场景组合推荐

从「我要做什么」出发，而不是从「哪个星多」出发。每个场景给出健康分最高、npm 已发布、近期活跃的一组候选与备选——客观信号排序，不接"最佳"叙事，不做付费置顶。覆盖 132 个推荐位，随每日快照刷新。

Start from "what do I want to do", not from "which has the most stars". Each scenario offers candidates and alternates with the highest health score, published to npm, and recently active — ranked by objective signals; we don't sell the "best" narrative and don't take paid placement. 132 recommendation slots, refreshed with the daily snapshot.

场景归属 = LLM 能力标签 ∪ 词汇桶（标注"LLM 生成，人工抽查"）；场景内排序 = 健康分 → npm 已发布 → 近 30 天活跃，星数仅作展示不参与排序。同样的数据在 /data/insights.json 开放，agent 可直接消费。

Scenario assignment = LLM capability tags ∪ keyword buckets (labeled "LLM-generated, human spot-checked"); ranking within a scenario = health score → published to npm → active in the last 30 days. Stars are display-only and never affect ranking. The same data is open at /data/insights.json for agents to consume directly.