last30days: the agent skill that reads people, not editors
last30days-skill has quietly become one of the most-starred agent skills on GitHub, past 54k stars and still climbing on the daily trending page. It does one thing with unusual focus: give your agent a real-time read on what actual communities are saying about a topic in the last thirty days, then synthesize it into a grounded summary. The pitch is that it searches people, not editors, meaning it goes to where real discussion happens instead of polished articles.
The source list is the reason it works. One skill pulls from Reddit with upvotes and comments, X, YouTube transcripts, TikTok, Instagram Reels, Hacker News, Polymarket prediction markets, GitHub, Bluesky, LinkedIn, Threads, Pinterest, arXiv, Techmeme and general web search, all filtered to the last thirty days. Instead of wiring up fifteen APIs yourself, your agent gets one skill that already speaks all of them and returns a synthesized answer with the receipts attached.
The distribution is the other half of the story. It installs across Claude Code, Cursor, Copilot, Gemini CLI, Claude Desktop as an MCP bundle, OpenClaw and fifty-plus other agent hosts, via npx skills add or the plugin marketplace, MIT licensed with no tracking. This is what a mature skill ecosystem looks like: one capability, written once, running everywhere an agent runs. The project is on v3.11.1 with 175 merged PRs since May, so it's not a weekend demo, it's an actively maintained piece of infrastructure.
The reason this matters for anyone building agents: research is the task everyone reinvents badly. Every agent eventually needs to know what the world thinks about something right now, and everyone hacks together their own brittle scrapers. A single well-maintained skill that does cross-platform, recency-filtered, community-grounded research is exactly the kind of shared building block the skills model was supposed to produce. Repo: https://github.com/mvanhorn/last30days-skill
← Back to all articles
The source list is the reason it works. One skill pulls from Reddit with upvotes and comments, X, YouTube transcripts, TikTok, Instagram Reels, Hacker News, Polymarket prediction markets, GitHub, Bluesky, LinkedIn, Threads, Pinterest, arXiv, Techmeme and general web search, all filtered to the last thirty days. Instead of wiring up fifteen APIs yourself, your agent gets one skill that already speaks all of them and returns a synthesized answer with the receipts attached.
The distribution is the other half of the story. It installs across Claude Code, Cursor, Copilot, Gemini CLI, Claude Desktop as an MCP bundle, OpenClaw and fifty-plus other agent hosts, via npx skills add or the plugin marketplace, MIT licensed with no tracking. This is what a mature skill ecosystem looks like: one capability, written once, running everywhere an agent runs. The project is on v3.11.1 with 175 merged PRs since May, so it's not a weekend demo, it's an actively maintained piece of infrastructure.
The reason this matters for anyone building agents: research is the task everyone reinvents badly. Every agent eventually needs to know what the world thinks about something right now, and everyone hacks together their own brittle scrapers. A single well-maintained skill that does cross-platform, recency-filtered, community-grounded research is exactly the kind of shared building block the skills model was supposed to produce. Repo: https://github.com/mvanhorn/last30days-skill
Comments