weflow-cli 1.15.0 → 1.16.0

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "weflow-cli",
-  "version": "1.15.0",
+  "version": "1.16.0",
   "description": "让 AI 读懂你的微信 — 公众号日报·聊天导出·知识库搜索·MCP Server",
   "type": "module",
   "bin": {

package/scripts/bonds.py

@@ -137,6 +137,14 @@ def shape_of(messages, now):
     span_days = max(1, ((last - min(stamps)) / 86400)) if len(stamps) > 1 else 1
     per_day = len(stamps) / span_days                    # 平时日均
     recent = sum(1 for t in stamps if (now - t) / 86400 <= 7)
+    # 最后一句原文。冷却名单里只写"已静默 45 天"没法决定要不要回 —— 补上这句话，
+    # 读者一眼就知道该说什么（2026-10-23：这是"复联提醒"缺的那一半）。
+    last_text = ''
+    for m in sorted(messages, key=lambda x: x.get('createTime') or 0, reverse=True):
+        text = ' '.join(str(m.get('content') or '').split())
+        if text:
+            last_text = text[:60]
+            break
     return {
         'messages': len(stamps),
         'activeDays': len(days),
@@ -145,6 +153,7 @@ def shape_of(messages, now):
         'recent7': recent,
         'mine': round(100.0 * sum(1 for m in messages if m.get('isSend')) / max(1, len(stamps))),
         'last': datetime.fromtimestamp(last, TZ).strftime('%Y-%m-%d') if last else '',
+        'lastText': last_text,
     }
 
 
@@ -253,6 +262,8 @@ def render(payload, limit=12):
             lines.append('  %-16s %5d 条 / %3d 个活跃天 / 已静默 %6.0f 天（最后一次 %s）%s'
                          % (p['name'][:16], p['messages'], p['activeDays'], p['silentDays'], p['last'],
                             '  ⚠只有微信这一条线' if 'only-wechat' in p['tags'] else ''))
+            if p.get('lastText'):
+                lines.append('      最后一句：%s' % p['lastText'])
     if view['heating']:
         lines.append('')
         lines.append('最近变热（近 7 天 ≥ 平时日均的 %.1f 倍）:' % HOT_RATIO)

package/scripts/daily_task.ps1

@@ -119,6 +119,42 @@ if (Test-Path $out) {
 $proc = Start-Process -FilePath $python -ArgumentList $allArgs -NoNewWindow -Wait -PassThru `
     -RedirectStandardOutput $out -RedirectStandardError $err
 
+# Fallback for the HTML stage.
+#
+# `pipeline.py` fetches, summarises, writes the day, then renders `index.html`. That last step
+# has hung twice now (2026-10-18: 11 hours with no output; 2026-10-23: the run died after the
+# README and never produced the page). Losing the page also loses the mail, because mailing
+# happens only on success - so a single slow step silently costs the whole morning report.
+#
+# The fetch and summaries are already on disk at that point (`.articles.json`), so rendering is
+# retried on its own. `$proc` is replaced by the retry's exit code, which is what lets the mail
+# and index steps below run on a day where only the renderer failed.
+if ($proc.ExitCode -ne 0) {
+    $dayDir = Join-Path $root ('output\biz-daily\' + $targetDay)
+    $art = Join-Path $dayDir '.articles.json'
+    $idx = Join-Path $dayDir 'index.html'
+    $htmlScript = Join-Path $PSScriptRoot 'generate_html.py'
+    # Only retry when the data is there and the page is missing or older than the data:
+    # a page newer than `.articles.json` means rendering already happened this run.
+    $needsRender = (Test-Path $art) -and (Test-Path $htmlScript)
+    if ($needsRender -and (Test-Path $idx)) {
+        $needsRender = (Get-Item $art).LastWriteTime -gt (Get-Item $idx).LastWriteTime
+    }
+    if ($needsRender) {
+        Add-Content -Path (Join-Path $logDir 'daily-task.log') -Value `
+            ('===== {0}  [{1}] {2} pipeline exit {3}, retrying HTML only =====' -f `
+                (Get-Date -Format 'yyyy-MM-dd HH:mm:ss'), $mode, $targetDay, $proc.ExitCode) -Encoding UTF8
+        $retryProc = Start-Process -FilePath $python `
+            -ArgumentList @($htmlScript, '--date', $targetDay) `
+            -NoNewWindow -Wait -PassThru `
+            -RedirectStandardOutput (Join-Path $logDir ('html-retry-{0}{1}.log' -f $stamp, $suffix)) `
+            -RedirectStandardError (Join-Path $logDir ('html-retry-{0}{1}.err.log' -f $stamp, $suffix))
+        if ($retryProc.ExitCode -eq 0) {
+            $proc = $retryProc
+        }
+    }
+}
+
 if ($proc.ExitCode -eq 0 -and $rest.Count -eq 0) {
     # Marker for the noon pass only; the day pass is judged by generated_at above.
     if ($mode -eq 'noon') {

package/scripts/install_tasks.ps1

@@ -60,24 +60,38 @@ $settings = New-ScheduledTaskSettingsSet -StartWhenAvailable -MultipleInstances
 $psExe = 'powershell.exe'
 $common = '-NoProfile -ExecutionPolicy Bypass -File ' + $script
 
+# Launch through wscript.exe, not powershell.exe directly.
+#
+# A scheduled action that runs powershell.exe is a *console* program: Windows creates the console
+# window before PowerShell can hide it, so every run flashes a black box - and while a game is in
+# the foreground that is a real interruption. wscript.exe is a GUI process with no console at all,
+# and run_hidden.vbs starts PowerShell with window style 0, so nothing appears.
+$wscript = Join-Path $env:SystemRoot 'System32\wscript.exe'
+$vbs = Join-Path $PSScriptRoot 'run_hidden.vbs'
+if (-not (Test-Path $vbs)) {
+    Write-Host "Missing launcher: $vbs" -ForegroundColor Red
+    exit 1
+}
+$action = '//B //Nologo "' + $vbs + '"'
+
 # End of day: yesterday. The at-logon trigger is the catch-up: this machine slept for
 # 37 hours and StartWhenAvailable did NOT actually start the missed run (NumberOfMissedRuns=2,
 # no run record), so "it will catch up by itself" is not something to rely on.
 Register-ScheduledTask -TaskName 'WeFlow Daily' `
-    -Action (New-ScheduledTaskAction -Execute $psExe -Argument $common) `
+    -Action (New-ScheduledTaskAction -Execute $wscript -Argument $action) `
     -Trigger @((New-ScheduledTaskTrigger -Daily -At '00:05'),
                (New-ScheduledTaskTrigger -AtLogOn)) `
     -Settings $settings `
-    -Description 'WeChat daily report for yesterday (00:05 + at logon, idempotent)' -Force | Out-Null
+    -Description 'WeChat daily report for yesterday (00:05 + at logon, idempotent, no window)' -Force | Out-Null
 
 # Noon: today so far. Deliberately NO at-logon trigger: a morning logon would produce a
 # very thin "noon" report and write the .noon_done marker, which would then make the real
 # 13:00 run skip itself (the guard keys on that marker).
 Register-ScheduledTask -TaskName 'WeFlow Daily Noon' `
-    -Action (New-ScheduledTaskAction -Execute $psExe -Argument ($common + ' noon')) `
+    -Action (New-ScheduledTaskAction -Execute $wscript -Argument ($action + ' noon')) `
     -Trigger (New-ScheduledTaskTrigger -Daily -At '13:00') `
     -Settings $settings `
-    -Description 'WeChat daily report for today so far (13:00)' -Force | Out-Null
+    -Description 'WeChat daily report for today so far (13:00, no window)' -Force | Out-Null
 
 foreach ($name in @('WeFlow Daily', 'WeFlow Daily Noon')) {
     $t = Get-ScheduledTask -TaskName $name

package/scripts/leads.py

@@ -0,0 +1,179 @@
+"""线索雷达：把聊天里"可能是生意"的对话挑出来，并指回原话。
+
+为什么它值得存在：微信里最容易丢的不是重要消息，是**混在闲聊里的机会** —— 有人问价、
+有人问能不能做、有人留了联系方式，然后被下一条消息压过去。它和 `promises` 用同一个骨架：
+规则先捞候选、再到同会话的后续对话里找"有没有跟上"，**每条结论都带原话**，不做黑箱判断。
+
+默认不调模型。信号分五类（联系方式 / 报价 / 采购 / 合作 / 需求），按命中类型加权排序，
+高分的往往是"问了价 + 留了联系方式"这种最该马上回的组合。
+
+用法：
+    python scripts/leads.py --days 30
+    python scripts/leads.py --days 90 --min-score 3
+    python scripts/leads.py --days 30 --json
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import pathlib
+import re
+import sys
+from collections import defaultdict
+from datetime import datetime, timezone, timedelta
+
+SCRIPT_DIR = pathlib.Path(__file__).resolve().parent
+sys.path.insert(0, str(SCRIPT_DIR))
+
+from _utils import load_config, decrypt_lock  # noqa: E402
+from promises import load_messages, same_account  # noqa: E402
+import nt_decrypt as nd  # noqa: E402
+
+TZ = timezone(timedelta(hours=8))
+
+#: 信号类型 → (正则, 权重)。权重表示"这条信号有多像生意"，不是"多重要"。
+SIGNALS = {
+    '联系方式': (re.compile(r'(微信|vx|wx|威信|手机号|电话|联系我|加我)|1[3-9]\d{9}'
+                          r'|[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}', re.I), 3),
+    '报价': (re.compile(r'(多少钱|报价|预算|费用|价格|单价|收费|怎么算钱|贵不贵)'), 3),
+    '采购': (re.compile(r'(采购|下单|合同|发票|订购|样品|打款|付款)'), 2),
+    '合作': (re.compile(r'(合作|对接|渠道|代理|分销|外包|接单|项目合作)'), 2),
+    '需求': (re.compile(r'(需要.{0,8}(做|开发|方案|系统|服务|设计)|能不能做|可以做吗'
+                      r'|有没有.{0,4}(做|卖|提供)|求推荐|想找.{0,6}(做|人))'), 2),
+}
+#: 我跟上了的迹象（在同一条线索之后的同会话消息里找）
+FOLLOWED = re.compile(r'(已加|加了|私聊|发你了|发您了|发过去|安排|稍后|好的.{0,4}(我|马上)|收到)')
+#: 已经明确黄掉的迹象
+DEAD = re.compile(r'(太贵|不考虑|算了|暂时不需要|下次吧|再看看)')
+
+MERGE_WINDOW = 3600      # 同一会话里这一秒数内的信号合成一条线索
+
+
+#: 广告 / 招聘 / 促销：这些恰好塞满"合作、采购、报价、联系方式"这些词，但**不是线索**。
+#: 2026-10-23 第一版没滤它们，30 天里 8 条"线索"全是校招宣讲和电商促销，没有一条真生意。
+NOISE = re.compile(
+    r'(招聘|校招|秋招|春招|宣讲会|双选会|简历|offer|求职|面试|直播|预告'
+    r'|优惠|券|折扣|秒杀|包邮|抽奖|福利|返现|下单链接|u\.jd\.com|拼团|预售|凑单)', re.I)
+#: 真线索几乎都带**请求或提问**。纯陈述句（广告、通知）不进来。
+ASK = re.compile(r'(吗|呢|？|\?|能否|可以吗|可不可以|麻烦|请问|想找|有没有|多少钱|怎么收费|求推荐)')
+
+
+def _score(text):
+    """(信号类型, 分数)。**先过噪点与提问两道闸，再算信号分。**
+
+    匹配前先剥掉 `@昵称`：群昵称里常带手机号（`@某人13800000000`），不剥的话每一句
+    亲戚群闲聊都会被判成"留了联系方式"（2026-10-23 实测，误报全来自这里）。
+    """
+    bare = re.sub(r'@[^\s]{1,40}\s*', '', text)
+    # URL 里的长数字串会命中手机号规则（`?id=4713032079053994316` 这类），先剥链接。
+    bare = re.sub(r'https?://\S+', '', bare)
+    if len(bare) > 300 or NOISE.search(bare) or not ASK.search(bare):
+        return [], 0
+    hit = []
+    score = 0
+    for name, (rx, weight) in SIGNALS.items():
+        if rx.search(bare):
+            hit.append(name)
+            score += weight
+    return hit, score
+
+
+def main():
+    ap = argparse.ArgumentParser(description='线索雷达：聊天里的报价/合作/需求信号')
+    ap.add_argument('--days', type=int, default=30)
+    ap.add_argument('--min-score', type=int, default=2, help='最低分（默认 2）')
+    ap.add_argument('--limit', type=int, default=20)
+    ap.add_argument('--json', action='store_true')
+    args = ap.parse_args()
+
+    cfg = load_config()
+    db = cfg.get('ntDbPath') or ''
+    my_wxid = cfg.get('wxid') or ''
+    if not db:
+        raise SystemExit('配置里没有 ntDbPath')
+    key = decrypt_lock(cfg.get('ntKey', '') or '')
+    salt = decrypt_lock(cfg.get('ntSalt', '') or '')
+    pw = decrypt_lock(cfg.get('favPassphrase', '') or cfg.get('decryptKey', '') or '')
+    conns = nd.connect_message_shards(db, key, salt, pw)
+    if not conns:
+        raise SystemExit('打不开消息分片')
+
+    try:
+        by_talker = load_messages(conns, args.days, my_wxid)
+        name_map = {}
+        try:
+            from network import load_graph
+            people, _rooms, _members = load_graph(cfg)
+            name_map = {p['username']: (p['remark'] or p['nick'] or p['username'])
+                        for p in people.values()}
+        except Exception as exc:
+            print(f'[WARN] 读联系人失败（{type(exc).__name__}），显示 id', file=sys.stderr)
+    finally:
+        for conn in conns:
+            try:
+                conn.close()
+            except Exception:
+                pass
+
+    leads = []
+    for talker, msgs in by_talker.items():
+        hits = []
+        for i, (ts, is_me, who, text) in enumerate(msgs):
+            if is_me:
+                continue                      # 线索来自别人说的，不是我自己发的
+            kinds, score = _score(text)
+            if score < args.min_score:
+                continue
+            hits.append({'i': i, 'ts': ts, 'text': text, 'kinds': kinds, 'score': score})
+
+        # 同一会话里时间挨近的信号合成一条线索，避免"问价"和"留手机号"被拆成两条。
+        merged = []
+        for h in hits:
+            if merged and h['ts'] - merged[-1]['last'] <= MERGE_WINDOW:
+                cur = merged[-1]
+                cur['last'] = h['ts']
+                cur['score'] = max(cur['score'], h['score']) + 1
+                cur['kinds'] = sorted(set(cur['kinds']) | set(h['kinds']))
+                if h['text'] not in cur['parts']:
+                    cur['parts'].append(h['text'])
+                cur['idx'].append(h['i'])
+            else:
+                merged.append({'first': h['ts'], 'last': h['ts'], 'score': h['score'],
+                               'kinds': list(h['kinds']), 'parts': [h['text']],
+                               'idx': [h['i']]})
+
+        for m in merged:
+            after = msgs[m['idx'][-1] + 1: m['idx'][-1] + 25]
+            replied = any(is_me for _, is_me, _w, _t in after)
+            followed = any(is_me and FOLLOWED.search(t) for _, is_me, _w, t in after)
+            dead = any(DEAD.search(t) for _, _me, _w, t in after)
+            state = ('已结束' if dead else '已跟进' if followed else '我回过' if replied else '没回')
+            leads.append({
+                'time': datetime.fromtimestamp(m['first'], tz=TZ).strftime('%Y-%m-%d %H:%M'),
+                'talker': name_map.get(talker, talker),
+                'kinds': m['kinds'],
+                'score': m['score'],
+                'said': ' ／ '.join(p[:80] for p in m['parts'][:3]),
+                'state': state,
+            })
+
+    leads.sort(key=lambda x: (-x['score'], x['time']), reverse=False)
+    leads = leads[:args.limit]
+
+    if args.json:
+        print(json.dumps({'success': True, 'days': args.days, 'leads': leads},
+                         ensure_ascii=False, indent=2))
+        return
+
+    if not leads:
+        print(f'最近 {args.days} 天里没有达到分数 {args.min_score} 的线索信号')
+        return
+    print(f'最近 {args.days} 天的线索信号 {len(leads)} 条（分数越高越像生意）：\n')
+    for it in leads:
+        print(f"[{it['score']}分 · {'/'.join(it['kinds'])} · {it['state']}] {it['time']}  {it['talker']}")
+        print(f"    {it['said']}")
+        print()
+
+
+if __name__ == '__main__':
+    main()