tt-help-cli-ycl 1.4.4 → 1.4.6

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tt-help-cli-ycl",
-  "version": "1.4.4",
+  "version": "1.4.6",
   "description": "TikTok user & video data scraper - extract ttSeller, verified, locationCreated from HTML source",
   "type": "module",
   "bin": {

package/src/cli/tag.js

@@ -66,12 +66,28 @@ function getBrowserMemory(port = null) {
         cmd =
           'ps aux | grep "[m]sedge" | awk \'{sum += $6} END {printf "%d", sum/1024}\'';
       }
+    } else if (platform === "win32") {
+      // Windows: 用 PowerShell 查 msedge 进程内存（WorkingSet64 → MB）
+      // 按端口筛选时匹配命令行参数 --remote-debugging-port=PORT
+      if (port) {
+        cmd = `powershell -NoProfile -Command "Get-CimInstance Win32_Process -Filter 'Name LIKE \\\"msedge%\\\"' | Where-Object { $_.CommandLine -match 'remote-debugging-port=${port}' } | ForEach-Object { $_.WorkingSet64 } | Measure-Object -Sum | Select-Object -ExpandProperty Sum"`;
+      } else {
+        cmd = `powershell -NoProfile -Command "Get-CimInstance Win32_Process -Filter 'Name LIKE \\\"msedge%\\\"' | ForEach-Object { $_.WorkingSet64 } | Measure-Object -Sum | Select-Object -ExpandProperty Sum"`;
+      }
     } else {
       return null;
     }
-    const result = execSync(cmd, { timeout: 3000 }).toString().trim();
-    const mb = parseInt(result, 10);
-    return mb > 0 ? mb : null;
+    const result = execSync(cmd, { timeout: 5000 }).toString().trim();
+    if (platform === "win32") {
+      // PowerShell 返回 WorkingSet64（字节），需除以 1048576 得 MB
+      const bytes = parseInt(result, 10);
+      const mb = bytes > 0 ? Math.round(bytes / 1048576) : 0;
+      return mb > 0 ? mb : null;
+    } else {
+      // macOS/Linux: ps RSS 是 KB，awk 已除以 1024 得 MB
+      const mb = parseInt(result, 10);
+      return mb > 0 ? mb : null;
+    }
   } catch {
     return null;
   }

package/src/lib/browser/cdp.js

@@ -97,6 +97,23 @@ function killEdgeProcesses(targetDir, port) {
         } else {
           command = "true";
         }
+      } else if (port) {
+        // 仅按端口筛选，不误杀其他端口的 Edge 实例
+        let pids = "";
+        try {
+          pids = execSync(
+            `ps aux | grep "[M]icrosoft Edge" | grep -v "Helper\\|crashpad" | grep "remote-debugging-port=${port}" | awk '{print $2}'`,
+          )
+            .toString()
+            .trim();
+        } catch (e) {
+          pids = "";
+        }
+        if (pids) {
+          command = `kill -9 ${pids} 2>/dev/null; rm -f ~/Library/Caches/Microsoft\\ Edge/Singleton*; true`;
+        } else {
+          command = "true";
+        }
       } else {
         command =
           'killall -9 "Microsoft Edge" 2>/dev/null; rm -f ~/Library/Caches/Microsoft\\ Edge/Singleton*; true';
@@ -143,6 +160,8 @@ function killEdgeProcesses(targetDir, port) {
       if (targetDir) {
         const escapedDir = targetDir.replace(/[-\/\\^$*+?.()|[\]{}]/g, "\\$&");
         command = `ps aux | grep msedge | grep -v Helper | grep -E -- '--user-data-dir=${escapedDir}($|[^A-Za-z0-9_])' | awk '{print $2}' | xargs -r kill -9 2>/dev/null; true`;
+      } else if (port) {
+        command = `ps aux | grep msedge | grep -v Helper | grep "remote-debugging-port=${port}" | awk '{print $2}' | xargs -r kill -9 2>/dev/null; true`;
       } else {
         command = "pkill -9 -f msedge 2>/dev/null; true";
       }

package/src/watch/db-tags.js

@@ -360,6 +360,13 @@ export function getRecentDiscoverLogs(country, limit = 10) {
 /**
  * 获取最近 N 轮生成的所有标签（去重），用于检测重复生成
  */
+export function getAllExistingTagNames() {
+  const db = getDb();
+  if (!db) return new Set();
+  const rows = db.prepare("SELECT tag FROM tags").all();
+  return new Set(rows.map((r) => r.tag));
+}
+
 export function getRecentGeneratedTags(country, lookback = 10) {
   const db = getDb();
   if (!db) return [];
@@ -493,6 +500,7 @@ export function getCountryHealth(country) {
   let recentDeadCount = 0;
   let recentTotalGenerated = 0;
   let recentEmptyRounds = 0;
+  let consecutiveLowProductiveRounds = 0; // 连续 productive=0 的轮数
 
   for (const log of recentLogs) {
     const tagsGen = JSON.parse(log.tags_generated || "[]");
@@ -504,10 +512,20 @@ export function getCountryHealth(country) {
       const tagResults = db
         .prepare(`SELECT status FROM tags WHERE tag IN (${placeholders})`)
         .all(...tagsGen);
-      recentProductiveCount += tagResults.filter(
+      const productiveInRound = tagResults.filter(
         (t) => t.status === "productive",
       ).length;
+      recentProductiveCount += productiveInRound;
       recentDeadCount += tagResults.filter((t) => t.status === "dead").length;
+      // 连续低效轮计数：productive=0 则累加，否则重置
+      if (productiveInRound === 0) {
+        consecutiveLowProductiveRounds++;
+      } else {
+        consecutiveLowProductiveRounds = 0;
+      }
+    } else {
+      // 空轮也算低效轮
+      consecutiveLowProductiveRounds++;
     }
   }
 
@@ -560,6 +578,14 @@ export function getCountryHealth(country) {
     );
   }
 
+  // 连续低效轮警告：连续多轮 productive=0，需要强制换方向
+  if (consecutiveLowProductiveRounds >= 2) {
+    healthStatus = healthStatus === "critical" ? "critical" : "warning";
+    issues.push(
+      `consecutive_low_productive_rounds=${consecutiveLowProductiveRounds}: 连续${consecutiveLowProductiveRounds}轮无有效产出，必须切换商品分类方向`,
+    );
+  }
+
   if (lowScored.length > productive.length * 2) {
     issues.push(
       `low_scored=${lowScored.length} vs productive=${productive.length}: 低分tag远多于有效tag，应清理`,
@@ -582,6 +608,7 @@ export function getCountryHealth(country) {
     recentProductiveRatio: recentProductiveRatio.toFixed(1),
     recentDeadRatio: recentDeadRatio.toFixed(1),
     recentEmptyRounds,
+    consecutiveLowProductiveRounds,
     saturation: saturation.toFixed(0),
     healthStatus,
     issues,
@@ -1025,11 +1052,13 @@ export function shouldTriggerReflection(country, lookback = 5, threshold = 1) {
   let totalProductive = 0;
   let totalDead = 0;
   let emptyRounds = 0;
+  let consecutiveLowProductiveRounds = 0; // 连续 productive ≤ 1 的轮数
 
   for (const log of recentLogs) {
     const tagsGen = JSON.parse(log.tags_generated || "[]");
     if (tagsGen.length === 0) {
       emptyRounds++;
+      consecutiveLowProductiveRounds++;
       continue;
     }
 
@@ -1044,21 +1073,33 @@ export function shouldTriggerReflection(country, lookback = 5, threshold = 1) {
     const dead = tagResults.filter((t) => t.status === "dead").length;
     totalProductive += productive;
     totalDead += dead;
+
+    // 连续低效轮计数：productive ≤ 1 则累加，否则重置
+    if (productive <= 1) {
+      consecutiveLowProductiveRounds++;
+    } else {
+      consecutiveLowProductiveRounds = 0;
+    }
   }
 
   const avgProductive = totalProductive / lookback;
-  const shouldReflect = avgProductive <= threshold;
+  // 触发条件：平均 productive ≤ threshold，或连续 ≥2 轮 productive ≤ 1
+  const shouldReflect =
+    avgProductive <= threshold || consecutiveLowProductiveRounds >= 2;
 
   return {
     shouldReflect,
     reason: shouldReflect
-      ? `最近${lookback}轮平均productive=${avgProductive.toFixed(1)} ≤ ${threshold}`
+      ? consecutiveLowProductiveRounds >= 2
+        ? `连续${consecutiveLowProductiveRounds}轮productive≤1，需要反思换方向`
+        : `最近${lookback}轮平均productive=${avgProductive.toFixed(1)} ≤ ${threshold}`
       : `最近${lookback}轮平均productive=${avgProductive.toFixed(1)} > ${threshold}`,
     recentRounds: recentLogs,
     totalProductive,
     totalDead,
     emptyRounds,
     avgProductive,
+    consecutiveLowProductiveRounds,
   };
 }
 

package/src/watch/tag-service.js

@@ -148,6 +148,8 @@ export function buildDiscoverPrompt(
   health,
   nicheRecommendation,
   hardRules = [],
+  existingCount = 0,
+  forceNicheSwitch = null,
 ) {
   const lang = getLang(country);
   const langNames = {
@@ -288,6 +290,18 @@ ${nicheRecommendation.uncovered.length > 0 ? `Focus your tags on the UNCOVERED n
 `;
   }
 
+  // ====== 自闭环：强制方向切换提示 ======
+  let forceNicheSwitchHint = "";
+  if (forceNicheSwitch) {
+    forceNicheSwitchHint = `
+=== 🔄 MANDATORY DIRECTION SWITCH ===
+⚠️ CRITICAL: The previous direction "${forceNicheSwitch.from}" has been exhausted — ${forceNicheSwitch.reason}.
+You MUST NOT generate tags in the "${forceNicheSwitch.from}" direction anymore.
+You MUST switch to the "${forceNicheSwitch.to}" direction: ${forceNicheSwitch.description}.
+Generate ONLY tags related to this new niche. Do NOT fall back to old patterns.
+`;
+  }
+
   return `You are discovering TikTok hashtags used by people who sell things in ${country}.
 
 ⛔ CRITICAL RULE: ONLY suggest hashtags that you are CONFIDENT actually exist on TikTok with real posts.
@@ -321,7 +335,7 @@ ${hardRules.map((r) => `  ${r.ruleType === "hard" ? "⛔" : "💡"} ${r.ruleText
     : ""
 }
 
-${productiveHint}${deadHint}${repeatedDeadHint}${langPatternHint}${errorHint}${userHint}${strategyReview}${healthHint}${nicheHint}
+${productiveHint}${deadHint}${repeatedDeadHint}${langPatternHint}${errorHint}${userHint}${strategyReview}${healthHint}${nicheHint}${forceNicheSwitchHint}
 
 ## Quality check before responding
 
@@ -337,7 +351,7 @@ Return ONLY a JSON object with three fields:
 }
 
 Rules:
-- Brainstorm broadly — generate 12-15 distinct, diverse tag ideas covering different categories (selling verbs, product types, community tags, marketplace terms, seasonal trends)
+- Brainstorm broadly — generate at least ${Math.max(count * 3, 20)} distinct, diverse tag ideas covering different categories (selling verbs, product types, community tags, marketplace terms, seasonal trends). We already have ${existingCount} existing tags in our database, so many of your suggestions will be filtered out as duplicates — generate MORE than you think we need to ensure enough survive filtering.
 - 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
@@ -433,8 +447,57 @@ export async function discoverTagsForCountry(
     }
   }
 
-  // 合并用户 prompt 和自动 niche prompt
-  const effectiveUserPrompt = autoNichePrompt || userPrompt;
+  // ====== 自闭环 Step 2b: 连续低效轮强制方向切换 ======
+  // 商品分类方向定义（用于强制切换）
+  const NICHE_CATEGORIES = {
+    fashion:
+      "clothing, shoes, accessories, vintage, secondhand, thrift, sneakers, jewelry",
+    food: "bakery, food delivery, restaurant, cooking, local food, cafe, patisserie, boulangerie",
+    handmade_crafts:
+      "handmade, DIY, crafts, ceramics, candles, resin art, macrame, pottery, weaving",
+    home_living:
+      "home decoration, furniture, interior design, cleaning, moving, renovation",
+    electronics_tech:
+      "electronics, gadgets, phones, computers, audio, gaming, chargers, accessories",
+    pets_animals:
+      "pet supplies, dog accessories, cat food, animal care, grooming, veterinary",
+    beauty_health:
+      "beauty, cosmetics, skincare, hair care, wellness, perfume, makeup, spa",
+    books_collectibles:
+      "books, collectibles, vinyl records, trading cards, antiques, coins, stamps",
+    services_professional:
+      "repair, cleaning, photography, florist, tutoring, legal, consulting, moving",
+    marketplace_local:
+      "local market, marketplace, shop, boutique, online store, flea market, bazaar",
+  };
+
+  let forceNicheSwitch = null;
+  if (health && health.consecutiveLowProductiveRounds >= 2) {
+    // 连续 ≥2 轮 productive=0，强制切换方向
+    // 找出当前 productive tag 覆盖最多的 niche（即当前方向）
+    const currentNiche =
+      nicheRecommendation?.saturated?.[0]?.niche || "marketplace_local";
+    // 找出覆盖最少的 niche 作为新方向
+    const targetNiche = nicheRecommendation?.uncovered?.[0]?.niche || "food";
+    forceNicheSwitch = {
+      from: currentNiche,
+      to: targetNiche,
+      reason: `连续 ${health.consecutiveLowProductiveRounds} 轮无有效产出`,
+      description: NICHE_CATEGORIES[targetNiche] || targetNiche,
+    };
+    console.error(
+      `[tag-service] 🔄 ${country} 强制方向切换: ${currentNiche} → ${targetNiche} (原因: 连续${health.consecutiveLowProductiveRounds}轮低效)`,
+    );
+  }
+
+  // 合并用户 prompt、自动 niche prompt 和强制方向切换
+  let effectiveUserPrompt = userPrompt;
+  if (forceNicheSwitch) {
+    // 强制方向切换优先级最高，覆盖 autoNichePrompt
+    effectiveUserPrompt = `MANDATORY DIRECTION CHANGE: You MUST switch from "${forceNicheSwitch.from}" to "${forceNicheSwitch.to}" niche. ${forceNicheSwitch.reason}. Focus on: ${forceNicheSwitch.description}. Do NOT generate tags in the "${forceNicheSwitch.from}" direction anymore — it has been exhausted with no productive results.`;
+  } else if (autoNichePrompt) {
+    effectiveUserPrompt = autoNichePrompt;
+  }
 
   // ====== 自闭环 Step 3: 动态温度调整 ======
   let temperature = 0.7; // 默认
@@ -444,15 +507,22 @@ export async function discoverTagsForCountry(
     } else if (health.healthStatus === "warning") {
       temperature = 0.5; // 较保守
     }
+    // 连续低效轮时进一步降温，让 LLM 更保守
+    if (health.consecutiveLowProductiveRounds >= 2) {
+      temperature = Math.min(temperature, 0.3);
+    }
   }
 
   // ====== 原有 discover 流程 ======
   // 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（应用层去重用）
-  const allTags = store.getTagsByCountry(country, 0);
-  const allExisting = new Set(allTags.map((t) => t.tag));
+  // 全局去重：获取所有已存在的 tag 名（不限国家、不限状态），避免跨国家重复入库
+  const { getAllExistingTagNames } = await import("./db-tags.js");
+  const allExisting = getAllExistingTagNames();
+  console.error(
+    `[tag-service] ${country} 全局去重集合: ${allExisting.size} 个已存在 tag`,
+  );
 
   // 获取最近生成的标签（用于检测重复生成）
   const { getRecentGeneratedTags } = await import("./db-tags.js");
@@ -482,6 +552,8 @@ export async function discoverTagsForCountry(
     health,
     nicheRecommendation,
     hardRules,
+    allExisting.size,
+    forceNicheSwitch,
   );
   console.error(
     `[tag-service] LLM discover ${country} round=${round} (lang=${getLang(country)}, count=${count}, temp=${temperature}, health=${health?.healthStatus || "unknown"})`,
@@ -492,9 +564,9 @@ export async function discoverTagsForCountry(
   const parsed = parseDiscoverResponse(content);
   const { strategy, tags } = parsed;
   const niche = parsed.niche || null;
-  const unique = [...new Set(tags)]
-    .filter((t) => !allExisting.has(t))
-    .slice(0, count);
+  const deduped = [...new Set(tags)]; // LLM 自身去重
+  const filtered = deduped.filter((t) => !allExisting.has(t)); // 全局去重
+  const unique = filtered.slice(0, count);
 
   console.error(
     `[tag-service] LLM strategy: ${strategy?.substring(0, 150) || "(none)"}`,
@@ -503,8 +575,14 @@ export async function discoverTagsForCountry(
     console.error(`[tag-service] LLM niche focus: ${niche}`);
   }
   console.error(
-    `[tag-service] LLM generated ${unique.length} tags: ${unique.join(", ")}`,
+    `[tag-service] ${country} 生成链路: LLM原始=${tags.length} → 自去重=${deduped.length} → 全局去重=${filtered.length} → 截取=${unique.length}`,
   );
+  if (filtered.length < count) {
+    const dupCount = deduped.length - filtered.length;
+    console.error(
+      `[tag-service] ⚠️ ${country} 去重后仅剩 ${filtered.length} 个（${dupCount} 个与已有 ${allExisting.size} 个 tag 重复），不足目标 ${count} 个`,
+    );
+  }
 
   // 去重写入数据库
   const inserted = [];