tt-help-cli-ycl 1.3.87 → 1.3.90

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tt-help-cli-ycl",
-  "version": "1.3.87",
+  "version": "1.3.90",
   "description": "TikTok user & video data scraper - extract ttSeller, verified, locationCreated from HTML source",
   "type": "module",
   "bin": {

package/src/cli/auto.js

@@ -232,6 +232,13 @@ export async function handleAuto(options) {
           displayName: Array.isArray(f) ? f[1] : null,
           guessedLocation,
         })),
+        discoveredRecommended: (result.discoveredRecommended || []).map(
+          (f) => ({
+            handle: Array.isArray(f) ? f[0] : f,
+            displayName: Array.isArray(f) ? f[1] : null,
+            guessedLocation,
+          }),
+        ),
       };
       await apiPost(`${serverUrl}/api/job/${username}`, payload);
       console.error("  已提交");

package/src/cli/explore.js

@@ -143,7 +143,9 @@ export async function handleExplore(options) {
     console.error(`CDP 端口: ${cdpOptions.port}, 用户编号: ${userId}`);
     console.error(`浏览器配置: ${path.basename(cdpOptions.userDataDir)}`);
 
-    const { apiGet, apiPost } = createApiClient({ meta: { port: cdpOptions.port } });
+    const { apiGet, apiPost } = createApiClient({
+      meta: { port: cdpOptions.port },
+    });
 
     await apiGet(`${serverUrl}/api/stats`);
 
@@ -508,7 +510,8 @@ export async function handleExplore(options) {
         if (result.hasFollowData && result.keepFollow) {
           const totalFollows =
             (result.discoveredFollowing || []).length +
-            (result.discoveredFollowers || []).length;
+            (result.discoveredFollowers || []).length +
+            (result.discoveredRecommended || []).length;
           if (totalFollows > 0) {
             lastFollowSuccessTime = Date.now();
           }
@@ -528,6 +531,13 @@ export async function handleExplore(options) {
             displayName: Array.isArray(f) ? f[1] : null,
             guessedLocation,
           })),
+          discoveredRecommended: (result.discoveredRecommended || []).map(
+            (f) => ({
+              handle: Array.isArray(f) ? f[0] : f,
+              displayName: Array.isArray(f) ? f[1] : null,
+              guessedLocation,
+            }),
+          ),
           processed: result.processed,
           hasFollowData: result.hasFollowData,
           keepFollow: result.keepFollow,

package/src/cli/refresh.js

@@ -155,7 +155,9 @@ export async function handleRefresh(options) {
       );
     }
 
-    const { apiGet, apiPost } = createApiClient({ meta: { port: cdpOptions.port } });
+    const { apiGet, apiPost } = createApiClient({
+      meta: { port: cdpOptions.port },
+    });
 
     // 连接服务器验证
     await apiGet(`${serverUrl}/api/stats`);
@@ -545,6 +547,13 @@ export async function handleRefresh(options) {
           displayName: Array.isArray(f) ? f[1] : null,
           guessedLocation,
         })),
+        discoveredRecommended: (result.discoveredRecommended || []).map(
+          (f) => ({
+            handle: Array.isArray(f) ? f[0] : f,
+            displayName: Array.isArray(f) ? f[1] : null,
+            guessedLocation,
+          }),
+        ),
         processed: result.processed,
         hasFollowData: result.hasFollowData,
         keepFollow: result.keepFollow,

package/src/cli/tag.js

@@ -5,7 +5,7 @@ import {
   DEFAULT_TARGET_LOCATIONS,
   isLocationInList,
 } from "../lib/target-locations.js";
-import { discoverTags, recordProductiveTag } from "../lib/tag-discover.js";
+import { discoverTags } from "../lib/tag-discover.js";
 import { server as cfgServer } from "../lib/constants.js";
 
 const ALL_COUNTRIES = DEFAULT_TARGET_LOCATIONS;
@@ -157,9 +157,18 @@ async function processTag(
         const countries = [
           ...new Set(videos.map((v) => v.locationCreated).filter(Boolean)),
         ];
-        for (const c of countries) {
-          recordProductiveTag(tag, c, pushResult.added);
-        }
+        // 通过 API 上报到服务端，由服务端写入数据库
+        try {
+          await fetch(`${serverUrl}/api/tags/productive`, {
+            method: "POST",
+            headers: { "Content-Type": "application/json" },
+            body: JSON.stringify({
+              tag,
+              countries,
+              pushedUsers: pushResult.added,
+            }),
+          });
+        } catch {}
         process.stderr.write(
           `  已记录标签 #${tag} (${countries.join(",")}, ${pushResult.added} 用户)\n`,
         );
@@ -221,7 +230,7 @@ export async function handleDiscover(parsed) {
     process.exit(1);
   }
 
-  const baseUrl = serverUrl || "http://127.0.0.1:3000";
+  const baseUrl = serverUrl || DEFAULT_SERVER;
 
   for (const country of countries) {
     const params = new URLSearchParams({ country, count: String(count) });
@@ -452,8 +461,8 @@ export async function handleScoreAll(parsed) {
   log("");
 
   let totalScored = 0;
-  let lastDiscoverTime = 0;
-  const DISCOVER_COOLDOWN = 5 * 60 * 1000; // 5 分钟冷却
+  let emptyRounds = 0; // 连续无任务的轮数
+  const DISCOVER_AFTER_EMPTY = 3; // 连续 3 轮无任务时触发 discover
 
   // 复用 TikTokScraper 实例，避免每次 enrich 都启动/关闭 headless 浏览器
   const enrichScraper = new TikTokScraper({ poolSize: 3 });
@@ -467,10 +476,12 @@ export async function handleScoreAll(parsed) {
       const tagsRes = await fetch(`${baseUrl}/api/tags?status=new&limit=1`);
       const tagsData = await tagsRes.json();
       if (!tagsData.tags || tagsData.tags.length === 0) {
-        // 自动发现：无任务时自动生成标签
-        if (autoDiscover && Date.now() - lastDiscoverTime > DISCOVER_COOLDOWN) {
+        emptyRounds++;
+
+        // 自动发现：连续 N 轮无任务时自动生成标签
+        if (autoDiscover && emptyRounds >= DISCOVER_AFTER_EMPTY) {
           log(
-            `🔍 无待打分标签，自动为 ${targetCountries.length} 个国家生成标签...`,
+            `🔍 连续 ${emptyRounds} 轮无待打分标签，自动为 ${targetCountries.length} 个国家生成标签...`,
           );
           for (const country of targetCountries) {
             try {
@@ -485,16 +496,19 @@ export async function handleScoreAll(parsed) {
               log(`  ${country}: 请求失败 (${e.message})`);
             }
           }
-          lastDiscoverTime = Date.now();
+          emptyRounds = 0; // 重置计数器
           // 等 3 秒让服务端处理完
           await new Promise((r) => setTimeout(r, 3000));
           continue;
         }
-        log(`⏳ 暂无待打分标签，10 秒后重试...`);
+        log(`⏳ 暂无待打分标签（连续 ${emptyRounds} 轮），10 秒后重试...`);
         await new Promise((r) => setTimeout(r, 10000));
         continue;
       }
 
+      // 有任务了，重置计数器
+      emptyRounds = 0;
+
       const tag = tagsData.tags[0].tag.replace(/^#+/, "").trim().toLowerCase();
       const startTime = Date.now();
 
@@ -684,6 +698,7 @@ export async function handleTag(parsed) {
     const discoverCount = typeof discover === "number" ? discover : 10;
     const generatedTags = await discoverTags(targetLocations, {
       count: discoverCount,
+      serverUrl,
     });
     finalTags = [...new Set([...finalTags, ...generatedTags])];
     process.stderr.write(`  共 ${finalTags.length} 个标签待处理\n\n`);

package/src/lib/browser/page.js

@@ -94,7 +94,13 @@ export async function isLoggedIn(page) {
       console.error(
         `  [登录检测] DOM 无法判断，刷新页面后重试 (${attempt}/${DOM_CHECK_RETRIES})...`,
       );
-      await page.reload({ waitUntil: "domcontentloaded" });
+      try {
+        await page.reload({ waitUntil: "domcontentloaded", timeout: 30000 });
+      } catch (reloadErr) {
+        console.error(
+          `  [登录检测] 页面刷新失败: ${reloadErr.message}，跳过重试继续检测...`,
+        );
+      }
     }
   }
   // 重试后仍无法判断，信任 Cookie
@@ -125,9 +131,16 @@ export async function safeCheckLogin(page) {
   for (let round = 2; round <= SAFE_CHECK_ROUNDS; round++) {
     await new Promise((r) => setTimeout(r, SAFE_CHECK_INTERVAL));
     // 重新导航到 TikTok 页面，确保页面状态刷新
-    await page.goto("https://www.tiktok.com", {
-      waitUntil: "domcontentloaded",
-    });
+    try {
+      await page.goto("https://www.tiktok.com", {
+        waitUntil: "domcontentloaded",
+        timeout: 30000,
+      });
+    } catch (gotoErr) {
+      console.error(
+        `[安全登录检测] 第 ${round} 轮: 页面导航失败: ${gotoErr.message}，直接在当前页面检测...`,
+      );
+    }
     loggedIn = await isLoggedIn(page);
     if (loggedIn) {
       console.error(`[安全登录检测] 第 ${round} 轮: 已登录 ✓`);

package/src/lib/tag-discover.js

@@ -1,150 +1,113 @@
-import { readFileSync, writeFileSync, existsSync } from "fs";
-import { resolve, dirname } from "path";
-import { fileURLToPath } from "url";
-
-const __dirname = dirname(fileURLToPath(import.meta.url));
-const TAGS_FILE = resolve(
-  __dirname,
-  "..",
-  "..",
-  "data",
-  "productive-tags.json",
-);
-
-function loadTags() {
-  try {
-    if (existsSync(TAGS_FILE)) {
-      return JSON.parse(readFileSync(TAGS_FILE, "utf-8"));
-    }
-  } catch {}
-  return { tags: [], lastUpdated: null };
+/**
+ * Tag 发现（CLI 模式）
+ *
+ * 使用 tag-service 的公共函数（LLM 调用、prompt 组装、解析）。
+ * 历史 tag 数据通过 API 从服务端获取，不再读写 productive-tags.json。
+ */
+import {
+  COUNTRY_LANG,
+  getLang,
+  callLLM,
+  normalizeTag,
+  parseTagsFromResponse,
+  buildDiscoverPrompt,
+} from "../watch/tag-service.js";
+
+const DEFAULT_SERVER = "http://127.0.0.1:3000";
+
+/**
+ * 从服务端获取某国的历史 tag（正样本 + 负样本 + 全部已存在）
+ */
+async function fetchTagHistory(serverUrl, country) {
+  const baseUrl = serverUrl || DEFAULT_SERVER;
+
+  const productivePromise = fetch(
+    `${baseUrl}/api/tags/history?country=${country}&type=productive`,
+  )
+    .then((r) => r.json())
+    .then((data) => data.tags || [])
+    .catch(() => []);
+
+  const deadPromise = fetch(
+    `${baseUrl}/api/tags/history?country=${country}&type=dead`,
+  )
+    .then((r) => r.json())
+    .then((data) => data.tags || [])
+    .catch(() => []);
+
+  // 获取所有已存在的 tag（防止重复生成）
+  const allPromise = fetch(
+    `${baseUrl}/api/tags/history?country=${country}&type=all`,
+  )
+    .then((r) => r.json())
+    .then((data) => data.tags || [])
+    .catch(() => []);
+
+  const [productive, dead, allExisting] = await Promise.all([
+    productivePromise,
+    deadPromise,
+    allPromise,
+  ]);
+  return { productive, dead, allExisting: allExisting.map((t) => t.tag) };
 }
 
-function saveTags(data) {
-  const dir = dirname(TAGS_FILE);
-  if (!existsSync(dir)) {
-    const { mkdirSync } = require("fs");
-    mkdirSync(dir, { recursive: true });
+/**
+ * 为单个国家生成 tag（CLI 模式，通过 API 获取历史数据）
+ */
+async function discoverTagsForCountryCli(
+  country,
+  count = 4,
+  userPrompt = null,
+  serverUrl = null,
+) {
+  if (!COUNTRY_LANG[country]) {
+    return { country, error: `不支持的国家代码: ${country}` };
   }
-  writeFileSync(TAGS_FILE, JSON.stringify(data, null, 2), "utf-8");
-}
-
-export function getProductiveTags() {
-  return loadTags().tags;
-}
-
-export function recordProductiveTag(tag, country, userCount) {
-  const data = loadTags();
-  const existing = data.tags.find((t) => t.tag === tag);
-  if (existing) {
-    if (!existing.countries.includes(country)) {
-      existing.countries.push(country);
-    }
-    existing.userCount += userCount;
-    existing.lastUsed = new Date().toISOString();
-  } else {
-    data.tags.push({
-      tag,
-      countries: [country],
-      userCount,
-      firstSeen: new Date().toISOString(),
-      lastUsed: new Date().toISOString(),
-    });
-  }
-  data.lastUpdated = new Date().toISOString();
-  saveTags(data);
-}
 
-async function callLLM(prompt) {
-  const apiKey = process.env.APIKEY || "";
-  const { fetch } = await import("undici");
-
-  const response = await fetch(
-    "http://82.156.52.214:18000/v1/chat/completions",
-    {
-      method: "POST",
-      headers: {
-        "Content-Type": "application/json",
-        Authorization: `Bearer ${apiKey}`,
-      },
-      body: JSON.stringify({
-        model: "zc-fast",
-        messages: [{ role: "user", content: prompt }],
-        max_tokens: 1024,
-        temperature: 0.7,
-      }),
-    },
-  );
-
-  const result = await response.json();
-  const content = result.choices?.[0]?.message?.content || "";
-  return content;
-}
+  // 从服务端获取历史 tag
+  const history = await fetchTagHistory(serverUrl, country);
 
-function normalizeTag(t) {
-  return t.replace(/^#+/, "").trim().toLowerCase();
-}
-
-function parseTagsFromResponse(content) {
-  try {
-    const parsed = JSON.parse(content);
-    if (Array.isArray(parsed)) {
-      return parsed.map(normalizeTag).filter((t) => t && t.length >= 2);
-    }
-    if (Array.isArray(parsed.tags)) {
-      return parsed.tags.map(normalizeTag).filter((t) => t && t.length >= 2);
-    }
-  } catch {}
-
-  const lines = content.split(/[\n,]+/);
-  const tags = [];
-  for (const line of lines) {
-    const cleaned = normalizeTag(line.replace(/^[-\d.\s]+/, ""));
-    if (cleaned && /^[a-z0-9_]+$/.test(cleaned) && cleaned.length >= 2) {
-      tags.push(cleaned);
-    }
-  }
-  return tags;
-}
-
-export async function discoverTags(countries, options = {}) {
-  const { language = "auto", count = 10 } = options;
-
-  const productiveTags = getProductiveTags();
-  const countryStr = Array.isArray(countries)
-    ? countries.join(", ")
-    : countries;
-  const langHint =
-    language === "auto" ? "" : `Tags should be in ${language} language.`;
-
-  const historyHint =
-    productiveTags.length > 0
-      ? `Previously productive tags for these countries: ${productiveTags
-          .filter((t) => t.countries.some((c) => countries.includes(c)))
-          .map((t) => `#${t.tag}`)
-          .join(", ")}. Generate new ones, don't repeat these.`
-      : "";
-
-  const prompt = `Generate ${count} TikTok hashtags (lowercase, no spaces, no # symbol) that are likely to be used by online sellers, shop owners, e-commerce merchants, and small businesses in these countries: ${countryStr}.
-
-Requirements:
-- Focus on tags that sellers/merchants actually use to promote their products
-- Include local language commerce tags (sell, shop, store, online, vendor, etc. in the local language)
-- Mix broad commerce tags with country-specific tags
-${langHint}
-${historyHint}
-
-Return ONLY a JSON array of tag strings, nothing else. Example: ["ventas","tiendaonline","vender"]`;
+  // 使用统一的 prompt 组装
+  const prompt = buildDiscoverPrompt(country, count, history, userPrompt);
 
   process.stderr.write(
-    `  [LLM] 正在生成 ${count} 个标签 (目标: ${countryStr})...\n`,
+    `  [LLM] 正在生成 ${count} 个标签 (国家: ${country}, 语言: ${getLang(country)})...\n`,
   );
   const content = await callLLM(prompt);
   const tags = parseTagsFromResponse(content);
-
   const unique = [...new Set(tags)].slice(0, count);
+
   process.stderr.write(
     `  [LLM] 生成 ${unique.length} 个标签: ${unique.join(", ")}\n`,
   );
   return unique;
 }
+
+/**
+ * 批量为多个国家生成 tag（兼容旧接口）
+ * @param {string|string[]} countries - 国家代码或数组
+ * @param {object} options
+ * @param {number} [options.count=10] - 每个国家生成的 tag 数量
+ * @param {string} [options.serverUrl] - 服务端地址
+ * @param {string} [options.prompt] - 用户自定义提示
+ */
+export async function discoverTags(countries, options = {}) {
+  const { count = 10, serverUrl, prompt: userPrompt } = options;
+
+  const countryList = Array.isArray(countries) ? countries : [countries];
+  const allTags = [];
+
+  for (const country of countryList) {
+    const tags = await discoverTagsForCountryCli(
+      country,
+      count,
+      userPrompt,
+      serverUrl,
+    );
+    allTags.push(...tags);
+  }
+
+  return allTags;
+}
+
+export { discoverTagsForCountryCli };

package/src/scraper/explore-core.js

@@ -35,6 +35,7 @@ async function processExplore(page, username, options, log) {
     discoveredGuessAuthors: [],
     discoveredFollowing: [],
     discoveredFollowers: [],
+    discoveredRecommended: [],
     collectedVideos: 0,
     processed: false,
     hasFollowData: false,
@@ -205,19 +206,18 @@ async function processExplore(page, username, options, log) {
               log(
                 `  商家用户，关注采集: ${effectiveMaxFollowing}, 粉丝采集: ${effectiveMaxFollowers}`,
               );
-            const { following, followers } = await extractFollowAndFollowers(
-              page,
-              {
+            const { following, followers, recommended } =
+              await extractFollowAndFollowers(page, {
                 maxFollowing: effectiveMaxFollowing,
                 maxFollowers: effectiveMaxFollowers,
                 log,
-              },
-            );
+              });
             result.discoveredFollowing = following || [];
             result.discoveredFollowers = followers || [];
+            result.discoveredRecommended = recommended || [];
             result.hasFollowData = true;
             log(
-              `  关注: ${result.discoveredFollowing.length}, 粉丝: ${result.discoveredFollowers.length}`,
+              `  关注: ${result.discoveredFollowing.length}, 粉丝: ${result.discoveredFollowers.length}, 推荐: ${result.discoveredRecommended.length}`,
             );
           } catch (e) {
             log(`  关注/粉丝提取失败: ${e.message}`);

package/src/scraper/modules/follow-extractor.js

@@ -2,7 +2,7 @@ import { delay, getDelayConfig } from "./page-helpers.js";
 import { scrollAndCollect } from "./scroll-collector.js";
 import { extractUniqueId, toProfileUrl } from "../../lib/url.js";
 
-const FILTER_WORDS = ["主页", "已关注", "粉丝", "推荐"];
+const FILTER_WORDS = ["主页", "已关注", "粉丝"];
 
 const FOLLOW_TRIGGER_SELECTORS = [
   "[data-e2e=following]",
@@ -11,6 +11,8 @@ const FOLLOW_TRIGGER_SELECTORS = [
   '[data-e2e*="following"]',
 ];
 
+const RECOMMEND_TAB_TEXTS = ["推荐", "Suggested", "Recommended"];
+
 async function waitForFollowTrigger(page, timeout = 15000) {
   await page
     .waitForFunction(
@@ -187,7 +189,7 @@ async function closeFollowModal(page) {
 
 function createUserCollectFn() {
   return (container) => {
-    const FILTER_WORDS = ["主页", "已关注", "粉丝", "推荐"];
+    const FILTER_WORDS = ["主页", "已关注", "粉丝"];
     const modal = document.querySelector("[class*=eyhy6180]");
     const root = modal || document;
     const users = [];
@@ -239,12 +241,55 @@ async function extractFollowAndFollowers(page, options = {}) {
   const followers = await extractUsersFromModal(page, maxFollowers);
   log(`  粉丝: ${followers.length}`);
 
+  // ===== 3. 采集推荐 =====
+  let recommended = [];
+  if (following.length > 0 || followers.length > 0) {
+    try {
+      await delay(500, 1500);
+      await clickRecommendTab(page);
+      await delay(500, 1500);
+      recommended = await scrollAndCollect(page, {
+        container: "[class*=DivUserListContainer]",
+        findScrollable: false,
+        collectFn: createUserCollectFn(),
+        uniqueKey: (u) => u.handle,
+        maxItems: 50,
+        staleThreshold: 2,
+      });
+      if (log) log(`    推荐: ${recommended.length}`);
+    } catch (e) {
+      if (log) log(`    推荐采集失败: ${e.message}`);
+    }
+  }
+
   await closeFollowModal(page);
 
   return {
     following: following.map((u) => [u.handle, u.displayName]),
     followers: followers.map((u) => [u.handle, u.displayName]),
+    recommended: recommended.map((u) => [u.handle, u.displayName]),
   };
 }
 
+async function clickRecommendTab(page) {
+  await page.evaluate(() => {
+    const tabs = document.querySelectorAll("[class*=DivTabItem]");
+    for (const tab of tabs) {
+      const text = (tab.textContent || "").trim();
+      if (
+        text.includes("推荐") ||
+        text.includes("Suggested") ||
+        text.includes("Recommended")
+      ) {
+        tab.click();
+        return;
+      }
+    }
+    throw new Error("未找到推荐 Tab");
+  });
+  await page.waitForSelector("[class*=DivUserListContainer]", {
+    timeout: 30000,
+  });
+}
+
 export { extractFollowAndFollowers };

package/src/watch/data-store.js

@@ -3786,6 +3786,17 @@ Standards: 90-100=clear match, 70-89=likely, 50-69=possible, 20-49=low, 0-19=unl
             (typeof f === "object" && f.guessedLocation) || guessedLocation,
         };
       }),
+      ...(result.discoveredRecommended || []).map((f) => {
+        const handle = Array.isArray(f) ? f[0] : f.handle || "";
+        const name = Array.isArray(f) ? f[1] : f.displayName || null;
+        return {
+          uniqueId: handle.replace(/^@/, ""),
+          nickname: name,
+          sources: ["recommended"],
+          guessedLocation:
+            (typeof f === "object" && f.guessedLocation) || guessedLocation,
+        };
+      }),
     ].filter((u) => u.uniqueId);
 
     // 先对 discovered 内部去重，再用 uidIndex 批量判断
@@ -3880,6 +3891,7 @@ Standards: 90-100=clear match, 70-89=likely, 50-69=possible, 20-49=low, 0-19=unl
         "discoveredGuessAuthors",
         "discoveredFollowing",
         "discoveredFollowers",
+        "discoveredRecommended",
         "uniqueId",
         "sources",
         "topRecentVideo", // 单独处理，不进入通用循环

package/src/watch/server.js

@@ -1125,6 +1125,51 @@ export function startWatchServer(
         return;
       }
 
+      // GET /api/tags/history?country=ES&type=productive|dead — CLI 模式获取历史 tag
+      if (req.method === "GET" && routePath === "/api/tags/history") {
+        const country = params.country || null;
+        const type = params.type || "productive";
+
+        if (!country) {
+          sendJSON(res, 400, { error: "缺少 country 参数" });
+          return;
+        }
+
+        let tags;
+        if (type === "dead") {
+          tags = store.getDeadTags(country);
+        } else if (type === "all") {
+          tags = store.getTagsByCountry(country, 0);
+        } else {
+          tags = store.getTagsByCountry(country, 50);
+        }
+
+        sendJSON(res, 200, { tags, total: tags.length });
+        return;
+      }
+
+      // POST /api/tags/productive — CLI 模式上报 productive tag
+      if (req.method === "POST" && routePath === "/api/tags/productive") {
+        try {
+          const body = await readBody(req);
+          const { tag, countries, pushedUsers } = body || {};
+
+          if (!tag || !countries || countries.length === 0) {
+            sendJSON(res, 400, { error: "tag 和 countries 不能为空" });
+            return;
+          }
+
+          // 将 productive 信息写入数据库（更新已有 tag 或插入新 tag）
+          for (const c of countries) {
+            store.insertTag(tag, [c], "cli-productive");
+          }
+          sendJSON(res, 200, { ok: true });
+        } catch (e) {
+          sendJSON(res, 500, { error: e.message });
+        }
+        return;
+      }
+
       if (
         req.method === "GET" &&
         (routePath === "/" || routePath === "/index.html")

package/src/watch/tag-service.js

@@ -6,7 +6,7 @@
  */
 
 // 国家 → 语言映射
-const COUNTRY_LANG = {
+export const COUNTRY_LANG = {
   CZ: "cs",
   GR: "el",
   HU: "hu",
@@ -22,16 +22,16 @@ const COUNTRY_LANG = {
   AT: "de",
 };
 
-const LLM_URL = "http://82.156.52.214:18000/v1/chat/completions";
-const LLM_MODEL = "zc-fast";
+export const LLM_URL = "http://82.156.52.214:18000/v1/chat/completions";
+export const LLM_MODEL = "zc-fast";
 
-function getLang(country) {
+export function getLang(country) {
   return COUNTRY_LANG[country] || "en";
 }
 
 // ====== LLM 调用 ======
 
-async function callLLM(prompt) {
+export async function callLLM(prompt) {
   const apiKey = process.env.APIKEY || "";
   const { fetch } = await import("undici");
 
@@ -53,11 +53,11 @@ async function callLLM(prompt) {
   return result.choices?.[0]?.message?.content || "";
 }
 
-function normalizeTag(t) {
+export function normalizeTag(t) {
   return t.replace(/^#+/, "").trim().toLowerCase();
 }
 
-function parseTagsFromResponse(content) {
+export function parseTagsFromResponse(content) {
   try {
     const parsed = JSON.parse(content);
     if (Array.isArray(parsed)) {
@@ -81,7 +81,7 @@ function parseTagsFromResponse(content) {
 
 // ====== Prompt 组装 ======
 
-function buildDiscoverPrompt(country, count, history, userPrompt) {
+export function buildDiscoverPrompt(country, count, history, userPrompt) {
   const lang = getLang(country);
   const langNames = {
     cs: "Czech",
@@ -98,18 +98,18 @@ function buildDiscoverPrompt(country, count, history, userPrompt) {
   };
   const langName = langNames[lang] || lang;
 
-  // 正样本：该国高分 tag
+  // 正样本：该国高分 tag（只给 LLM 看效果，不给模板）
   const productive = history.productive || [];
   const productiveHint =
     productive.length > 0
-      ? `\nHigh-performing tags for ${country}: ${productive.map((t) => t.tag).join(", ")}. Generate new tags in similar patterns.`
+      ? `\nTags that already worked well for ${country}: ${productive.map((t) => t.tag).join(", ")}. These are examples of what works — explore DIFFERENT directions, not variations of these.`
       : "";
 
   // 负样本：该国 dead tag
   const dead = history.dead || [];
   const deadHint =
     dead.length > 0
-      ? `\nAvoid these tags and similar patterns (they found no matching users): ${dead.map((t) => t.tag).join(", ")}.`
+      ? `\nTags that failed for ${country} (found no matching users): ${dead.map((t) => t.tag).join(", ")}. Avoid these and similar patterns.`
       : "";
 
   // 死因分析
@@ -118,20 +118,35 @@ function buildDiscoverPrompt(country, count, history, userPrompt) {
   ];
   const errorHint =
     errorPatterns.length > 0
-      ? `\nReasons previous tags failed: ${errorPatterns.join("; ")}. Avoid generating tags likely to have same issues.`
+      ? `\nWhy previous tags failed: ${errorPatterns.join("; ")}. Avoid tags likely to have same issues.`
+      : "";
+
+  // 已存在的所有 tag（防止重复生成）
+  const allExisting = history.allExisting || [];
+  const existingHint =
+    allExisting.length > 0
+      ? `\nTags already in database (DO NOT generate these again): ${allExisting.slice(-50).join(", ")}.`
       : "";
 
   const userHint = userPrompt
     ? `\nAdditional focus: ${userPrompt}. Generate tags specifically for this niche.`
     : "";
 
-  return `Generate ${count} TikTok hashtags in ${langName} language for e-commerce sellers, shop owners, and small business merchants in ${country}.
+  return `You are discovering TikTok hashtags used by people who sell things in ${country}.
+
+Your goal: Find hashtags that real sellers in ${country} actually use — any kind of tag they might use. Think broadly:
+- Who they are (seller, shop owner, entrepreneur, artisan...)
+- What they sell (shoes, clothes, jewelry, food, pets, furniture...)
+- How they sell (online, handmade, second-hand, local pickup...)
+- Product-specific tags (sneakers, dresses, cakes, necklaces...)
+
+All tags must be in ${langName} language (or widely used in ${country}).
+Generate ${count} tags that are ALL DIFFERENT from each other and from any existing tags.
 
-Requirements:
-- Tags must be in ${langName} language (or widely used in ${country})
-- Focus on tags that sellers/merchants actually use to promote their products
-- Include local language commerce tags (sell, shop, store, online, vendor, etc.)
-- Prefer specific/niche tags over generic ones (e.g., "vendozapatos" not "vender")${productiveHint}${deadHint}${errorHint}${userHint}
+Rules:
+- Each tag should explore a DIFFERENT angle — don't just swap country suffixes
+- Prefer specific and niche tags over generic ones (e.g., "vendozapatos" beats "vender")
+- Do NOT generate tags that already exist${productiveHint}${deadHint}${errorHint}${existingHint}${userHint}
 
 Return ONLY a JSON array of tag strings, nothing else. Example: ["ventas","tiendaonline","vender"]`;
 }
@@ -151,7 +166,10 @@ export async function discoverTagsForCountry(
   // 读取历史打分记录
   const productive = store.getTagsByCountry(country, 50);
   const dead = store.getDeadTags(country);
-  const history = { productive, dead };
+  // 获取该国所有已存在的 tag 名（防止重复生成）
+  const allTags = store.getTagsByCountry(country, 0);
+  const allExisting = allTags.map((t) => t.tag);
+  const history = { productive, dead, allExisting };
 
   // 组装 prompt 并调用 LLM
   const prompt = buildDiscoverPrompt(country, count, history, userPrompt);