AI News Buddy

Problem

General news feeds rank items by popularity, but rarely explain why a story matters to a reader. Users often face clickbait titles, mixed source quality, stale articles, and overly generic recommendations. This skill turns news into a personal briefing: it filters by profession, city, interests, and recent decision points, then connects each item to a concrete action. Each item should land on an information point that the reader can actually use.

How It Works

It first reads existing memory files quietly for cold start, then asks a few questions to build a local profile. During delivery, it prefers tencent-news-cli, falling back to web_search with a full four-digit year and topic:news. Results are limited to Chinese sites, deduplicated, filtered against the T4 blacklist, and checked for current-year or relative-time evidence before 5–8 items are selected. The default card includes card_title, summary, how_brief, and source_link; expanding an item adds insight, what, and how.

Limits

It is not a substitute for in-depth reporting, and it should not be used for real-time financial or medical decisions. Finance and health notes are information summaries only. The profile is stored locally in profile.json and is not exposed to the user. “换一批” means not interested and does not boost that dimension. Source filtering reduces risk but does not guarantee zero errors. If results are thin or sources are unavailable, the skill should say there is nothing fresh right now.