Wewrite Wechat Ai Publishing

# WeWrite — WeChat AI Publishing Skill

> Skill by [ara.so](https://ara.so) — Daily 2026 Skills collection.

WeWrite is a full-pipeline AI skill for producing WeChat Official Account (公众号) articles end-to-end: hotspot fetching → topic selection → writing → SEO → AI image generation → formatting → draft box publishing. It runs as a Claude Code skill (via `SKILL.md`) but every component also works standalone.

---

## Installation

### As a Claude Code Skill

```bash
# Clone the repo
git clone https://github.com/oaker-io/wewrite.git ~/.claude/skills/wewrite

# Or copy into an existing project
cp -r wewrite ~/.claude/skills/wewrite
```

### Python Dependencies

```bash
cd wewrite
pip install -r requirements.txt
```

### Configuration

```bash
cp config.example.yaml config.yaml
```

Edit `config.yaml`:

```yaml
wechat:
  appid: "${WECHAT_APPID}"         # WeChat Official Account App ID
  secret: "${WECHAT_SECRET}"       # WeChat Official Account Secret

image_gen:
  provider: "doubao"               # "doubao" or "openai"
  doubao_api_key: "${DOUBAO_API_KEY}"
  openai_api_key: "${OPENAI_API_KEY}"

output_dir: "./output"
```

Set environment variables instead of hardcoding secrets:

```bash
export WECHAT_APPID="wx1234567890abcdef"
export WECHAT_SECRET="your_secret_here"
export DOUBAO_API_KEY="your_doubao_key"
export OPENAI_API_KEY="sk-..."
```

---

## Triggering the Full Pipeline

Once installed as a Claude Code skill, simply say:

```
用 demo 的配置写一篇公众号文章
```

Claude will execute all steps automatically using `clients/demo/style.yaml` as the client profile.

You can also specify a client:

```
用 clients/tech-blog 的风格，围绕今日热点写一篇公众号文章，发布到草稿箱
```

---

## Pipeline Steps & Scripts

### 1. Fetch Hotspots

Scrapes real-time trending topics from Weibo, Toutiao, and Baidu.

```bash
python3 scripts/fetch_hotspots.py --limit 20
python3 scripts/fetch_hotspots.py --limit 10 --json   # JSON output
```

Output example:
```json
[
  {"rank": 1, "title": "DeepSeek R2 发布", "source": "weibo", "heat": 98200},
  {"rank": 2, "title": "A股科技板块大涨", "source": "baidu", "heat": 75300}
]
```

### 2. SEO Keyword Analysis

Queries Baidu and 360 search suggestions to score keywords.

```bash
python3 scripts/seo_keywords.py "AI大模型" "科技股"
python3 scripts/seo_keywords.py --json "ChatGPT" "人工智能"
```

Python usage:

```python
from scripts.seo_keywords import analyze_keywords

results = analyze_keywords(["AI大模型", "大语言模型", "GPT-5"])
for kw in results:
    print(f"{kw['keyword']}: score={kw['score']}, volume={kw['estimated_volume']}")
```

### 3. Topic Selection

Claude reads `references/topic-selection.md` and generates 10 candidate topics scored on:
- **热度** (trending heat)
- **契合度** (client fit)
- **差异化** (differentiation)

### 4. Framework & Writing

Claude reads `references/frameworks.md` (5 frameworks) and `references/writing-guide.md` (de-AI style rules), then writes the article adapted to the client's tone.

### 5. AI Image Generation

```bash
# Cover image (recommended 900×383)
python3 toolkit/image_gen.py \
  --prompt "科技感封面，蓝色光线，未来感" \
  --output output/cover.png \
  --size cover

# Inline content image
python3 toolkit/image_gen.py \
  --prompt "程序员在办公室工作，现代风格插画" \
  --output output/img1.png \
  --provider openai
```

Python usage:

```python
from toolkit.image_gen import generate_image

path = generate_image(
    prompt="AI机器人与人类握手，科技感插画",
    output_path="output/cover.png",
    size="cover",          # "cover" (900x383) or "content" (800x600)
    provider="doubao"      # "doubao" or "openai"
)
print(f"Generated: {path}")
```

### 6. Formatting — Markdown → WeChat HTML

WeChat requires inline styles. The convert