crawlpro-win 1.0.0

package/LICENSE

@@ -0,0 +1,19 @@
+Proprietary License
+
+Copyright (c) 2026 Xi'an Wenyao Network Information Technology Co., Ltd.
+All rights reserved.
+
+CrawlPro and its accompanying software, binaries, documentation, and other
+materials are proprietary and confidential. Possession of a copy does not
+grant ownership or any intellectual-property rights.
+
+Use of CrawlPro is permitted only under a valid license or written agreement
+issued by Xi'an Wenyao Network Information Technology Co., Ltd. Except where
+that license or agreement expressly permits it, copying, modification,
+distribution, sublicensing, sale, reverse engineering, decompilation, or
+creation of derivative works is prohibited.
+
+Third-party components remain subject to their respective license terms.
+
+For licensing enquiries, contact Xi'an Wenyao Network Information Technology
+Co., Ltd. through https://www.wenyaotech.com/.

package/README.md

@@ -0,0 +1,129 @@
+# CrawlPro
+
+English | Simplified Chinese is included as `README_CN.md` in the npm package.
+
+CrawlPro is a proprietary Agent Skill for finding and organizing publicly
+visible Douyin customer signals. It supports keyword-based collection, customer
+workbooks, assisted follow-up, user-authorized replies, and Excel/HTML analysis.
+
+Developed by **Xi'an Wenyao Network Information Technology Co., Ltd.**
+
+[Company Website](https://www.wenyaotech.com/)
+
+## Main Capabilities
+
+- Collect public videos, comments, and profiles by keyword.
+- Filter comments into customer-intent levels and export customer workbooks.
+- Draft replies or send them only after explicit user authorization.
+- Support human-confirmed assisted replies and follow-up status updates.
+- Generate cleaned Excel workbooks and HTML customer reports.
+- Keep keyword catalogs and runtime settings local to the user's computer.
+
+## Supported Platforms
+
+- macOS on Apple Silicon (`arm64`)
+- Windows 10 or 11 (`x64`)
+
+Intel Mac, Windows ARM64, and Linux are not currently supported.
+
+## Requirements
+
+- Node.js 18 or later and npm 9 or later for installation
+- Hermes, OpenClaw, Claude Code, or Codex
+- Microsoft Edge or Google Chrome; Safari is not supported
+- A local MongoDB or Redis instance for keyword catalogs
+- A valid CrawlPro License Key
+
+Python is not required when using the packaged executable.
+
+## Install
+
+```bash
+npx --yes crawlpro
+```
+
+The unified installer detects the operating system and architecture, installs
+only the matching platform package, detects installed agents, and asks where to
+install CrawlPro.
+
+Install into one, several, or all detected agents:
+
+```bash
+npx --yes crawlpro --agent hermes
+npx --yes crawlpro --agent claude,codex
+npx --yes crawlpro --all --yes
+```
+
+Inspect detection or preview changes without writing files:
+
+```bash
+npx --yes crawlpro --list
+npx --yes crawlpro --all --dry-run
+```
+
+The platform-specific commands remain available:
+
+```bash
+npx --yes crawlpro-mac
+npx --yes crawlpro-win
+```
+
+Updates replace only packaged release files. Existing `license.json`,
+`app/settings.json`, output files, browser profiles, and local databases are
+preserved.
+
+## First Run
+
+Start a new Agent session after installation, then ask the Agent to use
+CrawlPro. Before browser workflows, CrawlPro checks whether Edge or Chrome is
+available and whether Douyin is logged in.
+
+The packaged commands are:
+
+```text
+macOS:   ./main check
+Windows: .\main.exe check
+```
+
+If the browser cannot be started, use `start_chrome_debug.sh` on macOS or
+`start_chrome_debug.bat` on Windows, complete login, and run the check again.
+
+## License Activation
+
+CrawlPro requires a valid License Key. To purchase or request one, contact the
+CrawlPro team through the [company website](https://www.wenyaotech.com/).
+
+The License Key is not included in the npm package. Enter it only in the local
+CrawlPro activation prompt. The resulting device-bound `license.json` must not
+be shared or committed to source control.
+
+## Responsible Use and Security
+
+- Process only publicly visible information and follow applicable laws and
+  platform terms.
+- Sending replies requires explicit user authorization and user-defined limits.
+- Never place License Keys, API keys, customer data, or local configuration in
+  Agent conversations, logs, or source control.
+- Do not modify or bypass CrawlPro license validation or platform safeguards.
+- The npm packages contain no user credentials or runtime customer data.
+
+## Maintainer Commands
+
+```bash
+npm test
+npm run pack:check
+
+npm publish --workspace packages/macos --registry=https://registry.npmjs.org/
+npm publish --workspace packages/windows --registry=https://registry.npmjs.org/
+npm publish --workspace packages/cli --registry=https://registry.npmjs.org/
+```
+
+Publish both platform packages before publishing the unified `crawlpro` entry
+package.
+
+## License
+
+CrawlPro is distributed under a Proprietary License. The complete license is
+included in the npm package as the `LICENSE` file.
+
+Copyright © 2026 Xi'an Wenyao Network Information Technology Co., Ltd.

package/README_CN.md

@@ -0,0 +1,123 @@
+# CrawlPro
+
+[English](https://www.npmjs.com/package/crawlpro) | 中文
+
+CrawlPro 是一套专有 Agent Skill，用于从抖音公开可见内容中查找和整理客户
+信号，支持关键词采集、客户表格、辅助跟进、用户授权回复以及 Excel/HTML
+客户分析。
+
+本项目由 **西安文鳐网络信息科技有限责任公司** 开发。
+
+[公司官网](https://www.wenyaotech.com/)
+
+## 核心能力
+
+- 按关键词采集公开可见的视频、评论和用户资料。
+- 对评论进行客户意向分级，并导出客户工作簿。
+- 草拟回复；只有获得用户明确授权后才执行发送。
+- 支持人工确认的辅助回复和客户跟进状态更新。
+- 生成清洗后的 Excel 工作簿和 HTML 客户报告。
+- 关键词库和运行配置保存在用户本机。
+
+## 支持平台
+
+- Apple Silicon（`arm64`）macOS
+- Windows 10 或 Windows 11（`x64`）
+
+暂不支持 Intel Mac、Windows ARM64 和 Linux。
+
+## 使用要求
+
+- 安装时需要 Node.js 18 或更高版本、npm 9 或更高版本
+- Hermes、OpenClaw、Claude Code 或 Codex
+- Microsoft Edge 或 Google Chrome；不支持 Safari
+- 本地 MongoDB 或 Redis，用于保存关键词库
+- 有效的 CrawlPro License Key
+
+使用预编译程序时不需要安装 Python。
+
+## 安装
+
+```bash
+npx --yes crawlpro
+```
+
+统一安装器会判断操作系统和 CPU 架构，只安装匹配的平台包，然后检测已经
+安装的 Agent，并询问需要把 CrawlPro 安装到哪里。
+
+安装到一个、多个或全部 Agent：
+
+```bash
+npx --yes crawlpro --agent hermes
+npx --yes crawlpro --agent claude,codex
+npx --yes crawlpro --all --yes
+```
+
+查看检测结果或预演安装，不写入文件：
+
+```bash
+npx --yes crawlpro --list
+npx --yes crawlpro --all --dry-run
+```
+
+仍然可以直接使用平台包：
+
+```bash
+npx --yes crawlpro-mac
+npx --yes crawlpro-win
+```
+
+更新只会替换发布文件，已有的 `license.json`、`app/settings.json`、输出文件、
+浏览器资料和本地数据库不会被删除。
+
+## 首次使用
+
+安装后请在对应 Agent 中新建会话，并要求 Agent 使用 CrawlPro。开始浏览器
+工作流前，CrawlPro 会检查 Edge 或 Chrome 是否可用，以及抖音是否已登录。
+
+程序命令为：
+
+```text
+macOS:   ./main check
+Windows: .\main.exe check
+```
+
+浏览器无法启动时，macOS 运行 `start_chrome_debug.sh`，Windows 运行
+`start_chrome_debug.bat`；完成登录后再次执行检查。
+
+## License 激活
+
+使用 CrawlPro 需要有效的 License Key。如需购买或申请，请通过
+[公司官网](https://www.wenyaotech.com/)联系 CrawlPro 团队。
+
+License Key 不包含在 npm 安装包中。请仅在本机 CrawlPro 激活提示中输入。
+激活后生成的 `license.json` 与设备绑定，不要转发或提交到源代码仓库。
+
+## 合规和安全说明
+
+- 只处理公开可见的信息，并遵守适用法律和平台规则。
+- 发送回复必须获得用户明确授权，并使用用户设定的数量限制。
+- 不要在 Agent 对话、日志或源码中提供 License Key、API Key、客户数据或
+  本地配置。
+- 不要修改或绕过 CrawlPro License 校验及平台安全限制。
+- npm 安装包不包含用户凭证或运行时客户数据。
+
+## 维护者命令
+
+```bash
+npm test
+npm run pack:check
+
+npm publish --workspace packages/macos --registry=https://registry.npmjs.org/
+npm publish --workspace packages/windows --registry=https://registry.npmjs.org/
+npm publish --workspace packages/cli --registry=https://registry.npmjs.org/
+```
+
+必须先发布两个平台包，再发布统一入口包 `crawlpro`。
+
+## License
+
+CrawlPro 使用 Proprietary License，完整许可证条款包含在 npm 安装包的
+`LICENSE` 文件中。
+
+Copyright © 2026 西安文鳐网络信息科技有限责任公司

package/SKILL.md

@@ -0,0 +1,279 @@
+---
+name: crawlpro
+description: Operate the packaged CrawlPro application for Douyin login check, customer collection, keyword catalogs, follow-up status, and Excel/HTML customer analysis, plus drafting comment replies within user-set limits. Use when the user asks to run CrawlPro workflows such as check, collect, reply, semi-reply, follow-up, analyze, or config. Verifies browser login before browser workflows, confirms required limits before execution, processes only publicly visible data, and never delivers private settings or internal state files.
+---
+
+# CrawlPro agent workflow
+
+Use only the packaged application in the current release directory:
+
+```text
+Windows: ./main.exe
+macOS:   ./main
+```
+
+For source development, the equivalent command is `uv run python main.py`.
+
+The included browser launcher is also platform-specific:
+
+```text
+Windows: start_chrome_debug.bat
+macOS:   ./start_chrome_debug.sh
+```
+
+## Core rules
+
+1. Determine the requested workflow before running commands.
+2. Ask for missing keywords, limits and send mode.
+3. Run `APP check` before login, collection or reply workflows.
+4. Do not continue browser work unless `APP check` exits `0`.
+5. Never claim a reply was sent unless the application confirms it.
+6. Never claim a file exists without checking it.
+7. Never attach `app/settings.json`, `license.json`, `active_collection.json`, `reply_tracking.json`, browser profiles or API keys.
+8. Deliver only generated `.xlsx`, `.html`, `.log` when explicitly requested, or the application-generated ZIP.
+
+`APP` below means the platform command shown above.
+
+## Command router
+
+| Request | Command |
+|---|---|
+| Browser/login check | `APP check` |
+| Guided login | `APP login` |
+| Collect target customers | `APP collect` |
+| Draft/send replies within limits | `APP reply` |
+| Assisted reply (manual confirm) | `APP semi-reply` |
+| Follow-up status | `APP follow-up` |
+| Customer analysis | `APP analyze` |
+| Settings and keywords | `APP config` |
+| View model-generated keyword content | `APP keywords` |
+| Initialize or verify database | `APP init-db` |
+
+Interactive commands require a real terminal. If no PTY is available, ask the user to open the application and select the matching menu item.
+
+## Browser pre-flight
+
+Run before every browser workflow:
+
+```text
+APP check
+```
+
+| Exit | Meaning | Action |
+|---|---|---|
+| `0` | Browser connected and logged in | Continue |
+| `1` | Browser connected but logged out | Ask the user to scan QR, then rerun |
+| `2` | Browser unavailable | Ask the user to run the included browser launcher, then rerun |
+
+CrawlPro uses Edge first and Chrome as fallback on CDP port `19553`.
+Safari cannot be used because this workflow requires the Chromium CDP interface.
+On macOS, ask the user to install Edge or Chrome when neither is available.
+
+## Customer collection
+
+Ask for:
+
+1. Search keyword or comma-separated keywords;
+2. latest-first search: `yes` or `no` (`no` means hot/most-liked);
+3. videos per keyword;
+4. comments to retain per video (default: all; a positive number sets a cap);
+5. maximum target customers;
+6. maximum profiles;
+7. whether to use an existing keyword bundle or generate one with the model.
+
+If existing defaults are requested:
+
+- read numeric limits from `app/settings.json` when it exists;
+- obtain active keywords and generated content through `APP keywords`;
+- do not expect keywords inside `settings.json`.
+
+Run:
+
+```text
+APP collect "keywordA,keywordB" --videos 10 --leads 50 --profiles 50
+APP collect "keywordA" --comments 20  # optional per-video cap
+APP collect "keywordA" --sort hot
+```
+
+Every keyword is searched independently. Duplicate video IDs are removed.
+Only publicly visible comments and profile data are processed. Intent grading
+prioritizes explicit request actions and rejects praise-only comments. Count
+repeated values by distinct users, not raw comment rows. Any number used by at
+least three users under one video is treated as that video's spoken code; two
+users require surrounding request evidence. Normalize repeated text for
+punctuation and platform emoji tags. Repeated request phrases are high intent,
+ambiguous repeated phrases are medium-intent candidates, and repeated
+praise/reactions remain low. All signals are scoped to one video.
+
+Treat store/address, on-site service, delivery coverage, and supported-city
+questions as high intent; bare creator-location questions as medium; and
+filming-location or scenery questions as low.
+
+Expected primary output:
+
+```text
+output/Test/collections/<keyword-group>/target_customs.xlsx
+```
+
+The collection ZIP contains only final customer workbooks, cleaned analysis workbooks, HTML reports, and `.log` files. Runtime JSON, search tables, per-video comment tables, and per-user detail tables stay out of the bundle.
+
+## Reply
+
+Ask for:
+
+1. keyword set;
+2. live search or `target_customs.xlsx` source;
+3. latest-first search (`yes`/`no`), videos per keyword, and comments per video when live search is selected;
+4. whether to draft replies only or send them (sending requires explicit user authorization and confirmation);
+5. model use: `yes` or `no`;
+6. daily reply limit;
+7. allowed comment age;
+8. whether to collect profiles.
+
+Commands:
+
+```text
+APP reply "keywordA"
+APP reply "keywordA" --no-reply
+APP reply "keywordA" --profile
+APP reply "keywordA" --from-leads
+APP reply "keywordA" --sort hot
+```
+
+Before sending, state the selected keyword, source and reply limit, and confirm the user has authorized sending. Sending happens only within the user-set daily limit and must not bypass platform interaction limits. Live search is the default and has no Excel input. `--from-leads` reads only `target_customs.xlsx`. `activate_customs.xlsx` is an optional profile/reply-status output, and `all_comment_users.xlsx` is never a sending source. Stop if selectors are missing or sending cannot be confirmed.
+
+## Assisted reply
+
+Run:
+
+```text
+APP semi-reply
+```
+
+This workflow requires a person at the computer. It reads only the active keyword group's `target_customs.xlsx`, filters customers, opens the related comment and either sends after confirmation or fills the text for manual sending. It does not read `activate_customs.xlsx` or `all_comment_users.xlsx`.
+
+## Follow-up
+
+Run:
+
+```text
+APP follow-up
+```
+
+The user marks rows as 未跟进 (not followed up), 已私信 (DM sent), 已加微信 (WeChat added) or 已成交 (deal closed). The workflow writes back to Excel. If Excel/WPS locks the file, ask the user to close it and retry.
+
+## Analysis
+
+Ask for:
+
+1. `target_customs.xlsx`, `activate_customs.xlsx`, `all_comment_users.xlsx`, or an explicit Excel path;
+2. brief or full terminal statistics;
+3. whether the generated Excel/HTML should be delivered.
+
+Commands:
+
+```text
+APP analyze
+APP analyze --source target
+APP analyze --source profiles
+APP analyze --source comments
+APP analyze --brief
+APP analyze --file "path/to/customer_table.xlsx" --full
+APP evaluate-intent "path/to/labeled_comments.xlsx"
+```
+
+Without a source flag, an interactive terminal offers all three customer tables. `--daily` remains a compatibility alias for `--source profiles`; `--csv` remains a compatibility alias for `--file`, but the input is an Excel workbook.
+
+Each schema is normalized from its real fields. The formal `activate_customs.xlsx` includes intent and reply status; compatible older internal profile data may omit intent, which stays unlabelled. Comment-user workbooks map highest intent, aggregated comments, latest comment time and source videos.
+
+Generated files are placed beside the source workbook under `analysis/`:
+
+```text
+analysis/<source>_clean.xlsx
+analysis/<source>_report.html
+```
+
+The cleaned workbook and HTML contain only customer delivery fields: user identity, contact information, intent, source video, intent comment, comment time and IP region. The HTML is a simple customer table (user, Douyin ID, user-homepage link, contact method, contact detail, intent, video link, video title, intent comment, comment time, IP region), sorted by intent level (high → mid) then newest comment first, with a region filter dropdown — not a dashboard.
+
+The analysis ZIP contains the cleaned workbook, the HTML report, and the same main tables/logs found in the keyword directory.
+
+## Intent evaluation
+
+Use `APP evaluate-intent <labeled_comments.xlsx>` only when the user supplies a
+manually labeled workbook with `评论内容` and `预期意向` columns. The command
+uses the production classifier, including video-scoped repeated codes when
+`视频ID` is present. It makes no model call, spends no tokens, and writes no
+copy of the labeled data.
+
+## Settings and keyword catalog
+
+`app/settings.json` contains only English application settings:
+
+- `llm`;
+- `database`;
+- `parameters`.
+
+Keywords, intent words, reply-intent words and reply templates are stored with English field names in the password-free MongoDB/Redis `keywords` collection. `APP init-db` also migrates compatible legacy Chinese field names without regenerating content.
+
+Commands:
+
+```text
+APP init-db
+APP config
+APP config "keywordA,keywordB"
+APP config --reset
+APP keywords
+```
+
+Ask for confirmation before `APP config --reset` because it overwrites application settings. API keys must be entered privately in the application, never in chat.
+
+## Output and delivery
+
+Main customer outputs:
+
+```text
+target_customs.xlsx
+activate_customs.xlsx
+all_comment_users.xlsx
+analysis/<source>_clean.xlsx
+analysis/<source>_report.html
+deliverables/*.zip
+```
+
+Internal files such as `active_collection.json` and `reply_tracking.json` are necessary runtime state, not delivery files.
+
+After success, report:
+
+1. command and keyword set;
+2. approved limits;
+3. processed/success/failure counts;
+4. whether real replies were sent;
+5. exact existing output paths;
+6. remaining user action.
+
+Use the printed `DELIVERABLE_PATH=...` value for ZIP delivery. Do not search for or construct a guessed ZIP path.
+
+## Recovery
+
+| Problem | Action |
+|---|---|
+| Browser exit `1` | Ask for QR login and rerun check |
+| Browser exit `2` | Run platform browser launcher and retry |
+| Search returns no data | Verify login and keywords |
+| Page disconnect | Keep browser open, rerun check, retry once |
+| Model failure | Ask whether to retry or continue without generated content |
+| Reply selector missing | Stop sending and report probable layout change |
+| `send_unconfirmed` | Do not count as sent |
+| Excel locked | Ask user to close Excel/WPS and retry |
+| Analysis source missing/empty | Stop and report the exact expected path |
+| Database unavailable | Verify the local service and URL, run `APP init-db`, and report its exact error |
+| macOS exit `137` / `Killed: 9` | Stop retrying; compare `uname -m` with `file ./main`, then report the output of `codesign --verify --deep --strict --verbose=4 ./main` |
+
+## Safety
+
+- Follow Douyin terms of service and applicable privacy and data-protection laws.
+- Process only publicly visible comment and profile data, and only for user-authorized private-domain communication.
+- Keep collection and reply volume conservative; never bypass platform interaction limits or anti-automation measures.
+- Do not use customer data for harassment, fraud, spam, or unlawful resale.
+- Password-free database URLs are restricted to loopback hosts or a local Redis Unix socket; do not attempt to bypass this validation.
+- Never expose private settings, API keys, licensing data, or browser profiles.

package/bin/crawlpro-win.cjs

@@ -0,0 +1,376 @@
+#!/usr/bin/env node
+
+'use strict';
+
+const childProcess = require('node:child_process');
+const fs = require('node:fs');
+const os = require('node:os');
+const path = require('node:path');
+const readline = require('node:readline');
+
+const SKILL_NAME = 'crawlpro';
+const PACKAGE_ROOT = path.resolve(__dirname, '..');
+const PLATFORM = process.env.CRAWLPRO_INSTALL_TEST_PLATFORM || process.platform;
+const ARCH = process.env.CRAWLPRO_INSTALL_TEST_ARCH || process.arch;
+const INSTALL_HOME = process.env.CRAWLPRO_INSTALL_HOME || os.homedir();
+const LOCAL_APP_DATA = process.env.CRAWLPRO_INSTALL_LOCALAPPDATA ||
+  process.env.LOCALAPPDATA || path.join(INSTALL_HOME, 'AppData', 'Local');
+
+const SOURCE_FILES = [
+  ['SKILL.md', 'SKILL.md'],
+  ['README.md', 'README.md'],
+  ['README_CN.md', 'README_CN.md'],
+  ['main.exe', 'main.exe'],
+  ['start_chrome_debug.bat', 'start_chrome_debug.bat'],
+];
+
+const CANONICAL_FILES = [
+  'SKILL.md',
+  'README.md',
+  'README_CN.md',
+  'main.exe',
+  'start_chrome_debug.bat',
+];
+
+function hermesHome() {
+  return process.env.HERMES_HOME || path.join(INSTALL_HOME, '.hermes');
+}
+
+const AGENTS = {
+  hermes: {
+    label: 'Hermes Agent',
+    command: 'hermes',
+    detectPaths: () => [
+      hermesHome(),
+      path.join(LOCAL_APP_DATA, 'Programs', 'Hermes'),
+    ],
+    destination: () => path.join(hermesHome(), 'skills', SKILL_NAME),
+  },
+  openclaw: {
+    label: 'OpenClaw',
+    command: 'openclaw',
+    detectPaths: () => [path.join(INSTALL_HOME, '.openclaw')],
+    destination: () => path.join(INSTALL_HOME, '.openclaw', 'skills', SKILL_NAME),
+  },
+  claude: {
+    label: 'Claude Code',
+    command: 'claude',
+    detectPaths: () => [path.join(INSTALL_HOME, '.claude')],
+    destination: () => path.join(INSTALL_HOME, '.claude', 'skills', SKILL_NAME),
+  },
+  codex: {
+    label: 'Codex',
+    command: 'codex',
+    detectPaths: () => [
+      path.join(INSTALL_HOME, '.codex'),
+      path.join(LOCAL_APP_DATA, 'Programs', 'Codex'),
+    ],
+    destination: () => path.join(INSTALL_HOME, '.agents', 'skills', SKILL_NAME),
+  },
+};
+
+const AGENT_ORDER = ['hermes', 'openclaw', 'claude', 'codex'];
+
+function usage() {
+  console.log(`CrawlPro Windows Skill 安装器
+
+用法：
+  npx crawlpro
+  npx crawlpro --agent hermes,codex
+  npx crawlpro --all --yes
+
+平台包直接调用：
+  npx crawlpro-win
+
+选项：
+  --agent <名称>  安装到指定 Agent；可用逗号分隔多个名称
+                  支持 hermes、openclaw、claude、codex
+  --all           安装到检测到的全部 Agent
+  --yes, -y       更新已有安装时不再询问
+  --dry-run       只显示操作，不写入文件
+  --list          只显示检测结果
+  --help, -h      显示帮助`);
+}
+
+function parseArgs(argv) {
+  const options = {
+    requestedAgents: '',
+    all: false,
+    yes: false,
+    dryRun: false,
+    list: false,
+    help: false,
+  };
+
+  for (let index = 0; index < argv.length; index += 1) {
+    const argument = argv[index];
+    if (argument === '--agent') {
+      const value = argv[index + 1];
+      if (!value || value.startsWith('--')) {
+        throw new Error('--agent 后需要 Agent 名称。');
+      }
+      options.requestedAgents = value;
+      index += 1;
+    } else if (argument.startsWith('--agent=')) {
+      options.requestedAgents = argument.slice('--agent='.length);
+    } else if (argument === '--all') {
+      options.all = true;
+    } else if (argument === '--yes' || argument === '-y') {
+      options.yes = true;
+    } else if (argument === '--dry-run') {
+      options.dryRun = true;
+    } else if (argument === '--list') {
+      options.list = true;
+    } else if (argument === '--help' || argument === '-h') {
+      options.help = true;
+    } else {
+      throw new Error(`未知参数：${argument}`);
+    }
+  }
+
+  if (options.requestedAgents && options.all) {
+    throw new Error('--agent 和 --all 不能同时使用。');
+  }
+  return options;
+}
+
+function commandExists(command) {
+  try {
+    const result = childProcess.spawnSync('where.exe', [command], {
+      stdio: 'ignore',
+      windowsHide: true,
+    });
+    return result.status === 0;
+  } catch {
+    return false;
+  }
+}
+
+function agentDetected(key) {
+  const agent = AGENTS[key];
+  return commandExists(agent.command) || agent.detectPaths().some((entry) => fs.existsSync(entry));
+}
+
+function printDetection() {
+  console.log('CrawlPro 支持的 Agent：');
+  for (const key of AGENT_ORDER) {
+    const status = agentDetected(key) ? '已检测' : '未检测';
+    const suffix = agentDetected(key) ? `  ${AGENTS[key].destination()}` : '';
+    console.log(`  [${status}] ${AGENTS[key].label}${suffix}`);
+  }
+}
+
+function normalizeAgent(value) {
+  const normalized = value.trim().toLowerCase();
+  const aliases = {
+    hermes: 'hermes',
+    'hermes-agent': 'hermes',
+    openclaw: 'openclaw',
+    'open-claw': 'openclaw',
+    claude: 'claude',
+    'claude-code': 'claude',
+    claudecode: 'claude',
+    codex: 'codex',
+    'openai-codex': 'codex',
+  };
+  return aliases[normalized] || null;
+}
+
+function parseRequestedAgents(raw) {
+  const selected = [];
+  for (const item of raw.replaceAll('，', ',').split(',')) {
+    if (!item.trim()) continue;
+    const key = normalizeAgent(item);
+    if (!key) throw new Error(`不支持的 Agent：${item.trim()}`);
+    if (!selected.includes(key)) selected.push(key);
+  }
+  return selected;
+}
+
+function ask(question) {
+  const terminal = readline.createInterface({ input: process.stdin, output: process.stdout });
+  return new Promise((resolve) => {
+    terminal.question(question, (answer) => {
+      terminal.close();
+      resolve(answer.trim());
+    });
+  });
+}
+
+async function selectInteractively(detected) {
+  if (!process.stdin.isTTY || !process.stdout.isTTY) {
+    throw new Error('当前不是交互式终端，请使用 --agent 或 --all。');
+  }
+  console.log('\n请选择安装目标，可输入一个或多个编号（例如 1,3）：');
+  detected.forEach((key, index) => console.log(`  ${index + 1}) ${AGENTS[key].label}`));
+  console.log('  a) 全部');
+  console.log('  q) 退出');
+  const answer = (await ask('选择：')).replaceAll('，', ',').replaceAll(/\s/g, '');
+  if (/^q$/i.test(answer)) return [];
+  if (/^a$/i.test(answer)) return [...detected];
+
+  const selected = [];
+  for (const value of answer.split(',')) {
+    if (!/^\d+$/.test(value)) throw new Error(`无效选项：${value}`);
+    const index = Number(value) - 1;
+    if (index < 0 || index >= detected.length) throw new Error(`无效选项：${value}`);
+    if (!selected.includes(detected[index])) selected.push(detected[index]);
+  }
+  return selected;
+}
+
+function verifyPackage() {
+  for (const [sourceRelative] of SOURCE_FILES) {
+    const source = path.join(PACKAGE_ROOT, sourceRelative);
+    if (!fs.statSync(source, { throwIfNoEntry: false })?.isFile()) {
+      throw new Error(`npm 包不完整，缺少 ${sourceRelative}。`);
+    }
+  }
+}
+
+function installationExists(destination) {
+  return CANONICAL_FILES.some((relative) => fs.existsSync(path.join(destination, relative)));
+}
+
+function timestamp() {
+  const now = new Date();
+  const pad = (value) => String(value).padStart(2, '0');
+  return `${now.getFullYear()}${pad(now.getMonth() + 1)}${pad(now.getDate())}-` +
+    `${pad(now.getHours())}${pad(now.getMinutes())}${pad(now.getSeconds())}-${process.pid}`;
+}
+
+function backupExistingFiles(key, destination) {
+  const backupRoot = path.join(
+    LOCAL_APP_DATA,
+    'CrawlProInstaller',
+    'backups',
+    key,
+    timestamp(),
+  );
+  let copied = 0;
+  for (const relative of CANONICAL_FILES) {
+    const source = path.join(destination, relative);
+    if (!fs.existsSync(source)) continue;
+    const target = path.join(backupRoot, relative);
+    fs.mkdirSync(path.dirname(target), { recursive: true });
+    fs.copyFileSync(source, target);
+    copied += 1;
+  }
+  if (copied > 0) console.log(`  旧版本已备份：${backupRoot}`);
+  return backupRoot;
+}
+
+function atomicCopy(source, destination) {
+  fs.mkdirSync(path.dirname(destination), { recursive: true });
+  const temporary = `${destination}.crawlpro-new-${process.pid}`;
+  const previous = `${destination}.crawlpro-old-${process.pid}`;
+  fs.copyFileSync(source, temporary);
+
+  let movedPrevious = false;
+  try {
+    if (fs.existsSync(destination)) {
+      fs.renameSync(destination, previous);
+      movedPrevious = true;
+    }
+    fs.renameSync(temporary, destination);
+    if (movedPrevious) fs.rmSync(previous, { force: true });
+  } catch (error) {
+    fs.rmSync(temporary, { force: true });
+    if (movedPrevious && !fs.existsSync(destination) && fs.existsSync(previous)) {
+      fs.renameSync(previous, destination);
+    }
+    if (error && ['EBUSY', 'EPERM', 'EACCES'].includes(error.code)) {
+      throw new Error(`无法更新 ${destination}；请关闭 CrawlPro 和对应 Agent 后重试。`);
+    }
+    throw error;
+  }
+}
+
+async function installAgent(key, options) {
+  const agent = AGENTS[key];
+  const destination = agent.destination();
+  console.log(`\n安装到 ${agent.label}：${destination}`);
+
+  if (installationExists(destination) && !options.yes && !options.dryRun) {
+    if (!process.stdin.isTTY || !process.stdout.isTTY) {
+      throw new Error('目标已存在；非交互模式更新请加 --yes。');
+    }
+    const answer = await ask('  已存在 CrawlPro，备份并更新？[y/N] ');
+    if (!/^y$/i.test(answer)) {
+      console.log('  已跳过。');
+      return;
+    }
+  }
+
+  if (options.dryRun) {
+    console.log('  [dry-run] 将保留现有授权、配置和输出数据，并更新发布文件。');
+    return;
+  }
+
+  if (installationExists(destination)) backupExistingFiles(key, destination);
+  for (const [sourceRelative, destinationRelative] of SOURCE_FILES) {
+    atomicCopy(
+      path.join(PACKAGE_ROOT, sourceRelative),
+      path.join(destination, destinationRelative),
+    );
+  }
+
+  for (const required of ['main.exe', 'start_chrome_debug.bat']) {
+    if (!fs.statSync(path.join(destination, required), { throwIfNoEntry: false })?.isFile()) {
+      throw new Error(`发布文件校验失败：${required}`);
+    }
+  }
+  console.log('  安装完成。现有配置和生成数据未被删除。');
+}
+
+async function main() {
+  const options = parseArgs(process.argv.slice(2));
+  if (options.help) {
+    usage();
+    return;
+  }
+  verifyPackage();
+  if (PLATFORM !== 'win32' || ARCH !== 'x64') {
+    throw new Error(`crawlpro-win 仅支持 Windows x64；当前系统为 ${PLATFORM} ${ARCH}。`);
+  }
+
+  const detected = AGENT_ORDER.filter(agentDetected);
+  if (options.list) {
+    printDetection();
+    return;
+  }
+
+  let selected;
+  if (options.requestedAgents) {
+    selected = parseRequestedAgents(options.requestedAgents);
+    for (const key of selected) {
+      if (!agentDetected(key)) {
+        console.error(`警告：未检测到 ${AGENTS[key].label}，仍按显式参数继续安装。`);
+      }
+    }
+  } else if (options.all) {
+    selected = [...detected];
+  } else {
+    printDetection();
+    if (detected.length === 0) {
+      throw new Error('没有检测到受支持的 Agent。安装 Agent 后重试，或使用 --agent 显式指定目标。');
+    }
+    selected = await selectInteractively(detected);
+  }
+
+  if (selected.length === 0) {
+    if (!options.requestedAgents && !options.all) console.log('已取消。');
+    else throw new Error('没有可安装的目标。');
+    return;
+  }
+
+  for (const key of selected) await installAgent(key, options);
+  console.log(options.dryRun
+    ? '\ndry-run 完成，没有写入文件。'
+    : '\nCrawlPro 安装完成。请在对应 Agent 中新建会话，让 Skill 清单重新加载。');
+}
+
+main().catch((error) => {
+  console.error(`错误：${error.message}`);
+  process.exitCode = 1;
+});

package/package.json

@@ -0,0 +1,49 @@
+{
+  "name": "crawlpro-win",
+  "version": "1.0.0",
+  "description": "Install the prebuilt CrawlPro Skill for Windows x64 into Hermes, OpenClaw, Claude Code, or Codex.",
+  "license": "SEE LICENSE IN LICENSE",
+  "author": "Xi'an Wenyao Network Information Technology Co., Ltd.",
+  "homepage": "https://www.wenyaotech.com/",
+  "bin": {
+    "crawlpro-win": "bin/crawlpro-win.cjs"
+  },
+  "files": [
+    "bin/crawlpro-win.cjs",
+    "main.exe",
+    "start_chrome_debug.bat",
+    "SKILL.md",
+    "README.md",
+    "README_CN.md",
+    "LICENSE"
+  ],
+  "os": [
+    "win32"
+  ],
+  "cpu": [
+    "x64"
+  ],
+  "engines": {
+    "node": ">=18",
+    "npm": ">=9"
+  },
+  "scripts": {
+    "prepack": "node ../../tooling/sync-shared.cjs windows",
+    "prepublishOnly": "npm test",
+    "test": "node ../../tooling/sync-shared.cjs windows && node --check bin/crawlpro-win.cjs && node test/npm-installer-smoke.cjs"
+  },
+  "publishConfig": {
+    "access": "public",
+    "registry": "https://registry.npmjs.org/"
+  },
+  "keywords": [
+    "agent-skill",
+    "claude-code",
+    "codex",
+    "crawlpro",
+    "douyin",
+    "hermes-agent",
+    "openclaw",
+    "windows"
+  ]
+}

package/start_chrome_debug.bat

@@ -0,0 +1,63 @@
+@echo off
+chcp 65001 >nul
+echo ============================================
+echo   Start Chromium debug mode on port 19553
+echo ============================================
+echo.
+
+rem Prefer Microsoft Edge, then fall back to Google Chrome.
+set "BROWSER="
+set "BROWSER_NAME="
+set "PROFILE_NAME="
+if exist "C:\Program Files (x86)\Microsoft\Edge\Application\msedge.exe" (
+    set "BROWSER=C:\Program Files (x86)\Microsoft\Edge\Application\msedge.exe"
+    set "BROWSER_NAME=Microsoft Edge"
+    set "PROFILE_NAME=.douyin_edge_profile"
+)
+if not defined BROWSER if exist "C:\Program Files\Microsoft\Edge\Application\msedge.exe" (
+    set "BROWSER=C:\Program Files\Microsoft\Edge\Application\msedge.exe"
+    set "BROWSER_NAME=Microsoft Edge"
+    set "PROFILE_NAME=.douyin_edge_profile"
+)
+if not defined BROWSER if exist "%LOCALAPPDATA%\Microsoft\Edge\Application\msedge.exe" (
+    set "BROWSER=%LOCALAPPDATA%\Microsoft\Edge\Application\msedge.exe"
+    set "BROWSER_NAME=Microsoft Edge"
+    set "PROFILE_NAME=.douyin_edge_profile"
+)
+if not defined BROWSER if exist "C:\Program Files\Google\Chrome\Application\chrome.exe" (
+    set "BROWSER=C:\Program Files\Google\Chrome\Application\chrome.exe"
+    set "BROWSER_NAME=Google Chrome"
+    set "PROFILE_NAME=.douyin_chrome_profile"
+)
+if not defined BROWSER if exist "C:\Program Files (x86)\Google\Chrome\Application\chrome.exe" (
+    set "BROWSER=C:\Program Files (x86)\Google\Chrome\Application\chrome.exe"
+    set "BROWSER_NAME=Google Chrome"
+    set "PROFILE_NAME=.douyin_chrome_profile"
+)
+if not defined BROWSER if exist "%LOCALAPPDATA%\Google\Chrome\Application\chrome.exe" (
+    set "BROWSER=%LOCALAPPDATA%\Google\Chrome\Application\chrome.exe"
+    set "BROWSER_NAME=Google Chrome"
+    set "PROFILE_NAME=.douyin_chrome_profile"
+)
+
+if not defined BROWSER (
+    echo Microsoft Edge or Google Chrome was not found.
+    echo Install a Chromium browser that supports --remote-debugging-port=19553.
+    pause
+    exit /b 1
+)
+
+echo Browser: %BROWSER_NAME%
+echo Executable: %BROWSER%
+
+rem Keep separate persistent profiles for Edge and Chrome.
+set "DEBUG_PROFILE=%~dp0%PROFILE_NAME%"
+if not exist "%DEBUG_PROFILE%" mkdir "%DEBUG_PROFILE%"
+
+start "" "%BROWSER%" --remote-debugging-port=19553 --user-data-dir="%DEBUG_PROFILE%" --new-window "https://www.douyin.com/"
+
+echo.
+echo %BROWSER_NAME% started. Scan the Douyin QR code on the first run.
+echo Run main.exe check again after login.
+echo.
+pause