How to use the WeChat Article Writing Team agent skill

What is an agent skill and what does this one do?

An agent skill is a folder containing a SKILL.md file that tells an AI coding agent how to handle a specific kind of task. When the skill is installed and matches your request, the agent follows the instructions in that file.

This skill, named wechat-article, is generated from the source file skills/wechat-article.md in the ai-berkshire repository. It works in four stages: the agent first researches the topic with parallel sub-agents, then drafts the article in an 'author' role, then runs separate 'editor' and 'reader' reviews to tighten structure and check readability, and finally produces a markdown file. The skill also covers formatting rules such as LaTeX formulas, plain-language explanations, and extracting figures from research paper PDFs.

• Research: parallel research sub-agents gather material, with web search where needed.
• Draft: an author role writes the long-form article.
• Review: editor and reader roles review the draft in parallel.
• Output: a final markdown file, with formulas in LaTeX and figures extracted from PDFs where the topic is a paper.

What you need before installing

The skill is not standalone. The source says it depends on the ai-berkshire repository: it calls shared tools in the repository's tools/ folder (for example tools/financial_rigor.py) and expects the rules in AGENTS.md. If you do not have that checkout, the workflow will not run as intended.

The skill also expects an agent with sub-agent and shell support, web search access for the research stages, and shell tools such as pdftoppm and Python with PIL for figure extraction.

• Claude Code or a Codex-style agent with sub-agent and shell support.
• The ai-berkshire repository checkout, including its tools/ folder and AGENTS.md.
• Web search access for research.
• pdftoppm and Python with PIL, only needed if you extract figures from PDFs.

Install the wechat-article skill

Download the ZIP for the skill from this page and unzip it. The skill folder contains SKILL.md. Then place it in one of the following locations.

For a personal install available across your projects, put the unzipped folder in ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ folder.

If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/ (the skill folder with SKILL.md inside).
• Project only: .claude/skills/ inside your project folder.
• Claude apps: Settings → Capabilities → Skills, then upload.

How to trigger the skill with a natural request

You do not need a special command. Describe the topic in Chinese and ask for a WeChat public-account article. The source gives examples of topic inputs such as 大模型OPD技术解读, Qwen3技术报告解读 and 为什么巴菲特不买科技股.

The agent will ask you to confirm details if you leave them out. Its defaults are a reader with some technical background but not expert knowledge, a medium depth with formulas explained clearly, a length of 3000-4000 Chinese characters, a conversational style, and PDF and figure download turned on.

• Example: 帮我写一篇公众号文章：大模型OPD技术解读
• Example: 做一篇 Qwen3技术报告解读 的公众号文章，需要从论文里提取图表
• Example: 为什么巴菲特不买科技股，写成公众号长文，中深度

What the output looks like and where it is saved

The final result is a markdown file. The source sets file naming and storage rules: technology topics go in reports/AI产业研究/ and are named 公众号-{主题关键词}-{YYYYMMDD}.md. Investment topics go in reports/{公司名}/ with the name {公司名}-公众号-{YYYYMMDD}.md. General topics go in reports/ with the same pattern as technology topics.

For paper-based articles, figures are cropped from the PDF and stored under assets/{主题简称}/, and referenced with markdown image syntax. Formulas use LaTeX, with $...$ inline and $$...$$ for display.

• Technology: reports/AI产业研究/公众号-{主题关键词}-{YYYYMMDD}.md
• Investment: reports/{公司名}/{公司名}-公众号-{YYYYMMDD}.md
• General: reports/公众号-{主题关键词}-{YYYYMMDD}.md
• Figures: assets/{主题简称}/fig{序号}-{描述}.png

Safety notes: scripts and network access

This skill runs scripts and uses the network. Its workflow involves shell commands, research through web search, and rendering PDF pages to images. Review the commands the agent proposes before approving them, especially those that write files or download content.

The source states that the agent should check the research data itself. It asks for sourced data and says that figures without a source should be labelled as estimates. The skill does not remove the need for fact-checking.

• The skill runs shell commands and performs web searches.
• Check the output for data errors or estimates before you publish.
• Only install and run the skill from a source you trust, and review the repository's tools/ scripts it calls.

Tips for better results

Give the agent a specific topic and, if you have one, a paper or document to work from. The source says paper-based explainers are where figure extraction applies. Confirm the reader and length up front, because the defaults only apply if you do not change them.

Expect a review cycle. The workflow is designed so that the editor and reader feedback shape the final draft, with the source saying reader feedback wins when editor and reader disagree.

• State the reader, depth and length in your request.
• For paper explainers, give the paper name or PDF so figures can be extracted.
• Verify every number before publishing, since the source asks for sourced data but does not guarantee it.