Xiaohongshu Operations Split Into Five Skills; Commenters Shouted 'Account Ban' First
Someone on GitHub has strung topic discovery, viral-post breakdown, cover creation, image cards, and publishing into an AI pipeline. The toolchain is impressively complete, but there’s also plenty of account-ban feedback. First, look at which steps are worth using, and which step you shouldn’t test with your main account.
Someone on GitHub has split Xiaohongshu operations into an AI Skill chain: finding topics, breaking down viral posts, deciding covers, making cards, and publishing. The main post lists five repos, with star counts ranging from 1.5k to 107k. Commenters didn’t rush to bookmark it; they first asked: would you dare use it? My account got banned the next day.
First, some background. A Skill can be understood as a set of operating instructions and a toolkit for an AI Agent, not a standalone app. MCP is a protocol that lets models like Claude call external tools. Some of the repos below are Skills, and some are MCP services; all of them automate a segment of the Xiaohongshu operations workflow.
Finding Topics: Agent-Reach
Agent-Reach solves the problem of CLI Agents not being able to reach logged-in platforms, video subtitles, and Chinese platforms. It supports Bilibili, V2EX, GitHub, YouTube, and web pages, with optional Xiaohongshu, X, Reddit, Boss Zhipin, Xueqiu, etc. It is read-only: no publishing, commenting, or liking. For platforms that require login, it uses a dedicated browser; the first time, it copies your login for that site from Chrome, and it’s recommended to use a secondary account. To install, just send the Agent the address of install.md in the repo, and it will run the rest itself. Note: the package named agent-reach on PyPI is not this project—don’t install the wrong one.
The star-count controversy also needs mentioning. The main post says Agent-Reach has about 107k stars; commenters pointed out that the real star count is under 10k, and that 107k counts the entire GitHub star pool. Treat this number as unverified for now. Agent-Reach itself is read-only, so the risk is low, and it’s suitable as an information entry point.
Breaking Down Viral Posts and Diagnosing Accounts: xiaohongshu-ops-skill
This Skill works with Openclaw and can analyze the home recommendation feed, account positioning, topic inspiration, knowledge base, auto-publishing, auto-replies, downloading target notes, and replicating viral posts. It is based on browser automation CDP; scan the QR code to log in the first time, with no repeated verification afterward.
The repo demonstrates Feishu-based auto-publishing and auto-replying to comments.
Viral-post replication lets you enter a link, analyze the viral factors, and generate a similar note, including images and text.
The most useful parts are account diagnosis and recommendation-feed analysis. A commenter said breaking down viral posts can also diagnose why your own account’s data is poor; this step is the most practical. Auto-publishing and auto-replies are in the same Skill, and that’s where the risk is too. Someone said outright that their account was banned the next day.
Cover Design: xhs-visual-director-skill
A visual director that plans and generates Xiaohongshu image-text posts. It first asks 10 questions, then gives three style options, a 6–8 page image-text structure, and a unified visual template; it generates one sample image first, and after confirmation generates the whole set. Suitable for people who can write but can’t make images.
The repo emphasizes avoiding a cheap AI-template feel; covers are 3:4 and readable on mobile. Netizens asked about the image-generation success rate; the repo doesn’t provide data, so you’ll have to run it yourself with a topic. This Skill doesn’t touch publishing, so the risk is low.
Making Cards: baoyu-xhs-images
The repo has 27k stars; it splits an article into 1–10 images. It solves the manual labor of turning long articles into Xiaohongshu image-text posts.
Publishing: xiaohongshu-mcp
16.7k stars; connect it to Claude and you can publish directly. But commenters report having been banned long ago. This is the highest-risk step in the whole chain. Xiaohongshu has platform rules for automated publishing, and accounts may be restricted. If you want to try it, use a secondary account. Don’t use your main account.
Verdict
The truly valuable part of this chain is the first half: finding topics, breaking down viral posts, diagnosing accounts, and visual planning. It pulls people out of information gathering and layout. The second half—publishing and auto-replies—takes on risk for your own account. One netizen said treating AI as an outsourced contractor is fine, but treating it as a stand-in is not. Another said the chain is all packaged up, and the remaining difficulty isn’t whether you know how to use it—it’s whether you have anything worth being promoted. Tools solve process, not content. The persona and real experience behind a Xiaohongshu account are what drive repeat engagement. AI can imitate viral-post structures, but it can’t imitate why you deserve attention. If you want to try it, run the first half first, and use a secondary account for publishing. Don’t hand your main account to an intern with very fast hands.
发布时间: 2026-11-05 23:32