Publish updated WeChat Radar

2026-09-08 16:58:30 +09:00 · 2026-05-26 09:10:04 +08:00
parent 8abb29e13e
commit de7ec99fed
44 changed files with 2900 additions and 746 deletions
@@ -98,6 +98,7 @@ export interface DashboardLinkHighlight {
  title: string;
  url: string;
  domain: string;
+  source: string;
  score: number;
  verdict: string;
  count: number;
@@ -149,7 +150,7 @@ export function buildDashboardIntelligence(
  groupNames = new Map<string, string>(),
 ): DashboardIntelligence {
  date = resolveIntelligenceDate(date);
-  const key = `dashboard-intelligence:${date}:v11`;
+  const key = `dashboard-intelligence:${date}:v14`;
  const cached = cache.get(key) as DashboardIntelligence | undefined;
  if (cached) return cached;

@@ -174,6 +175,37 @@ export function buildDashboardIntelligence(
       LIMIT ?`,
    )
    .all(minusDays(date, 7), date, 9000) as MessageSignalRow[];
+  const linkRows = db()
+    .prepare(
+      `SELECT
+         ml.url,
+         ml.canonical_url,
+         ml.title,
+         ml.domain,
+         ml.source,
+         ml.confidence,
+         ml.time,
+         ml.chatroom_id,
+         m.content
+       FROM message_links ml
+       JOIN messages m
+         ON m.chatroom_id = ml.chatroom_id
+        AND m.local_id = ml.local_id
+       WHERE ml.date = ?
+       ORDER BY ml.timestamp DESC
+       LIMIT 600`,
+    )
+    .all(date) as Array<{
+    url: string;
+    canonical_url: string;
+    title: string | null;
+    domain: string;
+    source: string;
+    confidence: number;
+    time: string;
+    chatroom_id: string;
+    content: string;
+  }>;

  const candidates: DashboardSignalItem[] = [];
  const opportunities: DashboardOpportunityItem[] = [];
@@ -203,6 +235,7 @@ export function buildDashboardIntelligence(
      groups: Set<string>;
      last_seen: string;
      snippets: string[];
+      sources: Set<string>;
    }
  >();

@@ -261,28 +294,30 @@ export function buildDashboardIntelligence(
    if (TOOL_RE.test(clean)) source.tool_count++;
    sourceMap.set(sourceKey, source);

-    for (const url of extractUrls(row.content)) {
-      const domain = domainOf(url);
-      if (!domain) continue;
-      const kind = isArticleUrl(url) ? 'article' : isToolUrl(url, clean) ? 'tool' : null;
-      if (!kind) continue;
-      const key = normalizeUrlKey(url);
-      const bucket = linkBuckets.get(key) ?? {
-        kind,
-        title: titleFromLinkContext(row.content, url),
-        url,
-        domain,
-        count: 0,
-        groups: new Set<string>(),
-        last_seen: row.time,
-        snippets: [],
-      };
-      bucket.count++;
-      bucket.groups.add(row.chatroom_id);
-      bucket.last_seen = bucket.last_seen > row.time ? bucket.last_seen : row.time;
-      if (clean) bucket.snippets.push(clean.slice(0, 80));
-      linkBuckets.set(key, bucket);
-    }
+  }
+
+  for (const row of linkRows) {
+    const kind = isArticleUrl(row.canonical_url) ? 'article' : isToolUrl(row.canonical_url, row.content) ? 'tool' : null;
+    if (!kind) continue;
+    const key = normalizeUrlKey(row.canonical_url);
+    const clean = cleanContent(row.content);
+    const bucket = linkBuckets.get(key) ?? {
+      kind,
+      title: (row.title || titleFromLinkContext(row.content, row.url)).slice(0, 80),
+      url: row.url,
+      domain: row.domain || domainOf(row.canonical_url),
+      count: 0,
+      groups: new Set<string>(),
+      last_seen: row.time,
+      snippets: [],
+      sources: new Set<string>(),
+    };
+    bucket.count++;
+    bucket.groups.add(row.chatroom_id);
+    bucket.sources.add(row.source);
+    bucket.last_seen = bucket.last_seen > row.time ? bucket.last_seen : row.time;
+    if (clean) bucket.snippets.push(clean.slice(0, 80));
+    linkBuckets.set(key, bucket);
  }

  const mustRead = candidates
@@ -430,6 +465,7 @@ function buildLinkHighlights(
      groups: Set<string>;
      last_seen: string;
      snippets: string[];
+      sources: Set<string>;
    }
  >,
 ): DashboardLinkHighlight[] {
@@ -441,6 +477,7 @@ function buildLinkHighlights(
        title: item.title,
        url: item.url,
        domain: item.domain,
+        source: preferredLinkSource(item.sources),
        score,
        verdict: verdictForLink(item.kind, item.count, item.groups.size, item.snippets.join(' ')),
        count: item.count,
@@ -452,6 +489,13 @@ function buildLinkHighlights(
    .slice(0, MAX_LINK_HIGHLIGHTS);
 }

+function preferredLinkSource(sources: Set<string>): string {
+  if (sources.has('wechat_raw')) return 'wechat_raw';
+  if (sources.has('public_search')) return 'public_search';
+  if (sources.has('manual')) return 'manual';
+  return Array.from(sources)[0] ?? 'plain_url';
+}
+
 function buildPeopleRadar(
  sourceMap: Map<
    string,
@@ -602,14 +646,6 @@ function minusDays(date: string, days: number): string {
  return `${d.getFullYear()}-${String(d.getMonth() + 1).padStart(2, '0')}-${String(d.getDate()).padStart(2, '0')}`;
 }

-function extractUrls(content: string): string[] {
-  const decoded = decodeHtml(content);
-  return Array.from(decoded.matchAll(URL_GLOBAL_RE))
-    .map((m) => cleanUrl(m[0]))
-    .filter(Boolean)
-    .slice(0, 6);
-}
-
 function cleanUrl(raw: string): string {
  return raw
    .replace(/[),，。；;!?！？、\]}>]+$/g, '')
@@ -645,7 +681,7 @@ function isArticleUrl(raw: string): boolean {
    if (host === 'mp.weixin.qq.com') return true;
    if ((host === 'x.com' || host === 'twitter.com') && /\/status\/\d{12,}/.test(u.pathname)) return true;
    if (host === 'youtube.com' || host === 'youtu.be') return true;
-    return /zhihu|toutiao|sohu|163\.com|qq\.com|medium\.com|substack\.com|juejin\.cn/i.test(host);
+    return /zhihu|toutiao|sohu|163\.com|qq\.com|medium\.com|substack\.com|juejin\.cn|podcasts\.apple\.com|podscan\.fm/i.test(host);
  } catch {
    return false;
  }
@@ -84,6 +84,34 @@ function migrate(d: Database.Database) {
    CREATE INDEX IF NOT EXISTS idx_messages_timestamp ON messages(timestamp DESC);
    CREATE INDEX IF NOT EXISTS idx_messages_sender ON messages(sender);

+    CREATE TABLE IF NOT EXISTS message_links (
+      chatroom_id TEXT NOT NULL,
+      local_id INTEGER NOT NULL,
+      date TEXT NOT NULL,
+      sender TEXT NOT NULL,
+      time TEXT NOT NULL,
+      timestamp INTEGER NOT NULL,
+      url TEXT NOT NULL,
+      canonical_url TEXT NOT NULL,
+      title TEXT,
+      description TEXT,
+      domain TEXT NOT NULL,
+      source TEXT NOT NULL,
+      raw_kind TEXT NOT NULL,
+      confidence REAL NOT NULL DEFAULT 1,
+      created_at INTEGER NOT NULL,
+      PRIMARY KEY (chatroom_id, local_id, canonical_url)
+    );
+
+    CREATE INDEX IF NOT EXISTS idx_message_links_date
+      ON message_links(date, timestamp DESC);
+    CREATE INDEX IF NOT EXISTS idx_message_links_canonical
+      ON message_links(canonical_url);
+    CREATE INDEX IF NOT EXISTS idx_message_links_domain
+      ON message_links(domain);
+    CREATE INDEX IF NOT EXISTS idx_message_links_source
+      ON message_links(source);
+
    CREATE TABLE IF NOT EXISTS topics (
      id INTEGER PRIMARY KEY AUTOINCREMENT,
      date TEXT NOT NULL,
@@ -115,6 +143,9 @@ function migrate(d: Database.Database) {
      PRIMARY KEY (date, version)
    );

+    CREATE INDEX IF NOT EXISTS idx_link_intelligence_cache_generated
+      ON link_intelligence_cache(generated_at DESC);
+
    CREATE TABLE IF NOT EXISTS sync_state (
      chatroom_id TEXT PRIMARY KEY,
      last_synced_at INTEGER NOT NULL,
@@ -136,30 +167,50 @@ function migrate(d: Database.Database) {
  ensureColumn(d, 'sync_state', 'total_chunks', 'INTEGER NOT NULL DEFAULT 0');
 }

-function ensureColumn(d: Database.Database, table: string, name: string, definition: string) {
+function ensureColumn(
+  d: Database.Database,
+  table: string,
+  name: string,
+  definition: string,
+) {
  const rows = d.prepare(`PRAGMA table_info(${table})`).all() as Array<{ name: string }>;
  if (rows.some((r) => r.name === name)) return;
  d.prepare(`ALTER TABLE ${table} ADD COLUMN ${name} ${definition}`).run();
 }

-const SEED_VERSION = 'wechat_radar_v1_2026_05_24';
+const SEED_VERSION = 'qiaomu_v2_2026_05_23';

 const DEFAULT_GROUPS: Array<{ name: string; color: string; emoji: string }> = [
-  { name: 'AI / Coding', color: '#7dd3a8', emoji: '💻' },
-  { name: 'Tools', color: '#f59e0b', emoji: '🛠️' },
-  { name: 'Articles', color: '#06b6d4', emoji: '📚' },
-  { name: 'Business', color: '#10b981', emoji: '💼' },
-  { name: 'Events', color: '#f97316', emoji: '📅' },
-  { name: 'Research', color: '#a855f7', emoji: '🔬' },
-  { name: 'Lifestyle', color: '#fb7185', emoji: '🏠' },
+  { name: 'AI产品蝗虫团', color: '#ef4444', emoji: '🐝' },
+  { name: '自营/读者群', color: '#22c55e', emoji: '🌟' },
+  { name: 'WaytoAGI', color: '#06b6d4', emoji: '🛸' },
+  { name: 'HowOneAI', color: '#0ea5e9', emoji: '🚀' },
+  { name: 'Vibe Coding · 编程', color: '#6366f1', emoji: '💻' },
+  { name: 'AIGC · 内容创作', color: '#ec4899', emoji: '🎨' },
+  { name: 'AI 学术', color: '#a855f7', emoji: '🎓' },
+  { name: 'AI 商业 · 营销', color: '#10b981', emoji: '💰' },
+  { name: 'AI 工具用户群', color: '#f59e0b', emoji: '🛠️' },
+  { name: '付费社区', color: '#eab308', emoji: '💎' },
+  { name: 'AI 圈社交', color: '#8b5cf6', emoji: '🤖' },
+  { name: '大佬 · 媒体圈', color: '#f97316', emoji: '📰' },
+  { name: '行业活动', color: '#22d3ee', emoji: '🎯' },
+  { name: '生活 · 兴趣', color: '#fb7185', emoji: '🏘️' },
 ];

 function seed(d: Database.Database) {
-  const meta = d.prepare("SELECT value FROM meta WHERE key = 'seed_version'").get() as { value: string } | undefined;
+  const meta = d
+    .prepare("SELECT value FROM meta WHERE key = 'seed_version'")
+    .get() as { value: string } | undefined;
+
  if (meta?.value === SEED_VERSION) return;

+  // Check if any groups have user tags — if so, leave them alone (additive seed).
  const tagged = d.prepare('SELECT COUNT(*) AS n FROM group_tags').get() as { n: number };
-  if (tagged.n === 0) d.prepare('DELETE FROM groups').run();
+
+  if (tagged.n === 0) {
+    // Safe to wipe and re-seed.
+    d.prepare('DELETE FROM groups').run();
+  }

  const now = Date.now();
  const insertOrIgnore = d.prepare(
@@ -169,5 +220,7 @@ function seed(d: Database.Database) {
    DEFAULT_GROUPS.forEach((g, i) => insertOrIgnore.run(g.name, g.color, g.emoji, i, now));
  })();

-  d.prepare("INSERT OR REPLACE INTO meta (key, value) VALUES ('seed_version', ?)").run(SEED_VERSION);
+  d.prepare(
+    "INSERT OR REPLACE INTO meta (key, value) VALUES ('seed_version', ?)",
+  ).run(SEED_VERSION);
 }
@@ -2,35 +2,83 @@ import type { GroupRow } from './groups';

 export function classifyGroupHeuristic(name: string, summary: string, groups: GroupRow[]) {
  const text = `${name} ${summary}`.toLowerCase();
-  const lookup = (target: string) => groups.find((g) => g.name.toLowerCase().includes(target.toLowerCase()));
+  const lookup = (target: string) => groups.find((g) => g.name.includes(target));

-  if (/vibe.?coding|coding|代码|编程|developer|dev|cli|mcp|skills?|github|开源|agent|gpt|claude|llm/i.test(text)) {
-    const t = lookup('AI / Coding');
-    if (t) return { group_id: t.id, group_name: t.name, reason: 'AI / Coding keywords' };
+  if (/蝗虫团|huangchong/i.test(name)) {
+    const t = lookup('蝗虫');
+    if (t) return { group_id: t.id, group_name: t.name, reason: '蝗虫团系列' };
  }
-  if (/工具|产品|插件|内测|api|chrome|notion|obsidian|飞书|workflow|workspace|效率/i.test(text)) {
-    const t = lookup('Tools');
-    if (t) return { group_id: t.id, group_name: t.name, reason: 'Tools / product keywords' };
+  if (/自营|公众号读者|读者群|用户群|粉丝群/i.test(name)) {
+    const t = lookup('自营/读者群');
+    if (t) return { group_id: t.id, group_name: t.name, reason: '自营 / 读者群' };
  }
-  if (/文章|公众号|日报|newsletter|读者|知识库|教程|报告|访谈|paper|论文/i.test(text)) {
-    const t = lookup('Articles');
-    if (t) return { group_id: t.id, group_name: t.name, reason: 'Articles / knowledge keywords' };
+  if (/waytoagi|通往agi|通往ai|通往 ai/i.test(name)) {
+    const t = lookup('WaytoAGI');
+    if (t) return { group_id: t.id, group_name: t.name, reason: 'WaytoAGI 系列' };
  }
-  if (/商业|营销|增长|seo|geo|销售|客户|采购|团购|合作|商务|创业|投资/i.test(text)) {
-    const t = lookup('Business');
-    if (t) return { group_id: t.id, group_name: t.name, reason: 'Business keywords' };
+  if (/howoneai|howone/i.test(name)) {
+    const t = lookup('HowOneAI');
+    if (t) return { group_id: t.id, group_name: t.name, reason: 'HowOneAI 系列' };
  }
-  if (/活动|报名|直播|大会|线下|分享会|训练营|课程|会议|meetup|workshop/i.test(text)) {
-    const t = lookup('Events');
-    if (t) return { group_id: t.id, group_name: t.name, reason: 'Event keywords' };
+  if (
+    /vibe.?coding|vibecoding|vibe first|cherry studio|cli|claude.?skills|clawdbot|codepilot|mcp|cola|geoflow|refly|camel|eigent|thinkinai|skills|all in cli/i.test(
+      text,
+    )
+  ) {
+    const t = lookup('Vibe Coding');
+    if (t) return { group_id: t.id, group_name: t.name, reason: '编程 / Skills / CLI' };
  }
-  if (/研究|学术|论文|paper|模型|实验|benchmark|评测/i.test(text)) {
-    const t = lookup('Research');
-    if (t) return { group_id: t.id, group_name: t.name, reason: 'Research keywords' };
+  if (/学术|论文|paper|未来硅世界|研究室|nixy|simonlin|博文视点|knowledge|灵枭/i.test(text)) {
+    const t = lookup('AI 学术');
+    if (t) return { group_id: t.id, group_name: t.name, reason: '学术 / 论文' };
  }
-  if (/生活|阅读|运动|小区|邻里|钓鱼|健身|跑步|英语|校友|投资主题/i.test(text)) {
-    const t = lookup('Lifestyle');
-    if (t) return { group_id: t.id, group_name: t.name, reason: 'Lifestyle keywords' };
+  if (/seo|geo|商业化|营销|kol|gaidn|adg|vip|生财|appsail|tutti|商业|broker|出版|收付款|社交新品|社群/i.test(text)) {
+    const t = lookup('AI 商业');
+    if (t) return { group_id: t.id, group_name: t.name, reason: '商业 / 营销 / KOL' };
+  }
+  if (/aigc|图|视频|音乐|spy|拍我ai|ai媒体|ai音视频|创意|graceful|创作|graphic|listenhub|notetomp|youmind|短视频|video|music|illustrat/i.test(text)) {
+    const t = lookup('AIGC');
+    if (t) return { group_id: t.id, group_name: t.name, reason: 'AIGC / 内容创作' };
+  }
+  if (/vip|烟花|修饼|传术师|生财有术|兔子ai|ai领导力|早鸟|内测|种子用户|订阅用户|私董|学员|api 渠道|一人公司|训练营|课程/i.test(text)) {
+    const t = lookup('付费社区');
+    if (t) return { group_id: t.id, group_name: t.name, reason: '付费 / 内测 / VIP' };
+  }
+  if (/用户群|用户中文|内测群|jackywine|mindcode|remio|cherry|camel|refly|cola|geoflow|hosi|aigocode|appsail|tutti|api/i.test(text)) {
+    const t = lookup('AI 工具用户群');
+    if (t) return { group_id: t.id, group_name: t.name, reason: '工具用户群' };
+  }
+  if (/神的孩子|明人明言|先行者|agi bar|智能体成精|life hacker|超级玩家|未来趋势|agent橘|新物种|创造营|不息为体|未知书社|新互联网/i.test(text)) {
+    const t = lookup('AI 圈社交');
+    if (t) return { group_id: t.id, group_name: t.name, reason: 'AI 圈社交' };
+  }
+  if (/donews|何夕|辛亥|对接群|百度世界|央馆|火山方舟|43talks|tgo|商务|媒体/i.test(text)) {
+    const t = lookup('大佬');
+    if (t) return { group_id: t.id, group_name: t.name, reason: '大佬 / 媒体圈' };
+  }
+  if (/活动现场|聚餐|筹备组|聚会|开播|直播|线下|大会|分享会|黑客松|日历/i.test(text)) {
+    const t = lookup('行业活动');
+    if (t) return { group_id: t.id, group_name: t.name, reason: '一次性活动群' };
+  }
+  if (/钓友|路亚|果粉|大家庭|班级|邻里|小区|羽毛球|健身|跑步|徒步|阅读|共读|英语|校友|歌友|篮球/i.test(text)) {
+    const t = lookup('生活');
+    if (t) return { group_id: t.id, group_name: t.name, reason: '生活 / 兴趣' };
+  }
+  if (/粉丝|fans|读者/i.test(text)) {
+    const t = lookup('AI 圈社交');
+    if (t) return { group_id: t.id, group_name: t.name, reason: '粉丝团 / 读者群' };
+  }
+  if (/财经|股票|投资|基金|币圈|crypto|trade/i.test(text)) {
+    const t = lookup('AI 商业');
+    if (t) return { group_id: t.id, group_name: t.name, reason: '财经 / 投资' };
+  }
+  if (/x boost|twitter|推特|x kol/i.test(text)) {
+    const t = lookup('AIGC');
+    if (t) return { group_id: t.id, group_name: t.name, reason: 'X / 推特运营' };
+  }
+  if (/ai|agent|gpt|claude|llm|coding|开源/i.test(text)) {
+    const t = lookup('AI 圈社交');
+    if (t) return { group_id: t.id, group_name: t.name, reason: '通用 AI（兜底）' };
  }
  return null;
 }
@@ -13,7 +13,7 @@ const TITLE_FETCH_TIMEOUT_MS = 1400;
 const MAX_TITLE_GENERATION_ITEMS = 80;
 const CODEX_TIMEOUT_MS = Number(process.env.WECHAT_RADAR_LINK_CODEX_TIMEOUT_MS ?? 180_000);
 const CODEX_MODEL = process.env.WECHAT_RADAR_CODEX_MODEL;
-const LINK_INTELLIGENCE_CACHE_VERSION = 'v6';
+const LINK_INTELLIGENCE_CACHE_VERSION = 'v8';
 const LINK_INTELLIGENCE_CACHE_TTL_SECONDS = 60 * 60 * 24;

 const TOOL_HINT_RE =
@@ -27,6 +27,10 @@ const ARTICLE_HOSTS = [
  'page.om.qq.com',
  'www.163.com',
  'mparticle.uc.cn',
+  'podcasts.apple.com',
+  'open.spotify.com',
+  'podcasters.spotify.com',
+  'podscan.fm',
 ];

 const TOOL_HOST_HINTS = [
@@ -61,6 +65,12 @@ interface MessageLinkRow {
  time: string;
  timestamp: number;
  type: string;
+  url: string;
+  canonical_url: string;
+  link_title: string | null;
+  domain: string;
+  source: string;
+  confidence: number;
 }

 export interface LinkIntelligenceItem {
@@ -80,6 +90,7 @@ export interface LinkIntelligenceItem {
    time: string;
    local_id: number;
    snippet: string;
+    source: string;
  }>;
  dedupe_key?: string;
 }
@@ -134,46 +145,6 @@ function decodeHtmlEntities(s: string): string {
    .replace(/&#39;/g, "'");
 }

-function cleanUrl(raw: string): string {
-  return decodeHtmlEntities(raw)
-    .replace(/[),，。；;!?！？、\]}>]+$/g, '')
-    .replace(/\.{3,}$/g, '')
-    .trim();
-}
-
-function normalizeUrl(raw: string): string | null {
-  if (raw.includes('...') || raw.includes('…')) return null;
-  try {
-    const u = new URL(cleanUrl(raw));
-    u.hash = '';
-    for (const key of ['utm_source', 'utm_medium', 'utm_campaign', 'utm_term', 'utm_content']) {
-      u.searchParams.delete(key);
-    }
-    return u.toString();
-  } catch {
-    return null;
-  }
-}
-
-function extractUrls(content: string): string[] {
-  const decoded = decodeHtmlEntities(content);
-  const urls = new Set<string>();
-
-  for (const m of decoded.matchAll(/imgsourceurl="([^"]+)"/g)) {
-    try {
-      urls.add(decodeURIComponent(m[1]));
-    } catch {
-      urls.add(m[1]);
-    }
-  }
-
-  for (const m of decoded.matchAll(/https?:\/\/[^\s<>"']+/g)) {
-    urls.add(cleanUrl(m[0]));
-  }
-
-  return Array.from(urls).filter(Boolean);
-}
-
 function domainOf(url: string): string {
  try {
    return new URL(url).hostname.replace(/^www\./, '');
@@ -532,11 +503,26 @@ export async function getDailyLinkIntelligence(

  const rows = db()
    .prepare(
-      `SELECT chatroom_id, local_id, sender, content, time, timestamp, type
-       FROM messages
-       WHERE date = ?
-         AND content LIKE '%http%'
-       ORDER BY timestamp DESC
+      `SELECT
+         m.chatroom_id,
+         m.local_id,
+         m.sender,
+         m.content,
+         m.time,
+         m.timestamp,
+         m.type,
+         ml.url,
+         ml.canonical_url,
+         ml.title AS link_title,
+         ml.domain,
+         ml.source,
+         ml.confidence
+       FROM message_links ml
+       JOIN messages m
+         ON m.chatroom_id = ml.chatroom_id
+        AND m.local_id = ml.local_id
+       WHERE ml.date = ?
+       ORDER BY ml.timestamp DESC
       LIMIT ?`,
    )
    .all(date, MAX_MESSAGES) as MessageLinkRow[];
@@ -548,10 +534,7 @@ export async function getDailyLinkIntelligence(
  const buckets = new Map<string, LinkIntelligenceItem>();

  for (const row of rows) {
-    for (const raw of extractUrls(row.content)) {
-      const canonical = normalizeUrl(raw);
-      if (!canonical) continue;
-
+      const canonical = row.canonical_url;
      const kind: LinkKind | null = isArticleLink(canonical)
        ? 'article'
        : isToolLink(canonical, row.content)
@@ -568,6 +551,7 @@ export async function getDailyLinkIntelligence(
        time: row.time,
        local_id: row.local_id,
        snippet: cleanSnippet(row.content),
+        source: row.source,
      };

      if (existing) {
@@ -581,10 +565,10 @@ export async function getDailyLinkIntelligence(
      } else {
        buckets.set(key, {
          kind,
-          url: raw,
+          url: row.url,
          canonical_url: canonical,
-          title: titleFromContext(row.content, raw),
-          domain: domainOf(canonical),
+          title: row.link_title || titleFromContext(row.content, row.url),
+          domain: row.domain || domainOf(canonical),
          count: 1,
          group_count: 1,
          first_seen: row.time,
@@ -592,7 +576,6 @@ export async function getDailyLinkIntelligence(
          sources: [source],
        });
      }
-    }
  }

  const sortItems = (kind: LinkKind, limit = MAX_ITEMS_PER_KIND) =>
@@ -0,0 +1,290 @@
+import { db } from './db';
+import type { MessageRow } from './messages-store';
+
+export type MessageLinkSource = 'wechat_raw' | 'plain_url' | 'public_search' | 'manual';
+
+export interface ParsedMessageLink {
+  url: string;
+  canonical_url: string;
+  title: string | null;
+  description: string | null;
+  domain: string;
+  source: MessageLinkSource;
+  raw_kind: string;
+  confidence: number;
+}
+
+type LinkInput = Pick<
+  MessageRow,
+  'chatroom_id' | 'local_id' | 'date' | 'sender' | 'content' | 'time' | 'timestamp'
+>;
+
+export function decodeHtmlEntities(s: string): string {
+  return s
+    .replace(/&amp;/g, '&')
+    .replace(/&lt;/g, '<')
+    .replace(/&gt;/g, '>')
+    .replace(/&quot;/g, '"')
+    .replace(/&#39;/g, "'")
+    .replace(/&#x([0-9a-f]+);/gi, (_, hex: string) => String.fromCodePoint(Number.parseInt(hex, 16)))
+    .replace(/&#(\d+);/g, (_, num: string) => String.fromCodePoint(Number.parseInt(num, 10)));
+}
+
+export function cleanUrl(raw: string): string {
+  return decodeHtmlEntities(raw)
+    .replace(/[),，。；;!?！？、\]}>]+$/g, '')
+    .replace(/\.{3,}$/g, '')
+    .trim();
+}
+
+export function normalizeUrl(raw: string): string | null {
+  if (!raw || raw.includes('...') || raw.includes('…')) return null;
+  try {
+    const u = new URL(cleanUrl(raw));
+    u.hash = '';
+    for (const key of ['utm_source', 'utm_medium', 'utm_campaign', 'utm_term', 'utm_content']) {
+      u.searchParams.delete(key);
+    }
+    return u.toString();
+  } catch {
+    return null;
+  }
+}
+
+export function domainOf(url: string): string {
+  try {
+    return new URL(url).hostname.replace(/^www\./, '');
+  } catch {
+    return '';
+  }
+}
+
+function isWechatArticleUrl(url: string): boolean {
+  try {
+    const u = new URL(cleanUrl(url));
+    return u.hostname === 'mp.weixin.qq.com' && (/^\/s\/?/.test(u.pathname) || u.searchParams.has('__biz'));
+  } catch {
+    return false;
+  }
+}
+
+function tagText(content: string, tag: string): string {
+  const text = content.match(new RegExp(`<${tag}[^>]*>([\\s\\S]*?)<\\/${tag}>`, 'i'))?.[1] ?? '';
+  return decodeHtmlEntities(text)
+    .replace(/<!\[CDATA\[([\s\S]*?)\]\]>/g, '$1')
+    .replace(/\s+/g, ' ')
+    .trim();
+}
+
+function attrValues(content: string, attr: string): string[] {
+  return Array.from(content.matchAll(new RegExp(`${attr}=["']([^"']+)["']`, 'gi')))
+    .map((m) => decodeHtmlEntities(m[1]).trim())
+    .filter(Boolean);
+}
+
+function titleFromContext(content: string, url: string): string | null {
+  const xmlTitle = tagText(content, 'title');
+  if (xmlTitle) return xmlTitle.slice(0, 160);
+
+  const decoded = decodeHtmlEntities(content).replace(/<\?xml[\s\S]+?<\/msg>/g, ' ');
+  const lines = decoded
+    .split(/\n+/)
+    .map((line) =>
+      line
+        .replace(url, '')
+        .replace(/https?:\/\/\S+/g, '')
+        .replace(/\[引用\]/g, '')
+        .replace(/^\s*↳\s*/, '')
+        .replace(/^\s*\[链接\]\s*/, '')
+        .trim(),
+    )
+    .filter((line) => line.length >= 4 && line.length <= 120);
+
+  return lines.find((line) => !/^[@#\d\s:：-]+$/.test(line))?.slice(0, 160) ?? null;
+}
+
+export function extractMessageLinks(content: string): ParsedMessageLink[] {
+  const decoded = decodeHtmlEntities(content);
+  const hasXml = /<msg[\s>]|<appmsg[\s>]/i.test(decoded);
+  const xmlTitle = tagText(decoded, 'title') || null;
+  const xmlDescription = tagText(decoded, 'des') || tagText(decoded, 'digest') || null;
+  const candidates: Array<{ url: string; source: MessageLinkSource; raw_kind: string; confidence: number }> = [];
+
+  for (const tag of ['url', 'lowurl']) {
+    const url = tagText(decoded, tag);
+    if (url && isWechatArticleUrl(url)) {
+      candidates.push({ url, source: 'wechat_raw', raw_kind: `appmsg_${tag}`, confidence: 1 });
+    }
+  }
+
+  for (const value of attrValues(decoded, 'url')) {
+    if (!isWechatArticleUrl(value)) continue;
+      candidates.push({
+        url: value,
+        source: hasXml ? 'wechat_raw' : 'plain_url',
+        raw_kind: hasXml ? 'appmsg_attr_url' : 'plain_attr_url',
+        confidence: hasXml ? 0.98 : 0.9,
+      });
+  }
+
+  for (const m of decoded.matchAll(/https?:\/\/[^\s<>"']+/g)) {
+    const rawUrl = cleanUrl(m[0]);
+    const article = isWechatArticleUrl(rawUrl);
+    const source: MessageLinkSource = hasXml && article ? 'wechat_raw' : 'plain_url';
+    candidates.push({
+      url: rawUrl,
+      source,
+      raw_kind: hasXml && article ? 'appmsg_url_text' : 'plain_url',
+      confidence: article ? 0.96 : 0.9,
+    });
+  }
+
+  const out = new Map<string, ParsedMessageLink>();
+  for (const c of candidates) {
+    const canonical = normalizeUrl(c.url);
+    if (!canonical) continue;
+    const domain = domainOf(canonical);
+    if (!domain) continue;
+    const existing = out.get(canonical);
+    if (existing && existing.confidence >= c.confidence) continue;
+    out.set(canonical, {
+      url: cleanUrl(c.url),
+      canonical_url: canonical,
+      title: xmlTitle ?? titleFromContext(decoded, c.url),
+      description: xmlDescription,
+      domain,
+      source: c.source,
+      raw_kind: c.raw_kind,
+      confidence: c.confidence,
+    });
+  }
+
+  return Array.from(out.values());
+}
+
+const upsertMessageLink = () =>
+  db().prepare(`
+    INSERT INTO message_links (
+      chatroom_id, local_id, date, sender, time, timestamp,
+      url, canonical_url, title, description, domain, source, raw_kind, confidence, created_at
+    )
+    VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+    ON CONFLICT(chatroom_id, local_id, canonical_url) DO UPDATE SET
+      url = excluded.url,
+      title = COALESCE(excluded.title, message_links.title),
+      description = COALESCE(excluded.description, message_links.description),
+      domain = excluded.domain,
+      source = excluded.source,
+      raw_kind = excluded.raw_kind,
+      confidence = excluded.confidence
+  `);
+
+export function upsertLinksForMessage(message: LinkInput): number {
+  const links = extractMessageLinks(message.content);
+  if (links.length === 0) return 0;
+  const stmt = upsertMessageLink();
+  let changed = 0;
+  for (const link of links) {
+    const r = stmt.run(
+      message.chatroom_id,
+      message.local_id,
+      message.date,
+      message.sender ?? '',
+      message.time ?? '',
+      message.timestamp ?? 0,
+      link.url,
+      link.canonical_url,
+      link.title,
+      link.description,
+      link.domain,
+      link.source,
+      link.raw_kind,
+      link.confidence,
+      Date.now(),
+    );
+    changed += r.changes;
+  }
+  return changed;
+}
+
+export function upsertResolvedLinkForMessage(input: {
+  chatroom_id: string;
+  local_id: number;
+  url: string;
+  title?: string | null;
+  description?: string | null;
+  source: Extract<MessageLinkSource, 'public_search' | 'manual'>;
+  confidence?: number;
+}): { ok: boolean; error?: string } {
+  const message = db()
+    .prepare(
+      `SELECT chatroom_id, local_id, sender, content, time, timestamp, type, date
+       FROM messages
+       WHERE chatroom_id = ? AND local_id = ?`,
+    )
+    .get(input.chatroom_id, input.local_id) as MessageRow | undefined;
+
+  if (!message) return { ok: false, error: 'message not found' };
+
+  const canonical = normalizeUrl(input.url);
+  if (!canonical) return { ok: false, error: 'invalid url' };
+
+  const domain = domainOf(canonical);
+  if (!domain) return { ok: false, error: 'invalid domain' };
+
+  upsertMessageLink().run(
+    message.chatroom_id,
+    message.local_id,
+    message.date,
+    message.sender ?? '',
+    message.time ?? '',
+    message.timestamp ?? 0,
+    cleanUrl(input.url),
+    canonical,
+    input.title?.trim() || titleFromContext(message.content, input.url),
+    input.description?.trim() || null,
+    domain,
+    input.source,
+    input.source,
+    input.confidence ?? (input.source === 'manual' ? 0.95 : 0.72),
+    Date.now(),
+  );
+
+  return { ok: true };
+}
+
+export function backfillMessageLinks(since?: string, until?: string): { scanned: number; links: number } {
+  const clauses = ["(content LIKE '%http%' OR content LIKE '%<url>%' OR content LIKE '%imgsourceurl=%')"];
+  const params: string[] = [];
+  if (since) {
+    clauses.push('date >= ?');
+    params.push(since);
+  }
+  if (until) {
+    clauses.push('date <= ?');
+    params.push(until);
+  }
+
+  const rows = db()
+    .prepare(
+      `SELECT chatroom_id, local_id, sender, content, time, timestamp, type, date
+       FROM messages
+       WHERE ${clauses.join(' AND ')}
+       ORDER BY timestamp DESC`,
+    )
+    .all(...params) as MessageRow[];
+
+  let links = 0;
+  const tx = db().transaction(() => {
+    db()
+      .prepare(
+        `DELETE FROM message_links
+         WHERE source = 'wechat_raw'
+           AND canonical_url NOT LIKE '%://mp.weixin.qq.com/%'`,
+      )
+      .run();
+    for (const row of rows) links += upsertLinksForMessage(row);
+  });
+  tx();
+  return { scanned: rows.length, links };
+}
@@ -1,4 +1,5 @@
 import { db } from './db';
+import { upsertLinksForMessage } from './message-links';
 import type { WxMessage } from './wx-types';

 export interface MessageRow extends WxMessage {
@@ -42,6 +43,15 @@ export function bulkInsertMessages(chatroomId: string, messages: WxMessage[]): n
        m.type ?? '',
        dateOfMessage(m),
      );
+      upsertLinksForMessage({
+        chatroom_id: chatroomId,
+        local_id: m.local_id,
+        sender: m.sender ?? '',
+        content: m.content ?? '',
+        time: m.time ?? '',
+        timestamp: m.timestamp ?? 0,
+        date: dateOfMessage(m),
+      });
      if (r.changes > 0) inserted++;
    }
  });
@@ -42,19 +42,35 @@ type TopicWithMembers = {
 };

 function cleanContent(s: string): string {
-  return s
+  const text = s
    .replace(/\[图片\]\s*local_id=\d+/g, '')
-    .replace(/\[引用\][^\n]*\n?/g, '')
    .replace(/\[小程序\][^\n]*/g, '')
-    .replace(/↳\s*[^\n]*/g, '')
    .replace(/<\?xml[\s\S]+?\?>[\s\S]*?<\/msg>/g, '')
+    .replace(/\[(?:链接|链接\/文件)\]\s*(?:当前(?:微信)?版本不支持展示该内容，请升级至?最新(?:版|版本)|当前版本不支持展示该内容，请升级至最新版本)[。.]?/gi, ' ')
+    .replace(/当前(?:微信)?版本不支持展示该内容，请升级至?最新(?:版|版本)[。.]?/gi, ' ')
    .replace(/https?:\/\/\S+/g, ' ')
+    .replace(/\b[a-f0-9]{24,}\b/gi, ' ')
+    .replace(/\b\d{12,}\b/g, ' ')
    .trim();
+
+  const parts = text
+    .split(/↳|\\n|\n/)
+    .map((part) =>
+      part
+        .replace(/^\[引用\]\s*/g, '')
+        .replace(/^\[(?:链接|链接\/文件|图片|视频|表情)\]\s*/g, '')
+        .replace(/\s+/g, ' ')
+        .trim(),
+    )
+    .filter((part) => part.length >= MIN_MESSAGE_LENGTH && !isPlaceholderTitle(part));
+
+  return parts[0] ?? text.replace(/\s+/g, ' ').trim();
 }

 // 这些消息整体就是占位符 / wrapper，无实质内容
 const PLACEHOLDER_PATTERNS = [
-  /^\[链接\]\s*当前版本不支持/,
+  /^\[链接\]\s*当前(?:微信)?版本不支持/,
+  /^当前(?:微信)?版本不支持展示该内容，请升级至?最新(?:版|版本)[。.]?$/,
  /^\[文件\]\s*[^\s]+\.\w+\s*$/,
  /^\[视频\]\s*$/,
  /^\[音频\]\s*$/,
@@ -70,7 +86,16 @@ const PLACEHOLDER_PATTERNS = [

 function isPlaceholderOnly(content: string): boolean {
  if (!content) return true;
-  return PLACEHOLDER_PATTERNS.some((p) => p.test(content));
+  return PLACEHOLDER_PATTERNS.some((p) => p.test(content)) || isPlaceholderTitle(content);
+}
+
+function isPlaceholderTitle(value: string): boolean {
+  const text = value
+    .replace(/^\[(?:链接|链接\/文件|图片|视频|表情)\]\s*/g, '')
+    .replace(/\s+/g, ' ')
+    .trim();
+  if (!text) return true;
+  return /^(?:当前(?:微信)?版本不支持展示该内容，请升级至?最新(?:版|版本)|当前版本不支持|请升级至最新版本)[。.]?$/i.test(text);
 }

 function loadCandidateMessages(date: string): SourceMsg[] {
@@ -237,6 +262,8 @@ function buildExtractionPrompt(
 - 合并同一事件、产品、工具、论文、观点、问题及其追问/回应/转述。
 - 优先保留跨群出现的话题；同一群内高密度连续讨论也可以保留。
 - 忽略问候、纯闲聊、广告、无上下文碎片、纯占位内容和过泛的「AI 很火」类讨论。
+- 严禁把「当前版本不支持展示该内容」「当前微信版本不支持展示该内容」「请升级至最新版本」这类微信占位文案当作话题标题或摘要。
+- 遇到链接/视频/小程序占位时，只能根据前后文里真正有人讨论的对象命名；没有可读上下文就丢弃。
 - 每个话题至少包含 ${MIN_MESSAGES_PER_TOPIC} 条消息。
 - 最多输出 ${maxTopics} 个话题，按重要性排序。
 - title 用 8-15 个汉字，优先写产品名/事件名/讨论焦点。
@@ -268,6 +295,7 @@ function buildMergePrompt(date: string, drafts: LlmTopic[], maxTopics: number):
 任务要求：
 - 合并语义相同或强相关的话题草稿，message_ids 取并集。
 - 删除过泛、重复、证据不足的话题。
+- 删除微信升级提示、资源封面、头像链接、无语义数字串等占位话题。
 - 每个最终话题至少包含 ${MIN_MESSAGES_PER_TOPIC} 条消息。
 - 最多输出 ${maxTopics} 个最终话题，按重要性排序。
 - title 用 8-15 个汉字，summary 用 1-2 句中文。
@@ -294,7 +322,7 @@ function normalizeTopics(rawTopics: LlmTopic[], messageMap: Map<string, SourceMs
    seenSignatures.add(signature);

    out.push({
-      title: (raw.title || members[0].content.slice(0, 16) || '未命名话题').slice(0, 80),
+      title: cleanTopicTitle(raw.title, members),
      summary: (raw.summary || '').slice(0, 400),
      members,
      groupSet: new Set(members.map((m) => m.chatroom_id)),
@@ -304,6 +332,17 @@ function normalizeTopics(rawTopics: LlmTopic[], messageMap: Map<string, SourceMs
  return out.sort((a, b) => b.members.length - a.members.length).slice(0, MAX_TOPICS_TO_SAVE);
 }

+function cleanTopicTitle(title: string, members: SourceMsg[]): string {
+  const cleaned = cleanContent(title || '');
+  if (cleaned && !isPlaceholderTitle(cleaned)) return cleaned.slice(0, 80);
+  const candidate = members
+    .slice()
+    .sort((a, b) => b.content.length - a.content.length)
+    .map((m) => cleanContent(m.content))
+    .find((text) => text.length >= 8 && !isPlaceholderTitle(text));
+  return (candidate || '未命名话题').slice(0, 80);
+}
+
 async function aggregateWithCodex(
  date: string,
  messages: SourceMsg[],
@@ -9,21 +9,21 @@ const glob = (fsp as unknown as {
 }).glob;

 const WX_CACHE_ROOT = join(
-  /*turbopackIgnore: true*/ homedir(),
+  homedir(),
  'Library/Containers/com.tencent.xinWeChat/Data/Documents/xwechat_files',
 );

 let _userDirCache: string | null = null;

-/** 找到当前微信用户目录（取最近修改的） */
+/** 找到当前登录微信用户目录（取最近修改的） */
 function findUserDir(): string | null {
-  if (_userDirCache && existsSync(/*turbopackIgnore: true*/ _userDirCache)) return _userDirCache;
-  if (!existsSync(/*turbopackIgnore: true*/ WX_CACHE_ROOT)) return null;
-  const entries = readdirSync(/*turbopackIgnore: true*/ WX_CACHE_ROOT, { withFileTypes: true })
+  if (_userDirCache && existsSync(_userDirCache)) return _userDirCache;
+  if (!existsSync(WX_CACHE_ROOT)) return null;
+  const entries = readdirSync(WX_CACHE_ROOT, { withFileTypes: true })
    .filter((e) => e.isDirectory() && e.name !== 'all_users' && e.name !== 'Backup')
    .map((e) => {
-      const p = join(/*turbopackIgnore: true*/ WX_CACHE_ROOT, e.name);
-      return { p, mtime: statSync(/*turbopackIgnore: true*/ p).mtimeMs };
+      const p = join(WX_CACHE_ROOT, e.name);
+      return { p, mtime: statSync(p).mtimeMs };
    })
    .sort((a, b) => b.mtime - a.mtime);
  if (entries.length === 0) return null;
@@ -33,9 +33,9 @@ function findUserDir(): string | null {

 /** 列出按月分的子目录，按时间倒序（最近月份优先） */
 function listMonthDirs(userDir: string): string[] {
-  const cacheDir = join(/*turbopackIgnore: true*/ userDir, 'cache');
-  if (!existsSync(/*turbopackIgnore: true*/ cacheDir)) return [];
-  return readdirSync(/*turbopackIgnore: true*/ cacheDir)
+  const cacheDir = join(userDir, 'cache');
+  if (!existsSync(cacheDir)) return [];
+  return readdirSync(cacheDir)
    .filter((d) => /^\d{4}-\d{2}$/.test(d))
    .sort((a, b) => b.localeCompare(a));
 }
@@ -49,7 +49,7 @@ export interface ResolvedImage {
 /** 检测文件 magic bytes */
 function detectFormat(path: string): ResolvedImage['format'] {
  try {
-    const fd = readFileSync(/*turbopackIgnore: true*/ path, { flag: 'r' });
+    const fd = readFileSync(path, { flag: 'r' });
    const h = fd.subarray(0, 4);
    if (h[0] === 0xff && h[1] === 0xd8) return 'jpeg';
    if (h[0] === 0x89 && h[1] === 0x50 && h[2] === 0x4e && h[3] === 0x47) return 'png';
@@ -69,8 +69,8 @@ async function buildMonthIndex(userDir: string, month: string): Promise<void> {
  if (existing) return existing;

  const p = (async () => {
-    const monthRoot = join(/*turbopackIgnore: true*/ userDir, 'cache', month, 'Message');
-    if (!existsSync(/*turbopackIgnore: true*/ monthRoot)) {
+    const monthRoot = join(userDir, 'cache', month, 'Message');
+    if (!existsSync(monthRoot)) {
      monthIndexCache.set(month, new Map());
      return;
    }
@@ -89,13 +89,13 @@ async function buildMonthIndex(userDir: string, month: string): Promise<void> {
    };

    for await (const p of glob('*/ImageTemp/*hd_temp_convert', { cwd: monthRoot })) {
-      consider(join(/*turbopackIgnore: true*/ monthRoot, String(p)), 'hd');
+      consider(join(monthRoot, String(p)), 'hd');
    }
    for await (const p of glob('*/ImageTemp/*mid_temp_convert', { cwd: monthRoot })) {
-      consider(join(/*turbopackIgnore: true*/ monthRoot, String(p)), 'mid');
+      consider(join(monthRoot, String(p)), 'mid');
    }
    for await (const p of glob('*/Thumb/*thumb.jpg', { cwd: monthRoot })) {
-      consider(join(/*turbopackIgnore: true*/ monthRoot, String(p)), 'thumb');
+      consider(join(monthRoot, String(p)), 'thumb');
    }
    monthIndexCache.set(month, idx);
  })();