Full-process automation from trend collection to content generation to multi-platform publishing. Daniel only needs 2 minutes a day to select topics.
Core Workflow
热点采集(10+平台) → AI选题评分 → 推送Top10给用户
↓
用户选择主题(或自定义)
↓
多平台内容生成(LLM)
↓
草稿审核 → 确认发布
↓
自动发布 → 数据追踪
Use Cases
✅ USE when:
- "What are today's trends" / "Collect today's trending topics"
- "Help me generate content" / "Write a Xiaohongshu article"
- "Topic scoring" / "Recommend what to write today"
- "Publish to Xiaohongshu/WeChat/Twitter"
- "Run the Content Factory pipeline" / "Run the full process once"
- "Check today's drafts" / "Review content"
❌ DON'T use when:
- GEO optimization (AI search ranking) → use geo-agent
- Pure SEO keyword optimization → use the SEO skill
- One-off writing (no pipeline needed) → use content-creator
Data Directory
data/
├── hotpool/ # 每日热点池 (YYYY-MM-DD.json)
├── topics/ # 评分选题 (YYYY-MM-DD.json)
├── drafts/ # 生成草稿 (YYYY-MM-DD/)
├── reviewed/ # 审核通过
├── published/ # 已发布记录
├── config/ # 运行配置
│ └── sources.json # 采集源配置
├── templates/ # 平台模板
│ ├── xiaohongshu.md
│ ├── wechat.md
│ └── twitter.md
└── assets/ # 图片等素材
Script Description
| Script | Function | Dependency |
|---|---|---|
scripts/aggregator/fetch_all.py |
10+ platform trend collection | curl, python3 |
scripts/topic_scorer.py |
AI topic scoring (Top10) | LLM API (DeepSeek/GLM) |
scripts/content_generator.py |
Multi-platform content generation | LLM API |
scripts/draft_reviewer.py |
Draft review push | Telegram API |
scripts/auto_publisher.py |
Automatic publishing | playwright (optional) |
scripts/topic_presenter.py |
Topic card push | Telegram API |
scripts/run_daily.sh |
Full-process chaining | bash |
scripts/paths.py |
Path configuration (portable) | - |
Collection Sources (10+)
| Platform | Method | Content Type |
|---|---|---|
| Bilibili Trending | API | Videos/Updates |
| GitHub Trending | API | Open-source projects |
| API | Discussions/News | |
| YouTube | API | Videos |
| Weibo Hot Search | API | Social trends |
| Zhihu Hot List | API | In-depth discussions |
| Toutiao | API | News and information |
| Douyin | API | Short videos |
| Twitter/X | Syndication | KOL updates |
| LinuxDo | API | Tech communities |
Install Dependencies
cd ~/clawd/skills/content-factory
pip install -r requirements.txt
Full Pipeline
# 手动执行全流程
bash scripts/run_portable.sh
# 或分步执行
python3 scripts/aggregator/fetch_all.py # Step 1: 热点采集
python3 scripts/topic_scorer.py # Step 2: AI评分
python3 scripts/topic_presenter.py # Step 3: 推送选题
python3 scripts/content_generator.py --top 3 # Step 4: 内容生成
python3 scripts/draft_reviewer.py --all # Step 5: 草稿审核
# python3 scripts/auto_publisher.py # Step 6: 发布(需确认)
Working with Ralph CEO Loop
The Content Factory can iterate continuously through the Ralph CEO Loop:
- Xiaodata: trend collection + data cleaning
- Xiaoresearch: in-depth topic research
- Xiaocontent: content generation + copy optimization
- Xiaopm: process coordination + quality acceptance
- Xiaomarket: publishing strategy + channel optimization
Configuration
Collection source configuration (data/config/sources.json)
You can enable/disable each platform and configure followed accounts, channels, etc.
LLM Configuration
Scoring and generation use an OpenAI-compatible API:
- Default: DeepSeek (scoring) + GLM-5 (generation)
- Via environment variable
LLM_API_KEY/LLM_BASE_URLoverride