tt-help-cli-ycl 1.4.0 → 1.4.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tt-help-cli-ycl",
-  "version": "1.4.0",
+  "version": "1.4.2",
   "description": "TikTok user & video data scraper - extract ttSeller, verified, locationCreated from HTML source",
   "type": "module",
   "bin": {

package/src/cli/tag.js

@@ -629,10 +629,8 @@ export async function handleScoreAll(parsed) {
 
   await setupPageRequestBlocking(page);
 
-  const portPoolStart = Math.max(
-    1,
-    parseInt(process.env.TAG_SCOREALL_PORT_POOL_START || "7222", 10) || 7222,
-  );
+  // 端口池起始以 CLI --port / 环境变量 / 默认 7222 为准
+  const portPoolStart = cdpPort;
   const portPoolSize = Math.max(
     2,
     parseInt(process.env.TAG_SCOREALL_PORT_POOL_SIZE || "10", 10) || 10,
@@ -656,7 +654,7 @@ export async function handleScoreAll(parsed) {
   );
   const memHeapRecycleMb = Math.max(
     128,
-    parseInt(process.env.TAG_SCOREALL_RECYCLE_HEAP_MB || "320", 10) || 320,
+    parseInt(process.env.TAG_SCOREALL_RECYCLE_HEAP_MB || "512", 10) || 512,
   );
   const recycleCooldownMs = Math.max(
     0,
@@ -689,9 +687,9 @@ export async function handleScoreAll(parsed) {
     }
   }
 
-  async function recycleCdpSession(reason) {
+  async function recycleCdpSession(reason, { switchPort = true } = {}) {
     const oldPort = cdpPort;
-    if (switchPortOnRecycle) {
+    if (switchPort && switchPortOnRecycle) {
       cdpPort = pickNextPort(cdpPort);
       cdpOpts.port = cdpPort;
       clientMeta.port = cdpPort;
@@ -724,6 +722,7 @@ export async function handleScoreAll(parsed) {
       }
       await recycleCdpSession(
         `mem rss=${rssMB.toFixed(0)}MB heap=${heapMB.toFixed(0)}MB (threshold rss=${memRssRecycleMb} heap=${memHeapRecycleMb})`,
+        { switchPort: false },
       );
       return;
     }
@@ -732,7 +731,7 @@ export async function handleScoreAll(parsed) {
       totalScored > 0 &&
       totalScored % periodicRecycleEvery === 0
     ) {
-      await recycleCdpSession(`periodic every ${totalScored} tasks`);
+      await recycleCdpSession(`periodic every ${totalScored} tasks`, { switchPort: false });
     }
   }
 
@@ -987,7 +986,9 @@ export async function handleScoreAll(parsed) {
 
         if (is403) {
           consecutive403++;
-          log(`  🚫 HTTP 403 — 连续 ${consecutive403}/${CONSECUTIVE_403_BLOCKED} 次`);
+          log(
+            `  🚫 HTTP 403 — 连续 ${consecutive403}/${CONSECUTIVE_403_BLOCKED} 次`,
+          );
         } else {
           consecutive403 = 0;
         }
@@ -995,8 +996,7 @@ export async function handleScoreAll(parsed) {
         log(`  ❌ 失败: ${e.message}`);
         try {
           // 使用当前正在处理的 tag（tag 变量在 try 块开头已赋值，catch 中仍可访问）
-          const failedTag =
-            typeof tag === "string" && tag ? tag : "unknown";
+          const failedTag = typeof tag === "string" && tag ? tag : "unknown";
           await reportToServer(
             baseUrl,
             {

package/src/lib/constants.js

@@ -260,8 +260,17 @@ const HELP_TEXT = [
   "    选项:",
   "      --countries <CSV>         目标国家，逗号分隔（默认 13 个欧洲国家）",
   "      -s, --server <URL>        服务端地址（默认 http://127.0.0.1:3000）",
+  "      --port <N>                起始 CDP 端口，端口池默认为 N~N+9（默认 7222）",
+  "      --discover                开启自动发现：连续 3 轮无新标签时自动生成",
+  "    行为:",
+  "      连续 3 次 HTTP 403 → 判断账号被封 → 等 2 分钟后自动切换到池中另一端口",
+  "    环境变量:",
+  "      TAG_SCOREALL_PORT_POOL_SIZE=N  端口池大小（默认 10）",
+  "      TAG_SCOREALL_RECYCLE_RSS_MB=N  内存回收阈值 RSS MB（默认 900）",
+  "      TAG_SCOREALL_RECYCLE_HEAP_MB=N 内存回收阈值 Heap MB（默认 512）",
   "    示例: tt-help tag score-all",
   "          tt-help tag score-all --countries ES -s http://127.0.0.1:3001",
+  "          tt-help tag score-all --port 9222 -s http://127.0.0.1:3001 --discover",
   "",
   "  config  [show|set|unset|reset]",
   "    config                    查看当前配置",

package/src/lib/tag-discover.js

@@ -66,16 +66,26 @@ async function discoverTagsForCountryCli(
 
   // 从服务端获取历史 tag
   const history = await fetchTagHistory(serverUrl, country);
-
-  // 使用统一的 prompt 组装
-  const prompt = buildDiscoverPrompt(country, count, history, userPrompt);
+  const { productive, dead, allExisting: existingSet } = history;
+  const existingArr = Array.isArray(existingSet) ? existingSet : [];
+
+  // 使用统一的 prompt 组装（不传 allExisting，由应用层去重）
+  const prompt = buildDiscoverPrompt(
+    country,
+    count,
+    { productive, dead },
+    userPrompt,
+  );
 
   process.stderr.write(
-    `  [LLM] 正在生成 ${count} 个标签 (国家: ${country}, 语言: ${getLang(country)})...\n`,
+    `  [LLM] 正在生成标签 (国家: ${country}, 语言: ${getLang(country)}, 目标: ${count} 个)...\n`,
   );
   const content = await callLLM(prompt);
   const tags = parseTagsFromResponse(content);
-  const unique = [...new Set(tags)].slice(0, count);
+  // 两阶段筛选：先内部去重，再过滤已存在，最后取 count 个
+  const unique = [...new Set(tags)]
+    .filter((t) => !existingArr.includes(t))
+    .slice(0, count);
 
   process.stderr.write(
     `  [LLM] 生成 ${unique.length} 个标签: ${unique.join(", ")}\n`,

package/src/watch/db-tags.js

@@ -22,11 +22,19 @@ export function insertTag(tag, countries, source = "llm") {
     return { inserted: false, error: "invalid tag" };
   }
   try {
+    // ON CONFLICT: 合并 countries 数组，避免跨国家共享 tag 时信息丢失
     const result = db
       .prepare(
-        "INSERT OR IGNORE INTO tags (tag, countries, source) VALUES (?, ?, ?)",
+        `INSERT INTO tags (tag, countries, source) VALUES (@tag, @countries, @source)
+         ON CONFLICT(tag) DO UPDATE SET countries = (
+           SELECT json_group_array(DISTINCT value) FROM (
+             SELECT value FROM json_each(countries)
+             UNION
+             SELECT value FROM json_each(@countries)
+           )
+         )`,
       )
-      .run(normalized, JSON.stringify(countries), source);
+      .run({ tag: normalized, countries: JSON.stringify(countries), source });
     return { inserted: result.changes > 0, tag: normalized };
   } catch (e) {
     return { inserted: false, error: e.message };

package/src/watch/tag-service.js

@@ -45,7 +45,7 @@ export async function callLLM(prompt) {
       model: LLM_MODEL,
       messages: [{ role: "user", content: prompt }],
       max_tokens: 1024,
-      temperature: 0.3,
+      temperature: 0.7,
     }),
   });
 
@@ -182,13 +182,6 @@ export function buildDiscoverPrompt(
       ? `\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(0, 50).join(", ")}.`
-      : "";
-
   const userHint = userPrompt
     ? `\nAdditional focus: ${userPrompt}. Generate tags specifically for this niche.`
     : "";
@@ -244,7 +237,7 @@ Think about the MOST COMMON hashtags that real sellers in ${country} use on TikT
 - Common seller identities (shop, boutique, small business, creator...)
 - Well-known community tags (support local, marketplace, second round...)
 
-${qualityWarning}${productiveHint}${deadHint}${errorHint}${existingHint}${userHint}${strategyReview}
+${qualityWarning}${productiveHint}${deadHint}${errorHint}${userHint}${strategyReview}
 
 ## Quality check before responding
 
@@ -255,15 +248,16 @@ For each tag you consider, ask yourself: "Have I actually seen this hashtag used
 Return ONLY a JSON object with two fields:
 {
   "strategy": "Your analysis (2-3 sentences). Acknowledge what worked/failed before and explain why your chosen tags are likely real and commonly used.",
-  "tags": ["tag1", "tag2", "tag3", "tag4"]
+  "tags": ["tag1", "tag2", "tag3", "tag4", "..."]
 }
 
 Rules:
+- Brainstorm broadly — generate 12-15 distinct, diverse tag ideas covering different categories (selling verbs, product types, community tags, marketplace terms, seasonal trends)
+- We will filter duplicates and existing tags on our end, so explore freely
 - QUALITY over creativity — 3 real, commonly used tags beat 4 invented ones
 - NEVER invent compound words or translate concepts into hashtags
 - Prefer simpler, more common tags over hyper-specific ones
-- Look for tags with broad appeal (many potential posters)
-- Do NOT generate tags that already exist`;
+- Look for tags with broad appeal (many potential posters)`;
 }
 
 // ====== discover: 单国家标签发现 ======
@@ -281,10 +275,10 @@ export async function discoverTagsForCountry(
   // 读取历史打分记录
   const productive = store.getTagsByCountry(country, 50);
   const dead = store.getDeadTags(country);
-  // 获取该国所有已存在的 tag 名（防止重复生成）
+  // 获取该国所有已存在的 tag 名（应用层去重用）
   const allTags = store.getTagsByCountry(country, 0);
-  const allExisting = allTags.map((t) => t.tag);
-  const history = { productive, dead, allExisting };
+  const allExisting = new Set(allTags.map((t) => t.tag));
+  const history = { productive, dead };
 
   // 读取历史策略效果（自学习）
   const { getDiscoverRound, getDiscoverEffectiveness } =
@@ -305,9 +299,11 @@ export async function discoverTagsForCountry(
   );
   const content = await callLLM(prompt);
 
-  // 解析响应：提取 strategy 和 tags
+  // 解析响应：提取 strategy 和 tags，筛选掉数据库中已存在的
   const { strategy, tags } = parseDiscoverResponse(content);
-  const unique = [...new Set(tags)].slice(0, count);
+  const unique = [...new Set(tags)]
+    .filter((t) => !allExisting.has(t))
+    .slice(0, count);
 
   console.error(
     `[tag-service] LLM strategy: ${strategy?.substring(0, 150) || "(none)"}`,