A research skill built around recency and engagement
/last30days is an agent skill that answers a problem every LLM user feels: model knowledge is months stale, and the people who actually track a fast-moving topic live on Reddit, X, YouTube, and Hacker News, not in last year’s training data. The skill searches those platforms plus TikTok, Polymarket, and GitHub in parallel, scores results by what real people engage with (upvotes, likes, transcripts, prediction-market money), and has an agent judge synthesize one grounded brief. Its framing is sharp: Google aggregates editors, this searches people.
The reason it exists as its own skill rather than a feature of one assistant is the walled-garden problem. Each platform is a separate API with its own auth and tokens, and no single hosted AI reaches all of them: ChatGPT has a Reddit deal but not X or TikTok, Gemini has YouTube but not Reddit, Claude has none natively. By bringing your own keys and browser sessions, an agent can search all of them at once and rank them against each other, which is the thing you cannot buy off the shelf.
Install
For Claude Code, with auto-updates via the marketplace:
/plugin marketplace add mvanhorn/last30days-skill
/plugin install last30daysFor Codex, Cursor, Copilot, Gemini CLI, or any Agent Skills host:
npx skills add mvanhorn/last30days-skill -gThe -g installs it globally for your user; drop it to scope per-project. It is zero-config to start: Reddit, Hacker News, Polymarket, and GitHub work immediately, and a setup wizard unlocks X, YouTube, and TikTok in about thirty seconds.
What you actually get back
A query like /last30days <person> returns what someone is doing this month, the joined-a-company, shipping-PRs, getting-debated-on-Reddit reality scattered across X posts, threads, transcripts, and commits, rather than a stale LinkedIn from two years ago. That is the use it nails: catching up on a person, product, or topic as of now, with the sources cited and weighted by engagement.
How the search actually runs
The mechanic underneath is parallel search plus an agent judge, and it is worth picturing because it explains both the speed and the limits. Rather than querying one source and stopping, the skill fans out across the connected platforms at once, pulls candidate results with their engagement signals attached, and then has an agent weigh them against each other before writing the brief. That cross-source comparison is the part a single search engine cannot do, because no one engine sees Reddit comments, X posts, and YouTube transcripts together. The tradeoff is that quality scales with how many platforms you have unlocked: on the zero-config sources you get a narrower read, and the synthesis is only as good as the breadth of what it had to compare.
The stance to go in aware of
The scoring philosophy is a genuine editorial choice, not a neutral fact, and it is worth holding at arm’s length. Ranking by upvotes, likes, and prediction-market money surfaces what is engaging, which is not the same as what is true or important; engagement rewards controversy and recency, and Polymarket odds reflect bettors, not verified outcomes. The skill is excellent for taking the pulse of a community fast, and it should be read as exactly that: a weighted snapshot of attention, not a fact-checked report. Treat the brief as leads to verify, not settled conclusions.
The practical caveat is the same one every cross-platform tool carries: you bring your own keys and sessions, and platform access is a moving target, which shows in an active tracker (143 open issues as of 2026-06).
last30days versus Agent Reach
| last30days-skill | Agent Reach | |
|---|---|---|
| Stars | 38,593 | 25,808 |
| Role | research skill that ranks and synthesizes | access toolbox for platforms |
| Output | one scored, synthesized brief | raw reads and searches |
| Scoring | by engagement, agent-judged | none, you decide |
Counts are from GitHub as of June 2026. Agent Reach is the lower-level toolkit that gives an agent the ability to read and search these platforms at all. last30days is the opinionated layer on top: it searches them in parallel, scores by engagement, and hands back a synthesized brief. They are complementary, plumbing versus an opinionated research workflow built on plumbing like it.
Related
For the platform-access toolkit this kind of skill relies on, see Agent Reach. For what else is climbing, see LLM tooling, the daily digest, and the weekly report.
FAQ
What does /last30days do? It searches Reddit, X, YouTube, HN, TikTok, Polymarket, and GitHub in parallel, scores results by real engagement, and synthesizes one current brief.
How do I install it? /plugin marketplace add mvanhorn/last30days-skill then /plugin install last30days, or npx skills add mvanhorn/last30days-skill -g for other hosts.
Do I need API keys? Reddit, HN, Polymarket, and GitHub work with zero config. A wizard unlocks X, YouTube, and TikTok with your own keys and sessions.
Is engagement the same as truth? No. Ranking by upvotes, likes, and prediction-market money surfaces what is engaging, not what is verified. Treat the brief as leads to check.