AI Tech News Briefing

Problem

Raw AI news feeds often mix stale posts, clickbait, PR copy, and duplicate reports across platforms. Keeping a daily briefing current, deduplicated, and source-transparent is hard when search results are noisy.

How It Works

The skill runs a fixed pipeline from source retrieval to ranked output:
- Two-stage retrieval: it first uses scripts/fetch_rss.js to aggregate the 36Kr public feed and RSS sources such as TMTPost, Leiphone, Sspai, Zhidongxi, and IT Home; web_search is used only as a fallback when coverage is insufficient.
- Strict time windows: the morning brief covers the previous day 20:00 through the current day 08:00, while the evening brief covers the current day 08:00 through 20:00.
- AI-domain filtering: it keeps topics such as large language models, AIGC, multimodal AI, AI chips, AI agents, machine learning, computer vision, NLP, and AI safety, while excluding entertainment, ads, rumors, and low-signal content.
- Deduplication and scoring: candidates are compared semantically by title, core event, actor, and key data; ranking weighs industry impact, freshness, source authority, and cross-platform attention, then emits a TOP 10 brief.

Boundaries

This skill is scoped to AI tech news, not general news aggregation. If fewer than 10 valid items exist in the window, it outputs the actual count and notes it. The final response is the briefing body, not a search or reasoning trace; because feed availability can vary, verify critical claims against the original source.