weflow-cli 1.13.1 → 1.14.1

package/scripts/daily_mail.py

@@ -0,0 +1,472 @@
+"""把当天日报按主题整理成一封邮件（AI / 文学 / 学术），SMTP 发出。
+
+为什么不发 HTML 阅读器：`output/biz-daily/<日期>/index.html` 的图片是**本地相对路径**，
+发到邮箱里会全部裂掉；内联 base64 又会让邮件变成好几 MB。所以正文走「文本 + 原文链接」，
+要带图的完整版仍在电脑上开阅读器。
+
+"每篇都由 AI 读过"不是这里做的：逐篇摘要（`【核心观点】/【关键细节】`）是 `biz_daily.py`
+在抓取阶段用决策模型生成的，本脚本只负责**按主题聚合 + 写每个主题的综述**（每个主题一次调用）。
+
+用法：
+    python scripts/daily_mail.py --date 2026-10-18
+    python scripts/daily_mail.py --date 2026-10-18 --to someone@example.com --dry-run
+"""
+from __future__ import annotations
+
+import argparse
+import html
+import json
+import os
+import pathlib
+import re
+import smtplib
+import sys
+from email.header import Header
+from email.mime.multipart import MIMEMultipart
+from email.mime.text import MIMEText
+from email.utils import formataddr, formatdate
+
+SCRIPT_DIR = pathlib.Path(__file__).resolve().parent
+sys.path.insert(0, str(SCRIPT_DIR))
+
+from _utils import get_api_key, load_config  # noqa: E402
+
+ROOT = SCRIPT_DIR.parent
+DEFAULT_TOPICS = ('AI', '学术')
+SMTP_CONFIG = pathlib.Path(os.path.expanduser('~/.weflow-issue-watch.json'))
+
+
+def articles_path(date: str) -> pathlib.Path:
+    return ROOT / 'output' / 'biz-daily' / date / '.articles.json'
+
+
+def load_articles(date: str, topics):
+    path = articles_path(date)
+    if not path.is_file():
+        raise SystemExit(f'没有这一天的数据：{path}')
+    data = json.loads(path.read_text(encoding='utf-8'))
+    arts = [a for a in (data.get('articles') or []) if a.get('topic') in topics and a.get('title')]
+    return data, arts
+
+
+def group_by_topic(articles, topics):
+    groups = {t: [] for t in topics}
+    for a in articles:
+        groups.setdefault(a.get('topic'), []).append(a)
+    for t in groups:
+        groups[t].sort(key=lambda a: (a.get('time') or '', a.get('relevanceScore') or 0), reverse=True)
+    return groups
+
+
+GITHUB_RE = re.compile(r'https?://github\.com/[\w\-./]+', re.I)
+#: 主题内分类的小圆点配色。同一条新闻流里给不同聚类换色，读起来不容易串行。
+PALETTE = ('#6366f1', '#8b5cf6', '#0ea5e9', '#f59e0b', '#ec4899', '#14b8a6')
+_TAG_RE = re.compile(r'【[^】]{1,12}】')
+
+
+def clean_summary(summary: str, limit=170) -> str:
+    """把 `【核心观点】…【关键细节】1. …；2. …` 压成**一句人话**。
+
+    邮件里那串标签是给机器分段的，对读者只是噪音：谁在手机上读日报，都不会去分辨
+    「核心观点」和「关键细节」的区别。所以只取【核心观点】那一段，标签一律去掉；
+    没有标签的老格式就整段压平。
+    """
+    text = summary or ''
+    m = re.search(r'【核心观点】(.+?)(?=【|$)', text, re.S)
+    if m:
+        text = m.group(1)
+    text = _TAG_RE.sub(' ', text)
+    text = ' '.join(text.split())
+    if len(text) <= limit:
+        return text
+    return text[:limit].rstrip('，；、,. ') + '…'
+
+
+def body_index(date: str) -> dict:
+    """url -> 那篇文章的 md 全文（用来从正文里找 GitHub 链接）。
+
+    按 frontmatter 的 `url` 建索引，而不是按文件名：文件名是 `sanitize_filename` 拼的，
+    标题里有 `/`、`:` 或超长名字时会被改写，按名字匹配会静默丢文章。
+    """
+    index = {}
+    for md in (ROOT / 'output' / 'biz-daily' / date).glob('*/*.md'):
+        try:
+            text = md.read_text(encoding='utf-8', errors='replace')
+        except OSError:
+            continue
+        m = re.search(r'^url:\s*"?([^"\n]+?)"?\s*$', text, re.M)
+        if m:
+            index[m.group(1)] = text
+    return index
+
+
+def github_links(text: str, limit=3) -> list:
+    if not text:
+        return []
+    seen, out = set(), []
+    for url in GITHUB_RE.findall(text):
+        url = url.rstrip('.,;)）')
+        key = url.rstrip('/').lower()
+        if key in seen or '/features' in key or '/pricing' in key:
+            continue
+        seen.add(key)
+        out.append(url)
+        if len(out) >= limit:
+            break
+    return out
+
+
+def _cluster_once(topic, items, api_key, want, strict):
+    """一次聚类调用。`strict=True` 用在"模型偷懒只给一组"的重试上。"""
+    # 只喂**编号 + 标题**：把摘要也塞进去会让输入膨胀到六七千字，模型反而更容易
+    # 直接吐一组（2026-10-21 那封：91 篇 AI 被分成单独一组「模型与产品」）。
+    numbered = '\n'.join(f'{i}. {a.get("title")}' for i, a in enumerate(items, 1))
+    demand = (
+        '**必须分成至少 5 组**。只给一组视为无效答案——这些文章明显可以按话题再细分。\n'
+        if strict else
+        f'分成 {want - 1}-{want + 1} 组。\n'
+    )
+    prompt = (
+        f'把下面 {len(items)} 条公众号文章按话题分组。\n'
+        + demand +
+        '组名用 4-8 个汉字，能概括这一组在讲什么（例如「模型与产品」「科研工具」「治理与风险」）。\n'
+        '只输出 JSON 数组，不要解释、不要代码围栏：\n'
+        '[{"name":"组名","items":[1,2,3]}, ...]\n'
+        '要求：每个序号出现且只出现一次。\n\n'
+        + numbered
+    )
+    try:
+        raw = __import__('_utils').call_ai(prompt, 'deepseek', api_key, max_tokens=1500) or ''
+    except Exception as exc:
+        print(f'[WARN] {topic} 聚类调用失败（{type(exc).__name__}: {exc}）', file=sys.stderr)
+        return None
+
+    data = None
+    try:
+        data = json.loads(raw)
+    except Exception:
+        import re as _re
+        m = _re.search(r'\[[\s\S]*\]', raw)
+        if m:
+            try:
+                data = json.loads(m.group(0))
+            except Exception:
+                data = None
+    if not isinstance(data, list) or not data:
+        print(f'[WARN] {topic} 聚类返回无法解析', file=sys.stderr)
+        return None
+
+    seen, clusters = set(), []
+    for group in data:
+        if not isinstance(group, dict):
+            continue
+        name = str(group.get('name') or '').strip()[:12] or '其他'
+        picked = []
+        for idx in group.get('items') or []:
+            try:
+                i = int(idx)
+            except Exception:
+                continue
+            if 1 <= i <= len(items) and i not in seen:
+                seen.add(i)
+                picked.append(items[i - 1])
+        if picked:
+            clusters.append((name, picked))
+    leftover = [a for i, a in enumerate(items, 1) if i not in seen]
+    if leftover:
+        if clusters and len(leftover) <= max(3, len(items) // 4):
+            clusters[-1][1].extend(leftover)
+        else:
+            clusters.append(('其他', leftover))
+    return clusters or None
+
+
+def cluster_items(topic: str, items, api_key: str, want=5):
+    """把同一主题下的文章再分一次类（让几十篇的段落不再是一根长肠子）。
+
+    为什么不用现成的 `tags`：实测 68 篇 AI 文章里有 **382 个不同标签**，频次最高的才 6 次——
+    按标签聚类会碎成一地。所以交给模型分一次。
+
+    **退化要重试**：2026-10-21 那天模型把 91 篇 AI 全塞进一组「模型与产品」，
+    版式当场退回上线前的样子（一根长肠子），而且完全不报错。现在"只给一组但篇数很多"
+    会被当成失败、用更硬的指令再要一次。
+
+    返回 `[(类名, [文章...])]`；任何一步失败都退回"一组=全部"，**绝不因此不发信**。
+    """
+    if len(items) < 8 or not api_key:
+        return [(topic, list(items))]
+
+    result = None
+    for strict in (False, True):
+        result = _cluster_once(topic, items, api_key, want, strict)
+        degenerate = result and len(result) == 1 and len(items) > 20
+        if result and not degenerate:
+            return result
+        if degenerate:
+            print(f'[WARN] {topic} 聚类退化成 1 组（{len(items)} 篇），{"再要求一次" if not strict else "仍失败"}',
+                  file=sys.stderr)
+    return result or [(topic, list(items))]
+
+
+def topic_digest(topic: str, items, api_key: str) -> str:
+    """每个主题一次调用：让它读完全部标题+摘要，写一段综述。失败就返回空串（不影响发信）。"""
+    if not api_key:
+        return ''
+    lines = [f'- {a.get("title")}｜{a.get("source")}｜{clean_summary(a.get("summary"), 90)}' for a in items]
+    prompt = (
+        f'下面是今天「{topic}」主题下 {len(items)} 篇公众号文章的标题与要点。\n'
+        '请用中文写一段 150-250 字的综述：今天这个主题主要在讨论什么、有哪些值得注意的进展或分歧、'
+        '其中哪一两篇值得优先精读（给出标题）。不要罗列全部篇目，不要空话，不要用列表。\n\n'
+        + '\n'.join(lines)
+    )
+    try:
+        return api_key and __import__('_utils').call_ai(prompt, 'deepseek', api_key, max_tokens=600) or ''
+    except Exception as exc:
+        print(f'[WARN] {topic} 综述生成失败（{type(exc).__name__}: {exc}），邮件里跳过这一段', file=sys.stderr)
+        return ''
+
+
+def render(date: str, groups, digests, bodies=None, clusters=None) -> tuple[str, str, int]:
+    bodies = bodies or {}
+    clusters = clusters or {}
+    total = sum(len(v) for v in groups.values())
+    # 主题配色：AI 靛蓝、学术 青绿。用 table + inline style —— 邮件客户端会剥离
+    # `<style>` 与 class，class 选择器在 QQ / Gmail / 企业微信里表现不一致。
+    accents = {'AI': ('#4f46e5', '#eef2ff'), '学术': ('#0d9488', '#ecfdf5')}
+
+    text_lines = [f'公众号日报 {date}', f'共 {total} 篇', '']
+    blocks: list[str] = []
+
+    for topic, items in groups.items():
+        if not items:
+            continue
+        accent, tint = accents.get(topic, ('#334155', '#f1f5f9'))
+        text_lines.append(f'=== {topic}（{len(items)} 篇）===')
+        digest = (digests or {}).get(topic) or ''
+        if digest:
+            text_lines.append(digest)
+        text_lines.append('')
+
+        # 主题分区：一条通栏色条 + 主题名，比一个小圆角标签更能"切开"版面。
+        # 档位名取自 `_utils.RELEVANCE_NAMES`：写死 `('高','中','低')` 会与判据层漂移，
+        # 而 `topic_taxonomy_test` 就是专门拦这件事的（它当场红过一次）。
+        from _utils import RELEVANCE_NAMES as _REL
+        hi, mid, lo = (sum(1 for a in items if a.get('relevance') == k) for k in (_REL[2], _REL[1], _REL[0]))
+        def pct(n):
+            return max(1, round(n * 100 / max(1, len(items))))
+        blocks.append(
+            '<tr><td style="padding:30px 28px 0;">'
+            f'<table role="presentation" width="100%" cellpadding="0" cellspacing="0"><tr>'
+            f'<td width="4" style="background:{accent};border-radius:2px;"></td>'
+            '<td style="padding-left:10px;">'
+            f'<span style="font-size:18px;font-weight:700;color:#111827;">{html.escape(topic)}</span>'
+            f'<span style="font-size:12px;color:#9aa3af;margin-left:8px;">{len(items)} 篇</span>'
+            '</td></tr></table>'
+            # 相关度比例条：不用看数字就知道今天"值得读的占多少"
+            f'<table role="presentation" width="100%" cellpadding="0" cellspacing="0" style="margin-top:10px;">'
+            '<tr>'
+            f'<td width="{pct(hi)}%" height="7" style="background:{accent};border-radius:4px 0 0 4px;"></td>'
+            f'<td width="{pct(mid)}%" height="7" style="background:{tint};"></td>'
+            f'<td width="{pct(lo)}%" height="7" style="background:#eef1f4;border-radius:0 4px 4px 0;"></td>'
+            '</tr></table>'
+            f'<div style="font-size:11.5px;color:#9aa3af;margin-top:6px;">'
+            f'<span style="color:{accent};font-weight:600;">高 {hi}</span> · 中 {mid} · 低 {lo}</div>'
+            '</td></tr>'
+        )
+        if digest:
+            blocks.append(
+                '<tr><td style="padding:14px 28px 0;">'
+                '<table role="presentation" width="100%" cellpadding="0" cellspacing="0" '
+                f'style="background:#f8fafc;border-left:3px solid {accent};border-radius:8px;">'
+                '<tr><td style="padding:14px 16px;font-size:14px;line-height:1.8;color:#3f4652;">'
+                f'{html.escape(digest)}</td></tr></table></td></tr>'
+            )
+
+        # 主题内再分一次类：68 篇挤成一排没人看，分成 4-6 组后每一段都有名字。
+        section_clusters = (clusters or {}).get(topic) or [(topic, list(items))]
+        for ci, (cname, citems) in enumerate(section_clusters):
+            dot = PALETTE[ci % len(PALETTE)]
+            if len(section_clusters) > 1:
+                blocks.append(
+                    '<tr><td style="padding:22px 28px 0;">'
+                    f'<span style="display:inline-block;width:7px;height:7px;border-radius:50%;'
+                    f'background:{dot};vertical-align:middle;"></span>'
+                    f'<span style="font-size:13.5px;font-weight:700;color:#374151;margin-left:7px;">'
+                    f'{html.escape(cname)}</span>'
+                    f'<span style="font-size:11.5px;color:#b3bac4;margin-left:6px;">{len(citems)}</span>'
+                    '</td></tr>'
+                )
+                text_lines.append(f'-- {cname}（{len(citems)}）--')
+
+            for a in citems:
+                i = items.index(a) + 1
+                title = a.get('title') or ''
+                source = a.get('source') or ''
+                when = (a.get('time') or '')[:5]
+                rel = a.get('relevance') or ''
+                url = a.get('url') or ''
+                repos = github_links(bodies.get(url, ''))
+                meta = ' · '.join(x for x in (source, when, rel) if x)
+                num = f'{i:02d}'
+                title_html = (f'<a href="{html.escape(url)}" style="color:#111827;text-decoration:none;">'
+                              f'{html.escape(title)}</a>' if url else html.escape(title))
+
+                # **颜色必须自己说明自己。** 上一版把序号徽章染成主题色来表示"高相关"，
+                # 但没有任何图例——用户看到截图的第一反应是"这边为什么有的是绿的"（2026-10-20）。
+                # 现在徽章统一成灰色，相关度改成 meta 行里一个**写着"高"的小色块**：
+                # 颜色解释自己，不需要图例。
+                if rel == _REL[2]:
+                    summary = clean_summary(a.get('summary'), 190)
+                    text_lines.append(f'{num}. ★ {title} — {meta}')
+                elif rel == _REL[1]:
+                    summary = clean_summary(a.get('summary'), 120)
+                    text_lines.append(f'{num}. {title} — {meta}')
+                else:
+                    summary = ''
+                    text_lines.append(f'{num}. {title} — {meta}')
+                if rel == _REL[2]:
+                    rel_html = (f'<span style="display:inline-block;background:{accent};color:#ffffff;'
+                                f'font-size:10.5px;font-weight:600;padding:1px 6px;border-radius:4px;">'
+                                f'{html.escape(rel)}</span>')
+                else:
+                    rel_html = f'<span style="color:#b3bac4;">{html.escape(rel)}</span>' if rel else ''
+                meta_html = ' · '.join(x for x in (
+                    html.escape(source), html.escape(when), rel_html) if x)
+                if summary:
+                    text_lines.append(f'    {summary}')
+                for repo in repos:
+                    text_lines.append(f'    GitHub: {repo}')
+                if url:
+                    text_lines.append(f'    {url}')
+                blocks.append(
+                    '<tr><td style="padding:10px 28px 0;">'
+                    '<table role="presentation" width="100%" cellpadding="0" cellspacing="0" '
+                    'style="background:#ffffff;border:1px solid #eef0f3;border-radius:12px;">'
+                    '<tr><td style="padding:14px 16px;">'
+                    '<table role="presentation" width="100%" cellpadding="0" cellspacing="0"><tr>'
+                    f'<td width="28" valign="top"><span style="display:inline-block;width:24px;height:24px;'
+                    'line-height:24px;text-align:center;border-radius:7px;background:#f1f5f9;'
+                    f'color:#94a3b8;font-size:11.5px;font-weight:700;">{num}</span></td>'
+                    '<td valign="top">'
+                    f'<div style="font-size:14.5px;font-weight:600;color:#111827;line-height:1.55;">{title_html}</div>'
+                    f'<div style="font-size:11.5px;color:#9aa3af;margin-top:4px;">{meta_html}</div>'
+                    + (f'<div style="font-size:13.5px;line-height:1.75;color:#4b5563;margin-top:8px;">'
+                       f'{html.escape(summary)}</div>' if summary else '')
+                    + ''.join(
+                        f'<div style="margin-top:8px;font-size:12.5px;">'
+                        f'<a href="{html.escape(repo)}" style="color:{accent};text-decoration:none;">'
+                        f'&#9679; {html.escape(repo.replace("https://github.com/", "").rstrip("/"))}</a></div>'
+                        for repo in repos
+                    )
+                    + '</td></tr></table></td></tr></table></td></tr>'
+                )
+        text_lines.append('')
+
+    text_lines.append('完整版（含配图）在电脑上打开：')
+    text_lines.append(f'  weflow-cli daily-server --date {date} --open')
+
+    stats = ' · '.join(f'{t} {len(v)} 篇' for t, v in groups.items() if v)
+    page = (
+        '<!doctype html><html><head><meta charset="utf-8">'
+        '<meta name="viewport" content="width=device-width,initial-scale=1">'
+        '<title>' + html.escape(f'公众号日报 {date}') + '</title></head>'
+        '<body style="margin:0;padding:0;background:#f4f5f7;">'
+        '<table role="presentation" width="100%" cellpadding="0" cellspacing="0" '
+        'style="background:#f4f5f7;"><tr><td align="center" style="padding:24px 12px;">'
+        '<table role="presentation" width="640" cellpadding="0" cellspacing="0" '
+        'style="width:100%;max-width:640px;background:#ffffff;border-radius:14px;'
+        'overflow:hidden;box-shadow:0 1px 3px rgba(0,0,0,.06);">'
+        '<tr><td style="background:#111827;padding:26px 28px;">'
+        '<div style="font-size:12px;letter-spacing:.12em;color:#9ca3af;font-weight:600;">'
+        'WEFLOW · 公众号日报</div>'
+        f'<div style="font-size:26px;font-weight:700;color:#ffffff;margin-top:8px;">{html.escape(date)}</div>'
+        f'<div style="font-size:13px;color:#c7cbd1;margin-top:8px;">{html.escape(stats)}</div>'
+        '</td></tr>'
+        + ''.join(blocks)
+        + '<tr><td style="padding:22px 28px 28px;">'
+        '<div style="font-size:12px;color:#9aa3af;line-height:1.8;">'
+        '完整版（含配图）在电脑上打开：<br>'
+        '<span style="font-family:ui-monospace,Consolas,Menlo,monospace;color:#6b7280;">'
+        f'weflow-cli daily-server --date {html.escape(date)} --open</span>'
+        '</div></td></tr>'
+        '</table></td></tr></table></body></html>'
+    )
+    return '\n'.join(text_lines), page, total
+
+
+def smtp_settings():
+    if not SMTP_CONFIG.is_file():
+        raise SystemExit(f'找不到发信凭据 {SMTP_CONFIG}（需要 smtp_user 与 smtp_auth）')
+    cfg = json.loads(SMTP_CONFIG.read_text(encoding='utf-8'))
+    if not cfg.get('smtp_user') or not cfg.get('smtp_auth'):
+        raise SystemExit(f'{SMTP_CONFIG} 里缺 smtp_user / smtp_auth')
+    return cfg['smtp_user'], cfg['smtp_auth']
+
+
+def send(user, auth, to, subject, text_body, html_body):
+    msg = MIMEMultipart('alternative')
+    msg['From'] = formataddr((str(Header('WeFlow CLI 日报', 'utf-8')), user))
+    msg['To'] = to
+    msg['Subject'] = Header(subject, 'utf-8')
+    msg['Date'] = formatdate(localtime=True)
+    msg.attach(MIMEText(text_body, 'plain', 'utf-8'))
+    msg.attach(MIMEText(html_body, 'html', 'utf-8'))
+    with smtplib.SMTP_SSL('smtp.qq.com', 465, timeout=45) as smtp:
+        smtp.login(user, auth)
+        refused = smtp.sendmail(user, [to], msg.as_string())
+    if refused:
+        raise SystemExit(f'被拒收：{refused}')
+
+
+def main():
+    ap = argparse.ArgumentParser(description='把某天的日报按主题整理成邮件并发送')
+    ap.add_argument('--date', required=True, help='YYYY-MM-DD')
+    ap.add_argument('--to', default='', help='收件人；缺省用配置 dailyMailTo，再缺省发给自己')
+    ap.add_argument('--topics', default=','.join(DEFAULT_TOPICS), help='逗号分隔的主题')
+    ap.add_argument('--label', default='', help='标题后缀，如「截至中午」——区分半日速览与次日完整版')
+    ap.add_argument('--no-digest', action='store_true', help='不生成每主题的 AI 综述（省钱、更快）')
+    ap.add_argument('--dry-run', action='store_true', help='只渲染，不发送')
+    args = ap.parse_args()
+
+    topics = [t.strip() for t in args.topics.split(',') if t.strip()]
+    data, articles = load_articles(args.date, set(topics))
+    groups = group_by_topic(articles, topics)
+    total = sum(len(v) for v in groups.values())
+    print(f'{args.date}: 命中 {total} 篇（{", ".join(f"{t}={len(v)}" for t, v in groups.items())}）')
+    if not total:
+        raise SystemExit('没有可发送的文章')
+
+    cfg = load_config()
+    # `--no-digest` 只关"每主题一段综述"；聚类是版式结构，仍要做（两件事分开开关，
+    # 否则想看新版式的人一加 `--no-digest` 就退回一根长肠子）。
+    api_key = get_api_key(cfg)
+    digests = {} if args.no_digest else {t: topic_digest(t, items, api_key) for t, items in groups.items() if items}
+    clusters = {t: cluster_items(t, items, api_key, want=5) for t, items in groups.items() if len(items) >= 8}
+    for t, cs in clusters.items():
+        print(f'  {t} 分了 {len(cs)} 组: ' + ' / '.join(f'{n}({len(v)})' for n, v in cs))
+
+    bodies = body_index(args.date)
+    with_repo = sum(1 for a in articles if github_links(bodies.get(a.get('url'), '')))
+    print(f'正文含 GitHub 链接: {with_repo} 篇')
+
+    text_body, html_body, total = render(args.date, groups, digests, bodies, clusters)
+    label = (args.label or '').strip()
+    suffix = f'（{label}）' if label else ''
+    subject = f'公众号日报 {args.date}{suffix}：{" / ".join(t for t in topics if groups.get(t))} {total} 篇'
+    if label:
+        text_body = f'【{label}】以下是到此刻为止收到的 {total} 篇。完整版会在次日凌晨按整天重发。\n\n' + text_body
+
+    user, auth = smtp_settings()
+    to = args.to or str(cfg.get('dailyMailTo') or '') or user
+    print(f'收件人: {to}  主题: {subject}  正文字符: {len(text_body)}')
+    if args.dry_run:
+        out = ROOT / 'output' / f'daily-mail-{args.date}.html'
+        out.write_text(html_body, encoding='utf-8')
+        print(f'[dry-run] 已渲染到 {out}')
+        return
+    send(user, auth, to, subject, text_body, html_body)
+    print(f'已发送到 {to}')
+
+
+if __name__ == '__main__':
+    main()

package/scripts/daily_task.ps1

@@ -127,6 +127,51 @@ if ($proc.ExitCode -eq 0 -and $rest.Count -eq 0) {
             Set-Content -Path (Join-Path $dayDir '.noon_done') -Value (Get-Date -Format 'yyyy-MM-dd HH:mm:ss') -Encoding UTF8
         }
     }
+
+    # Retrieval indexes lag behind unless something refreshes them: on 2026-10-22 the
+    # semantic index was still at 10-15 (7 days stale) and the phrase index at 10-17, so
+    # `search` simply could not find anything newer. Both run here, after the day is on
+    # disk, in their own logs -- a failure is recorded and never fails the day.
+    $searchScript = Join-Path $PSScriptRoot 'semantic_search.py'
+    if ($mode -eq 'day' -and (Test-Path $searchScript)) {
+        $idxOut = Join-Path $logDir ('index-{0}.log' -f $stamp)
+        $idxErr = Join-Path $logDir ('index-{0}.err.log' -f $stamp)
+        # `update` only embeds what is not in the index yet (4,061 new records on the first
+        # run after a week). `phrase-build` is local FTS, no API, so it rebuilds in full.
+        $idxProc = Start-Process -FilePath $python `
+            -ArgumentList @($searchScript, 'update', '--article-days', '30') `
+            -NoNewWindow -Wait -PassThru -RedirectStandardOutput $idxOut -RedirectStandardError $idxErr
+        $phraseProc = Start-Process -FilePath $python `
+            -ArgumentList @($searchScript, 'phrase-build') `
+            -NoNewWindow -Wait -PassThru `
+            -RedirectStandardOutput (Join-Path $logDir ('phrase-{0}.log' -f $stamp)) `
+            -RedirectStandardError (Join-Path $logDir ('phrase-{0}.err.log' -f $stamp))
+        if ($idxProc.ExitCode -ne 0 -or $phraseProc.ExitCode -ne 0) {
+            Add-Content -Path (Join-Path $logDir 'daily-task.log') -Value `
+                ('===== {0}  index update failed (update={1}, phrase={2}), see {3} =====' -f `
+                    (Get-Date -Format 'yyyy-MM-dd HH:mm:ss'), $idxProc.ExitCode, $phraseProc.ExitCode, $idxErr) -Encoding UTF8
+        }
+    }
+
+    # **Both passes mail now.** The noon pass used to be silent, which read as "it never
+    # ran" - the only way to tell it had done anything was to open the log. The two emails
+    # are clearly different: the noon one is labelled 截至中午 and says the complete version
+    # follows next morning. Mail is still its own step with its own log pair, so a mail
+    # failure never fails the day nor blocks the noon marker.
+    $mailScript = Join-Path $PSScriptRoot 'daily_mail.py'
+    $mailOut = Join-Path $logDir ('daily-mail-{0}{1}.log' -f $stamp, $suffix)
+    $mailErr = Join-Path $logDir ('daily-mail-{0}{1}.err.log' -f $stamp, $suffix)
+    if (Test-Path $mailScript) {
+        $mailArgs = @($mailScript, '--date', $targetDay)
+        if ($mode -eq 'noon') { $mailArgs += @('--label', '截至中午') }
+        $mailProc = Start-Process -FilePath $python -ArgumentList $mailArgs `
+            -NoNewWindow -Wait -PassThru -RedirectStandardOutput $mailOut -RedirectStandardError $mailErr
+        if ($mailProc.ExitCode -ne 0) {
+            Add-Content -Path (Join-Path $logDir 'daily-task.log') -Value `
+                ('===== {0}  [{1}] {2} mail failed (exit {3}), see {4} =====' -f `
+                    (Get-Date -Format 'yyyy-MM-dd HH:mm:ss'), $mode, $targetDay, $mailProc.ExitCode, $mailErr) -Encoding UTF8
+        }
+    }
 }
 
 exit $proc.ExitCode