Consumer Tech Frontier Brief

Problem

Consumer tech feeds are noisy: stale items, cross-platform duplicates, clickbait, and subjective commentary make it hard to get a clean same-day summary. This skill constrains the task to a fixed 12-hour brief: morning covers (previous day 20:00, current day 08:00], evening covers (current day 08:00, current day 20:00]. It keeps only allowed categories such as smart hardware, smart home, AR/VR/MR, consumer robots, drones, smartphones, and consumer chips, while excluding entertainment gossip, marketing ads, and unverified scoops.

How It Works

The skill uses a two-stage retrieval loop: it first runs node scripts/fetch_rss.js to pull structured items from the public 36Kr API and RSS feeds; if the script is unavailable, times out, or yields too few valid items, it falls back to web_search and verifies dates or key facts with web_extract. Items are filtered by time window and domain allowlist, then deduplicated semantically using title, actor, core action, and key data at threshold 0.6. Ranking weighs industry impact, freshness, source authority, and spread, then emits a table-free Markdown TOP10 brief. Each entry includes rank, title, description, source, publish_time, and score; rank 1-3 get 50-150 characters, while rank 4-10 get 50-100 characters.

Boundaries

The skill uses public web sources only, does not read local files, and does not collect personal data. If feeds are unreachable or the network is unstable, it may output fewer than 10 valid items and annotate total_valid. It is not suitable for non-consumer-tech topics, corporate fluff, pure opinion pieces, or unconfirmed rumors.