
    # 重建 stocks 列表，用新数据判断（不再用旧 is_stale 标记，因为已全部重评）
    stocks = []
    for (name, code, price, buy_low, buy_high, cur, is_stale) in all_candidates:
        # 重评后重新检查 actionability（用新 timing_signal）
        sig = code_data.get(code, {}).get("timing_signal", "")
        if not is_actionable(cur, sig):
            continue
        lot = lot_cost(code, price)
        ratio = lot / cash if cash > 0 else 999
        stocks.append((name, code, price, buy_low, buy_high, lot, ratio))

    # 加载portfolio获取持仓信息（A/H去重用）
    pf = {"holdings": []}
    try:
        with open(PORTFOLIO_PATH) as f:
            pf = json.load(f)
    except Exception:
        pass

    stocks.sort(key=lambda s: (
        0 if len(str(s[1])) == 6 else 1,
        -code_data.get(s[1], {}).get("rr_ratio", 0)
    ))

    # 只展示有清晰操作信号的个股
    # timing_signal 必须是明确操作方向：买入/加仓/观望/关注/信号不充分
    # 行业描述（行业偏弱/行业偏强/大盘变盘等）不是操作信号，一律跳过
    VALID_SIGNALS = {"买入", "加仓", "观望", "关注", "信号不充分"}
    SKIP_KEYWORDS = ["等企稳", "信号不充分"]

    actionable = []
    for s in stocks:
        sig = code_data.get(s[1], {}).get("timing_signal", "")
        if not sig:
            continue
        # 跳过非操作信号
        if any(kw in sig for kw in SKIP_KEYWORDS):
            continue
        # 中性信号跳过
        stripped = sig.strip()
        if not stripped or stripped.lower() in ("", "neutral", "持有", "深套持有", "弱势持有"):
            continue
        # 信号必须含买入/加仓才推荐——其他非操作信号跳过
        if not any(kw in sig for kw in ["买入", "加仓"]):
            continue
        # 趋势检查：必须不是空头排列（价格在MA5以下且MA5<MA10）
        trend = fetch_trend_data(s[1])
        if trend:
            if not trend['uptrend']:
                # 空头排列或弱势震荡，不推荐
                continue
        # (如果趋势数据获取失败，放行—不因数据问题错杀)

        # A/H跨市场去重：同一公司已有持仓，不推荐另一市场的品种
        name_s1 = code_data.get(s[1], {}).get("name", "") or s[0]
        skip_ah = False
        for h in pf.get("holdings", []):
            if h["code"] == s[1]:
                continue  # 同一代码，不跳过
            if h.get("name", "") == name_s1:
                skip_ah = True
                break
        if skip_ah:
            continue  # 同一公司已在另一市场持有，不推荐

        # 市场时段检查：不在交易时段内的市场不推买入建议
        if not market_is_open(s[1]):
            continue

        actionable.append(s)

    if not actionable:
        return 0  # 无操作信号 → 静默，不推

    # 加载基本面缓存（PE等）
    fund_cache = {}
    try:
        with open("/home/hmo/web-dashboard/data/multi_tf_cache.json") as f:
            mtf = json.load(f)
        for code, v in mtf.items():
            fund_cache[code] = v.get("fundamentals", {})
    except Exception:
        pass

    # 仓位计算：从holding.xls导入的portfolio.json读取总资产和现金
    n = len(actionable)
    total_assets = 0
    available_cash = 0
    try:
        with open("/home/hmo/web-dashboard/data/portfolio.json") as f:
            pf = json.load(f)
        available_cash = pf.get("cash_available", pf.get("cash", 0)) or 0
        # 直接取 portfolio.json 的总资产（导入时已做港币→人民币换算）
        total_assets = pf.get("total_assets", 0) or 0
        if total_assets <= 0:
            # fallback: 手动算
            for h in pf.get("holdings", []):
                mv = h.get("shares", 0) * h.get("price", 0)
                if len(str(h.get("code", ""))) <= 5:  # 港股
                    mv *= 0.866
                total_assets += mv
            total_assets += available_cash
    except Exception:
        total_assets = available_cash * 5  # fallback

    # 加载策略树模块（获取当前情景+分支评估）
    st = None
    scenario_id = ""
    scenario_label = ""
    try:
        import importlib.util
        spec = importlib.util.spec_from_file_location("st_module", "/home/hmo/MoFin/strategy_tree.py")
        st = importlib.util.module_from_spec(spec)
        spec.loader.exec_module(st)
        sc = st.detect_scenario()
        scenario_id = sc.get("id", "")
        scenario_label = sc.get("label", "")
    except Exception:
        pass

    def calc_position(lot_cost, rr, market_factor, cat, code=""):
        # 理论推荐仓位(% of 总资产) — 仅基于RR+市场+品种，不受现金限制
        if rr >= 5:
            theo_pct = 25
        elif rr >= 3:
            theo_pct = 18
        elif rr >= 2:
            theo_pct = 12
        else:
            theo_pct = 8
        if "偏弱" in market_factor:
            theo_pct = int(theo_pct * 0.8)
        elif "偏强" in market_factor:
            theo_pct = int(theo_pct * 1.15)
        if cat in ("蓝筹", "白马"):
            theo_pct = int(theo_pct * 1.2)
        elif cat in ("题材", "短线"):
            theo_pct = int(theo_pct * 0.6)
        elif cat in ("高波动", "成长"):
            theo_pct = int(theo_pct * 0.85)
        theo_pct = max(5, min(30, theo_pct))

        # 当前建议仓位：理论占总资产% → 按现金锁死
        ideal_budget = total_assets * theo_pct / 100
        # 可操作N只时，现金分配不超过 available_cash / n * 1.5
        max_use_cash = (available_cash / max(n, 1)) * 1.5
        budget = min(ideal_budget, max_use_cash, available_cash)
        lots = int(budget / lot_cost) if lot_cost > 0 else 0

        if lots == 0 and lot_cost > 0 and budget > lot_cost * 0.8:
            # 预算覆盖超过80%的1手金额 → 至少1手（仅差一档）
            lots = 1

        lot_cost_total = lots * lot_cost
        if lots == 0:
            pct_actual = 0
        elif total_assets > 0:
            pct_actual = round(lot_cost_total / total_assets * 100)
        else:
            pct_actual = 0

        if lots == 0:
            details = f"预算不足1手（{budget:,.0f}/{lot_cost:,.0f}元）"
        else:
            if len(str(code)) == 5:
                hk_lot = hk_lot_size(code)
                shares = lots * hk_lot
            elif code.startswith("688"):
                shares = lots * 200
            else:
                shares = lots * 100
            details = f"{lots}手（{shares}股，{lot_cost_total:,.0f}元）"

        return theo_pct, pct_actual, details, lots, lot_cost_total

    # ── 换仓评估 ──────────────────────────────────────────────────────
    # score_future_outlook 从 stock_scorer 模块导入（6维评分）
    
    def evaluate_swap(lot_cost_target, rr, sig, tp, sl, name, code, price_in,
                      total_assets_in, cash_in, pf_in, cd_in):
        """现金不足时评估是否卖差票换推荐股。
        
        核心逻辑：
        - 已发生的亏损是沉没成本，不参与决策
        - 用6维评分法评估每个持仓的未来前景（基于决策系统既有数据）
        - 优先卖前景最差的票，保留前景好的票（无论当前盈亏%）
        - 卖港股→买A股需T+2到账，如果推荐此方案则标注延迟风险
        - 对目标票(RR>=3+买入信号)才有换仓资格
        
        返回（推荐文案str, 缺口float）或 (None, gap)
        """
        gap = lot_cost_target - cash_in
        # 目标票质量门槛
        if rr < 3.0 or gap <= 0 or gap > total_assets_in * 0.5:
            return None, gap
        if not any(kw in sig for kw in ["买入", "加仓", "建仓"]):
            return None, gap
        
        # 收集持仓数据 + 前景评分
        ph = []
        for h in pf_in.get("holdings", []):
            hs = h.get("shares", 0) or 0
            hp = h.get("price", 0) or 0
            hc = h.get("cost", 0) or 0
            if hs <= 0 or hp <= 0:
                continue
            hmv = hs * hp
            h_code = str(h.get("code", ""))
            if len(h_code) <= 5:
                hmv *= 0.866  # approximate HKD→CNY
            hpl_pct = (hp - hc) / hc * 100 if hc else 0
            
            # 6维全面评分（越低越差，越建议卖）
            fscore, _ = score_future_outlook(h_code, cd_in)
            
            ph.append({
                "code": h_code,
                "name": h.get("name", ""),
                "shares": hs,
                "price": hp,
                "cost": hc,
                "mv": round(hmv),
                "pl_pct": round(hpl_pct, 1),
                "score": fscore,
            })
        
        # 按前景评分升序（最差的排最前面）
        ph.sort(key=lambda x: x["score"])
        
        # 打印调试信息：所有持仓的前景评分
        # print(f"[SWAP_DEBUG] 前景评分(越低越差):", file=sys.stderr)
        # for x in ph[:10]:
        #     print(f"  {x['name']}({x['code']}) 评分{x['score']} 亏{x['pl_pct']}% 市值{x['mv']:,}", file=sys.stderr)
        
        # 只考虑评分<=0（前景差或中性偏弱）的作为减仓候选
        candidates = [h for h in ph if h["score"] <= 0]
        if not candidates:
            return None, gap
        
        # 贪心选评分最差的，凑够现金缺口（最多2只）
        selected = []
        cash_freed = 0
        for h in candidates:
            if cash_freed >= gap:
                break
            cash_freed += h["mv"]
            selected.append(h)
        
        if cash_freed < gap or len(selected) > 2:
            return None, gap
        
        # 计算目标票的预期涨幅
        if tp and tp > 0:
            target_gain_pct = (tp - price_in) / price_in * 100
        else:
            target_gain_pct = rr * 3
        
        # 构建推荐文案
        buy_is_a = not is_hk_stock(code)  # 目标是否是A股
        sell_parts = []
        sell_names = []
        settlement_warnings = []
        for h in selected:
            # 每个被选股票配一句"为什么卖它"
            reason = f"评分{h['score']}"
            if h['pl_pct'] <= -30:
                reason += "深套"
            elif h['pl_pct'] <= -15:
                reason += f"亏损{h['pl_pct']}%"
            sell_parts.append(f"{h['name']}({h['code']}) {h['shares']}股 亏{h['pl_pct']}% ({reason})")
            sell_names.append(h['name'])
            # 检查结算延迟：卖港股→买A股
            if is_hk_stock(h['code']) and buy_is_a:
                settlement_warnings.append(f"{h['name']}是港股通，卖出需T+2到账才能买A股")
        sell_desc = "；".join(sell_parts)
        
        new_budget = cash_in + cash_freed
        new_lots = int(new_budget / lot_cost_target) if lot_cost_target > 0 else 0
        if new_lots == 0:
            return None, gap
        if code.startswith("688"):
            new_shares = new_lots * 200
        elif len(code) <= 5:
            new_shares = new_lots * hk_lot_size(code)
        else:
            new_shares = new_lots * 100
        new_cost = new_lots * lot_cost_target
        new_pct = round(new_cost / total_assets_in * 100) if total_assets_in > 0 else 0
        
        text = (
            f"换仓建议：卖{sell_desc}"
            f"→腾{round(cash_freed):,}元"
            f"→买{name}({code}) {new_lots}手({new_shares}股,{round(new_cost):,}元)"
            f"占{new_pct}%仓位"
            f"（止损{sl}(-{round((price_in-sl)/price_in*100,1)}%)"
            f"止盈{tp}(+{round(target_gain_pct,1)}%)"
            f" RR={rr}）\n"
            f"    理由：{', '.join(sell_names)}评分最低，"
            f"继续持有无积极信号且技术偏弱；"
            f"换到有明确信号和止损的标的，预期收益更优。"
        )
        if settlement_warnings:
            text += "\n    ⚠️ " + " | ".join(settlement_warnings)
        return text, gap

    # 标准格式：每个可操作标的 — 大盘/行业/个股三面 + 仓位
    lines.append(f"【💡 操作建议】（当前{n}只自选可操作 | 总资产{total_assets:,.0f}元 现金{available_cash:,.0f}元）")
    for s in actionable:
        name, code, price, buy_low, buy_high, lot, ratio = s
        d = code_data.get(code, {})
        sl = d.get("stop_loss", 0)
        tp = d.get("take_profit", 0)
        rr = d.get("rr_ratio", 0)
        sig = d.get("timing_signal", "")
        sector = d.get("sector_context", "")
        tech = d.get("tech_snapshot", "")
        mtf_ctx = d.get("multi_tf_context", "")
        note = d.get("note", "")
        d_factors = d.get("signal_factors", [])
        cat = d.get("stock_category", "")

        # 提取技术位
        ss = {"强撑":"-", "弱撑":"-", "弱压":"-", "强压":"-"}
        for tag in ss:
            m = re.search(rf'{tag}:([\d.]+)', tech)
            if m:
                ss[tag] = m.group(1)

        # 基本面
        fund = fund_cache.get(code, {})
        pe = fund.get("pe", 0)
        eps = fund.get("eps", 0)
        pe_str = f"PE{pe:.0f}" if pe else ""
        eps_str = f"EPS{eps:.2f}" if eps else ""

        # 从 signal_factors 提取各维度
        def _match_factor(prefix):
            for f in d_factors:
                if f.startswith(prefix):
                    return f
            return ""

        market_factor = _match_factor("大盘")
        sector_factor = _match_factor("行业")
        value_factor = _match_factor("高估值") or _match_factor("低估值") or _match_factor("蓝筹") or pe_str or ""
        news_factor = _match_factor("消息")
        tech_factor = _match_factor("净利") or _match_factor("组合") or ""

        # 构建分析行
        parts = []
        if market_factor:
            parts.append(f"大盘{market_factor.replace('大盘','')}")
        if sector_factor:
            parts.append(f"行业{sector_factor.replace('行业','')}")
        if pe_str or value_factor:
            parts.append(value_factor or pe_str)
        if news_factor:
            parts.append(news_factor)
        if not parts:
            parts.append(sector or cat or "")

        analysis = " | ".join(p for p in parts if p)

        # 仓位计算
        theo_pct, actual_pct, details, lots, lot_cost_total = calc_position(
            lot, rr, market_factor, cat, code
        )

        pfx = "" if len(code) == 6 else "HK$"

        # 取分支动作类型
        branch_action = "hold"
        branch_rationale = ""
        if st and scenario_id:
            try:
                results = st.evaluate_branches(code, scenario_id, price, d.get("shares", 0), d.get("cost", 0))
                applicable = [r for r in results if r.get("applicable")]
                if applicable:
                    best = min(applicable, key=lambda r: r.get("priority", 999))
                    branch_action = best.get("action_type", "hold")
                    branch_rationale = best.get("rationale", "")
            except Exception:
                pass

        # 冷却检查：相同股+相同操作30分钟内不发
        cooled, elapsed, cd_key = in_cooldown(code, branch_action, cooldown)
        if cooled:
            continue
        
        # 策略质量过滤：只有正向/中性信号才推操作建议
        bad_keywords = ["偏弱", "弱势", "观望", "卖出", "回避", "回避"]
        if any(kw in sig for kw in bad_keywords):
            continue
        
        # 行业背景过滤：行业大跌时不在买入区推荐（即使个股信号好）
        if "大跌" in sector:
            continue

        # 换仓评估：现金不足时评估是否卖差票换推荐股
        swap_text = None
        if lots == 0:
            swap_text, _ = evaluate_swap(
                lot, rr, sig, tp, sl, name, code, price,
                total_assets, available_cash, pf, code_data
            )
        
        action_tag = "🛒" if (lots > 0 or swap_text) else "⚠️"
        
        lines.append(
            f"  {action_tag} {name}({code}) {pfx}{price:.2f} 买区{buy_low}~{buy_high} | "
            f"1手{lot:,.0f}元 RR={rr:.1f} 损{sl} 盈{tp}\n"
            f"    {analysis}\n"
            f"    技术{ss['强撑']}→{ss['弱撑']}→{ss['弱压']}→{ss['强压']} | 信号{sig}\n"
            f"    仓位：理论{theo_pct}%×总资产 | 建议{actual_pct}%（{details}）"
        )
        
        if mtf_ctx:
            lines[-1] += f"\n    均线{mtf_ctx}"
        
        if swap_text:
            lines[-1] += f"\n    {swap_text}"

        # 分支描述
        branch_line = ""
        if branch_action != "hold":
            branch_line = f"    【{scenario_label}→{branch_action}】{branch_rationale}"
        if branch_line:
            lines[-1] += f"\n{branch_line}"

        # 记录推送时间（冷却计时用）
        cooldown[cd_key] = now_ts

    save_cooldown(cooldown)

    # 修正可操作数量（剔除冷却跳过后的实际数量）
    actual_n = len(lines) - (1 if macro_line else 0) - 1  # 减去市场背景 + 操作建议标题
    if actual_n != n:
        # 更新操作建议行
        for i, ln in enumerate(lines):
            if "【💡 操作建议】" in ln:
                lines[i] = f"【💡 操作建议】（当前{actual_n}只自选可操作 | 总资产{total_assets:,.0f}元 现金{available_cash:,.0f}元）"
                break

    if actual_n <= 0:
        return 0  # 全部冷却中 → 静默，不推

    # ── T+2前瞻：扫描近期可能入买区的A股，提前准备现金 ──
    t2_lines = []
    try:
        dec_t2 = json.loads(open("/home/hmo/web-dashboard/data/decisions.json").read())
        for entry in dec_t2.get("decisions", []):
            if entry.get("status") == "closed" or entry.get("type") != "自选策略":
                continue
            ec = entry["code"]
            el = entry.get("entry_low", 0) or 0
            eh = entry.get("entry_high", 0) or 0
            ep = entry.get("price", 0) or 0
            if not eh or not ep or el <= 0:
                continue
            # A股+价格在买入区上方5%以内（即将进入买入区）
            if not is_hk_stock(ec) and el <= ep <= eh * 1.05 and ep > eh:
                anticipation_pct = (ep - eh) / eh * 100
                lot = lot_cost(ec, ep)
                if lot > available_cash:
                    # 现金不足 → 卖港股提前准备
                    ph = []
                    for h in pf.get("holdings", []):
                        hs = h.get("shares", 0) or 0
                        hp = h.get("price", 0) or 0
                        hc = h.get("cost", 0) or 0
                        if hs <= 0 or hp <= 0 or not is_hk_stock(h.get("code","")):
                            continue
                        sc = score_future_outlook(h.get("code",""), code_data)
                        ph.append((sc, h))
                    ph.sort(key=lambda x: x[0])
                    if ph:
                        worst = ph[0][1]
                        w_name = worst.get("name","?")
                        w_code = worst.get("code","")
                        w_price = worst.get("price",0)
                        w_shares = worst.get("shares",0)
                        w_value = w_price * w_shares
                        if w_value >= lot:
                            name_e = entry.get("name","")
                            t2_lines.append(
                                f"  ⏳ {name_e}({ec})距买入区仅{anticipation_pct:.0f}%，"
                                f"需{lot:,.0f}元。建议提前卖{w_name}({w_code})"
                                f"腾{w_value:,.0f}元(T+2到账后可用)"
                            )
    except:
        pass

    if t2_lines:
        lines.append("")
        lines.append("【⏳ 提前准备（T+2港股提前出清）】")
        lines.extend(t2_lines)

    lines.insert(0, f"【知微】自选买入提醒 {now} | 总资产{total_assets:,.0f}元")
    out = "\n".join(lines)
    print(out)
    push_to_xmpp(out)
    return 0


if __name__ == "__main__":
    sys.exit(main())