tt-help-cli-ycl 1.4.9 → 1.4.11

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "tt-help-cli-ycl",
-  "version": "1.4.9",
+  "version": "1.4.11",
   "description": "TikTok user & video data scraper - extract ttSeller, verified, locationCreated from HTML source",
   "type": "module",
   "bin": {

package/src/cli/comments.js

@@ -5,8 +5,9 @@
 //   手动模式: tt-help comments <URL> [N] [--save]            单视频评论提取
 // ────────────────────────────────────────
 
-import { chromium } from "playwright";
+import { writeFileSync, unlinkSync } from "fs";
 import os from "os";
+import path from "path";
 import { execSync } from "child_process";
 
 import { fetchUserCommentsAPI } from "../lib/api-interceptor-comment.js";
@@ -17,6 +18,8 @@ import {
 import {
   server as defaultServer,
   proxy as configuredProxy,
+  userId as configuredUserId,
+  saveUserId,
 } from "../lib/constants.js";
 import {
   DEFAULT_TARGET_LOCATIONS,
@@ -31,6 +34,7 @@ import {
   closeCaptcha,
 } from "../scraper/modules/captcha-handler.js";
 import { detectModals, closeAllModals } from "../lib/modal-handler.js";
+import { getMacOrUuid } from "../lib/mac-or-uuid.js";
 
 // ────────────────────────────────────────
 // Section 1: 常量 & 配置默认值
@@ -52,9 +56,9 @@ const CFG = {
   /** 连续验证码后切换端口 */
   maxConsecutiveCaptcha: 3,
   /** 内存回收 — 触发回收的 RSS(MB) 阈值 */
-  recycleRssMb: 900,
+  recycleRssMb: 2000,
   /** 内存回收 — 触发回收的 Heap(MB) 阈值 */
-  recycleHeapMb: 512,
+  recycleHeapMb: 2000,
   /** 内存回收 — 触发回收的 Edge 进程(MB) 阈值 */
   recycleEdgeMb: 2000,
   /** 内存回收 — 两次回收最小间隔(ms) */
@@ -63,6 +67,8 @@ const CFG = {
   periodicRecycleEvery: 0,
   /** 内存回收 — 回收时是否切换端口 */
   switchPortOnRecycle: true,
+  /** 连续任务处理错误阈值（达到后上报服务端） */
+  consecutiveTaskErrorsMax: 20,
 };
 
 /** 从环境变量覆写配置 */
@@ -95,6 +101,10 @@ function loadEnvConfig() {
   );
   CFG.switchPortOnRecycle =
     String(process.env.COMMENTS_SWITCH_PORT_ON_RECYCLE || "1") !== "0";
+  CFG.consecutiveTaskErrorsMax = Math.max(
+    1,
+    int("COMMENTS_CONSECUTIVE_TASK_ERRORS_MAX", CFG.consecutiveTaskErrorsMax),
+  );
 }
 
 // ────────────────────────────────────────
@@ -106,32 +116,82 @@ function formatMemoryUsage(mem = process.memoryUsage()) {
   return `rss:${mb(mem.rss)}MB heap:${mb(mem.heapUsed)}MB ext:${mb(mem.external)}MB ab:${mb(mem.arrayBuffers)}MB`;
 }
 
-function getBrowserMemory(port) {
+/**
+ * 获取 Edge 浏览器进程的内存占用（跨平台：macOS/Linux/Windows）
+ */
+function getBrowserMemory(port = null) {
   try {
     const platform = os.platform();
-    if (platform !== "darwin" && platform !== "linux") return null;
-    const procName = platform === "darwin" ? "Microsoft Edge" : "msedge";
-    const c = procName[0];
-    const rest = procName.slice(1);
     let cmd;
-    if (port) {
-      cmd = `ps aux | grep "[${c}]${rest}" | grep "remote-debugging-port=${port}" | awk '{sum += $6} END {printf "%d", sum/1024}'`;
-    } else {
-      cmd = `ps aux | grep "[${c}]${rest}" | awk '{sum += $6} END {printf "%d", sum/1024}'`;
+    if (platform === "darwin") {
+      if (port) {
+        cmd = `ps aux | grep "[M]icrosoft Edge" | grep "remote-debugging-port=${port}" | awk '{sum += $6} END {printf "%d", sum/1024}'`;
+      } else {
+        cmd =
+          'ps aux | grep "[M]icrosoft Edge" | awk \'{sum += $6} END {printf "%d", sum/1024}\'';
+      }
+      const result = execSync(cmd, { timeout: 5000 }).toString().trim();
+      const mb = parseInt(result, 10);
+      return mb > 0 ? mb : null;
+    } else if (platform === "linux") {
+      if (port) {
+        cmd = `ps aux | grep "[m]sedge" | grep "remote-debugging-port=${port}" | awk '{sum += $6} END {printf "%d", sum/1024}'`;
+      } else {
+        cmd =
+          'ps aux | grep "[m]sedge" | awk \'{sum += $6} END {printf "%d", sum/1024}\'';
+      }
+      const result = execSync(cmd, { timeout: 5000 }).toString().trim();
+      const mb = parseInt(result, 10);
+      return mb > 0 ? mb : null;
+    } else if (platform === "win32") {
+      const ps1Path = path.join(os.tmpdir(), `edge-mem-${Date.now()}.ps1`);
+      try {
+        let ps1Content;
+        if (port) {
+          ps1Content =
+            `$procs = Get-CimInstance Win32_Process -Filter "Name LIKE 'msedge%'" -ErrorAction SilentlyContinue; ` +
+            `$sum = 0; ` +
+            `foreach ($p in $procs) { ` +
+            `  if ($p.CommandLine -and $p.CommandLine -like "*--remote-debugging-port=${port}*") { ` +
+            `    $sum += $p.WorkingSetSize ` +
+            `  } ` +
+            `}; ` +
+            `Write-Output $sum`;
+        } else {
+          ps1Content =
+            `$procs = Get-CimInstance Win32_Process -Filter "Name LIKE 'msedge%'" -ErrorAction SilentlyContinue; ` +
+            `$sum = 0; ` +
+            `foreach ($p in $procs) { $sum += $p.WorkingSetSize }; ` +
+            `Write-Output $sum`;
+        }
+        writeFileSync(ps1Path, ps1Content);
+        cmd = `powershell.exe -NoProfile -ExecutionPolicy Bypass -File "${ps1Path}"`;
+        const result = execSync(cmd, { timeout: 5000 }).toString().trim();
+        try {
+          unlinkSync(ps1Path);
+        } catch {}
+        const bytes = parseInt(result, 10);
+        const mb = bytes > 0 ? Math.round(bytes / 1048576) : 0;
+        return mb > 0 ? mb : null;
+      } catch {
+        try {
+          unlinkSync(ps1Path);
+        } catch {}
+        return null;
+      }
     }
-    const mb = parseInt(execSync(cmd, { timeout: 5000 }).toString().trim(), 10);
-    return mb > 0 ? mb : null;
+    return null;
   } catch {
     return null;
   }
 }
 
-function formatMemoryWithBrowser(port) {
+function formatMemoryWithBrowser(port = null) {
   const nodeMem = formatMemoryUsage();
   const edgeMB = getBrowserMemory(port);
   if (!edgeMB) return nodeMem;
-  const label = port ? ` edge:${port}:${edgeMB}MB` : ` edge:${edgeMB}MB`;
-  return nodeMem + label;
+  const portLabel = port ? `:${port}` : "";
+  return `${nodeMem} edge${portLabel}:${edgeMB}MB`;
 }
 
 // ────────────────────────────────────────
@@ -175,7 +235,43 @@ function getPage(browser) {
 }
 
 // ────────────────────────────────────────
-// Section 4: CDP 会话管理（端口池/浏览器/页面/内存回收）
+// Section 4: 页面请求拦截（独立函数，供自动/手动模式共用）
+// ────────────────────────────────────────
+
+const _blockedPages = new WeakSet();
+
+async function setupPageBlocking(page) {
+  if (!page || _blockedPages.has(page)) return;
+  await page.route("**/*", (route) => {
+    const req = route.request();
+    const t = req.resourceType();
+    const url = req.url();
+
+    // 资源类型拦截
+    if (t === "image" || t === "stylesheet" || t === "media") {
+      return route.abort();
+    }
+
+    // 视频流拦截（TikTok 视频通过 fetch/xhr 加载 mp4 分片，
+    // resourceType 为 fetch/xhr，需按 URL 模式匹配）
+    // 真实 URL:
+    //   v16-webapp-prime.us.tiktok.com/video/tos/...?mime_type=video_mp4
+    //   v16m-webapp.tiktokcdn-us.com/.../video/tos/...?mime_type=video_mp4
+    if (
+      /webapp.*(tiktok\.com|tiktokcdn).*\/video\/tos\//i.test(url) ||
+      /mime_type=video_mp4/i.test(url) ||
+      /\.(mp4|m3u8|ts)(\?|$)/i.test(url)
+    ) {
+      return route.abort();
+    }
+
+    route.continue();
+  });
+  _blockedPages.add(page);
+}
+
+// ────────────────────────────────────────
+// Section 5: CDP 会话管理（端口池/浏览器/页面/内存回收）
 // ────────────────────────────────────────
 
 class CdpSessionManager {
@@ -187,7 +283,8 @@ class CdpSessionManager {
     this.browser = null;
     this.page = null;
     this.lastRecycleAt = 0;
-    this._blockedPages = new WeakSet();
+    this._lastMemCheck = 0;
+    this._shuttingDown = false;
   }
 
   pickNextPort() {
@@ -211,7 +308,7 @@ class CdpSessionManager {
   /** 获取或恢复 page */
   async _ensurePage() {
     let p = getPage(this.browser);
-    if (p) await this._setupBlocking(p);
+    if (p) await setupPageBlocking(p);
     return p;
   }
 
@@ -222,17 +319,6 @@ class CdpSessionManager {
     return this.page;
   }
 
-  /** 页面请求拦截：屏蔽图片/样式减少内存 */
-  async _setupBlocking(page) {
-    if (!page || this._blockedPages.has(page)) return;
-    await page.route("**/*", (route) => {
-      const t = route.request().resourceType();
-      if (t === "image" || t === "stylesheet") route.abort();
-      else route.continue();
-    });
-    this._blockedPages.add(page);
-  }
-
   /** 轻量回收：清空当前 tab */
   async recyclePage() {
     if (!this.page || this.page.isClosed()) {
@@ -278,8 +364,12 @@ class CdpSessionManager {
     this.lastRecycleAt = Date.now();
   }
 
-  /** 内存检查 + 按需回收 */
+  /** 内存检查 + 按需回收（每 30s 最多采样一次） */
   async maybeRecycleForMemory(processedCount) {
+    const now = Date.now();
+    if (now - this._lastMemCheck < 30000) return;
+    this._lastMemCheck = now;
+
     const mem = process.memoryUsage();
     const rssMB = mem.rss / 1024 / 1024;
     const heapMB = mem.heapUsed / 1024 / 1024;
@@ -334,6 +424,8 @@ class CdpSessionManager {
   }
 
   async shutdown() {
+    if (this._shuttingDown) return;
+    this._shuttingDown = true;
     if (this.page) await this.page.close().catch(() => {});
     await this.browser?.close().catch(() => {});
     killEdgeProcesses(null, this.port);
@@ -341,7 +433,7 @@ class CdpSessionManager {
 }
 
 // ────────────────────────────────────────
-// Section 5: 共享操作（视频页准备 & 评论提取 & 保存）
+// Section 6: 共享操作（视频页准备 & 评论提取 & 保存）
 // ────────────────────────────────────────
 
 async function prepareVideoPage(page, href) {
@@ -390,13 +482,12 @@ function extractUniqueAuthors(comments) {
   ];
 }
 
-async function saveAuthorsToServer(serverUrl, authors, guessedLocation) {
-  const { apiPost } = createApiClient({
-    checkStatus: true,
-    maxRetries: 2,
-    backoff: 2000,
-    log: true,
-  });
+async function saveAuthorsToServer(
+  apiPost,
+  serverUrl,
+  authors,
+  guessedLocation,
+) {
   return apiPost(`${serverUrl}/api/users`, {
     usernames: authors.map((a) => "@" + a),
     sources: ["comment"],
@@ -405,7 +496,7 @@ async function saveAuthorsToServer(serverUrl, authors, guessedLocation) {
 }
 
 // ────────────────────────────────────────
-// Section 6: 自动模式 — 循环从服务端取任务、抓评论、提交
+// Section 7: 自动模式 — 循环从服务端取任务、抓评论、提交
 // ────────────────────────────────────────
 
 async function runAutoMode(options) {
@@ -426,7 +517,14 @@ async function runAutoMode(options) {
 
   loadEnvConfig();
 
-  const { apiGet, apiPut } = createApiClient({
+  // ── 设备标识 ──
+  let userId = configuredUserId;
+  if (!userId) {
+    userId = await getMacOrUuid();
+    saveUserId(userId);
+  }
+
+  const { apiGet, apiPost, apiPut } = createApiClient({
     checkStatus: true,
     maxRetries: 2,
     backoff: 2000,
@@ -441,7 +539,8 @@ async function runAutoMode(options) {
     browserRestartCount = 0;
   let consecutiveErrors = 0,
     consecutive403 = 0,
-    consecutiveCaptchaCount = 0;
+    consecutiveCaptchaCount = 0,
+    consecutiveTaskErrors = 0;
 
   // ── CDP 会话 ──
   const session = new CdpSessionManager({
@@ -617,6 +716,16 @@ async function runAutoMode(options) {
               });
 
               if (captchaResult.detected) {
+                // 上报验证码错误
+                apiPost(`${serverUrl}/api/error-report`, {
+                  userId,
+                  username: "",
+                  errorType: "captcha",
+                  errorMessage: resultNote || "评论采集时出现验证码",
+                  stage: "comments",
+                  errorStack: "",
+                }).catch(() => {});
+
                 if (captchaResult.resolved) {
                   resultNote = "验证码已解决";
                   consecutiveCaptchaCount = 0;
@@ -648,6 +757,7 @@ async function runAutoMode(options) {
 
               if (authors.length > 0) {
                 const saveResult = await saveAuthorsToServer(
+                  apiPost,
                   serverUrl,
                   authors,
                   loc,
@@ -669,6 +779,7 @@ async function runAutoMode(options) {
             await apiPut(`${serverUrl}/api/comment-task/${id}`);
             processedCount++;
             consecutive403 = 0;
+            consecutiveTaskErrors = 0;
 
             // 每 10 个任务输出内存汇总
             if (totalTasks() % 10 === 0) {
@@ -678,6 +789,7 @@ async function runAutoMode(options) {
             }
           } catch (err) {
             errorCount++;
+            consecutiveTaskErrors++;
             const isBrowserClosed = isBrowserClosedError(err);
             const is403 =
               err.message &&
@@ -703,6 +815,22 @@ async function runAutoMode(options) {
             } else {
               console.error(`        ${err.message}`);
             }
+
+            // 连续任务错误达到阈值，汇报服务端
+            if (consecutiveTaskErrors >= CFG.consecutiveTaskErrorsMax) {
+              console.error(
+                `  ⚠️ 连续任务错误 ${consecutiveTaskErrors}/${CFG.consecutiveTaskErrorsMax} → 汇报服务端`,
+              );
+              apiPost(`${serverUrl}/api/error-report`, {
+                userId,
+                username: "",
+                errorType: "consecutive_task_errors",
+                errorMessage: `连续 ${consecutiveTaskErrors} 个任务处理失败 (port=${session.port})`,
+                stage: "comments",
+                errorStack: "",
+              }).catch(() => {});
+              consecutiveTaskErrors = 0;
+            }
           }
 
           // 任务间内存检查
@@ -710,6 +838,7 @@ async function runAutoMode(options) {
         }
       } catch (err) {
         errorCount++;
+        consecutiveTaskErrors++;
         if (isBrowserClosedError(err)) {
           browserRestartCount++;
           console.error(`  ⚠ 浏览器断开 (#${browserRestartCount}) → 重启`);
@@ -717,6 +846,22 @@ async function runAutoMode(options) {
         } else {
           console.error(`  × ${err.message}`);
         }
+
+        // 连续轮间错误达到阈值，汇报服务端
+        if (consecutiveTaskErrors >= CFG.consecutiveTaskErrorsMax) {
+          console.error(
+            `  ⚠️ 连续轮间错误 ${consecutiveTaskErrors}/${CFG.consecutiveTaskErrorsMax} → 汇报服务端`,
+          );
+          apiPost(`${serverUrl}/api/error-report`, {
+            userId,
+            username: "",
+            errorType: "consecutive_task_errors",
+            errorMessage: `连续 ${consecutiveTaskErrors} 个轮次处理失败 (port=${session.port})`,
+            stage: "comments",
+            errorStack: "",
+          }).catch(() => {});
+          consecutiveTaskErrors = 0;
+        }
       }
 
       // 轮间内存检查 + 等待
@@ -734,7 +879,7 @@ async function runAutoMode(options) {
 }
 
 // ────────────────────────────────────────
-// Section 7: 手动模式 — 单视频评论提取 + 输出/保存
+// Section 8: 手动模式 — 单视频评论提取 + 输出/保存
 // ────────────────────────────────────────
 
 async function runManualMode(url, options) {
@@ -742,17 +887,24 @@ async function runManualMode(url, options) {
     maxComments = CFG.maxComments,
     save: shouldSave,
     serverUrl = defaultServer,
+    basePort = 9222,
+    proxy: proxyOption,
   } = options;
+  const effectiveProxy = proxyOption || configuredProxy;
 
   let browser;
   try {
-    browser = await chromium.connectOverCDP("http://127.0.0.1:9222");
+    const { chromium } = await import("playwright");
+    browser = await chromium.connectOverCDP(`http://127.0.0.1:${basePort}`);
     const page = getPage(browser);
     if (!page) {
       console.error("未找到可用页面");
       process.exit(1);
     }
 
+    // 拦截图片/视频请求
+    await setupPageBlocking(page);
+
     console.error(`正在打开: ${url}`);
     await prepareVideoPage(page, url);
 
@@ -788,7 +940,14 @@ async function runManualMode(url, options) {
         if (guessedLocation) console.error(`猜测国家: ${guessedLocation}`);
 
         try {
+          const { apiPost } = createApiClient({
+            checkStatus: true,
+            maxRetries: 2,
+            backoff: 2000,
+            log: true,
+          });
           const saveResult = await saveAuthorsToServer(
+            apiPost,
             serverUrl,
             authors,
             guessedLocation,
@@ -834,7 +993,7 @@ async function runManualMode(url, options) {
 }
 
 // ────────────────────────────────────────
-// Section 8: 入口 — handleComments
+// Section 9: 入口 — handleComments
 // ────────────────────────────────────────
 
 export async function handleComments(options) {
@@ -866,5 +1025,7 @@ export async function handleComments(options) {
     maxComments: commentsMax,
     save: commentsSave,
     serverUrl: commentsServer || defaultServer,
+    basePort: commentsBasePort,
+    proxy: commentsProxy,
   });
 }

package/src/cli/tag.js

@@ -17,7 +17,14 @@ import { discoverTags } from "../lib/tag-discover.js";
 import {
   server as cfgServer,
   proxy as configuredProxy,
+  userId as configuredUserId,
+  saveUserId,
 } from "../lib/constants.js";
+import { getMacOrUuid } from "../lib/mac-or-uuid.js";
+import {
+  waitAndGetCaptcha,
+  closeCaptcha,
+} from "../scraper/modules/captcha-handler.js";
 
 const ALL_COUNTRIES = DEFAULT_TARGET_LOCATIONS;
 const DEFAULT_SERVER = cfgServer || "http://127.0.0.1:3000";
@@ -688,6 +695,13 @@ export async function handleScoreAll(parsed) {
   log(`📊 启动内存: ${formatMemoryWithBrowser(cdpPort)}`);
   log("");
 
+  // ── 设备标识 ──
+  let userId = configuredUserId;
+  if (!userId) {
+    userId = await getMacOrUuid();
+    saveUserId(userId);
+  }
+
   // 连接 CDP 浏览器
   const cdpOpts = { port: cdpPort };
   if (effectiveProxy) cdpOpts.proxyServer = effectiveProxy;
@@ -850,6 +864,8 @@ export async function handleScoreAll(parsed) {
 
   let totalScored = 0;
   let emptyRounds = 0; // 连续无任务的轮数
+  let consecutiveCaptcha = 0;
+  const MAX_CONSECUTIVE_CAPTCHA = 3;
   let consecutive403 = 0; // 连续 HTTP 403 次数，达 3 次认为被封
   const CONSECUTIVE_403_BLOCKED = 3; // 连续 3 次 403 认为账号被封
 
@@ -1098,6 +1114,57 @@ export async function handleScoreAll(parsed) {
         }
 
         log(`  ❌ 失败: ${e.message}`);
+
+        // ── 验证码检测 ──
+        const isCaptcha =
+          e.message &&
+          (e.message.includes("验证码") || e.message.includes("captcha"));
+
+        if (isCaptcha) {
+          consecutiveCaptcha++;
+          log(
+            `  🔐 验证码 (${consecutiveCaptcha}/${MAX_CONSECUTIVE_CAPTCHA})，等待用户手动解决...`,
+          );
+
+          // 上报验证码错误
+          try {
+            await fetch(`${baseUrl}/api/error-report`, {
+              method: "POST",
+              headers: buildClientHeaders(clientId, clientMeta, {
+                "Content-Type": "application/json",
+              }),
+              body: JSON.stringify({
+                userId,
+                username: "",
+                errorType: "captcha",
+                errorMessage: e.message,
+                stage: "tag-scoring",
+                errorStack: "",
+              }),
+            });
+          } catch {}
+
+          // 等待用户手动解决验证码（最多 3 分钟）
+          try {
+            const captchaResult = await waitAndGetCaptcha(page, {
+              waitMs: 180000,
+              pollInterval: 5000,
+              log: () => {},
+            });
+            if (captchaResult.detected && !captchaResult.resolved) {
+              await closeCaptcha(page).catch(() => {});
+            }
+          } catch {}
+
+          if (consecutiveCaptcha >= MAX_CONSECUTIVE_CAPTCHA) {
+            log(`  ♻️ 连续${MAX_CONSECUTIVE_CAPTCHA}次验证码 → 切换端口`);
+            await recycleCdpSession("连续验证码", { switchPort: true });
+            consecutiveCaptcha = 0;
+          }
+        } else {
+          consecutiveCaptcha = 0;
+        }
+
         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";

package/src/watch/data-store.js

@@ -500,6 +500,8 @@ export function createStore(filePath, options = {}) {
     llmScore: false,
     llmMinScore: 60,
     llmSampleSize: 100,
+    llmTagRatio: 0.4,
+    llmCommentRatio: 0.3,
     ...options.refillLlm,
   };
 
@@ -957,27 +959,30 @@ export function createStore(filePath, options = {}) {
       return { moved: 0 };
     }
 
-    // 如果启用 LLM 打分：先取 tag（一次性），再取非 tag 走 LLM 打分
+    // 如果启用 LLM 打分：tag 和 comment 免打分直接入围，其余走 LLM 打分
     if (useLlm && normalizedLocations && normalizedLocations.length > 0) {
       const llmTotal = options.llmTotal ?? 200; // 总条数
-      const llmTagRatio = options.llmTagRatio ?? 0.6; // tag 占比 60%
-      const llmTagLimit = Math.floor(llmTotal * llmTagRatio); // tag 上限 120
-      const llmNonTagTarget = llmTotal - llmTagLimit; // 非 tag 目标 80
+      const llmTagRatio = options.llmTagRatio ?? 0.4; // tag 占比 40%
+      const llmCommentRatio = options.llmCommentRatio ?? 0.3; // comment 占比 30%
+      const llmTagLimit = Math.floor(llmTotal * llmTagRatio); // tag 上限
+      const llmCommentLimit = Math.floor(llmTotal * llmCommentRatio); // comment 上限
+      const llmLLMTarget = llmTotal - llmTagLimit - llmCommentLimit; // LLM 打分目标
       const llmMinScore = options.llmMinScore ?? 60;
       const llmSampleSize = options.llmSampleSize ?? 100;
       const maxBatches = options.llmMaxBatches ?? 10;
 
       console.error(
-        `[data-store] LLM 打分开始: 总目标 ${llmTotal} 条，tag 最多 ${llmTagLimit} 条（一次性），非 tag 目标 ${llmNonTagTarget} 条（LLM 打分）`,
+        `[data-store] LLM 调度开始: 总目标 ${llmTotal} 条，tag ${llmTagLimit} 条，comment ${llmCommentLimit} 条（免打分），LLM 打分 ${llmLLMTarget} 条`,
       );
 
       // 返回 Promise，调用方需要 await
       return (async () => {
         const allTagQualified = [];
-        const allNonTagQualified = [];
+        const allCommentQualified = [];
+        const allLLMQualified = [];
         const allScores = [];
 
-        // ===== 第一步：一次性取所有 tag（全局，最多 llmTagLimit 条）=====
+        // ===== 第一步：一次性取 tag（全局，最多 llmTagLimit 条，免打分）=====
         let tagOffset = offsetStore.get("_tag") || 0;
         const tagGlobalCount =
           getDb()
@@ -1025,43 +1030,97 @@ export function createStore(filePath, options = {}) {
 
         offsetStore.set("_tag", tagOffset);
 
-        // ===== 第二步：按国家取非 tag，走 LLM 打分，直到合格数达到 llmNonTagTarget =====
+        // ===== 第二步：一次性取 comment（全局，最多 llmCommentLimit 条，免打分）=====
+        let commentOffset = offsetStore.get("_comment") || 0;
+        const commentGlobalCount =
+          getDb()
+            .prepare(
+              `SELECT COUNT(*) as c FROM raw_jobs WHERE ${whereSql} AND sources LIKE '%comment%' AND sources NOT LIKE '%tag%'`,
+            )
+            .get(...args)?.c || 0;
+
+        if (commentOffset >= commentGlobalCount) {
+          commentOffset = 0;
+          offsetStore.set("_comment", 0);
+        }
+
+        console.error(
+          `[data-store] Comment 全局共 ${commentGlobalCount} 条，从偏移量 ${commentOffset} 开始`,
+        );
+
+        while (
+          allCommentQualified.length < llmCommentLimit &&
+          commentOffset < commentGlobalCount
+        ) {
+          const batch = getDb()
+            .prepare(
+              `
+              SELECT * FROM raw_jobs WHERE ${whereSql} AND sources LIKE '%comment%' AND sources NOT LIKE '%tag%'
+              ORDER BY COALESCE(video_count, 0) DESC, created_at DESC
+              LIMIT ? OFFSET ?
+            `,
+            )
+            .all(
+              ...args,
+              Math.min(
+                llmSampleSize,
+                llmCommentLimit - allCommentQualified.length,
+              ),
+              commentOffset,
+            );
+
+          if (!batch.length) break;
+
+          allCommentQualified.push(...batch.map((s) => s.unique_id));
+          commentOffset += batch.length;
+
+          console.error(
+            `[data-store] Comment 本批 ${batch.length} 条，累计 ${allCommentQualified.length}/${llmCommentLimit}`,
+          );
+        }
+
+        offsetStore.set("_comment", commentOffset);
+
+        // ===== 第三步：按国家取非 tag 非 comment，走 LLM 打分，直到合格数达到 llmLLMTarget =====
         for (const location of normalizedLocations) {
-          if (allNonTagQualified.length >= llmNonTagTarget) break;
+          if (allLLMQualified.length >= llmLLMTarget) break;
 
-          const nonTagOffsetKey = `${location}:nonTag`;
-          let offset = offsetStore.get(nonTagOffsetKey) || 0;
+          const llmOffsetKey = `${location}:llm`;
+          let offset = offsetStore.get(llmOffsetKey) || 0;
 
           const locationArgs = [...args, location];
-          const nonTagCount =
+          const llmCount =
             getDb()
               .prepare(
-                `SELECT COUNT(*) as c FROM raw_jobs WHERE ${whereSql} AND guessed_location = ? AND (sources NOT LIKE '%tag%' OR sources IS NULL)`,
+                `SELECT COUNT(*) as c FROM raw_jobs WHERE ${whereSql} AND guessed_location = ? AND (sources NOT LIKE '%tag%' OR sources IS NULL) AND (sources NOT LIKE '%comment%' OR sources IS NULL)`,
               )
               .get(...locationArgs)?.c || 0;
 
-          if (nonTagCount === 0) {
-            console.error(`[data-store] 国家 ${location}: 无非 tag 数据，跳过`);
+          if (llmCount === 0) {
+            console.error(
+              `[data-store] 国家 ${location}: 无 LLM 打分候选数据，跳过`,
+            );
             continue;
           }
 
-          if (offset >= nonTagCount) {
+          if (offset >= llmCount) {
             offset = 0;
-            offsetStore.set(nonTagOffsetKey, 0);
+            offsetStore.set(llmOffsetKey, 0);
           }
 
           console.error(
-            `[data-store] 国家 ${location}: 非 tag 共 ${nonTagCount} 条，从偏移量 ${offset} 开始`,
+            `[data-store] 国家 ${location}: LLM 候选共 ${llmCount} 条，从偏移量 ${offset} 开始`,
           );
 
           for (let batch = 0; batch < maxBatches; batch++) {
-            if (allNonTagQualified.length >= llmNonTagTarget) break;
+            if (allLLMQualified.length >= llmLLMTarget) break;
 
             const samples = getDb()
               .prepare(
                 `
                 SELECT * FROM raw_jobs WHERE ${whereSql} AND guessed_location = ?
                   AND (sources NOT LIKE '%tag%' OR sources IS NULL)
+                  AND (sources NOT LIKE '%comment%' OR sources IS NULL)
                 ORDER BY COALESCE(video_count, 0) DESC, created_at DESC
                 LIMIT ? OFFSET ?
               `,
@@ -1075,24 +1134,28 @@ export function createStore(filePath, options = {}) {
               DEFAULT_TARGET_LOCATIONS,
             );
             const qualified = scores.filter((s) => s.score >= llmMinScore);
-            allNonTagQualified.push(...qualified.map((s) => s.uniqueId));
+            allLLMQualified.push(...qualified.map((s) => s.uniqueId));
             allScores.push(...scores);
 
             offset += samples.length;
-            offsetStore.set(nonTagOffsetKey, offset);
+            offsetStore.set(llmOffsetKey, offset);
 
             console.error(
-              `[data-store] ${location} 第 ${batch + 1} 批: 采样 ${samples.length} 条，本批合格 ${qualified.length} 条，非 tag 累计 ${allNonTagQualified.length}/${llmNonTagTarget}`,
+              `[data-store] ${location} 第 ${batch + 1} 批: 采样 ${samples.length} 条，本批合格 ${qualified.length} 条，LLM 累计 ${allLLMQualified.length}/${llmLLMTarget}`,
             );
           }
         }
 
         // ===== 最终结果 =====
-        const qualified = [...allTagQualified, ...allNonTagQualified];
+        const qualified = [
+          ...allTagQualified,
+          ...allCommentQualified,
+          ...allLLMQualified,
+        ];
 
         if (!qualified.length) {
           console.error(
-            `[data-store] LLM 打分后无符合条件的任务（tag: ${allTagQualified.length}, 非 tag: ${allNonTagQualified.length}）`,
+            `[data-store] LLM 调度后无符合条件的任务（tag: ${allTagQualified.length}, comment: ${allCommentQualified.length}, LLM: ${allLLMQualified.length}）`,
           );
           return {
             moved: 0,
@@ -1103,7 +1166,7 @@ export function createStore(filePath, options = {}) {
         }
 
         console.error(
-          `[data-store] LLM 打分完成: tag ${allTagQualified.length} 条 + 非 tag ${allNonTagQualified.length} 条 = 共 ${qualified.length} 条`,
+          `[data-store] LLM 调度完成: tag ${allTagQualified.length} + comment ${allCommentQualified.length} + LLM ${allLLMQualified.length} = 共 ${qualified.length} 条`,
         );
 
         // 移动符合条件的记录
@@ -1181,6 +1244,7 @@ export function createStore(filePath, options = {}) {
       WHERE __WHERE__
       ORDER BY
         CASE WHEN sources LIKE '%tag%' THEN 0 ELSE 1 END,
+        CASE WHEN sources LIKE '%comment%' THEN 0 ELSE 1 END,
         COALESCE(video_count, 0) DESC, created_at DESC
       LIMIT ?
     `;
@@ -1192,6 +1256,7 @@ export function createStore(filePath, options = {}) {
         WHERE __WHERE__
         ORDER BY
           CASE WHEN sources LIKE '%tag%' THEN 0 ELSE 1 END,
+          CASE WHEN sources LIKE '%comment%' THEN 0 ELSE 1 END,
           COALESCE(video_count, 0) DESC, created_at DESC
         LIMIT ?
       )