Real-Time News Aggregator

Problem

News, papers, podcasts, and essays are spread across sites, making it hard to compare hot items, sources, timestamps, and heat in one consistent format. Asking an LLM to summarize each item directly can introduce hallucinated details, missing timestamps, and inconsistent handling of proper nouns. news-aggregator-skill-v2 targets this information-aggregation workflow by turning multi-source fetching, templated writing, and report persistence into a fixed process.

How It Works

The core loop is straightforward:
- Use fetch_news.py to retrieve items from hackernews, github, weibo, huggingface, ai_newsletters, and 28 supported sources.
- Format every item with the unified report template in Simplified Chinese, preserving key fields such as time, heat, links, and analysis.
- Apply source-specific rules, such as keeping Hacker News discussion links, GitHub stars, Hugging Face upvotes, or exact Weibo heat text.
- Save the report under reports/YYYY-MM-DD/, then display the full content.
It also includes daily_briefing.py profiles like general, finance, tech, social, and ai_daily, which are useful for recurring briefings. The rules emphasize using only the fetched JSON data, keeping time mandatory, and marking supplemental items with ⚠️ when the original result set is too small.

Boundaries

This is closer to a fetch-and-summarize pipeline than a news database or real-time event tracker. Output is fixed to Simplified Chinese. Some sources, such as HF Papers and Ben's Bites, require an additional browser environment. For podcasts, essays, and newsletters, analysis quality depends on how complete the retrieved text is; if source data lacks time or context, the report can only mark it as Unknown Time or stay conservative based on available fields.