diff --git a/skills/post-to-xhs/SKILL.md b/skills/post-to-xhs/SKILL.md
new file mode 100644
index 0000000..f9f4d97
--- /dev/null
+++ b/skills/post-to-xhs/SKILL.md
@@ -0,0 +1,251 @@
+---
+name: post-to-xhs
+description: >
+  小红书内容发布技能。支持两种发布模式：(1) 上传图文模式 - 图片+短文；(2) 写长文模式 - 长篇文章+排版模板。
+  支持两种输入方式：用户提供完整内容和图片/图片URL，直接发布；或提供网页URL，自动提取内容和图片。
+  用户说"发长文"时使用长文模式，否则默认图文模式。
+---
+
+# 小红书内容发布
+
+根据用户输入自动判断发布方式和发布模式，简化发布流程。
+
+## 发布模式
+
+- **上传图文**（默认）：图片 + 短文，适合日常分享
+- **写长文**：长篇文章 + 排版模板选择，适合深度内容。用户明确说"发长文"时使用
+
+## 工作流程
+
+```
+用户输入
+    │
+    ├─ 完整内容 + 图片/图片URL → 判断模式 → 发布流程
+    │
+    └─ 网页 URL → WebFetch 提取内容和图片
+                      │
+                      ├─ 有图片 → 适当总结内容 → 判断模式 → 发布流程
+                      │
+                      └─ 无图片 → 提示用户手动下载图片
+                                  │
+                                  └─ 用户提供图片后 → 发布流程
+```
+
+## Step 1: 判断输入类型
+
+根据用户输入判断：
+
+- **完整内容模式**：用户提供了标题、正文内容、以及图片（本地路径或URL）
+- **URL 提取模式**：用户只提供了一个网页 URL
+
+如果不确定，询问用户。
+
+## Step 2: 处理内容
+
+### 完整内容模式
+
+直接使用用户提供的标题和正文，跳到 Step 3。
+
+### URL 提取模式
+
+1. 使用 WebFetch 提取网页内容
+2. 提取关键信息：标题、正文、图片URL
+3. 适当总结内容，保持：
+   - 关键信息完整
+   - 语言自然流畅
+   - 适合小红书阅读习惯
+
+#### 图片提取失败处理
+
+如果从网页中提取不到图片URL，或图片URL无法访问，**必须**：
+
+1. 告知用户图片提取失败
+2. 提供原网页链接，请用户手动访问
+3. 指导用户：
+   - 在浏览器中打开原网页
+   - 右键点击想要的图片 → "图片另存为" 或 "复制图片地址"
+   - 将保存的图片路径或复制的图片URL提供给我
+4. 等待用户提供图片后再继续发布流程
+
+**示例提示语**：
+```
+从网页中未能提取到可用的图片。请手动获取：
+
+1. 打开原文链接：[URL]
+2. 找到合适的配图，右键另存为本地，或复制图片地址
+3. 将图片路径或URL发给我
+
+拿到图片后我们继续发布。
+```
+
+## Step 3: 内容检查
+
+### 标题检查
+
+标题长度必须 ≤ 38，计算规则：
+- 中文字符和中文标点（《》、，。等）：每个计 2
+- 英文字母/数字/空格/ASCII标点：每个计 1
+
+如果超长，自动生成符合长度要求的新标题，保持语义一致。
+
+### 正文格式
+
+- 段落之间使用双换行分隔
+- 语言自然，避免机器翻译感
+- 简体中文
+
+## Step 4: 发布到小红书
+
+完整发布流程参考: [references/publish-workflow.md](references/publish-workflow.md)
+
+### 4.1 用户确认内容
+
+通过 `AskUserQuestion` 向用户展示即将发布的内容（标题、正文、图片），获得明确确认后再继续。
+
+### 4.2 选择发布模式
+
+通过 `AskUserQuestion` 让用户选择发布模式：
+
+- **无头模式**（推荐）：后台运行，速度快，无浏览器窗口。发布完成后直接报告结果。
+- **有窗口模式**：显示浏览器窗口，可以预览内容。需要用户确认后再点击发布。
+
+```
+AskUserQuestion 示例：
+问题：选择发布模式
+选项：
+  - 无头模式（推荐）：后台快速发布，无需预览
+  - 有窗口模式：显示浏览器，可预览确认
+```
+
+### 4.3 写入临时文件
+
+将标题和正文写入临时 UTF-8 文本文件。不要在 `python -c` 中内联中文文本。
+
+### 4.4 运行发布（根据模式分流）
+
+#### A. 上传图文模式（默认）
+
+根据用户选择的模式执行发布脚本：
+
+**无头模式**（添加 `--headless` 参数）：
+```bash
+python "C:\Users\admin\AI\.claude\skills\post-to-xhs\scripts\publish_pipeline.py" --headless --title-file title.txt --content-file content.txt --image-urls "URL1" "URL2"
+```
+
+**有窗口模式**（不添加 `--headless`）：
+```bash
+python "C:\Users\admin\AI\.claude\skills\post-to-xhs\scripts\publish_pipeline.py" --title-file title.txt --content-file content.txt --image-urls "URL1" "URL2"
+```
+
+**其他参数**：
+```bash
+# 发布到指定账号
+python ... --account myaccount ...
+
+# 使用本地图片
+python ... --images "C:\path\to\image.jpg"
+```
+
+处理输出：
+- `NOT_LOGGED_IN` (exit code 1) → 脚本自动切换到有窗口模式，提示用户扫码登录，确认后重新运行
+- `READY_TO_PUBLISH` (exit code 0) → 根据模式进入下一步
+- Exit code 2 → 报告错误
+
+#### B. 写长文模式
+
+**Step B.1 — 填写长文内容 + 一键排版：**
+
+```bash
+python "C:\Users\admin\AI\.claude\skills\post-to-xhs\scripts\cdp_publish.py" long-article --title-file title.txt --content-file content.txt
+```
+
+可选 `--images img1.jpg img2.jpg` 插入图片到编辑器中。
+
+输出中包含 `TEMPLATES: [...]` JSON 数组，为可用的排版模板名称列表。
+
+**Step B.2 — 让用户选择模板：**
+
+使用 `AskUserQuestion` 将模板名称作为选项展示给用户选择（从 TEMPLATES 输出中解析）。
+
+**Step B.3 — 选择模板：**
+
+```bash
+python "C:\Users\admin\AI\.claude\skills\post-to-xhs\scripts\cdp_publish.py" select-template --name "用户选择的模板名"
+```
+
+**Step B.4 — 点击下一步并填写发布页正文描述：**
+
+```bash
+python "C:\Users\admin\AI\.claude\skills\post-to-xhs\scripts\cdp_publish.py" click-next-step --content-file content.txt
+```
+
+注意：发布页有独立的正文描述编辑器，必须通过 `--content` 或 `--content-file` 传入内容填写。
+如果正文超过 1000 字，应压缩到 800 字左右再填入，保持语义不变。
+
+**Step B.5 — 用户预览确认并发布：** 进入下方 4.5 步骤。
+
+### 4.5 用户预览确认（仅有窗口模式 / 长文模式）
+
+**仅当用户选择有窗口模式或使用长文模式时**，使用 `AskUserQuestion` 请用户在浏览器中检查预览，确认后再发布。
+
+无头模式的图文发布跳过此步骤，直接进入 4.6。
+
+### 4.6 点击发布
+
+点击发布按钮：
+
+```bash
+python "C:\Users\admin\AI\.claude\skills\post-to-xhs\scripts\cdp_publish.py" click-publish
+```
+
+### 4.7 报告结果
+
+根据命令输出告知用户发布是否成功。
+
+## 重要提示
+
+- **绝不自动发布** - 必须获得用户确认
+- **图片要求** - 上传图文模式必须有图片；写长文模式图片可选
+- **长文模式** - 必须让用户选择模板，不要自动选择
+- **正文描述** - 长文模式的发布页有独立正文描述框，超过 1000 字需压缩到 800 字左右
+- **无头模式**：使用 `--headless` 参数自动化发布。如需登录，脚本自动切换到有窗口模式
+- 如果页面结构变化导致选择器失效，参考 `references/publish-workflow.md` 更新
+
+## 账号管理
+
+系统支持多个小红书账号，每个账号有独立的 Chrome profile。
+
+### 列出账号
+
+```bash
+python "C:\Users\admin\AI\.claude\skills\post-to-xhs\scripts\cdp_publish.py" list-accounts
+```
+
+### 添加账号
+
+```bash
+python "C:\Users\admin\AI\.claude\skills\post-to-xhs\scripts\cdp_publish.py" add-account myaccount --alias "我的账号"
+```
+
+### 登录
+
+```bash
+# 默认账号
+python "C:\Users\admin\AI\.claude\skills\post-to-xhs\scripts\cdp_publish.py" login
+
+# 指定账号
+python "C:\Users\admin\AI\.claude\skills\post-to-xhs\scripts\cdp_publish.py" --account myaccount login
+```
+
+### 切换账号
+
+```bash
+python "C:\Users\admin\AI\.claude\skills\post-to-xhs\scripts\cdp_publish.py" switch-account
+python "C:\Users\admin\AI\.claude\skills\post-to-xhs\scripts\cdp_publish.py" --account otheraccount switch-account
+```
+
+### 设置默认账号
+
+```bash
+python "C:\Users\admin\AI\.claude\skills\post-to-xhs\scripts\cdp_publish.py" set-default-account myaccount
+```
diff --git a/skills/post-to-xhs/config/accounts.json b/skills/post-to-xhs/config/accounts.json
new file mode 100644
index 0000000..8679cbe
--- /dev/null
+++ b/skills/post-to-xhs/config/accounts.json
@@ -0,0 +1,10 @@
+{
+  "default_account": "default",
+  "accounts": {
+    "default": {
+      "alias": "默认账号",
+      "profile_dir": "C:\\Users\\admin\\AppData\\Local\\Google\\Chrome\\XiaohongshuProfiles\\default",
+      "created_at": null
+    }
+  }
+}
diff --git a/skills/post-to-xhs/references/publish-workflow.md b/skills/post-to-xhs/references/publish-workflow.md
new file mode 100644
index 0000000..bdeddb1
--- /dev/null
+++ b/skills/post-to-xhs/references/publish-workflow.md
@@ -0,0 +1,297 @@
+# 小红书发布流程参考
+
+本文档描述通过 CDP（Chrome DevTools Protocol）自动发布内容到小红书创作者中心的完整流程。
+
+## 前置条件
+
+1. **Chrome 浏览器已安装** - 标准 Google Chrome
+2. **Python 依赖已安装** - `websockets`、`requests`
+3. **首次登录已完成** - 至少登录过一次小红书（cookie 持久化在专用 profile 中）
+
+## 流程概览
+
+**上传图文模式**:
+```
+生成文案 → 用户确认 → 启动 Chrome → 检查登录 → 导航发布页 → 上传图片 → 填写标题 → 填写正文 → 用户确认发布
+```
+
+**写长文模式**:
+```
+生成文案 → 用户确认 → 启动 Chrome → 检查登录 → 导航发布页 → 点击"写长文"tab → 点击"新的创作" → 填写标题 → 填写正文 → 一键排版 → 用户选择模板 → 下一步 → 填写发布页正文描述 → 用户确认发布
+```
+
+## 详细步骤
+
+### 1. 启动 / 连接 Chrome
+
+脚本: `scripts/chrome_launcher.py`
+
+- 检测 `127.0.0.1:9222` 端口是否已有 Chrome 实例
+- 若无，启动 Chrome 并附带以下参数:
+  - `--remote-debugging-port=9222`
+  - `--user-data-dir=%LOCALAPPDATA%/Google/Chrome/XiaohongshuProfile`
+  - `--no-first-run`
+  - `--no-default-browser-check`
+  - `--headless=new`（仅在无头模式下）
+- 等待端口就绪（最多 15 秒）
+
+**用户数据目录说明**: 使用独立的 `XiaohongshuProfile` 目录，与用户日常浏览器 profile 完全隔离，不会干扰正常使用。
+
+**无头模式说明**: 使用 `--headless` 参数启动时，Chrome 不会显示窗口，适合自动化发布。如需登录或切换账号，脚本会自动切换到有窗口模式。
+
+### 2. 检查登录状态
+
+脚本: `scripts/cdp_publish.py` → `check_login()`
+
+- 导航到 `https://creator.xiaohongshu.com`
+- 检查当前 URL 是否包含 "login"（被重定向到登录页）
+- 检查页面是否存在用户信息相关的 DOM 元素
+- 若未登录，提示用户在 Chrome 窗口中扫码登录
+
+### 3. 导航到发布页
+
+- 目标 URL: `https://creator.xiaohongshu.com/publish/publish`
+- 等待页面完全加载
+
+### 4. 上传图片
+
+脚本: `scripts/cdp_publish.py` → `_upload_images()`
+
+- 通过 CDP `DOM.querySelector` 定位 `input[type="file"]` 元素
+- 使用 CDP `DOM.setFileInputFiles` 命令设置文件路径
+- 等待图片上传和处理完成
+
+**图片来源**: 如果图片是 URL，先用 `scripts/image_downloader.py` 下载到临时目录，发布后自动清理。
+
+### 5. 填写标题
+
+脚本: `scripts/cdp_publish.py` → `_fill_title()`
+
+- 定位标题输入框
+- 设置 value 并触发 `input` 和 `change` 事件
+
+### 6. 填写正文
+
+脚本: `scripts/cdp_publish.py` → `_fill_content()`
+
+- 定位 contenteditable 编辑区域（TipTap/ProseMirror editor）
+- 将正文按段落拆分，包裹为 `<p>` 标签写入 innerHTML，段落之间插入 `<p><br></p>` 空行
+- 触发 `input` 事件
+
+### 7. 用户确认并发布
+
+- 脚本填写完成后暂停，提示用户在浏览器中检查预览
+- 用户确认后，脚本点击发布按钮
+- 或用户选择手动点击发布按钮
+
+## 写长文模式详细步骤
+
+### 1-2. 启动 Chrome 和检查登录
+
+同上传图文模式。
+
+### 3. 导航到发布页并点击"写长文"tab
+
+脚本: `scripts/cdp_publish.py` → `_click_long_article_tab()`
+
+- 导航到 `https://creator.xiaohongshu.com/publish/publish`
+- 在 `div.creator-tab` 中查找文本为"写长文"的 tab 并点击
+
+### 4. 点击"新的创作"
+
+脚本: `scripts/cdp_publish.py` → `_click_new_creation()`
+
+- 在页面中查找包含"新的创作"文本的元素并点击
+- 等待长文编辑器页面加载
+
+### 5. 填写长文标题
+
+脚本: `scripts/cdp_publish.py` → `_fill_long_title()`
+
+- 定位 `textarea.d-text[placeholder="输入标题"]` 元素
+- 使用 `HTMLTextAreaElement.prototype.value` 的 native setter 设置值
+- 触发 `input` 和 `change` 事件
+
+### 6. 填写长文正文
+
+同上传图文模式的正文填写（TipTap/ProseMirror 编辑器）。
+
+### 7. 一键排版
+
+脚本: `scripts/cdp_publish.py` → `_click_auto_format()`
+
+- 查找并点击"一键排版"按钮
+- 等待模板列表加载
+
+### 8. 模板选择
+
+脚本: `scripts/cdp_publish.py` → `get_template_names()` + `select_template(name)`
+
+- `get_template_names()` 从 `.template-card .template-title` 获取所有模板名称
+- `select_template(name)` 点击指定名称的模板卡片
+- 已选中的模板卡片 class 为 `template-card selected`
+
+### 9. 下一步并填写发布页描述
+
+脚本: `scripts/cdp_publish.py` → `click_next_and_prepare_publish(content)`
+
+- 点击"下一步"按钮进入发布预览页
+- 发布页有独立的正文描述编辑器（`div.tiptap.ProseMirror`），需要单独填入内容
+
+### 10. 用户确认并发布
+
+同上传图文模式。
+
+## DOM 选择器参考
+
+> **注意**: 小红书前端可能随时更新，以下选择器基于编写时的页面结构。如果自动化失败，需要在浏览器 DevTools 中重新抓取选择器，并更新 `cdp_publish.py` 中的 `SELECTORS` 字典。
+
+| 元素 | 主选择器 | 备选选择器 | 说明 |
+|---|---|---|---|
+| 图片上传 | `input.upload-input` | `input[type="file"]` | 隐藏的文件输入，通过 CDP 直接操作 |
+| 标题输入（图文） | `input[placeholder*="填写标题"]` | `input.d-text` | 图文模式标题输入框 |
+| 标题输入（长文） | `textarea.d-text[placeholder="输入标题"]` | - | 长文模式 textarea 标题 |
+| 正文编辑 | `div.tiptap.ProseMirror` | `div.ProseMirror[contenteditable="true"]` | TipTap/ProseMirror 富文本编辑器 |
+| 发布按钮 | 文本匹配"发布" | - | 通过遍历按钮文本定位 |
+| 写长文 tab | 文本匹配"写长文"（`div.creator-tab`） | - | 长文模式入口 |
+| 新的创作按钮 | 文本匹配"新的创作" | - | 长文编辑器入口 |
+| 一键排版按钮 | 文本匹配"一键排版" | - | 触发模板选择 |
+| 模板卡片 | `.template-card` | `.template-card.selected`（已选） | 排版模板列表 |
+| 模板名称 | `.template-card .template-title` | - | 模板卡片内的名称 span |
+| 下一步按钮 | 文本匹配"下一步" | - | 模板选择后进入发布页 |
+| 登录检测 | URL 包含 "login" | `.user-info, .creator-header` | 重定向检测 + DOM 元素检测 |
+
+## 选择器维护指南
+
+当小红书更新页面导致自动化失败时:
+
+1. 在 Chrome 中打开 `https://creator.xiaohongshu.com/publish/publish`
+2. 按 F12 打开开发者工具
+3. 使用元素选择器（Ctrl+Shift+C）定位目标元素
+4. 记录新的选择器
+5. 更新 `scripts/cdp_publish.py` 中 `SELECTORS` 字典对应的值
+
+## 错误处理
+
+| 错误 | 原因 | 解决方案 |
+|---|---|---|
+| Chrome 未启动 | 端口 9222 无响应 | 运行 `chrome_launcher.py` 或手动启动 Chrome |
+| 找不到 Chrome | 非标准安装路径 | 检查 Chrome 安装，或在脚本中指定路径 |
+| 未登录 | cookie 过期或首次使用 | 在 Chrome 窗口中扫码登录 |
+| 选择器失效 | 小红书页面更新 | 按上述维护指南更新选择器 |
+| 图片上传失败 | 文件路径错误或格式不支持 | 检查图片路径，确保格式为 jpg/png/webp |
+| 发布按钮找不到 | 页面未完全加载 | 增加等待时间或手动点击发布 |
+
+## CLI 用法
+
+所有脚本位于 `scripts/` 目录。
+
+### 方式 A: 统一 pipeline（推荐）
+
+```bash
+# 无头模式（推荐）- 无浏览器窗口，更快
+python publish_pipeline.py --headless --title "标题" --content "正文" --image-urls URL1 URL2
+
+# 无头模式 - 从文件读取标题和正文
+python publish_pipeline.py --headless --title-file title.txt --content-file body.txt --image-urls URL1
+
+# 有窗口模式 - 用于调试或首次登录
+python publish_pipeline.py --title "标题" --content "正文" --image-urls URL1 URL2
+
+# 使用本地图片文件
+python publish_pipeline.py --headless --title "标题" --content "正文" --images img1.jpg img2.jpg
+
+# 填写并自动发布
+python publish_pipeline.py --headless --title "标题" --content "正文" --image-urls URL1 --auto-publish
+```
+
+输出状态码:
+- 退出码 0 + `READY_TO_PUBLISH` = 表单已填写，等待确认
+- 退出码 0 + `PUBLISHED` = 已发布
+- 退出码 1 + `NOT_LOGGED_IN` = 未登录，需扫码（无头模式下会自动切换到有窗口模式）
+- 退出码 2 = 其他错误
+
+### 方式 B: 分步调用（图文模式）
+
+```bash
+# 1. 启动 Chrome（可选 --headless）
+python chrome_launcher.py
+python chrome_launcher.py --headless
+
+# 2. 检查登录（退出码 0=已登录, 1=未登录）
+python cdp_publish.py check-login
+python cdp_publish.py --headless check-login
+
+# 3. 填写表单
+python cdp_publish.py fill --title "标题" --content-file body.txt --images img1.jpg
+python cdp_publish.py --headless fill --title "标题" --content-file body.txt --images img1.jpg
+
+# 4. 用户确认后点击发布
+python cdp_publish.py click-publish
+
+# 或一步完成填写+发布
+python cdp_publish.py --headless publish --title "标题" --content-file body.txt --images img1.jpg
+```
+
+### 方式 C: 分步调用（长文模式）
+
+```bash
+# 1. 启动 Chrome
+python chrome_launcher.py
+
+# 2. 检查登录
+python cdp_publish.py check-login
+
+# 3. 填写长文 + 一键排版（输出包含 TEMPLATES JSON）
+python cdp_publish.py long-article --title-file title.txt --content-file content.txt
+
+# 4. 选择模板
+python cdp_publish.py select-template --name "模板名称"
+
+# 5. 下一步 + 填写发布页正文描述
+python cdp_publish.py click-next-step --content-file content.txt
+
+# 6. 用户确认后点击发布
+python cdp_publish.py click-publish
+```
+
+### 方式 D: Pipeline 长文模式
+
+```bash
+# 长文模式（图片可选）
+python publish_pipeline.py --mode long-article --title-file title.txt --content-file content.txt
+python publish_pipeline.py --mode long-article --title "标题" --content "正文" --images img1.jpg
+```
+
+### 账号管理
+
+```bash
+# 首次登录或 session 过期 - 打开浏览器扫码登录
+python cdp_publish.py login
+
+# 切换账号 - 清除 cookie 并打开登录页
+python cdp_publish.py switch-account
+
+# 关闭 Chrome
+python chrome_launcher.py --kill
+
+# 重启 Chrome（可选无头模式）
+python chrome_launcher.py --restart
+python chrome_launcher.py --restart --headless
+```
+
+### Claude Code 集成
+
+在 Claude Code 中通过 Bash 工具调用。推荐使用 pipeline 方式:
+
+1. 将中文标题和正文写入临时文本文件（UTF-8 编码）
+2. 调用 `publish_pipeline.py --headless` 传入文件路径和图片 URL
+3. 根据输出状态码处理结果：
+   - 未登录 → 脚本自动切换到有窗口模式，提示用户扫码
+   - 已填写 → 请用户确认预览
+4. 用户确认后调用 `cdp_publish.py click-publish` 发布
+
+**切换账号流程**:
+1. 调用 `cdp_publish.py switch-account`
+2. 等待用户扫码确认
+3. 继续正常发布流程
diff --git a/skills/post-to-xhs/scripts/account_manager.py b/skills/post-to-xhs/scripts/account_manager.py
new file mode 100644
index 0000000..0e27f15
--- /dev/null
+++ b/skills/post-to-xhs/scripts/account_manager.py
@@ -0,0 +1,309 @@
+"""
+Multi-account manager for Xiaohongshu publishing.
+
+Manages multiple Xiaohongshu accounts with separate Chrome profiles:
+- Each account has its own user-data-dir for cookie isolation
+- Accounts are stored in a JSON config file
+- Supports add/remove/list/switch operations
+
+Usage:
+    python account_manager.py list
+    python account_manager.py add <name> [--alias <alias>]
+    python account_manager.py remove <name>
+    python account_manager.py info <name>
+    python account_manager.py set-default <name>
+"""
+
+import json
+import os
+import sys
+import shutil
+from typing import Optional
+
+# Config file location
+CONFIG_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), "..", "config")
+ACCOUNTS_FILE = os.path.join(CONFIG_DIR, "accounts.json")
+
+# Base directory for account profiles
+PROFILES_BASE = os.path.join(os.environ.get("LOCALAPPDATA", os.path.expanduser("~")),
+                              "Google", "Chrome", "XiaohongshuProfiles")
+
+# Default account name (for backward compatibility)
+DEFAULT_PROFILE_NAME = "default"
+
+
+def _ensure_config_dir():
+    """Ensure the config directory exists."""
+    os.makedirs(CONFIG_DIR, exist_ok=True)
+
+
+def _load_accounts() -> dict:
+    """Load accounts from config file."""
+    _ensure_config_dir()
+    if os.path.exists(ACCOUNTS_FILE):
+        try:
+            with open(ACCOUNTS_FILE, "r", encoding="utf-8") as f:
+                return json.load(f)
+        except (json.JSONDecodeError, IOError):
+            pass
+    # Default structure
+    return {
+        "default_account": DEFAULT_PROFILE_NAME,
+        "accounts": {
+            DEFAULT_PROFILE_NAME: {
+                "alias": "默认账号",
+                "profile_dir": os.path.join(PROFILES_BASE, DEFAULT_PROFILE_NAME),
+                "created_at": None,
+            }
+        }
+    }
+
+
+def _save_accounts(data: dict):
+    """Save accounts to config file."""
+    _ensure_config_dir()
+    with open(ACCOUNTS_FILE, "w", encoding="utf-8") as f:
+        json.dump(data, f, ensure_ascii=False, indent=2)
+
+
+def get_profile_dir(account_name: Optional[str] = None) -> str:
+    """
+    Get the Chrome profile directory for a given account.
+
+    Args:
+        account_name: Account name. If None, uses the default account.
+
+    Returns:
+        Path to the Chrome user-data-dir for this account.
+    """
+    data = _load_accounts()
+
+    if account_name is None:
+        account_name = data.get("default_account", DEFAULT_PROFILE_NAME)
+
+    if account_name not in data["accounts"]:
+        # Fallback to default
+        account_name = DEFAULT_PROFILE_NAME
+        if account_name not in data["accounts"]:
+            # Create default account entry
+            data["accounts"][account_name] = {
+                "alias": "默认账号",
+                "profile_dir": os.path.join(PROFILES_BASE, account_name),
+                "created_at": None,
+            }
+            _save_accounts(data)
+
+    return data["accounts"][account_name]["profile_dir"]
+
+
+def get_default_account() -> str:
+    """Get the name of the default account."""
+    data = _load_accounts()
+    return data.get("default_account", DEFAULT_PROFILE_NAME)
+
+
+def set_default_account(account_name: str) -> bool:
+    """
+    Set the default account.
+
+    Returns True if successful, False if account doesn't exist.
+    """
+    data = _load_accounts()
+    if account_name not in data["accounts"]:
+        return False
+    data["default_account"] = account_name
+    _save_accounts(data)
+    return True
+
+
+def list_accounts() -> list[dict]:
+    """
+    List all registered accounts.
+
+    Returns a list of dicts with account info.
+    """
+    data = _load_accounts()
+    default = data.get("default_account", DEFAULT_PROFILE_NAME)
+    result = []
+    for name, info in data["accounts"].items():
+        result.append({
+            "name": name,
+            "alias": info.get("alias", ""),
+            "profile_dir": info.get("profile_dir", ""),
+            "is_default": name == default,
+        })
+    return result
+
+
+def add_account(name: str, alias: Optional[str] = None) -> bool:
+    """
+    Add a new account.
+
+    Args:
+        name: Unique account name (used as identifier)
+        alias: Display name / description
+
+    Returns True if added, False if name already exists.
+    """
+    data = _load_accounts()
+    if name in data["accounts"]:
+        return False
+
+    from datetime import datetime
+    profile_dir = os.path.join(PROFILES_BASE, name)
+    os.makedirs(profile_dir, exist_ok=True)
+
+    data["accounts"][name] = {
+        "alias": alias or name,
+        "profile_dir": profile_dir,
+        "created_at": datetime.now().isoformat(),
+    }
+    _save_accounts(data)
+    return True
+
+
+def remove_account(name: str, delete_profile: bool = False) -> bool:
+    """
+    Remove an account.
+
+    Args:
+        name: Account name to remove
+        delete_profile: If True, also delete the Chrome profile directory
+
+    Returns True if removed, False if not found or is default.
+    """
+    data = _load_accounts()
+    if name not in data["accounts"]:
+        return False
+
+    # Don't allow removing the default account if it's the only one
+    if name == data.get("default_account") and len(data["accounts"]) == 1:
+        return False
+
+    profile_dir = data["accounts"][name].get("profile_dir", "")
+    del data["accounts"][name]
+
+    # If we removed the default, set a new default
+    if name == data.get("default_account"):
+        data["default_account"] = next(iter(data["accounts"].keys()))
+
+    _save_accounts(data)
+
+    # Optionally delete the profile directory
+    if delete_profile and profile_dir and os.path.isdir(profile_dir):
+        try:
+            shutil.rmtree(profile_dir)
+        except Exception:
+            pass
+
+    return True
+
+
+def get_account_info(name: str) -> Optional[dict]:
+    """Get info for a specific account."""
+    data = _load_accounts()
+    if name not in data["accounts"]:
+        return None
+    info = data["accounts"][name].copy()
+    info["name"] = name
+    info["is_default"] = name == data.get("default_account")
+    return info
+
+
+def account_exists(name: str) -> bool:
+    """Check if an account exists."""
+    data = _load_accounts()
+    return name in data["accounts"]
+
+
+# ---------------------------------------------------------------------------
+# CLI
+# ---------------------------------------------------------------------------
+
+def main():
+    import argparse
+
+    parser = argparse.ArgumentParser(description="Xiaohongshu Account Manager")
+    sub = parser.add_subparsers(dest="command", required=True)
+
+    # list
+    sub.add_parser("list", help="List all accounts")
+
+    # add
+    p_add = sub.add_parser("add", help="Add a new account")
+    p_add.add_argument("name", help="Account name (unique identifier)")
+    p_add.add_argument("--alias", help="Display name / description")
+
+    # remove
+    p_rm = sub.add_parser("remove", help="Remove an account")
+    p_rm.add_argument("name", help="Account name to remove")
+    p_rm.add_argument("--delete-profile", action="store_true",
+                      help="Also delete the Chrome profile directory")
+
+    # info
+    p_info = sub.add_parser("info", help="Show account info")
+    p_info.add_argument("name", help="Account name")
+
+    # set-default
+    p_def = sub.add_parser("set-default", help="Set the default account")
+    p_def.add_argument("name", help="Account name to set as default")
+
+    # get-profile-dir (for internal use)
+    p_dir = sub.add_parser("get-profile-dir", help="Get profile directory for an account")
+    p_dir.add_argument("--account", help="Account name (default: default account)")
+
+    args = parser.parse_args()
+
+    if args.command == "list":
+        accounts = list_accounts()
+        if not accounts:
+            print("No accounts configured.")
+            return
+        print(f"{'Name':<20} {'Alias':<20} {'Default':<10}")
+        print("-" * 50)
+        for acc in accounts:
+            default_mark = "*" if acc["is_default"] else ""
+            print(f"{acc['name']:<20} {acc['alias']:<20} {default_mark:<10}")
+
+    elif args.command == "add":
+        if add_account(args.name, args.alias):
+            print(f"Account '{args.name}' added.")
+            print(f"Profile dir: {get_profile_dir(args.name)}")
+            print("\nTo log in to this account, run:")
+            print(f"  python cdp_publish.py --account {args.name} login")
+        else:
+            print(f"Error: Account '{args.name}' already exists.", file=sys.stderr)
+            sys.exit(1)
+
+    elif args.command == "remove":
+        if remove_account(args.name, args.delete_profile):
+            print(f"Account '{args.name}' removed.")
+        else:
+            print(f"Error: Cannot remove account '{args.name}'.", file=sys.stderr)
+            sys.exit(1)
+
+    elif args.command == "info":
+        info = get_account_info(args.name)
+        if info:
+            print(f"Name: {info['name']}")
+            print(f"Alias: {info.get('alias', '')}")
+            print(f"Profile dir: {info.get('profile_dir', '')}")
+            print(f"Default: {'Yes' if info.get('is_default') else 'No'}")
+            print(f"Created: {info.get('created_at', 'Unknown')}")
+        else:
+            print(f"Error: Account '{args.name}' not found.", file=sys.stderr)
+            sys.exit(1)
+
+    elif args.command == "set-default":
+        if set_default_account(args.name):
+            print(f"Default account set to '{args.name}'.")
+        else:
+            print(f"Error: Account '{args.name}' not found.", file=sys.stderr)
+            sys.exit(1)
+
+    elif args.command == "get-profile-dir":
+        print(get_profile_dir(args.account))
+
+
+if __name__ == "__main__":
+    main()
diff --git a/skills/post-to-xhs/scripts/cdp_publish.py b/skills/post-to-xhs/scripts/cdp_publish.py
new file mode 100644
index 0000000..3d47efe
--- /dev/null
+++ b/skills/post-to-xhs/scripts/cdp_publish.py
@@ -0,0 +1,1056 @@
+"""
+CDP-based Xiaohongshu publisher.
+
+Connects to a Chrome instance via Chrome DevTools Protocol to automate
+publishing articles on Xiaohongshu (RED) creator center.
+
+CLI usage:
+    # Basic commands (image-text mode)
+    python cdp_publish.py check-login [--headless] [--account NAME]
+    python cdp_publish.py fill --title "标题" --content "正文" --images img1.jpg [--headless] [--account NAME]
+    python cdp_publish.py publish --title "标题" --content "正文" --images img1.jpg [--headless] [--account NAME]
+    python cdp_publish.py click-publish [--headless] [--account NAME]
+
+    # Long article mode
+    python cdp_publish.py long-article --title "标题" --content "正文" [--images img1.jpg] [--account NAME]
+    python cdp_publish.py click-next-step [--account NAME]
+
+    # Account management
+    python cdp_publish.py login [--account NAME]           # open browser for QR login
+    python cdp_publish.py re-login [--account NAME]        # clear cookies and re-login same account
+    python cdp_publish.py switch-account [--account NAME]  # clear cookies + open login for new account
+    python cdp_publish.py list-accounts                    # list all configured accounts
+    python cdp_publish.py add-account NAME [--alias ALIAS] # add a new account
+    python cdp_publish.py remove-account NAME              # remove an account
+
+Library usage:
+    from cdp_publish import XiaohongshuPublisher
+
+    publisher = XiaohongshuPublisher()
+    publisher.connect()
+    publisher.check_login()
+    publisher.publish(
+        title="Article title",
+        content="Article body text",
+        image_paths=["/path/to/img1.jpg", "/path/to/img2.jpg"],
+    )
+"""
+
+import json
+import os
+import time
+import sys
+from typing import Any
+
+# Ensure UTF-8 output on Windows consoles
+if sys.platform == "win32":
+    os.environ.setdefault("PYTHONIOENCODING", "utf-8")
+    try:
+        sys.stdout.reconfigure(encoding="utf-8", errors="replace")
+        sys.stderr.reconfigure(encoding="utf-8", errors="replace")
+    except Exception:
+        pass
+
+import requests
+import websockets.sync.client as ws_client
+
+# ---------------------------------------------------------------------------
+# Configuration - centralised selectors and URLs for easy maintenance
+# ---------------------------------------------------------------------------
+
+CDP_HOST = "127.0.0.1"
+CDP_PORT = 9222
+
+# Xiaohongshu URLs
+XHS_CREATOR_URL = "https://creator.xiaohongshu.com/publish/publish"
+XHS_HOME_URL = "https://www.xiaohongshu.com"
+XHS_LOGIN_CHECK_URL = "https://creator.xiaohongshu.com"
+
+# DOM selectors (update these when Xiaohongshu changes their page structure)
+# Last verified: 2026-02
+SELECTORS = {
+    # "上传图文" tab - must click before uploading images
+    "image_text_tab": "div.creator-tab",
+    "image_text_tab_text": "上传图文",
+    # Upload area - the file input element for images (visible after clicking tab)
+    "upload_input": "input.upload-input",
+    "upload_input_alt": 'input[type="file"]',
+    # Title input field (visible after image upload)
+    "title_input": 'input[placeholder*="填写标题"]',
+    "title_input_alt": "input.d-text",
+    # Content editor area - TipTap/ProseMirror contenteditable div
+    "content_editor": "div.tiptap.ProseMirror",
+    "content_editor_alt": 'div.ProseMirror[contenteditable="true"]',
+    # Publish button
+    "publish_button_text": "发布",
+    # Login indicator - URL-based check (redirect to /login if not logged in)
+    "login_indicator": '.user-info, .creator-header, [class*="user"]',
+    # Long article mode
+    "long_article_tab_text": "写长文",
+    "new_creation_btn_text": "新的创作",
+    "long_title_input": 'textarea.d-text[placeholder="输入标题"]',
+    "auto_format_btn_text": "一键排版",
+    "next_step_btn_text": "下一步",
+    "template_card": ".template-card",
+}
+
+# Timing
+PAGE_LOAD_WAIT = 3  # seconds to wait after navigation
+TAB_CLICK_WAIT = 2  # seconds to wait after clicking tab
+UPLOAD_WAIT = 6  # seconds to wait after image upload for editor to appear
+ACTION_INTERVAL = 1  # seconds between actions
+AUTO_FORMAT_WAIT = 5  # seconds to wait after clicking auto-format
+TEMPLATE_WAIT = 10  # seconds max to wait for template cards to appear
+
+
+class CDPError(Exception):
+    """Error communicating with Chrome via CDP."""
+
+
+class XiaohongshuPublisher:
+    """Automates publishing to Xiaohongshu via CDP."""
+
+    def __init__(self, host: str = CDP_HOST, port: int = CDP_PORT):
+        self.host = host
+        self.port = port
+        self.ws = None
+        self._msg_id = 0
+
+    # ------------------------------------------------------------------
+    # CDP connection management
+    # ------------------------------------------------------------------
+
+    def _get_targets(self) -> list[dict]:
+        """Get list of available browser targets (tabs). Retries once on failure."""
+        url = f"http://{self.host}:{self.port}/json"
+        for attempt in range(2):
+            try:
+                resp = requests.get(url, timeout=5)
+                resp.raise_for_status()
+                return resp.json()
+            except Exception as e:
+                if attempt == 0:
+                    print(f"[cdp_publish] CDP connection failed ({e}), restarting Chrome...")
+                    from chrome_launcher import ensure_chrome
+                    ensure_chrome(self.port)
+                    time.sleep(2)
+                else:
+                    raise CDPError(f"Cannot reach Chrome on {self.host}:{self.port}: {e}")
+
+    def _find_or_create_tab(self, target_url_prefix: str = "") -> str:
+        """Find an existing tab matching the URL prefix, or return the first page tab."""
+        targets = self._get_targets()
+        pages = [t for t in targets if t.get("type") == "page"]
+
+        if target_url_prefix:
+            for t in pages:
+                if t.get("url", "").startswith(target_url_prefix):
+                    return t["webSocketDebuggerUrl"]
+
+        # Create a new tab
+        resp = requests.put(
+            f"http://{self.host}:{self.port}/json/new?{XHS_CREATOR_URL}",
+            timeout=5,
+        )
+        if resp.ok:
+            return resp.json().get("webSocketDebuggerUrl", "")
+
+        # Fallback: use first available page
+        if pages:
+            return pages[0]["webSocketDebuggerUrl"]
+
+        raise CDPError("No browser tabs available.")
+
+    def connect(self, target_url_prefix: str = ""):
+        """Connect to a Chrome tab via WebSocket."""
+        ws_url = self._find_or_create_tab(target_url_prefix)
+        if not ws_url:
+            raise CDPError("Could not obtain WebSocket URL for any tab.")
+
+        print(f"[cdp_publish] Connecting to {ws_url}")
+        self.ws = ws_client.connect(ws_url)
+        print("[cdp_publish] Connected to Chrome tab.")
+
+    def disconnect(self):
+        """Close the WebSocket connection."""
+        if self.ws:
+            self.ws.close()
+            self.ws = None
+
+    # ------------------------------------------------------------------
+    # CDP command helpers
+    # ------------------------------------------------------------------
+
+    def _send(self, method: str, params: dict | None = None) -> dict:
+        """Send a CDP command and return the result."""
+        if not self.ws:
+            raise CDPError("Not connected. Call connect() first.")
+
+        self._msg_id += 1
+        msg = {"id": self._msg_id, "method": method}
+        if params:
+            msg["params"] = params
+
+        self.ws.send(json.dumps(msg))
+
+        # Wait for the matching response
+        while True:
+            raw = self.ws.recv()
+            data = json.loads(raw)
+            if data.get("id") == self._msg_id:
+                if "error" in data:
+                    raise CDPError(f"CDP error: {data['error']}")
+                return data.get("result", {})
+            # else: it's an event, skip it
+
+    def _evaluate(self, expression: str) -> Any:
+        """Execute JavaScript in the page and return the result value."""
+        result = self._send("Runtime.evaluate", {
+            "expression": expression,
+            "returnByValue": True,
+            "awaitPromise": True,
+        })
+        remote_obj = result.get("result", {})
+        if remote_obj.get("subtype") == "error":
+            raise CDPError(f"JS error: {remote_obj.get('description', remote_obj)}")
+        return remote_obj.get("value")
+
+    def _navigate(self, url: str):
+        """Navigate the current tab to the given URL and wait for load."""
+        print(f"[cdp_publish] Navigating to {url}")
+        self._send("Page.enable")
+        self._send("Page.navigate", {"url": url})
+        time.sleep(PAGE_LOAD_WAIT)
+
+    # ------------------------------------------------------------------
+    # Login check
+    # ------------------------------------------------------------------
+
+    def check_login(self) -> bool:
+        """
+        Navigate to Xiaohongshu creator center and check if the user is logged in.
+
+        Returns True if logged in. If not logged in, prints instructions
+        and returns False.
+        """
+        self._navigate(XHS_LOGIN_CHECK_URL)
+        time.sleep(2)
+
+        # Check if we got redirected to a login page
+        current_url = self._evaluate("window.location.href")
+        print(f"[cdp_publish] Current URL: {current_url}")
+
+        if "login" in current_url.lower():
+            print(
+                "\n[cdp_publish] NOT LOGGED IN.\n"
+                "  Please scan the QR code in the Chrome window to log in,\n"
+                "  then run this script again.\n"
+            )
+            return False
+
+        print("[cdp_publish] Login confirmed.")
+        return True
+
+    def clear_cookies(self, domain: str = ".xiaohongshu.com"):
+        """
+        Clear all cookies for the given domain to force re-login.
+
+        Used when switching accounts.
+        """
+        print(f"[cdp_publish] Clearing cookies for {domain}...")
+        self._send("Network.enable")
+        self._send("Network.clearBrowserCookies")
+        # Also clear storage
+        self._send("Storage.clearDataForOrigin", {
+            "origin": "https://www.xiaohongshu.com",
+            "storageTypes": "cookies,local_storage,session_storage",
+        })
+        self._send("Storage.clearDataForOrigin", {
+            "origin": "https://creator.xiaohongshu.com",
+            "storageTypes": "cookies,local_storage,session_storage",
+        })
+        print("[cdp_publish] Cookies and storage cleared.")
+
+    def open_login_page(self):
+        """
+        Navigate to the Xiaohongshu login page for QR code scanning.
+
+        Used for initial login or after clearing cookies for account switch.
+        """
+        self._navigate(XHS_LOGIN_CHECK_URL)
+        time.sleep(2)
+        current_url = self._evaluate("window.location.href")
+        if "login" not in current_url.lower():
+            # Already logged in, navigate to login page explicitly
+            self._navigate("https://creator.xiaohongshu.com/login")
+            time.sleep(2)
+        print(
+            "\n[cdp_publish] Login page is open.\n"
+            "  Please scan the QR code in the Chrome window to log in.\n"
+        )
+
+    # ------------------------------------------------------------------
+    # Publishing actions
+    # ------------------------------------------------------------------
+
+    def _click_image_text_tab(self):
+        """Click the '上传图文' tab to switch to image+text publish mode."""
+        print("[cdp_publish] Clicking '上传图文' tab...")
+        tab_text = SELECTORS["image_text_tab_text"]
+        selector = SELECTORS["image_text_tab"]
+
+        clicked = self._evaluate(f"""
+            (function() {{
+                var tabs = document.querySelectorAll('{selector}');
+                for (var i = 0; i < tabs.length; i++) {{
+                    if (tabs[i].textContent.trim() === '{tab_text}') {{
+                        tabs[i].click();
+                        return true;
+                    }}
+                }}
+                return false;
+            }})();
+        """)
+
+        if not clicked:
+            raise CDPError(
+                f"Could not find '{tab_text}' tab. "
+                "The page structure may have changed."
+            )
+
+        print("[cdp_publish] Tab clicked, waiting for upload area...")
+        time.sleep(TAB_CLICK_WAIT)
+
+    def _upload_images(self, image_paths: list[str]):
+        """Upload images via the file input element."""
+        if not image_paths:
+            print("[cdp_publish] No images to upload, skipping.")
+            return
+
+        # Normalize paths (forward slashes for CDP)
+        normalized = [p.replace("\\", "/") for p in image_paths]
+
+        print(f"[cdp_publish] Uploading {len(image_paths)} image(s)...")
+
+        # Enable DOM domain
+        self._send("DOM.enable")
+
+        # Get the document root
+        doc = self._send("DOM.getDocument")
+        root_id = doc["root"]["nodeId"]
+
+        # Try primary selector, then fallback
+        node_id = 0
+        for selector in (SELECTORS["upload_input"], SELECTORS["upload_input_alt"]):
+            result = self._send("DOM.querySelector", {
+                "nodeId": root_id,
+                "selector": selector,
+            })
+            node_id = result.get("nodeId", 0)
+            if node_id:
+                break
+
+        if not node_id:
+            raise CDPError(
+                "Could not find file input element.\n"
+                "The page structure may have changed. Check references/publish-workflow.md."
+            )
+
+        # Use DOM.setFileInputFiles to set the files
+        self._send("DOM.setFileInputFiles", {
+            "nodeId": node_id,
+            "files": normalized,
+        })
+
+        print("[cdp_publish] Images uploaded. Waiting for editor to appear...")
+        time.sleep(UPLOAD_WAIT)
+
+    def _fill_title(self, title: str):
+        """Fill in the article title."""
+        print(f"[cdp_publish] Setting title: {title[:40]}...")
+        time.sleep(ACTION_INTERVAL)
+
+        for selector in (SELECTORS["title_input"], SELECTORS["title_input_alt"]):
+            found = self._evaluate(f"!!document.querySelector('{selector}')")
+            if found:
+                escaped_title = json.dumps(title)
+                self._evaluate(f"""
+                    (function() {{
+                        var el = document.querySelector('{selector}');
+                        var nativeSetter = Object.getOwnPropertyDescriptor(
+                            window.HTMLInputElement.prototype, 'value'
+                        ).set;
+                        el.focus();
+                        nativeSetter.call(el, {escaped_title});
+                        el.dispatchEvent(new Event('input', {{ bubbles: true }}));
+                        el.dispatchEvent(new Event('change', {{ bubbles: true }}));
+                    }})();
+                """)
+                print("[cdp_publish] Title set.")
+                return
+
+        raise CDPError("Could not find title input element.")
+
+    def _fill_content(self, content: str):
+        """Fill in the article body content using the TipTap/ProseMirror editor."""
+        print(f"[cdp_publish] Setting content ({len(content)} chars)...")
+        time.sleep(ACTION_INTERVAL)
+
+        for selector in (SELECTORS["content_editor"], SELECTORS["content_editor_alt"]):
+            found = self._evaluate(f"!!document.querySelector('{selector}')")
+            if found:
+                escaped = json.dumps(content)
+                self._evaluate(f"""
+                    (function() {{
+                        var el = document.querySelector('{selector}');
+                        el.focus();
+                        var text = {escaped};
+                        var paragraphs = text.split('\\n').filter(function(p) {{ return p.trim(); }});
+                        var html = [];
+                        for (var i = 0; i < paragraphs.length; i++) {{
+                            html.push('<p>' + paragraphs[i] + '</p>');
+                            if (i < paragraphs.length - 1) {{
+                                html.push('<p><br></p>');
+                            }}
+                        }}
+                        el.innerHTML = html.join('');
+                        el.dispatchEvent(new Event('input', {{ bubbles: true }}));
+                    }})();
+                """)
+                print("[cdp_publish] Content set.")
+                return
+
+        raise CDPError("Could not find content editor element.")
+
+    def _click_publish(self):
+        """Click the publish button (found by text content)."""
+        print("[cdp_publish] Clicking publish button...")
+        time.sleep(ACTION_INTERVAL)
+
+        btn_text = SELECTORS["publish_button_text"]
+        clicked = self._evaluate(f"""
+            (function() {{
+                // Strategy 1: search <button> elements by text
+                var buttons = document.querySelectorAll('button');
+                for (var i = 0; i < buttons.length; i++) {{
+                    var t = buttons[i].textContent.trim();
+                    if (t === '{btn_text}') {{
+                        buttons[i].click();
+                        return true;
+                    }}
+                }}
+                // Strategy 2: search d-button-content / d-text spans
+                var spans = document.querySelectorAll('.d-button-content .d-text, .d-button-content span');
+                for (var i = 0; i < spans.length; i++) {{
+                    if (spans[i].textContent.trim() === '{btn_text}') {{
+                        var el = spans[i].closest('button, [role="button"], .d-button, [class*="btn"], [class*="button"]');
+                        if (!el) el = spans[i].closest('.d-button-content');
+                        if (!el) el = spans[i];
+                        el.click();
+                        return true;
+                    }}
+                }}
+                return false;
+            }})();
+        """)
+
+        if clicked:
+            print("[cdp_publish] Publish button clicked.")
+        else:
+            raise CDPError(
+                "Could not find publish button. "
+                "Please click it manually in the browser."
+            )
+
+    # ------------------------------------------------------------------
+    # Long article actions
+    # ------------------------------------------------------------------
+
+    def _click_long_article_tab(self):
+        """Click the '写长文' tab to switch to long article mode."""
+        print("[cdp_publish] Clicking '写长文' tab...")
+        tab_text = SELECTORS["long_article_tab_text"]
+        selector = SELECTORS["image_text_tab"]  # same container: div.creator-tab
+
+        clicked = self._evaluate(f"""
+            (function() {{
+                var tabs = document.querySelectorAll('{selector}');
+                for (var i = 0; i < tabs.length; i++) {{
+                    if (tabs[i].textContent.trim() === '{tab_text}') {{
+                        tabs[i].click();
+                        return true;
+                    }}
+                }}
+                return false;
+            }})();
+        """)
+
+        if not clicked:
+            raise CDPError(
+                f"Could not find '{tab_text}' tab. "
+                "The page structure may have changed."
+            )
+
+        print("[cdp_publish] '写长文' tab clicked.")
+        time.sleep(TAB_CLICK_WAIT)
+
+    def _click_new_creation(self):
+        """Click the '新的创作' button to start a new long article."""
+        print("[cdp_publish] Clicking '新的创作' button...")
+        btn_text = SELECTORS["new_creation_btn_text"]
+
+        clicked = self._evaluate(f"""
+            (function() {{
+                // Search all elements for text match
+                var candidates = document.querySelectorAll(
+                    '.center span, .center div, .center button, .center a, '
+                    + 'button, [role="button"], [class*="btn"], [class*="creation"]'
+                );
+                for (var i = 0; i < candidates.length; i++) {{
+                    if (candidates[i].textContent.trim() === '{btn_text}') {{
+                        candidates[i].click();
+                        return true;
+                    }}
+                }}
+                return false;
+            }})();
+        """)
+
+        if not clicked:
+            raise CDPError(
+                f"Could not find '{btn_text}' button. "
+                "The page structure may have changed."
+            )
+
+        print("[cdp_publish] '新的创作' button clicked.")
+        time.sleep(PAGE_LOAD_WAIT)
+
+    def _fill_long_title(self, title: str):
+        """Fill in the long article title (textarea element)."""
+        print(f"[cdp_publish] Setting long article title: {title[:40]}...")
+        time.sleep(ACTION_INTERVAL)
+
+        selector = SELECTORS["long_title_input"]
+        found = self._evaluate(f"!!document.querySelector('{selector}')")
+        if not found:
+            raise CDPError(
+                f"Could not find long title textarea ('{selector}'). "
+                "The page structure may have changed."
+            )
+
+        escaped_title = json.dumps(title)
+        self._evaluate(f"""
+            (function() {{
+                var el = document.querySelector('{selector}');
+                var nativeSetter = Object.getOwnPropertyDescriptor(
+                    window.HTMLTextAreaElement.prototype, 'value'
+                ).set;
+                el.focus();
+                nativeSetter.call(el, {escaped_title});
+                el.dispatchEvent(new Event('input', {{ bubbles: true }}));
+                el.dispatchEvent(new Event('change', {{ bubbles: true }}));
+            }})();
+        """)
+        print("[cdp_publish] Long article title set.")
+
+    def _click_auto_format(self):
+        """Click the '一键排版' button."""
+        print("[cdp_publish] Clicking '一键排版' button...")
+        btn_text = SELECTORS["auto_format_btn_text"]
+
+        clicked = self._evaluate(f"""
+            (function() {{
+                var elems = document.querySelectorAll(
+                    'button, [role="button"], span, div, a, [class*="btn"]'
+                );
+                for (var i = 0; i < elems.length; i++) {{
+                    if (elems[i].textContent.trim() === '{btn_text}') {{
+                        elems[i].click();
+                        return true;
+                    }}
+                }}
+                return false;
+            }})();
+        """)
+
+        if not clicked:
+            raise CDPError(
+                f"Could not find '{btn_text}' button. "
+                "The page structure may have changed."
+            )
+
+        print("[cdp_publish] '一键排版' button clicked. Waiting for templates...")
+        time.sleep(AUTO_FORMAT_WAIT)
+
+    def _wait_for_templates(self) -> bool:
+        """Wait for template cards to appear after clicking auto-format."""
+        print("[cdp_publish] Waiting for template cards to load...")
+        selector = SELECTORS["template_card"]
+
+        for attempt in range(TEMPLATE_WAIT):
+            found = self._evaluate(
+                f"document.querySelectorAll('{selector}').length"
+            )
+            if found and found > 0:
+                print(f"[cdp_publish] Found {found} template card(s).")
+                return True
+            time.sleep(1)
+
+        print("[cdp_publish] Warning: No template cards found within timeout.")
+        return False
+
+    def get_template_names(self) -> list[str]:
+        """Get the list of available template names from the page."""
+        selector = SELECTORS["template_card"]
+        names = self._evaluate(f"""
+            (function() {{
+                var cards = document.querySelectorAll('{selector}');
+                var names = [];
+                for (var i = 0; i < cards.length; i++) {{
+                    var title = cards[i].querySelector('.template-title');
+                    names.push(title ? title.textContent.trim() : 'Template ' + i);
+                }}
+                return names;
+            }})();
+        """)
+        return names or []
+
+    def select_template(self, name: str) -> bool:
+        """Select a template by clicking the card with the matching name."""
+        print(f"[cdp_publish] Selecting template: {name}...")
+        selector = SELECTORS["template_card"]
+
+        clicked = self._evaluate(f"""
+            (function() {{
+                var cards = document.querySelectorAll('{selector}');
+                for (var i = 0; i < cards.length; i++) {{
+                    var title = cards[i].querySelector('.template-title');
+                    if (title && title.textContent.trim() === {json.dumps(name)}) {{
+                        cards[i].click();
+                        return true;
+                    }}
+                }}
+                return false;
+            }})();
+        """)
+
+        if clicked:
+            print(f"[cdp_publish] Template '{name}' selected.")
+            time.sleep(ACTION_INTERVAL)
+        else:
+            print(f"[cdp_publish] Warning: Template '{name}' not found.")
+
+        return bool(clicked)
+
+    def _click_next_step(self):
+        """Click the '下一步' button."""
+        print("[cdp_publish] Clicking '下一步' button...")
+        btn_text = SELECTORS["next_step_btn_text"]
+
+        clicked = self._evaluate(f"""
+            (function() {{
+                var elems = document.querySelectorAll(
+                    'button, [role="button"], span, div, a, [class*="btn"]'
+                );
+                for (var i = 0; i < elems.length; i++) {{
+                    if (elems[i].textContent.trim() === '{btn_text}') {{
+                        elems[i].click();
+                        return true;
+                    }}
+                }}
+                return false;
+            }})();
+        """)
+
+        if not clicked:
+            raise CDPError(
+                f"Could not find '{btn_text}' button. "
+                "The page structure may have changed."
+            )
+
+        print("[cdp_publish] '下一步' button clicked.")
+        time.sleep(PAGE_LOAD_WAIT)
+
+    def publish_long_article(
+        self,
+        title: str,
+        content: str,
+        image_paths: list[str] | None = None,
+    ) -> list[str]:
+        """
+        Execute the full long article publish workflow:
+        1. Navigate to creator publish page
+        2. Click '写长文' tab
+        3. Click '新的创作' button
+        4. Fill title (textarea)
+        5. Fill content (TipTap editor)
+        6. (Optional) Insert images into editor
+        7. Click '一键排版'
+        8. Wait for templates
+
+        Returns list of available template names for the caller to
+        present to the user for selection.
+
+        Args:
+            title: Article title
+            content: Article body text (paragraphs separated by newlines)
+            image_paths: Optional list of local file paths to images
+        """
+        if not self.ws:
+            raise CDPError("Not connected. Call connect() first.")
+
+        # Step 1: Navigate to publish page
+        self._navigate(XHS_CREATOR_URL)
+        time.sleep(2)
+
+        # Step 2: Click '写长文' tab
+        self._click_long_article_tab()
+
+        # Step 3: Click '新的创作'
+        self._click_new_creation()
+
+        # Step 4: Fill title
+        self._fill_long_title(title)
+
+        # Step 5: Fill content
+        self._fill_content(content)
+
+        # Step 6: Upload images into editor (if provided)
+        if image_paths:
+            print(f"[cdp_publish] Inserting {len(image_paths)} image(s) into editor...")
+            for img_path in image_paths:
+                normalized = img_path.replace("\\", "/")
+                self._evaluate(f"""
+                    (function() {{
+                        var editor = document.querySelector('{SELECTORS["content_editor"]}');
+                        if (!editor) return false;
+                        var img = document.createElement('img');
+                        img.src = 'file:///{normalized}';
+                        editor.appendChild(img);
+                        editor.dispatchEvent(new Event('input', {{ bubbles: true }}));
+                        return true;
+                    }})();
+                """)
+            time.sleep(ACTION_INTERVAL)
+
+        # Step 7: Click '一键排版'
+        self._click_auto_format()
+
+        # Step 8: Wait for templates and return names
+        self._wait_for_templates()
+        template_names = self.get_template_names()
+
+        print(
+            "\n[cdp_publish] Templates loaded.\n"
+            "  Available templates: " + ", ".join(template_names) + "\n"
+        )
+        return template_names
+
+    def click_next_and_prepare_publish(self, content: str = ""):
+        """After user selects a template, click '下一步' and fill the publish page description."""
+        self._click_next_step()
+
+        # The publish page has a separate content editor for the post description
+        if content:
+            time.sleep(ACTION_INTERVAL)
+            self._fill_content(content)
+
+        print(
+            "\n[cdp_publish] Ready to publish.\n"
+            "  Please review in the browser before confirming publish.\n"
+        )
+
+    # ------------------------------------------------------------------
+    # Main publish workflow (image-text mode)
+    # ------------------------------------------------------------------
+
+    def publish(
+        self,
+        title: str,
+        content: str,
+        image_paths: list[str] | None = None,
+    ):
+        """
+        Execute the full publish workflow:
+        1. Navigate to creator publish page
+        2. Click '上传图文' tab
+        3. Upload images (this triggers the editor to appear)
+        4. Fill title
+        5. Fill content
+
+        Args:
+            title: Article title
+            content: Article body text (paragraphs separated by newlines)
+            image_paths: List of local file paths to images to upload
+        """
+        if not self.ws:
+            raise CDPError("Not connected. Call connect() first.")
+
+        if not image_paths:
+            raise CDPError("At least one image is required to publish on Xiaohongshu.")
+
+        # Step 1: Navigate to publish page
+        self._navigate(XHS_CREATOR_URL)
+        time.sleep(2)
+
+        # Step 2: Click '上传图文' tab
+        self._click_image_text_tab()
+
+        # Step 3: Upload images (editor appears after upload)
+        self._upload_images(image_paths)
+
+        # Step 4: Fill title
+        self._fill_title(title)
+
+        # Step 5: Fill content
+        self._fill_content(content)
+
+        print(
+            "\n[cdp_publish] Content has been filled in.\n"
+            "  Please review in the browser before publishing.\n"
+        )
+
+
+
+# ---------------------------------------------------------------------------
+# CLI entry point
+# ---------------------------------------------------------------------------
+
+def main():
+    import argparse
+    from chrome_launcher import ensure_chrome, restart_chrome
+
+    parser = argparse.ArgumentParser(description="Xiaohongshu CDP Publisher")
+    parser.add_argument("--headless", action="store_true",
+                        help="Use headless Chrome (no GUI window)")
+    parser.add_argument("--account", help="Account name to use (default: default account)")
+    sub = parser.add_subparsers(dest="command", required=True)
+
+    # check-login
+    sub.add_parser("check-login", help="Check login status (exit 0=logged in, 1=not)")
+
+    # fill - fill form without clicking publish
+    p_fill = sub.add_parser("fill", help="Fill title/content/images without publishing")
+    p_fill.add_argument("--title", required=True)
+    p_fill.add_argument("--content", default=None)
+    p_fill.add_argument("--content-file", default=None, help="Read content from file")
+    p_fill.add_argument("--images", nargs="+", required=True)
+
+    # publish - fill form and click publish
+    p_pub = sub.add_parser("publish", help="Fill form and click publish")
+    p_pub.add_argument("--title", required=True)
+    p_pub.add_argument("--content", default=None)
+    p_pub.add_argument("--content-file", default=None, help="Read content from file")
+    p_pub.add_argument("--images", nargs="+", required=True)
+
+    # long-article - long article mode
+    p_long = sub.add_parser("long-article", help="Fill long article content with auto-format and template selection")
+    p_long.add_argument("--title", default=None)
+    p_long.add_argument("--title-file", default=None, help="Read title from file")
+    p_long.add_argument("--content", default=None)
+    p_long.add_argument("--content-file", default=None, help="Read content from file")
+    p_long.add_argument("--images", nargs="+", default=None, help="Optional image file paths")
+
+    # select-template - select a template by name
+    p_tpl = sub.add_parser("select-template", help="Select a long article template by name")
+    p_tpl.add_argument("--name", required=True, help="Template name to select")
+
+    # click-next-step - click next step button (for long article after template selection)
+    p_next = sub.add_parser("click-next-step", help="Click '下一步' button after template selection")
+    p_next.add_argument("--content", default=None, help="Post description text")
+    p_next.add_argument("--content-file", default=None, help="Read post description from file")
+
+    # click-publish - just click the publish button on current page
+    sub.add_parser("click-publish", help="Click publish button on already-filled page")
+
+    # login - open browser for QR code login (always headed)
+    sub.add_parser("login", help="Open browser for QR code login (always headed mode)")
+
+    # re-login - clear cookies and re-login the same account (always headed)
+    sub.add_parser("re-login", help="Clear cookies and re-login same account (always headed)")
+
+    # switch-account - clear cookies and open login page (always headed)
+    sub.add_parser("switch-account",
+                   help="Clear cookies and open login page for new account (always headed)")
+
+    # list-accounts - list all configured accounts
+    sub.add_parser("list-accounts", help="List all configured accounts")
+
+    # add-account - add a new account
+    p_add = sub.add_parser("add-account", help="Add a new account")
+    p_add.add_argument("name", help="Account name (unique identifier)")
+    p_add.add_argument("--alias", help="Display name / description")
+
+    # remove-account - remove an account
+    p_rm = sub.add_parser("remove-account", help="Remove an account")
+    p_rm.add_argument("name", help="Account name to remove")
+    p_rm.add_argument("--delete-profile", action="store_true",
+                      help="Also delete the Chrome profile directory")
+
+    # set-default-account - set default account
+    p_def = sub.add_parser("set-default-account", help="Set the default account")
+    p_def.add_argument("name", help="Account name to set as default")
+
+    args = parser.parse_args()
+    headless = args.headless
+    account = args.account
+
+    # Account management commands that don't need Chrome
+    if args.command == "list-accounts":
+        from account_manager import list_accounts
+        accounts = list_accounts()
+        if not accounts:
+            print("No accounts configured.")
+            return
+        print(f"{'Name':<20} {'Alias':<25} {'Default':<10}")
+        print("-" * 55)
+        for acc in accounts:
+            default_mark = "*" if acc["is_default"] else ""
+            print(f"{acc['name']:<20} {acc['alias']:<25} {default_mark:<10}")
+        return
+
+    elif args.command == "add-account":
+        from account_manager import add_account, get_profile_dir
+        if add_account(args.name, args.alias):
+            print(f"Account '{args.name}' added.")
+            print(f"Profile dir: {get_profile_dir(args.name)}")
+            print("\nTo log in to this account, run:")
+            print(f"  python cdp_publish.py --account {args.name} login")
+        else:
+            print(f"Error: Account '{args.name}' already exists.", file=sys.stderr)
+            sys.exit(1)
+        return
+
+    elif args.command == "remove-account":
+        from account_manager import remove_account
+        if remove_account(args.name, args.delete_profile):
+            print(f"Account '{args.name}' removed.")
+        else:
+            print(f"Error: Cannot remove account '{args.name}'.", file=sys.stderr)
+            sys.exit(1)
+        return
+
+    elif args.command == "set-default-account":
+        from account_manager import set_default_account
+        if set_default_account(args.name):
+            print(f"Default account set to '{args.name}'.")
+        else:
+            print(f"Error: Account '{args.name}' not found.", file=sys.stderr)
+            sys.exit(1)
+        return
+
+    # Commands that require Chrome - login/re-login/switch-account always headed
+    if args.command in ("login", "re-login", "switch-account"):
+        headless = False
+
+    if not ensure_chrome(headless=headless, account=account):
+        print("Failed to start Chrome. Exiting.")
+        sys.exit(1)
+
+    publisher = XiaohongshuPublisher()
+    try:
+        if args.command == "check-login":
+            publisher.connect()
+            logged_in = publisher.check_login()
+            if not logged_in and headless:
+                print(
+                    "[cdp_publish] Headless mode: cannot scan QR code.\n"
+                    "  Run with 'login' command or without --headless to log in."
+                )
+            sys.exit(0 if logged_in else 1)
+
+        elif args.command in ("fill", "publish"):
+            content = args.content
+            if args.content_file:
+                with open(args.content_file, encoding="utf-8") as f:
+                    content = f.read().strip()
+            if not content:
+                print("Error: --content or --content-file required.", file=sys.stderr)
+                sys.exit(1)
+
+            publisher.connect()
+            publisher.publish(title=args.title, content=content, image_paths=args.images)
+            print("FILL_STATUS: READY_TO_PUBLISH")
+
+            if args.command == "publish":
+                publisher._click_publish()
+                print("PUBLISH_STATUS: PUBLISHED")
+
+        elif args.command == "long-article":
+            title = args.title
+            if args.title_file:
+                with open(args.title_file, encoding="utf-8") as f:
+                    title = f.read().strip()
+            if not title:
+                print("Error: --title or --title-file required.", file=sys.stderr)
+                sys.exit(1)
+
+            content = args.content
+            if args.content_file:
+                with open(args.content_file, encoding="utf-8") as f:
+                    content = f.read().strip()
+            if not content:
+                print("Error: --content or --content-file required.", file=sys.stderr)
+                sys.exit(1)
+
+            publisher.connect()
+            template_names = publisher.publish_long_article(
+                title=title, content=content, image_paths=args.images,
+            )
+            # Print template names as JSON for programmatic consumption
+            print("TEMPLATES: " + json.dumps(template_names, ensure_ascii=False))
+            print("LONG_ARTICLE_STATUS: TEMPLATE_SELECTION")
+
+        elif args.command == "select-template":
+            publisher.connect(target_url_prefix="https://creator.xiaohongshu.com/publish")
+            if publisher.select_template(args.name):
+                print(f"TEMPLATE_SELECTED: {args.name}")
+            else:
+                print(f"Error: Template '{args.name}' not found.", file=sys.stderr)
+                sys.exit(1)
+
+        elif args.command == "click-next-step":
+            content = getattr(args, 'content', None)
+            if getattr(args, 'content_file', None):
+                with open(args.content_file, encoding="utf-8") as f:
+                    content = f.read().strip()
+            publisher.connect(target_url_prefix="https://creator.xiaohongshu.com/publish")
+            publisher.click_next_and_prepare_publish(content=content or "")
+            print("LONG_ARTICLE_STATUS: READY_TO_PUBLISH")
+
+        elif args.command == "click-publish":
+            publisher.connect(target_url_prefix="https://creator.xiaohongshu.com/publish")
+            publisher._click_publish()
+            print("PUBLISH_STATUS: PUBLISHED")
+
+        elif args.command == "login":
+            # Ensure headed mode for QR scanning
+            restart_chrome(headless=False, account=account)
+            publisher.connect()
+            publisher.open_login_page()
+            print("LOGIN_READY")
+
+        elif args.command == "re-login":
+            # Ensure headed mode, clear cookies, re-open login page for same account
+            restart_chrome(headless=False, account=account)
+            publisher.connect()
+            publisher.clear_cookies()
+            time.sleep(1)
+            publisher.open_login_page()
+            print("RE_LOGIN_READY")
+
+        elif args.command == "switch-account":
+            # Ensure headed mode, clear cookies, open login page
+            restart_chrome(headless=False, account=account)
+            publisher.connect()
+            publisher.clear_cookies()
+            time.sleep(1)
+            publisher.open_login_page()
+            print("SWITCH_ACCOUNT_READY")
+
+    finally:
+        publisher.disconnect()
+
+
+if __name__ == "__main__":
+    main()
diff --git a/skills/post-to-xhs/scripts/chrome_launcher.py b/skills/post-to-xhs/scripts/chrome_launcher.py
new file mode 100644
index 0000000..bf303f4
--- /dev/null
+++ b/skills/post-to-xhs/scripts/chrome_launcher.py
@@ -0,0 +1,296 @@
+"""
+Chrome launcher with CDP remote debugging support.
+
+Manages a dedicated Chrome instance for Xiaohongshu publishing:
+- Detects if Chrome is already listening on the debug port
+- Launches Chrome with a dedicated user-data-dir for login persistence
+- Waits for the debug port to become available
+- Supports headless mode for automated publishing without GUI
+- Supports switching between headless and headed mode (e.g. for login)
+- Supports multiple accounts with separate profile directories
+"""
+
+import os
+import sys
+import time
+import socket
+import subprocess
+import platform
+import signal
+from typing import Optional
+
+CDP_PORT = 9222
+PROFILE_DIR_NAME = "XiaohongshuProfile"
+STARTUP_TIMEOUT = 15  # seconds to wait for Chrome to start
+
+# Track the Chrome process we launched so we can kill it later
+_chrome_process: subprocess.Popen | None = None
+# Track the current account being used
+_current_account: Optional[str] = None
+
+
+def get_chrome_path() -> str:
+    """Find Chrome executable on Windows."""
+    candidates = []
+
+    # Standard install locations
+    for env_var in ("PROGRAMFILES", "PROGRAMFILES(X86)", "LOCALAPPDATA"):
+        base = os.environ.get(env_var, "")
+        if base:
+            candidates.append(os.path.join(base, "Google", "Chrome", "Application", "chrome.exe"))
+
+    for path in candidates:
+        if os.path.isfile(path):
+            return path
+
+    # Fallback: check PATH
+    import shutil
+    found = shutil.which("chrome") or shutil.which("chrome.exe")
+    if found:
+        return found
+
+    raise FileNotFoundError(
+        "Chrome not found. Please install Google Chrome or set its path manually."
+    )
+
+
+def get_user_data_dir(account: Optional[str] = None) -> str:
+    """
+    Return the Chrome profile directory path for a given account.
+
+    Args:
+        account: Account name. If None, uses the default account from account_manager.
+
+    Returns:
+        Path to the Chrome user-data-dir for this account.
+    """
+    try:
+        from account_manager import get_profile_dir
+        return get_profile_dir(account)
+    except ImportError:
+        # Fallback if account_manager not available
+        local_app_data = os.environ.get("LOCALAPPDATA", "")
+        if not local_app_data:
+            local_app_data = os.path.expanduser("~")
+        return os.path.join(local_app_data, "Google", "Chrome", PROFILE_DIR_NAME)
+
+
+def is_port_open(port: int, host: str = "127.0.0.1") -> bool:
+    """Check if a TCP port is accepting connections."""
+    with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
+        s.settimeout(1)
+        try:
+            s.connect((host, port))
+            return True
+        except (ConnectionRefusedError, socket.timeout, OSError):
+            return False
+
+
+def launch_chrome(port: int = CDP_PORT, headless: bool = False, account: Optional[str] = None) -> subprocess.Popen | None:
+    """
+    Launch Chrome with remote debugging enabled.
+
+    Args:
+        port: CDP remote debugging port.
+        headless: If True, launch Chrome in headless mode (no GUI window).
+        account: Account name to use. If None, uses the default account.
+
+    Returns the Popen object if a new process was started, or None if Chrome
+    was already running on the target port.
+    """
+    global _chrome_process, _current_account
+
+    if is_port_open(port):
+        print(f"[chrome_launcher] Chrome already running on port {port}.")
+        return None
+
+    chrome_path = get_chrome_path()
+    user_data_dir = get_user_data_dir(account)
+    _current_account = account
+
+    cmd = [
+        chrome_path,
+        f"--remote-debugging-port={port}",
+        f"--user-data-dir={user_data_dir}",
+        "--no-first-run",
+        "--no-default-browser-check",
+    ]
+
+    if headless:
+        cmd.append("--headless=new")
+
+    mode_label = "headless" if headless else "headed"
+    account_label = account or "default"
+    print(f"[chrome_launcher] Launching Chrome ({mode_label}, account: {account_label})...")
+    print(f"  executable : {chrome_path}")
+    print(f"  profile dir: {user_data_dir}")
+    print(f"  debug port : {port}")
+
+    proc = subprocess.Popen(
+        cmd,
+        stdout=subprocess.DEVNULL,
+        stderr=subprocess.DEVNULL,
+    )
+    _chrome_process = proc
+
+    # Wait for the debug port to become available
+    deadline = time.time() + STARTUP_TIMEOUT
+    while time.time() < deadline:
+        if is_port_open(port):
+            print(f"[chrome_launcher] Chrome is ready on port {port}.")
+            return proc
+        time.sleep(0.5)
+
+    print(
+        f"[chrome_launcher] WARNING: Chrome started but port {port} not responding "
+        f"after {STARTUP_TIMEOUT}s. It may still be initializing.",
+        file=sys.stderr,
+    )
+    return proc
+
+
+def kill_chrome(port: int = CDP_PORT):
+    """
+    Kill the Chrome instance on the given debug port.
+
+    Tries multiple strategies:
+    1. Send CDP Browser.close command via HTTP
+    2. Terminate the tracked subprocess
+    3. Kill by port on Windows (taskkill)
+    """
+    global _chrome_process
+
+    # Strategy 1: CDP Browser.close
+    try:
+        import requests
+        resp = requests.get(f"http://127.0.0.1:{port}/json/version", timeout=2)
+        if resp.ok:
+            ws_url = resp.json().get("webSocketDebuggerUrl")
+            if ws_url:
+                import websockets.sync.client as ws_client
+                ws = ws_client.connect(ws_url)
+                ws.send('{"id":1,"method":"Browser.close"}')
+                try:
+                    ws.recv(timeout=2)
+                except Exception:
+                    pass
+                ws.close()
+                print("[chrome_launcher] Sent Browser.close via CDP.")
+    except Exception:
+        pass
+
+    # Wait briefly for Chrome to shut down
+    time.sleep(1)
+
+    # Strategy 2: Terminate tracked subprocess
+    if _chrome_process and _chrome_process.poll() is None:
+        try:
+            _chrome_process.terminate()
+            _chrome_process.wait(timeout=5)
+            print("[chrome_launcher] Terminated tracked Chrome process.")
+        except Exception:
+            try:
+                _chrome_process.kill()
+            except Exception:
+                pass
+    _chrome_process = None
+
+    # Strategy 3: Windows taskkill by port (fallback)
+    if sys.platform == "win32" and is_port_open(port):
+        try:
+            result = subprocess.run(
+                ["netstat", "-ano"],
+                capture_output=True, text=True, timeout=5
+            )
+            for line in result.stdout.splitlines():
+                if f":{port}" in line and "LISTENING" in line:
+                    pid = line.strip().split()[-1]
+                    subprocess.run(
+                        ["taskkill", "/F", "/PID", pid],
+                        capture_output=True, timeout=5
+                    )
+                    print(f"[chrome_launcher] Killed process {pid} via taskkill.")
+                    break
+        except Exception:
+            pass
+
+    # Wait for port to be released
+    deadline = time.time() + 5
+    while time.time() < deadline:
+        if not is_port_open(port):
+            return
+        time.sleep(0.5)
+
+    if is_port_open(port):
+        print(f"[chrome_launcher] WARNING: port {port} still open after kill attempt.",
+              file=sys.stderr)
+
+
+def restart_chrome(port: int = CDP_PORT, headless: bool = False, account: Optional[str] = None) -> subprocess.Popen | None:
+    """
+    Kill the current Chrome instance and relaunch with the specified mode.
+
+    Useful for switching between headless and headed mode (e.g. when login
+    is needed during a headless session), or switching accounts.
+
+    Args:
+        port: CDP remote debugging port.
+        headless: If True, relaunch in headless mode.
+        account: Account name to use. If None, uses the default account.
+
+    Returns the Popen object for the new Chrome process.
+    """
+    account_label = account or "default"
+    print(f"[chrome_launcher] Restarting Chrome ({'headless' if headless else 'headed'}, account: {account_label})...")
+    kill_chrome(port)
+    time.sleep(1)
+    return launch_chrome(port, headless=headless, account=account)
+
+
+def ensure_chrome(port: int = CDP_PORT, headless: bool = False, account: Optional[str] = None) -> bool:
+    """
+    Ensure Chrome is running with remote debugging on the given port.
+
+    Args:
+        port: CDP remote debugging port.
+        headless: If True, launch in headless mode when starting a new instance.
+            If Chrome is already running, this parameter is ignored.
+        account: Account name to use. If None, uses the default account.
+
+    Returns True if Chrome is available, False otherwise.
+    """
+    if is_port_open(port):
+        return True
+    try:
+        launch_chrome(port, headless=headless, account=account)
+        return is_port_open(port)
+    except FileNotFoundError as e:
+        print(f"[chrome_launcher] Error: {e}", file=sys.stderr)
+        return False
+
+
+def get_current_account() -> Optional[str]:
+    """Get the name of the currently active account."""
+    return _current_account
+
+
+if __name__ == "__main__":
+    import argparse
+    parser = argparse.ArgumentParser(description="Chrome Launcher for CDP")
+    parser.add_argument("--headless", action="store_true", help="Launch in headless mode")
+    parser.add_argument("--kill", action="store_true", help="Kill the running Chrome instance")
+    parser.add_argument("--restart", action="store_true", help="Restart Chrome")
+    parser.add_argument("--account", help="Account name to use (default: default account)")
+    args = parser.parse_args()
+
+    if args.kill:
+        kill_chrome()
+        print("[chrome_launcher] Chrome killed.")
+    elif args.restart:
+        restart_chrome(headless=args.headless, account=args.account)
+        print("[chrome_launcher] Chrome restarted.")
+    elif ensure_chrome(headless=args.headless, account=args.account):
+        print("[chrome_launcher] Chrome is ready for CDP connections.")
+    else:
+        print("[chrome_launcher] Failed to start Chrome.", file=sys.stderr)
+        sys.exit(1)
diff --git a/skills/post-to-xhs/scripts/content.txt b/skills/post-to-xhs/scripts/content.txt
new file mode 100644
index 0000000..bb91837
--- /dev/null
+++ b/skills/post-to-xhs/scripts/content.txt
@@ -0,0 +1,5 @@
+微软向 Canary 通道推送了 Windows 11 Insider Preview Build 28020.1546 更新，补丁编号 KB5074176。
+
+本次更新为常规改进与修复，属于小幅迭代更新，没有重大功能变化。
+
+Canary 通道是 Windows Insider 最前沿的测试分支，适合愿意尝鲜和接受不稳定性的用户。
\ No newline at end of file
diff --git a/skills/post-to-xhs/scripts/image_downloader.py b/skills/post-to-xhs/scripts/image_downloader.py
new file mode 100644
index 0000000..17b6c5b
--- /dev/null
+++ b/skills/post-to-xhs/scripts/image_downloader.py
@@ -0,0 +1,141 @@
+"""
+Image downloader for Xiaohongshu publishing.
+
+Downloads images from URLs to a local temp directory for upload,
+and cleans up after publishing is complete.
+"""
+
+import os
+import sys
+import tempfile
+import shutil
+import uuid
+from urllib.parse import urlparse, unquote
+
+import requests
+
+DEFAULT_TIMEOUT = 30  # seconds per download
+TEMP_DIR_PREFIX = "xhs_images_"
+
+
+class ImageDownloader:
+    """Download images from URLs and manage a temporary directory for them."""
+
+    def __init__(self, temp_dir: str | None = None):
+        if temp_dir:
+            self.temp_dir = temp_dir
+            os.makedirs(self.temp_dir, exist_ok=True)
+            self._owns_dir = False
+        else:
+            self.temp_dir = tempfile.mkdtemp(prefix=TEMP_DIR_PREFIX)
+            self._owns_dir = True
+        self.downloaded_files: list[str] = []
+
+    def _guess_extension(self, url: str, content_type: str | None) -> str:
+        """Guess file extension from URL path or Content-Type header."""
+        # Try URL path first
+        path = urlparse(url).path
+        _, ext = os.path.splitext(unquote(path))
+        if ext and ext.lower() in (".jpg", ".jpeg", ".png", ".gif", ".webp", ".bmp"):
+            return ext.lower()
+
+        # Fall back to Content-Type
+        ct_map = {
+            "image/jpeg": ".jpg",
+            "image/png": ".png",
+            "image/gif": ".gif",
+            "image/webp": ".webp",
+            "image/bmp": ".bmp",
+        }
+        if content_type:
+            for mime, ext in ct_map.items():
+                if mime in content_type:
+                    return ext
+
+        return ".jpg"  # safe default
+
+    def download(self, url: str, referer: str | None = None) -> str:
+        """
+        Download a single image and return the local file path.
+
+        Args:
+            url: Image URL to download
+            referer: Optional Referer header. If None, auto-generates from URL domain.
+
+        Raises requests.RequestException on network errors.
+        """
+        # Build headers with Referer to bypass hotlink protection
+        parsed = urlparse(url)
+        if referer is None:
+            referer = f"{parsed.scheme}://{parsed.netloc}/"
+
+        headers = {
+            "Referer": referer,
+            "User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/120.0.0.0 Safari/537.36",
+        }
+
+        resp = requests.get(url, timeout=DEFAULT_TIMEOUT, stream=True, headers=headers)
+        resp.raise_for_status()
+
+        ext = self._guess_extension(url, resp.headers.get("Content-Type"))
+        filename = f"{uuid.uuid4().hex[:12]}{ext}"
+        filepath = os.path.join(self.temp_dir, filename)
+
+        with open(filepath, "wb") as f:
+            for chunk in resp.iter_content(chunk_size=8192):
+                f.write(chunk)
+
+        self.downloaded_files.append(filepath)
+        print(f"[image_downloader] Downloaded: {url}")
+        print(f"  -> {filepath} ({os.path.getsize(filepath)} bytes)")
+        return filepath
+
+    def download_all(self, urls: list[str]) -> list[str]:
+        """
+        Download multiple images. Returns list of local file paths.
+
+        Skips URLs that fail to download (logs the error, continues).
+        """
+        paths = []
+        for url in urls:
+            try:
+                path = self.download(url)
+                paths.append(path)
+            except Exception as e:
+                print(f"[image_downloader] Failed to download {url}: {e}", file=sys.stderr)
+        return paths
+
+    def cleanup(self):
+        """Remove all downloaded files and the temp directory."""
+        if self._owns_dir and os.path.isdir(self.temp_dir):
+            shutil.rmtree(self.temp_dir, ignore_errors=True)
+            print(f"[image_downloader] Cleaned up temp dir: {self.temp_dir}")
+        else:
+            for f in self.downloaded_files:
+                try:
+                    os.remove(f)
+                except OSError:
+                    pass
+            print(f"[image_downloader] Cleaned up {len(self.downloaded_files)} files.")
+        self.downloaded_files.clear()
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, *_):
+        self.cleanup()
+
+
+if __name__ == "__main__":
+    # Quick test: download URLs passed as command-line arguments
+    if len(sys.argv) < 2:
+        print("Usage: python image_downloader.py <url1> [url2] ...")
+        sys.exit(1)
+
+    dl = ImageDownloader()
+    paths = dl.download_all(sys.argv[1:])
+    print(f"\nDownloaded {len(paths)} image(s):")
+    for p in paths:
+        print(f"  {p}")
+    print(f"Temp dir: {dl.temp_dir}")
+    print("Files will remain until manually cleaned up.")
diff --git a/skills/post-to-xhs/scripts/publish_pipeline.py b/skills/post-to-xhs/scripts/publish_pipeline.py
new file mode 100644
index 0000000..d393efa
--- /dev/null
+++ b/skills/post-to-xhs/scripts/publish_pipeline.py
@@ -0,0 +1,238 @@
+"""
+Unified publish pipeline for Xiaohongshu.
+
+Single CLI entry point that orchestrates:
+  chrome_launcher → login check → image download → form fill → (optional) publish
+
+Usage:
+    # Fill form only (default) - review in browser before publishing
+    python publish_pipeline.py --title "标题" --content "正文" --image-urls URL1 URL2
+    python publish_pipeline.py --title-file t.txt --content-file body.txt --image-urls URL1
+
+    # Headless mode (no GUI window) - faster for automated publishing
+    python publish_pipeline.py --headless --title-file t.txt --content-file body.txt --image-urls URL1
+
+    # Publish to a specific account
+    python publish_pipeline.py --account myaccount --title "标题" --content "正文" --image-urls URL1
+
+    # Fill and auto-publish in one step
+    python publish_pipeline.py --title "标题" --content "正文" --image-urls URL1 --auto-publish
+
+    # Use local image files instead of URLs
+    python publish_pipeline.py --title "标题" --content "正文" --images img1.jpg img2.jpg
+
+    # Long article mode (images optional)
+    python publish_pipeline.py --mode long-article --title "标题" --content "正文"
+    python publish_pipeline.py --mode long-article --title "标题" --content "正文" --images img1.jpg
+
+Exit codes:
+    0 = success (READY_TO_PUBLISH or PUBLISHED)
+    1 = not logged in (NOT_LOGGED_IN) - headless auto-fallback will restart headed
+    2 = error (see stderr)
+"""
+
+import argparse
+import os
+import sys
+
+# Ensure UTF-8 output on Windows consoles
+if sys.platform == "win32":
+    os.environ.setdefault("PYTHONIOENCODING", "utf-8")
+    try:
+        sys.stdout.reconfigure(encoding="utf-8", errors="replace")
+        sys.stderr.reconfigure(encoding="utf-8", errors="replace")
+    except Exception:
+        pass
+
+# Add scripts dir to path so sibling modules can be imported
+SCRIPT_DIR = os.path.dirname(os.path.abspath(__file__))
+if SCRIPT_DIR not in sys.path:
+    sys.path.insert(0, SCRIPT_DIR)
+
+from chrome_launcher import ensure_chrome, restart_chrome
+from cdp_publish import XiaohongshuPublisher, CDPError
+from image_downloader import ImageDownloader
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="Xiaohongshu publish pipeline - unified entry point"
+    )
+
+    # Title
+    title_group = parser.add_mutually_exclusive_group(required=True)
+    title_group.add_argument("--title", help="Article title text")
+    title_group.add_argument("--title-file", help="Read title from UTF-8 file")
+
+    # Content
+    content_group = parser.add_mutually_exclusive_group(required=True)
+    content_group.add_argument("--content", help="Article body text")
+    content_group.add_argument("--content-file", help="Read content from UTF-8 file")
+
+    # Mode
+    parser.add_argument(
+        "--mode",
+        choices=["image-text", "long-article"],
+        default="image-text",
+        help="Publish mode: 'image-text' (default) or 'long-article'",
+    )
+
+    # Images (required for image-text, optional for long-article)
+    img_group = parser.add_mutually_exclusive_group(required=False)
+    img_group.add_argument(
+        "--image-urls", nargs="+", help="Image URLs to download"
+    )
+    img_group.add_argument(
+        "--images", nargs="+", help="Local image file paths"
+    )
+
+    # Publish mode
+    parser.add_argument(
+        "--auto-publish",
+        action="store_true",
+        default=False,
+        help="Click publish button after filling (default: fill only)",
+    )
+
+    # Headless mode
+    parser.add_argument(
+        "--headless",
+        action="store_true",
+        default=False,
+        help="Run Chrome in headless mode (no GUI). Auto-falls back to headed if login is needed.",
+    )
+
+    # Optional temp dir for downloaded images
+    parser.add_argument(
+        "--temp-dir",
+        default=None,
+        help="Directory for downloaded images (default: auto-created temp dir)",
+    )
+
+    # Account selection
+    parser.add_argument(
+        "--account",
+        default=None,
+        help="Account name to publish to (default: default account)",
+    )
+
+    args = parser.parse_args()
+    headless = args.headless
+    account = args.account
+
+    # --- Resolve title ---
+    if args.title_file:
+        with open(args.title_file, encoding="utf-8") as f:
+            title = f.read().strip()
+    else:
+        title = args.title
+
+    if not title:
+        print("Error: title is empty.", file=sys.stderr)
+        sys.exit(2)
+
+    # --- Resolve content ---
+    if args.content_file:
+        with open(args.content_file, encoding="utf-8") as f:
+            content = f.read().strip()
+    else:
+        content = args.content
+
+    if not content:
+        print("Error: content is empty.", file=sys.stderr)
+        sys.exit(2)
+
+    # --- Step 1: Ensure Chrome is running ---
+    mode_label = "headless" if headless else "headed"
+    account_label = account or "default"
+    print(f"[pipeline] Step 1: Ensuring Chrome is running ({mode_label}, account: {account_label})...")
+    if not ensure_chrome(headless=headless, account=account):
+        print("Error: Failed to start Chrome.", file=sys.stderr)
+        sys.exit(2)
+
+    # --- Step 2: Connect and check login ---
+    print("[pipeline] Step 2: Checking login status...")
+    publisher = XiaohongshuPublisher()
+    try:
+        publisher.connect()
+        logged_in = publisher.check_login()
+        if not logged_in:
+            publisher.disconnect()
+            if headless:
+                # Auto-fallback: restart Chrome in headed mode for QR login
+                print("[pipeline] Headless mode: not logged in. Switching to headed mode for login...")
+                restart_chrome(headless=False, account=account)
+                publisher.connect()
+                publisher.open_login_page()
+            print("NOT_LOGGED_IN")
+            sys.exit(1)
+    except CDPError as e:
+        print(f"Error: {e}", file=sys.stderr)
+        sys.exit(2)
+
+    # --- Step 3: Prepare images ---
+    image_paths = []
+    downloader = None
+
+    if args.image_urls:
+        print(f"[pipeline] Step 3: Downloading {len(args.image_urls)} image(s)...")
+        downloader = ImageDownloader(temp_dir=args.temp_dir)
+        image_paths = downloader.download_all(args.image_urls)
+        if not image_paths:
+            print("Error: All image downloads failed.", file=sys.stderr)
+            sys.exit(2)
+    elif args.images:
+        image_paths = args.images
+        # Verify local files exist
+        for p in image_paths:
+            if not os.path.isfile(p):
+                print(f"Error: Image file not found: {p}", file=sys.stderr)
+                sys.exit(2)
+        print(f"[pipeline] Step 3: Using {len(image_paths)} local image(s).")
+    elif args.mode == "image-text":
+        print("Error: Images are required for image-text mode. Use --image-urls or --images.", file=sys.stderr)
+        sys.exit(2)
+    else:
+        print("[pipeline] Step 3: No images (optional for long-article mode).")
+
+    # --- Step 4: Fill form ---
+    print("[pipeline] Step 4: Filling form...")
+    try:
+        if args.mode == "long-article":
+            publisher.publish_long_article(
+                title=title,
+                content=content,
+                image_paths=image_paths or None,
+            )
+            print("LONG_ARTICLE_STATUS: TEMPLATE_SELECTION")
+        else:
+            publisher.publish(title=title, content=content, image_paths=image_paths)
+            print("FILL_STATUS: READY_TO_PUBLISH")
+    except CDPError as e:
+        print(f"Error during form fill: {e}", file=sys.stderr)
+        if downloader:
+            downloader.cleanup()
+        sys.exit(2)
+
+    # --- Step 5: Publish (optional, image-text mode only) ---
+    if args.auto_publish and args.mode == "image-text":
+        print("[pipeline] Step 5: Clicking publish button...")
+        try:
+            publisher._click_publish()
+            print("PUBLISH_STATUS: PUBLISHED")
+        except CDPError as e:
+            print(f"Error clicking publish: {e}", file=sys.stderr)
+            if downloader:
+                downloader.cleanup()
+            sys.exit(2)
+
+    # --- Cleanup ---
+    publisher.disconnect()
+    if downloader:
+        downloader.cleanup()
+
+    print("[pipeline] Done.")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/skills/post-to-xhs/scripts/title.txt b/skills/post-to-xhs/scripts/title.txt
new file mode 100644
index 0000000..86da669
--- /dev/null
+++ b/skills/post-to-xhs/scripts/title.txt
@@ -0,0 +1 @@
+Win11 Build 28020 Canary通道更新
\ No newline at end of file