| name | last30days |
| description | AI agent-led search engine that researches any topic across Reddit, X (Twitter), YouTube, TikTok, Instagram, Hacker News, Polymarket, GitHub, and more — scored by real engagement signals (upvotes, likes, comments) rather than SEO. Produces rich HTML briefs with Best Takes, cross-source cluster merging, ELI5 mode, comparison mode, hiring signals, and competitor auto-discovery. Use when the user wants to research what people are saying about a topic, product, person, or trend in the last 30 days. Trigger keywords: /last30days, research, what are people saying, trending, sentiment, compare, ELI5, hiring signals, best takes, 调研, 最近30天, 大家都在说什么, 海外调研, AI新概念学习, 海外资讯, 社交媒体调研, 热门话题, 深度搜索, Reddit搜索, X搜索, 跨平台调研.
|
| metadata | {"category":"thinking","source":"https://github.com/mvanhorn/last30days-skill","references":["https://mp.weixin.qq.com/s/VDlBdkspV0SQNFJYLxOogQ"],"author":{"original":"mvanhorn","chinese_intro":"卡兹克"}} |
last30days — AI-Powered Topic Research
An AI agent-led search engine that researches any topic across 15+ platforms
scored by real engagement signals, not SEO or editorial curation.
Installation
claude plugin marketplace add mvanhorn/last30days-skill
claude plugin install last30days@last30days-skill
npx skills add mvanhorn/last30days-skill -g
Usage
/last30days <topic> [--mode <mode>] [--save]
Modes: research (default), compare, hiring, agent, eli5
What it searches
| Source | What it captures |
|---|
| Reddit | Thread scores, comment counts, subreddit communities |
| X (Twitter) | Tweet engagement, verified signals |
| YouTube | Video metadata, view counts |
| TikTok | Trending short-form content |
| Instagram | Visual trend signals |
| Hacker News | Tech community discussion scores |
| Polymarket | Real-money prediction market odds |
| GitHub | Repo stars, issue activity, person/org mode |
| Threads, Pinterest, Bluesky | Emerging platform signals |
| Perplexity | AI-powered web research |
| General Web | Brave/Exa/Serper search backends |
Key features
- Real engagement signals — upvotes, likes, comment counts, Polymarket odds
- Best Takes — extracts the most insightful comments/threads
- Cross-source cluster merging — deduplicates the same story across platforms
- ELI5 mode — simple explanations for complex topics
- Comparison mode — side-by-side analysis (e.g., "Cursor vs Copilot")
- Hiring signals — candidate/business evaluation patterns
- GitHub person-mode — research a developer's activity profile
- Shareable HTML briefs — self-contained reports with embedded media
Configuration
Config file at ~/.config/last30days/.env or .claude/last30days.env:
REDDIT_CLIENT_ID=...
REDDIT_CLIENT_SECRET=...
X_BEARER_TOKEN=...
SCRAPECREATORS_API_KEY=...
BLUESKY_APP_PASSWORD=...
PERPLEXITY_API_KEY=...
BRAVE_API_KEY=...
Output saved to ~/Documents/Last30Days/ (configurable via LAST30DAYS_MEMORY_DIR).
Architecture
skills/last30days/
├── SKILL.md # Canonical skill definition (10 LAWs + step-by-step)
└── scripts/
├── last30days.py # Main engine
└── lib/ # Source modules, search backends, scoring
中文介绍
以下内容整理自卡兹克的介绍文章:《开源项目简介:last30days》
这是什么?
给 last30days 一个话题,它让 AI agent 帮你扒一遍最近 30 天国外各大社交媒体上的真实讨论,然后综合成一份带引用的报告。
开源地址:https://github.com/mvanhorn/last30days-skill
和普通搜索的核心差异
第一,深度爬取,不只是看帖子标题。
Reddit 上一个帖子,它把评论带点赞数一起扒下来。X 上的推文,高赞回复也会纳入进去。就像你看小红书不会只看正文肯定要翻评论区一样,这个 Skill 把类似的考虑了进去,而且跨十几个平台一起翻。优质信息经常就埋在评论里,有时候评论比正文还值钱。
第二,按真实热度排序,不按 SEO。
Reddit 几千 upvote 的讨论,权重高于一篇没人看的博客。一个 3.6M 播放的 TikTok,权重高于一篇新闻稿。看到的排名是真人投出来的,不是哪篇博客做了 SEO 排上来的。
覆盖源
Reddit、X、YouTube、TikTok、Hacker News、GitHub、Polymarket 等十几个国外平台。
免费 vs 付费
- 零配置免费可用:Reddit、Hacker News、Polymarket、GitHub
- 简单配置后可用:X(登录浏览器账号或用 API key)、YouTube(装 yt-dlp)
- 免费源对日常学习新概念来说已经够用
适用场景
学习 AI 新概念新知识特别好用——国内很多信息是对海外资讯的搬运和二次咀嚼,直接用这个 Skill 从源头获取,信息更一手、更及时。
MIT License.