News Intelligence Radar

Problem

When technology, finance, and AI news arrive all at once, the bottleneck is usually not information volume but judgment: which items require action, which are duplicate reports, and which are irrelevant. News Intelligence Radar turns raw reports into prioritized decision input instead of a simple chronological news list.

How It Works

It ingests public sources such as Hacker News, GitHub Trending, 36Kr, Eastmoney, and ArXiv AI, then scores items using coverage across sources, heat, user-profile fit, and source credibility. Results are grouped into 🔴 Critical, 🟡 Watch, ⚪ Optional, and 🔇 Noise, making it easier to filter duplicates, marketing, and unrelated items.
- Semantic deduplication: multiple reports on the same event are merged while preserving different angles.
- User-profile adaptation: on first use it reads USER.md and MEMORY.md to extract focus areas; if none exist, it falls back to defaults such as AI, tech, and finance.
- Decision brief: important items include impact judgment and suggested next steps.
- Topic tracking integration: a tracking request can extract keywords and generate a tracking command template for ongoing monitoring.

Boundaries

The skill depends on requests and beautifulsoup4, with most other work done via the Python standard library. Some sources may fail due to network conditions, Cloudflare protection, or API changes; the system skips failed sources and continues. Stability for Weibo, Zhihu, and Wall Street CN depends on external access at runtime.