mcpField
X
xhs-batch

xhs-batch

github.com/chenxiachanProductivity42758 installsUpdated 11w ago
Repository

About xhs-batch

xhs-batch is a skill for extracting multiple Xiaohongshu posts from a supplied list of links. It follows the xhs extraction flow for each post, transcribes video speech, formats the result, and saves Markdown files in an Obsidian vault.

Key features

  1. 01Parses multiple post links
  2. 02Checks for the required cookies file
  3. 03Extracts post content and video transcripts
  4. 04Formats and saves Markdown notes
  5. 05Reports successful and failed extractions

Use cases

  1. 01Batch saving Xiaohongshu posts to Obsidian
  2. 02Transcribing video posts from a link list
  3. 03Reviewing extraction results across several posts

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/chenxiachan/xhs-claude-skills --skill xhs-batch

After installing, type/xhs-batchin Claude Code to confirm it loaded.

Details

From the project README

用户希望批量提取多个小红书帖子。请按以下步骤处理:

常量定义

  • Cookies 文件: ~/cookies.json
  • Obsidian 保存目录: ~/Documents/Obsidian Vault/xhs
  • Whisper 模型: mlx-community/whisper-large-v3-turbo

输入

用户提供的链接列表: $ARGUMENTS

流程

步骤 1:解析链接

从输入中提取所有小红书链接(支持多行、空格分隔、逗号分隔)。 每个链接提取帖子 ID 和 xsec_token。

步骤 2:检查 Cookies

检查 ~/cookies.json 是否存在。如不存在,按 /xhs 的步骤 0 引导用户导出。

步骤 3:逐个提取

对每个链接,执行 /xhs 的完整提取流程(步骤 2-4):

  • 请求页面 → 解析 INITIAL_STATE
  • 视频帖子做语音转录
  • 按 Peter Thiel 风格整理
  • 保存为 {YYYY-MM-DD} {短标题}.md

每个帖子之间间隔 3 秒,避免触发反爬。

步骤 4:汇总报告

全部完成后,输出简短汇总:

  • 成功/失败数量
  • 每个帖子的文件名和一句话摘要

Comments

Markdown supported