English | 中文

WeChat Article Pipeline

A Claude Code skill that handles the full production pipeline for WeChat public account articles — from raw input to a publish-ready package delivered to your phone via Telegram.

Philosophy: You spend your time on judgment. The pipeline handles research, writing, formatting, and delivery.

What You Get

For each article, the skill produces:

• 📄 Full article draft — structured WeChat-format Markdown, with sourced data
• 🐦 X/Twitter thread — 5 tweets in Zara Zhang "Party not Stage" style
• 🖥 Multi-channel HTML preview — WeChat / X / Xiaohongshu / Zhishi Xingqiu in one page
• 📦 Source folder — every claim traced to government docs, official earnings, or primary reporting
• 📱 Telegram delivery — full package pushed to your phone, ready to copy-paste and publish

How It Works

1. You send a raw input — a tweet link, podcast note, AI conversation excerpt, or book quote
2. The skill identifies the source type and framework tag automatically
3. Research phase: Tavily searches verify every number against primary sources
4. Article written in a fixed structure: hook → story → hidden angle → framework analysis → philosophical note → 3 questions for bosses → CTA
5. X thread written in Zara Zhang style: join conversations, don't lecture
6. HTML preview generated for all four channels
7. Everything pushed to your Telegram — you review, copy, publish

Input Types

The skill is built around a "four-quarter time allocation" framework:

Source	Signal	Output
Twitter/X tweet	Link or screenshot	Research-backed article with hidden angle
Podcast note	Show name + episode	Core insight → business decision language
AI dialogue	Description of your Claude/GPT conversation	Institutional AI framework (unique content)
Book note	Title + chapter + excerpt	Eastern philosophy × AI governance (moat content)

Quick Start

1. Install

git clone https://github.com/flychicken067/ivan-article.git ~/.claude/skills/ivan-article
cd ~/.claude/skills/ivan-article
pip install requests

2. Set Up Telegram (optional but recommended)

cp .env.example .env
# Fill in TELEGRAM_BOT_TOKEN and PROJECT_ROOT
python3 telegram/get_chat_id.py   # sends your chat_id after you message your bot
python3 telegram/send.py --test   # verify connection

To create a Telegram bot: message @BotFather → /newbot

3. Use

In Claude Code:

/ivan-article

Here's today's input:
[paste tweet / podcast note / AI dialogue / book excerpt]

Or invoke directly:

今天的素材是：[内容]

Output Structure

The skill writes to your project directory:

output/
├── articles/
│   ├── article_08_medvi.md               ← Full article
│   ├── article_08_medvi_preview.html     ← Multi-channel preview
│   └── article_08_sources/
│       ├── 00_INDEX.md                   ← Source index + verification table
│       ├── 01_primary_source.md
│       └── ...
├── tweets/
│   └── article_08_thread.md              ← X thread (5 tweets)
└── publish-log.md                        ← Running record of all articles

Source Verification Standards

Every number in every article must trace to a ⭐⭐⭐ or higher source:

Rating	Source Type
⭐⭐⭐⭐⭐	Government documents, official earnings reports, primary journalism, peer-reviewed papers
⭐⭐⭐⭐	Law firm analyses, specialist trade press
⭐⭐⭐	Major business media, first-person podcast quotes
⭐⭐	Secondary summaries, personal blogs
⭐	Unverifiable — flagged ⚠️, not used

Article Structure

Every article follows the same skeleton:

1. Hook (3–5 lines) — one fact that makes the reader need the next sentence
2. Story — enough detail to make it real
3. Hidden angle — what the mainstream coverage missed, backed by documents
4. Framework analysis — Three Bills (维护税/聚焦税/版权税) or organizational architecture
5. Philosophical note — Tao Te Ching / Sun Tzu × AI governance (≤100 words)
6. Three questions for bosses — not suggestions, interrogations
7. CTA — fixed format
8. Sources — numbered list

X Thread Style

Built on Zara Zhang's "Party not Stage" framework:

• No numbering (not 1/5, 2/5...)
• Enter existing conversations, don't start lectures
• Personal voice ("In enterprise AI consulting, the question I get most is...")
• Each tweet stands alone
• Link only in the last tweet

Requirements

• Claude Code (or compatible agent)
• Python 3.8+ with requests library
• Telegram bot (optional, for mobile delivery)
• Tavily MCP or web search capability for research phase

Configuration

# .env
TELEGRAM_BOT_TOKEN=your_token_here
TELEGRAM_CHAT_ID=your_chat_id_here
PROJECT_ROOT=/path/to/your/project

Telegram Scripts

Script	Purpose
telegram/send.py --article 08	Push full article package to phone
telegram/send.py --test	Verify connection
telegram/send.py "message"	Send any text
telegram/listen.py	Listen for inputs from your phone
telegram/get_chat_id.py	One-time setup: get your chat ID

Sample Output

See examples/article_09_sample.md for a complete real output — Article 09 produced by this skill in a single session.

What's included in that example:

• Full 6-section article (李圆方 numbered style)
• Complete 5-tweet X thread (Zara Zhang "Party not Stage")
• Phase-by-phase production log showing exactly what the skill did

Stats: Article 08 from the same pipeline: 88 reads, 10 forwards (historical high).

Article hook from Article 08:

Matthew Gallagher, 41, launched a company from his living room in September 2024. $20,000. Two months. Zero employees.

2025 full-year revenue: $401 million. Net margin: 16.2%.

The New York Times published a glowing profile on April 2, 2026.

Six weeks earlier, the FDA had sent a formal warning letter.

Article 09 opening (full example in /examples):

最近国内 Claude Code 圈子里最火的两件事，都跟"蒸馏人"有关。

女娲.skill 自己在文档里写了一句话：「蒸馏不了直觉。那种面对从没见过的问题时的'我觉得这个方向是对的'，不在文字里。」

这句话是今天这篇文章的起点。

License

MIT

---

Built by @flychicken067