feat: KOC 资源导入后自动爬取主页资料并调用 LLM 打标签

- 新增 structured_text_process 打标客户端（文本构造、标签解析清洗、MCP 调用）
- 写库后后台打标（tagUntaggedAccounts），单条新增 / 批量导入 / 每日补全三个入口统一
- 标签只补空，UPDATE 带 tags 为空守卫，不覆盖已有标签
- 新增 AI_TAGGING_TOOL 配置项（留空关闭），更新配置示例与文档
- 新增 16 个打标测试用例，全量 131 个测试通过

.dev.vars.example

@@ -8,6 +8,9 @@ KOC_MCP_API_KEY=replace-with-a-separate-long-random-secret
 AI_TOOL_CENTER_MCP_URL=https://middle-aitool.gbotai.cn/mcp
 AI_TOOL_CENTER_MCP_KEY=replace-with-mcp-key
 
+# 账号 LLM 自动打标（可选）：工具中心 structured_text_process 工具，留空 = 关闭自动打标
+AI_TAGGING_TOOL=
+
 # Feishu custom app credentials. Keep the secret out of source control.
 FEISHU_APP_ID=cli_xxxxxxxxxxxxxxxxx
 FEISHU_APP_SECRET=replace-with-feishu-app-secret

.env.self-hosted.example

@@ -24,6 +24,9 @@ FEISHU_APP_SECRET=
 AI_TOOL_CENTER_MCP_URL=
 AI_TOOL_CENTER_MCP_KEY=
 
+# 账号 LLM 自动打标（可选）：AI_TOOL_CENTER 的 structured_text_process 工具，留空 = 关闭自动打标。
+AI_TAGGING_TOOL=
+
 # 企业微信通知（临期催办 + 管理员汇总）。群机器人只需 webhook；KOC 侧催办还需 corp/agent/secret，
 # 并在后台 partners 编辑里把 wecom_external_user_id 填好。
 WECOM_ROBOT_WEBHOOK=

CLAUDE.md

@@ -84,6 +84,7 @@ ADMIN_ALLOWED_EMAIL    — ChatGPT email allowed for admin access
 ADMIN_INTERNAL_TOKEN   — Shared secret for API-to-API auth
 AI_TOOL_CENTER_MCP_URL — MCP endpoint for XHS data collection
 AI_TOOL_CENTER_MCP_KEY — MCP API key
+AI_TAGGING_TOOL        — 账号 LLM 自动打标：工具中心 structured_text_process，留空关闭（可选）
 FEISHU_APP_ID          — Feishu custom app credentials
 FEISHU_APP_SECRET      — Feishu app secret
 WECOM_CORP_ID          — 企业微信企业 ID（自建应用消息推送用，可选）

app/api/resources-import/route.ts

@@ -4,7 +4,7 @@ import {
   parseResourceImportFile,
   RESOURCE_IMPORT_MAX_BYTES,
   RESOURCE_IMPORT_MAX_ROWS,
-  resourceImportMissingFields,
+  needsResourceEnrichment,
   type ResourceImportRow,
 } from "../../../lib/resource-import";
 import {
@@ -12,6 +12,7 @@ import {
   enrichRows,
   loadAccounts,
   mergeExistingFields,
+  tagUntaggedAccounts,
   writeAnalyzedRows,
   RESOURCE_IMPORT_SYNC_ENRICH_ROWS,
   type AnalyzedRow,
@@ -49,7 +50,7 @@ function deferredEnrichmentCount(rows: ResourceImportRow[]) {
     (row) =>
       row.errors.length === 0 &&
       row.profileUrl &&
-      resourceImportMissingFields(row).length > 0,
+      needsResourceEnrichment(row),
   ).length;
 }
 
@@ -60,7 +61,7 @@ async function enrichImportedRowsInBackground(rows: ResourceImportRow[]) {
     (row) =>
       row.errors.length === 0 &&
       row.profileUrl &&
-      resourceImportMissingFields(row).length > 0,
+      needsResourceEnrichment(row),
   );
   if (missingRows.length === 0) return;
   const enriched = await enrichRows(missingRows, accounts.results);
@@ -132,6 +133,10 @@ export async function POST(request: Request) {
       );
     }
     await writeAnalyzedRows(importableRows);
+    runInBackground(
+      tagUntaggedAccounts(importableRows),
+      "resource account tagging",
+    );
     if (deferredEnrichment > 0) {
       const importableSourceRows = rows.filter((row) =>
         importableRowNumbers.has(row.rowNumber),

app/api/resources-insert/route.ts

@@ -1,9 +1,11 @@
 import { isManagerRequest, managerForbidden } from "../../../lib/user-auth";
+import { runInBackground } from "../../../lib/background";
 import { buildManualRow, type ManualResourceInput } from "../../../lib/resource-import";
 import {
   analyzeRows,
   enrichRows,
   loadAccounts,
+  tagUntaggedAccounts,
   writeAnalyzedRows,
 } from "../../../lib/resource-write";
 
@@ -48,6 +50,10 @@ export async function POST(request: Request) {
       );
     }
     await writeAnalyzedRows(analyzed);
+    runInBackground(
+      tagUntaggedAccounts(analyzed),
+      "resource account tagging",
+    );
     const result = analyzed[0];
     const outcome: InsertResult = {
       action: result.action === "update" ? "update" : "create",

examples/koc-resources-sample.csv

@@ -0,0 +1,9 @@
+账号链接,账号昵称,小红书号,IP属地,粉丝数,性别,简介,标签,合作来源
+https://www.xiaohongshu.com/user/profile/5ff0e6410000000001008401,小红薯爱分享,xhs_88012345,上海,1.2万,女,爱美食爱生活的上海小姐姐，合作私信~,"美食,探店,生活方式",媒介推荐
+https://www.xiaohongshu.com/user/profile/5ff0e6410000000001008402,阿杰的穿搭日记,xhs_7712ab,广东,8600,男,男生穿搭 | 平价好物 | 每周更新,"穿搭,男装,平价好物",历史合作
+https://www.xiaohongshu.com/user/profile/5ff0e6410000000001008403,琳琳妈妈育儿记,xhs_6633cd,浙江,3.5万,女,二胎妈妈，分享育儿干货和母婴好物,"母婴,育儿,好物分享",蒲公英
+https://www.xiaohongshu.com/user/profile/5ff0e6410000000001008404,健身教练大凯,xhs_5544ef,北京,10万+,男,NSCA认证教练，运动补剂科普,"健身,运动,补剂",自荐
+https://www.xiaohongshu.com/user/profile/5ff0e6410000000001008405,甜品研究社,xhs_4455gh,四川,2.1万,女,烘焙教程 | 甜品测评 | 成都探店,"烘焙,甜品,探店",媒介推荐
+https://www.douyin.com/user/ms4y9x873kq1,数码测评君,dy_3321ij,江苏,5.8万,男,手机数码深度测评，避坑指南,"数码,测评,科技",星图
+https://www.xiaohongshu.com/user/profile/5ff0e6410000000001008407,旅行摄影的小舟,xhs_2266mn,云南,6400,女,风光摄影 | 旅行vlog | 云南旅拍,"旅行,摄影",历史合作
+https://www.xiaohongshu.com/user/profile/5ff0e6410000000001008408,美妆小课堂Amy,xhs_1177op,上海,15万,女,化妆品成分党，专注国货美妆测评,"美妆,护肤,国货",蒲公英

examples/koc-resources-sample.jsonl

@@ -0,0 +1,8 @@
+{"profileUrl": "https://www.xiaohongshu.com/user/profile/5ff0e6410000000001008401", "nickname": "小红薯爱分享", "publicAccountId": "xhs_88012345", "ipLocation": "上海", "followers": "1.2万", "gender": "女", "bio": "爱美食爱生活的上海小姐姐，合作私信~", "tags": "美食,探店,生活方式", "cooperationSource": "媒介推荐"}
+{"profileUrl": "https://www.xiaohongshu.com/user/profile/5ff0e6410000000001008402", "nickname": "阿杰的穿搭日记", "publicAccountId": "xhs_7712ab", "ipLocation": "广东", "followers": "8600", "gender": "男", "bio": "男生穿搭 | 平价好物 | 每周更新", "tags": "穿搭,男装,平价好物", "cooperationSource": "历史合作"}
+{"profileUrl": "https://www.xiaohongshu.com/user/profile/5ff0e6410000000001008403", "nickname": "琳琳妈妈育儿记", "publicAccountId": "xhs_6633cd", "ipLocation": "浙江", "followers": "3.5万", "gender": "女", "bio": "二胎妈妈，分享育儿干货和母婴好物", "tags": "母婴,育儿,好物分享", "cooperationSource": "蒲公英"}
+{"profileUrl": "https://www.xiaohongshu.com/user/profile/5ff0e6410000000001008404", "nickname": "健身教练大凯", "publicAccountId": "xhs_5544ef", "ipLocation": "北京", "followers": "10万+", "gender": "男", "bio": "NSCA认证教练，运动补剂科普", "tags": "健身,运动,补剂", "cooperationSource": "自荐"}
+{"profileUrl": "https://www.xiaohongshu.com/user/profile/5ff0e6410000000001008405", "nickname": "甜品研究社", "publicAccountId": "xhs_4455gh", "ipLocation": "四川", "followers": "2.1万", "gender": "女", "bio": "烘焙教程 | 甜品测评 | 成都探店", "tags": "烘焙,甜品,探店", "cooperationSource": "媒介推荐"}
+{"profileUrl": "https://www.douyin.com/user/ms4y9x873kq1", "nickname": "数码测评君", "publicAccountId": "dy_3321ij", "ipLocation": "江苏", "followers": "5.8万", "gender": "男", "bio": "手机数码深度测评，避坑指南", "tags": "数码,测评,科技", "cooperationSource": "星图"}
+{"profileUrl": "https://www.xiaohongshu.com/user/profile/5ff0e6410000000001008407", "nickname": "旅行摄影的小舟", "publicAccountId": "xhs_2266mn", "ipLocation": "云南", "followers": "6400", "gender": "女", "bio": "风光摄影 | 旅行vlog | 云南旅拍", "tags": "旅行,摄影", "cooperationSource": "历史合作"}
+{"profileUrl": "https://www.xiaohongshu.com/user/profile/5ff0e6410000000001008408", "nickname": "美妆小课堂Amy", "publicAccountId": "xhs_1177op", "ipLocation": "上海", "followers": "15万", "gender": "女", "bio": "化妆品成分党，专注国货美妆测评", "tags": "美妆,护肤,国货", "cooperationSource": "蒲公英"}

lib/account-enrichment-service.ts

@@ -1,9 +1,12 @@
 import {
+  generateAccountTagsFromMcp,
   resolveXhsPublicAccountDetails,
   resolveAccountProfileFromMcp,
   resolveProfileDetailsFromMcp,
+  type AccountTaggingEnv,
   type CollectionMcpConfig,
 } from "./mcp-collection-client";
+import { getRuntimeEnv } from "./runtime-env";
 import { hashText } from "./mvp-db";
 import type { DatabaseClient } from "./database";
 
@@ -260,6 +263,7 @@ export async function backfillAccountProfiles(
   mcpConfig: CollectionMcpConfig,
   limit = 10,
 ) {
+  const taggingEnv: AccountTaggingEnv = getRuntimeEnv();
   const rows = await db
     .prepare(
       `SELECT
@@ -334,6 +338,7 @@ export async function backfillAccountProfiles(
     const isDemoAccount = row.platform_uid?.startsWith("xhs-");
     let attemptedThisRow = false;
     let publicProfileUpdated = false;
+    let tagsUpdated = false;
     if (
       row.platform === "小红书" &&
       !isDemoAccount &&
@@ -341,7 +346,8 @@ export async function backfillAccountProfiles(
       (!row.public_account_id ||
         Number(row.followers ?? 0) === 0 ||
         !row.gender ||
-        !row.bio)
+        !row.bio ||
+        !row.tags?.trim())
     ) {
       attempted += 1;
       attemptedThisRow = true;
@@ -402,6 +408,39 @@ export async function backfillAccountProfiles(
           .run();
         publicProfileUpdated = true;
       }
+      if (
+        row.account_id &&
+        !row.tags?.trim() &&
+        (details.bio ||
+          details.providerTags.length > 0 ||
+          details.recentNoteTitles.length > 0)
+      ) {
+        const tags = await generateAccountTagsFromMcp(
+          {
+            platform: "小红书",
+            nickname: row.nickname || details.nickname || "",
+            bio: details.bio,
+            gender: details.gender,
+            ipLocation: details.ipLocation ?? "",
+            followers: details.followers,
+            providerTags: details.providerTags,
+            recentNoteTitles: details.recentNoteTitles,
+          },
+          mcpConfig,
+          taggingEnv,
+        ).catch(() => []);
+        if (tags.length > 0) {
+          await db
+            .prepare(
+              `UPDATE accounts
+               SET tags = CASE WHEN ? != '' THEN ? ELSE tags END
+               WHERE id = ?`,
+            )
+            .bind(tags.join(","), tags.join(","), row.account_id)
+            .run();
+          tagsUpdated = true;
+        }
+      }
       if (
         (details.redId || row.public_account_id) &&
         Number(details.followers ?? row.followers ?? 0) > 0
@@ -410,7 +449,7 @@ export async function backfillAccountProfiles(
         continue;
       }
       if (attempted >= Math.max(1, Math.min(25, limit))) {
-        if (publicProfileUpdated) updated += 1;
+        if (publicProfileUpdated || tagsUpdated) updated += 1;
         continue;
       }
     }
@@ -430,7 +469,7 @@ export async function backfillAccountProfiles(
       row.platform_uid?.startsWith("pending-") ||
       Boolean(noteId && row.platform_uid === noteId));
     if (!needsEnrichment || !row.publish_url) {
-      if (publicProfileUpdated) updated += 1;
+      if (publicProfileUpdated || tagsUpdated) updated += 1;
       continue;
     }
 
@@ -443,9 +482,9 @@ export async function backfillAccountProfiles(
         row.nickname || "待识别账号",
         mcpConfig,
       );
-      if (result.updated || publicProfileUpdated) updated += 1;
+      if (result.updated || publicProfileUpdated || tagsUpdated) updated += 1;
     } catch {
-      if (publicProfileUpdated) updated += 1;
+      if (publicProfileUpdated || tagsUpdated) updated += 1;
       else failed += 1;
     }
   }

lib/mcp-collection-client.ts

@@ -1484,3 +1484,154 @@ export async function resolveXhsProfileDetailsFromMcp(
     ? lastError
     : new Error("账号主页数据采集服务暂时不可用");
 }
+
+export type AccountTaggingInput = {
+  platform: string;
+  nickname: string;
+  bio: string;
+  gender: string;
+  ipLocation: string;
+  followers: number | null;
+  providerTags: string[];
+  recentNoteTitles: string[];
+};
+
+export type AccountTaggingEnv = {
+  AI_TAGGING_TOOL?: string;
+};
+
+export function buildAccountTagText(input: AccountTaggingInput): string {
+  const facts: string[] = [`平台：${input.platform.trim() || "小红书"}`];
+  if (input.nickname.trim()) facts.push(`昵称：${input.nickname.trim()}`);
+  if (input.bio.trim()) facts.push(`简介：${input.bio.trim()}`);
+  if (input.gender.trim()) facts.push(`性别：${input.gender.trim()}`);
+  if (input.ipLocation.trim()) {
+    facts.push(`IP属地：${input.ipLocation.trim()}`);
+  }
+  if (input.followers !== null && input.followers > 0) {
+    facts.push(`粉丝数：${input.followers}`);
+  }
+  if (input.providerTags.length > 0) {
+    facts.push(`平台标签：${input.providerTags.slice(0, 5).join("、")}`);
+  }
+  if (input.recentNoteTitles.length > 0) {
+    const titles = input.recentNoteTitles
+      .slice(0, 20)
+      .map((title) => `《${title}》`)
+      .join("");
+    facts.push(`最近笔记标题：${titles}`);
+  }
+  return facts.join("\n");
+}
+
+function sanitizeTagList(items: unknown[]): string[] {
+  const tags: string[] = [];
+  for (const item of items) {
+    if (typeof item !== "string") continue;
+    const tag = item.trim().replace(/^#+|#+$/g, "").trim();
+    if (!tag || tag.length > 12) continue;
+    if (!tags.includes(tag)) tags.push(tag);
+  }
+  return tags.slice(0, 5);
+}
+
+export function parseAccountTags(text: string): string[] {
+  const stripped = text
+    .trim()
+    .replace(/^```(?:json)?\s*/i, "")
+    .replace(/\s*```\s*$/, "");
+  const candidates = [stripped];
+  const arrayMatch = stripped.match(/\[[\s\S]*\]/);
+  if (arrayMatch) candidates.push(arrayMatch[0]);
+  for (const candidate of candidates) {
+    let parsed: unknown;
+    try {
+      parsed = JSON.parse(candidate);
+    } catch {
+      continue;
+    }
+    if (!Array.isArray(parsed)) continue;
+    const tags = sanitizeTagList(parsed);
+    if (tags.length > 0) return tags;
+  }
+  return [];
+}
+
+function parseTagsFromToolPayload(payload: unknown, depth = 0): string[] {
+  if (typeof payload === "string") return parseAccountTags(payload);
+  if (Array.isArray(payload)) {
+    const direct = sanitizeTagList(payload);
+    if (direct.length > 0) return direct;
+    if (depth < 2) {
+      for (const item of payload) {
+        const tags = parseTagsFromToolPayload(item, depth + 1);
+        if (tags.length > 0) return tags;
+      }
+    }
+    return [];
+  }
+  const record = asRecord(payload);
+  if (record && depth < 2) {
+    for (const key of [
+      "tags",
+      "labels",
+      "result",
+      "data",
+      "response",
+      "output",
+      "text",
+      "content",
+      "message",
+    ]) {
+      if (key in record) {
+        const tags = parseTagsFromToolPayload(record[key], depth + 1);
+        if (tags.length > 0) return tags;
+      }
+    }
+  }
+  return [];
+}
+
+export async function generateAccountTagsFromMcp(
+  input: AccountTaggingInput,
+  config: CollectionMcpConfig,
+  taggingEnv: AccountTaggingEnv,
+  fetchImpl: typeof fetch = fetch,
+): Promise<string[]> {
+  const toolName = String(taggingEnv.AI_TAGGING_TOOL ?? "").trim();
+  if (!toolName || !config.endpoint.trim()) return [];
+  try {
+    const endpoint = buildMcpUrl(config);
+    const timeoutMs = Math.max(5_000, config.timeoutMs ?? 60_000);
+    const sessionId = await createMcpSession(fetchImpl, endpoint, timeoutMs);
+    const result = await invokeMcpTool(
+      fetchImpl,
+      endpoint,
+      sessionId,
+      timeoutMs,
+      toolName,
+      {
+        objective: "为KOC账号生成运营标签",
+        text: buildAccountTagText(input),
+        rules: [
+          "最多生成 5 个标签",
+          "每个标签不超过 8 个字",
+          "从内容领域、人设风格、受众画像等维度概括",
+          "只依据文本中给到的资料，不要编造",
+        ],
+        output_schema: {
+          type: "object",
+          properties: {
+            tags: { type: "array", items: { type: "string" } },
+          },
+          required: ["tags"],
+          additionalProperties: false,
+        },
+      },
+    );
+    if (result.isError) return [];
+    return parseTagsFromToolPayload(result.payload);
+  } catch {
+    return [];
+  }
+}

lib/resource-import.ts

@@ -279,6 +279,21 @@ export function resourceImportMissingFields(
   return missing;
 }
 
+export function needsResourceEnrichment(
+  row: Pick<
+    ResourceImportRow,
+    | "nickname"
+    | "publicAccountId"
+    | "ipLocation"
+    | "followersResolved"
+    | "gender"
+    | "bio"
+    | "tags"
+  >,
+) {
+  return resourceImportMissingFields(row).length > 0 || row.tags.length === 0;
+}
+
 function normalizeRows(rows: string[][]) {
   const header = findHeader(rows);
   if (!header) {

lib/resource-write.ts

@@ -1,8 +1,10 @@
 import { ensureSchema, getRawDb } from "./mvp-db";
 import {
+  generateAccountTagsFromMcp,
   resolveCollectionMcpConfig,
   resolveProfileDetailsFromMcp,
   resolveXhsPublicAccountDetails,
+  type AccountTaggingEnv,
   type CollectionMcpBindings,
 } from "./mcp-collection-client";
 import { getRuntimeEnv } from "./runtime-env";
@@ -135,9 +137,8 @@ export async function enrichRows(
   rows: ResourceImportRow[],
   accounts: AccountRow[],
 ) {
-  const mcpConfig = resolveCollectionMcpConfig(
+  const env = getRuntimeEnv() as unknown as CollectionMcpBindings;
-    getRuntimeEnv() as unknown as CollectionMcpBindings,
+  const mcpConfig = resolveCollectionMcpConfig(env);
-  );
   const baselineRows = mergeExistingFields(rows, accounts);
   return mapConcurrent(baselineRows, 4, async (baseline) => {
     const row = baseline;
@@ -216,6 +217,89 @@ export async function enrichRows(
   });
 }
 
+type UntaggedAccountRow = {
+  id: string;
+  platform: string;
+  platform_uid: string;
+  profile_url: string;
+  nickname: string;
+};
+
+export async function tagUntaggedAccounts(
+  rows: Array<Pick<AnalyzedRow, "platform" | "platformUid">>,
+) {
+  const env = getRuntimeEnv() as unknown as CollectionMcpBindings &
+    AccountTaggingEnv;
+  const toolName = String(env.AI_TAGGING_TOOL ?? "").trim();
+  const refs = new Set(
+    rows
+      .filter((row) => row.platformUid)
+      .map((row) => `${row.platform}:${row.platformUid}`),
+  );
+  if (!toolName || refs.size === 0) return;
+
+  const db = getRawDb();
+  const uids = [...new Set([...refs].map((key) => key.slice(key.indexOf(":") + 1)))];
+  const candidates: UntaggedAccountRow[] = [];
+  for (let index = 0; index < uids.length; index += 100) {
+    const chunk = uids.slice(index, index + 100);
+    const placeholders = chunk.map(() => "?").join(", ");
+    const result = await db
+      .prepare(
+        `SELECT id, platform, platform_uid, profile_url, nickname
+         FROM accounts
+         WHERE (tags = '' OR tags IS NULL)
+           AND profile_url IS NOT NULL AND profile_url != ''
+           AND platform_uid IN (${placeholders})`,
+      )
+      .bind(...chunk)
+      .all<UntaggedAccountRow>();
+    candidates.push(...result.results);
+  }
+  const targets = candidates.filter((account) =>
+    refs.has(`${account.platform}:${account.platform_uid}`),
+  );
+  if (targets.length === 0) return;
+
+  const mcpConfig = resolveCollectionMcpConfig(env);
+  await mapConcurrent(targets, 4, async (account) => {
+    const details = await resolveProfileDetailsFromMcp(
+      account.profile_url,
+      account.platform === "抖音" ? "抖音" : "小红书",
+      mcpConfig,
+    ).catch(() => null);
+    if (
+      !details ||
+      (!details.bio &&
+        details.providerTags.length === 0 &&
+        details.recentNoteTitles.length === 0)
+    ) {
+      return;
+    }
+    const tags = await generateAccountTagsFromMcp(
+      {
+        platform: account.platform,
+        nickname: details.nickname?.trim() || account.nickname || "",
+        bio: details.bio,
+        gender: details.gender,
+        ipLocation: details.ipLocation ?? "",
+        followers: details.followers,
+        providerTags: details.providerTags,
+        recentNoteTitles: details.recentNoteTitles,
+      },
+      mcpConfig,
+      { AI_TAGGING_TOOL: toolName },
+    ).catch(() => []);
+    if (tags.length === 0) return;
+    await db
+      .prepare(
+        `UPDATE accounts SET tags = ? WHERE id = ? AND (tags = '' OR tags IS NULL)`,
+      )
+      .bind(tags.join(","), account.id)
+      .run();
+  });
+}
+
 export function analyzeRows(
   rows: ResourceImportRow[],
   accountRows: AccountRow[],

lib/runtime-env.ts

@@ -18,6 +18,7 @@ export type RuntimeEnv = {
   FEISHU_APP_SECRET?: string;
   AI_TOOL_CENTER_MCP_URL?: string;
   AI_TOOL_CENTER_MCP_KEY?: string;
+  AI_TAGGING_TOOL?: string;
   COLLECTION_MCP_URL?: string;
   COLLECTION_MCP_KEY?: string;
   WECOM_CORP_ID?: string;

tests/account-tagging.test.mjs

@@ -0,0 +1,221 @@
+import assert from "node:assert/strict";
+import test from "node:test";
+import {
+  buildAccountTagText,
+  generateAccountTagsFromMcp,
+  parseAccountTags,
+} from "../lib/mcp-collection-client.ts";
+import { needsResourceEnrichment } from "../lib/resource-import.ts";
+
+function sse(payload, headers = {}) {
+  return new Response(`event: message\ndata: ${JSON.stringify(payload)}\n\n`, {
+    status: 200,
+    headers: {
+      "content-type": "text/event-stream",
+      ...headers,
+    },
+  });
+}
+
+function toolEnvelope(payload, isError = false) {
+  return {
+    jsonrpc: "2.0",
+    id: "tool-call",
+    result: {
+      isError,
+      content: [{ type: "text", text: JSON.stringify(payload) }],
+    },
+  };
+}
+
+function createFakeTaggingMcp(toolResult, isError = false) {
+  const calls = [];
+  const fetchImpl = async (url, init) => {
+    const body = JSON.parse(init.body);
+    calls.push({ url: String(url), body });
+    if (body.method === "initialize") {
+      return sse(
+        {
+          jsonrpc: "2.0",
+          id: body.id,
+          result: {
+            protocolVersion: "2025-03-26",
+            capabilities: {},
+            serverInfo: { name: "ai-tool-center", version: "test" },
+          },
+        },
+        { "mcp-session-id": "session-test" },
+      );
+    }
+    if (body.method === "notifications/initialized") {
+      return new Response("", { status: 202 });
+    }
+    if (body.method === "tools/call") {
+      return sse(toolEnvelope(toolResult, isError));
+    }
+    throw new Error(`Unexpected MCP method: ${body.method}`);
+  };
+  return { fetchImpl, calls };
+}
+
+const config = { endpoint: "https://mcp.test/mcp", key: "test-key" };
+const taggingEnv = { AI_TAGGING_TOOL: "generate_tags" };
+const input = {
+  platform: "小红书",
+  nickname: "小美的日常",
+  bio: "分享护肤和生活",
+  gender: "女",
+  ipLocation: "上海",
+  followers: 1200,
+  providerTags: ["美妆"],
+  recentNoteTitles: ["平价护肤分享"],
+};
+
+test("parseAccountTags 解析纯 JSON 数组", () => {
+  assert.deepEqual(parseAccountTags('["美妆","学生党"]'), ["美妆", "学生党"]);
+});
+
+test("parseAccountTags 剥离代码块围栏", () => {
+  assert.deepEqual(parseAccountTags('```json\n["美食"]\n```'), ["美食"]);
+  assert.deepEqual(parseAccountTags('```\n["旅行"]\n```'), ["旅行"]);
+});
+
+test("parseAccountTags 从夹杂文字中提取数组", () => {
+  const text = '好的，以下是标签：["旅行","摄影"] 希望有帮助';
+  assert.deepEqual(parseAccountTags(text), ["旅行", "摄影"]);
+});
+
+test("parseAccountTags 清洗与过滤", () => {
+  assert.deepEqual(parseAccountTags('["#美食","探店#"]'), ["美食", "探店"]);
+  assert.deepEqual(parseAccountTags('["美妆","美妆","穿搭"]'), [
+    "美妆",
+    "穿搭",
+  ]);
+  assert.deepEqual(
+    parseAccountTags('["这个标签实在太长了超过十二个字","护肤"]'),
+    ["护肤"],
+  );
+  assert.deepEqual(
+    parseAccountTags('[1, null, "母婴", true]'),
+    ["母婴"],
+  );
+});
+
+test("parseAccountTags 超过 5 个截断", () => {
+  const text = '["一","二","三","四","五","六"]';
+  assert.deepEqual(parseAccountTags(text), ["一", "二", "三", "四", "五"]);
+});
+
+test("parseAccountTags 无法解析时返回空数组", () => {
+  assert.deepEqual(parseAccountTags("无法生成标签"), []);
+  assert.deepEqual(parseAccountTags(""), []);
+  assert.deepEqual(parseAccountTags('{"tags":"美妆"}'), []);
+});
+
+test("buildAccountTagText 包含关键字段", () => {
+  const text = buildAccountTagText(input);
+  assert.ok(text.includes("小美的日常"));
+  assert.ok(text.includes("分享护肤和生活"));
+  assert.ok(text.includes("粉丝数：1200"));
+  assert.ok(text.includes("美妆"));
+  assert.ok(text.includes("平价护肤分享"));
+});
+
+test("buildAccountTagText 粉丝数为 0 或空时省略", () => {
+  assert.ok(!buildAccountTagText({ ...input, followers: 0 }).includes("粉丝数"));
+  assert.ok(
+    !buildAccountTagText({ ...input, followers: null }).includes("粉丝数"),
+  );
+});
+
+test("generateAccountTagsFromMcp 工具名为空时不发起请求", async () => {
+  const fetchImpl = async () => {
+    throw new Error("should not be called");
+  };
+  const tags = await generateAccountTagsFromMcp(
+    input,
+    config,
+    {},
+    fetchImpl,
+  );
+  assert.deepEqual(tags, []);
+});
+
+test("generateAccountTagsFromMcp 正常返回标签数组", async () => {
+  const { fetchImpl, calls } = createFakeTaggingMcp(["美妆", "探店", "好物分享"]);
+  const tags = await generateAccountTagsFromMcp(
+    input,
+    config,
+    taggingEnv,
+    fetchImpl,
+  );
+  assert.deepEqual(tags, ["美妆", "探店", "好物分享"]);
+  const toolCall = calls.find((call) => call.body.method === "tools/call");
+  assert.equal(toolCall.body.params.name, "generate_tags");
+  const args = toolCall.body.params.arguments;
+  assert.equal(args.objective, "为KOC账号生成运营标签");
+  assert.ok(args.text.includes("小美的日常"));
+  assert.ok(Array.isArray(args.rules) && args.rules.length > 0);
+  assert.equal(args.output_schema.properties.tags.type, "array");
+  assert.ok(String(toolCall.url).includes("key=test-key"));
+});
+
+test("generateAccountTagsFromMcp 支持对象包裹的返回", async () => {
+  const { fetchImpl } = createFakeTaggingMcp({ tags: ["母婴", "育儿"] });
+  const tags = await generateAccountTagsFromMcp(
+    input,
+    config,
+    taggingEnv,
+    fetchImpl,
+  );
+  assert.deepEqual(tags, ["母婴", "育儿"]);
+});
+
+test("generateAccountTagsFromMcp 工具报错时返回空数组", async () => {
+  const { fetchImpl } = createFakeTaggingMcp({ message: "boom" }, true);
+  const tags = await generateAccountTagsFromMcp(
+    input,
+    config,
+    taggingEnv,
+    fetchImpl,
+  );
+  assert.deepEqual(tags, []);
+});
+
+test("generateAccountTagsFromMcp 网络异常时返回空数组", async () => {
+  const fetchImpl = async () => {
+    throw new Error("network down");
+  };
+  const tags = await generateAccountTagsFromMcp(
+    input,
+    config,
+    taggingEnv,
+    fetchImpl,
+  );
+  assert.deepEqual(tags, []);
+});
+
+const completeRow = {
+  nickname: "小美",
+  publicAccountId: "xhs-123",
+  ipLocation: "上海",
+  followersResolved: true,
+  gender: "女",
+  bio: "分享生活",
+  tags: ["美妆"],
+};
+
+test("needsResourceEnrichment 资料齐全且有标签时为 false", () => {
+  assert.equal(needsResourceEnrichment(completeRow), false);
+});
+
+test("needsResourceEnrichment 资料齐全但无标签时为 true", () => {
+  assert.equal(needsResourceEnrichment({ ...completeRow, tags: [] }), true);
+});
+
+test("needsResourceEnrichment 缺资料且有标签时为 true", () => {
+  assert.equal(
+    needsResourceEnrichment({ ...completeRow, bio: "" }),
+    true,
+  );
+});