/last30days: AI agent for real-time social research
mvanhorn/last30days-skill
/last30days is an AI agent skill that searches Reddit, X, YouTube, Hacker News, Polymarket, and other social sources in parallel, then has an AI agent synthesize the results into one grounded brief. Reach for it if you want a single command that surfaces what people are actually saying about a topic across platforms Google doesn't touch; skip it if you'd rather not manage API keys or browser cookie access for the paid sources.
Understanding the /last30days Agent
/last30days is an AI agent skill, installed as a plugin into hosts like Claude Code, Codex, Cursor, and other Agent Skills-compatible tools, that researches a topic across Reddit, X, YouTube, Hacker News, Polymarket, and the web at once. Instead of returning a list of links, an AI agent judge reads across the sources and writes a single synthesized brief, ranked by what people actually engaged with rather than by editorial placement.
The Challenge of Current Information
Asking one AI assistant about a person or topic only covers whatever that assistant's search integration reaches, and per the README no single one reaches everything: ChatGPT has a Reddit deal though it cannot search X or TikTok, Gemini offers YouTube but not Reddit, and Claude has none of these natively. Each platform is what the README calls its own walled garden with a separate API, tokens, and auth, so getting a current read on a person or topic otherwise means checking each site by hand.
Key Capabilities and Data Sources
- ✓Searches configured sources in parallel and scores results by real engagement (upvotes, likes, view counts, betting odds) instead of SEO relevance, per the README.
- ✓Reddit, Hacker News, Polymarket, and GitHub work with zero config out of the box, per the README's "Zero config" claim.
- ✓A setup wizard can turn on additional sources including X, YouTube, TikTok, arXiv, and Techmeme.
- ✓Installs across many AI agent hosts through the open Agent Skills CLI, described in the README as "50+ Agent Skills hosts," including Codex, Cursor, Copilot, and Gemini CLI.
- ✓A `--hiring-signals` mode reads a company's job and careers pages as cited evidence of where it's hiring.
- ✓A discovery mode sweeps Reddit, Hacker News, and Digg listings and returns a ranked list of trending topics instead of a single brief.
- ✓Supports an `--as-of` flag for historical lookback and a `--competitors` mode for auto-discovered side-by-side comparisons.
- ✓Can emit a self-contained, shareable HTML brief via `--emit=html`.
Practical Applications of /last30days
- •Research a person before a meeting — the README's example pulls together someone's recent job moves, GitHub activity, and social discussion instead of a stale LinkedIn profile.
- •Read a company's hiring signals with `--hiring-signals`, treating current job postings as evidence of where it's investing.
- •Run discovery mode to find a topic before it peaks, sweeping Reddit, Hacker News, and Digg listings for a ranked list of trending topics.
- •Compare rival tools or projects side by side, pulling live GitHub star counts instead of relying on a stale blog post.
- •Check a destination before a trip using recent community reports on wait times, closures, and construction.
Installing the /last30days Skill
Claude Code is the README's recommended path: run `/plugin marketplace add mvanhorn/last30days-skill` then `/plugin install last30days`; the marketplace then handles updates automatically, or force a check with `claude plugin update last30days@last30days-skill`. For Codex, Cursor, Copilot, Gemini CLI, and other Agent Skills-compatible hosts, install with `npx skills add mvanhorn/last30days-skill -g` (the `-g` flag installs globally instead of per project). The README also documents Grok's Build CLI (`grok plugin install mvanhorn/last30days-skill`), claude.ai web (upload a downloaded `.skill` file with "Code execution and file creation" enabled), Claude Desktop (drag a downloaded `.mcpb` bundle into Settings > Extensions and paste API keys), and OpenClaw (`clawhub install last30days-official`).
Strengths
- ✓Reddit, Hacker News, Polymarket, and GitHub work with zero config and no API key, per the README.
- ✓Ranks results by real engagement — upvotes, likes, betting odds — instead of editorial placement or SEO.
- ✓Reaches platforms a single AI assistant typically can't; the README notes no one model natively covers Reddit, X, and TikTok together.
- ✓Installs across many AI agent hosts through the open Agent Skills CLI rather than locking you into one assistant.
- ✓Ships a `--preflight` command that shows what it will read or write before touching cookies or files.
Requirements and Considerations
- △Sources beyond the zero-config set — X, YouTube, TikTok, LinkedIn, and others — need the setup wizard, API keys, or browser cookie access, so the full source list isn't available out of the box.
- △Multiple install paths can target the same host (the Claude Code marketplace plugin vs. the `npx skills` install), and the README warns they don't dedupe — running both leaves two `/last30days` entries.
- △Some sources are opt-in and only auto-enable when a companion CLI (for arXiv, Techmeme, or Digg) is on PATH, adding setup steps for full coverage.
- △Several sources depend on browser cookie extraction, which ties research to your local, logged-in browser session rather than a portable credential.
Beyond Traditional Search Engines
Frequently Asked Questions
Per the README, /last30days installs on Claude Code, Grok's Build CLI, Claude Desktop, claude.ai web, OpenClaw, and any of the "50+ Agent Skills hosts," including Codex, Cursor, Copilot, and Gemini CLI.
Not all of them — the README lists Reddit, Hacker News, Polymarket, and GitHub as zero-config sources, while sources like X, YouTube, and TikTok need a setup wizard or API keys.
/last30days scores results by real engagement — upvotes, likes, view counts, and Polymarket odds backed by real money — rather than by SEO relevance, and an AI agent judge synthesizes the results into one brief.
/last30days supports historical lookback through an `--as-of` flag, according to the README's changelog.
/last30days is open source under the MIT license, hosted on GitHub at mvanhorn/last30days-skill.
The README documents an active changelog of new sources, security hardening, and expanded platform support since the v3 pipeline, including community-contributed pull requests, without stating a fixed release cadence.
How to use
Run the command with a topic or person's name, e.g. `/last30days Peter Steinberger`, and the agent researches across its configured sources and returns one synthesized brief. Asking an open question instead of naming a topic, e.g. `/last30days what's exploding in AI agents?`, switches to discovery mode and returns a ranked list of trending topics instead. Flags extend the base command: `--hiring-signals` for a company's hiring signals, `--as-of` for historical lookback, `--competitors` for an auto-discovered comparison, `--github-user=<name>` for GitHub person-mode, and `--emit=html` for a shareable HTML brief. On folder-mode hosts, the README recommends running the bundled `scripts/last30days.py --preflight` before first use to see what it plans to read or write before it touches any cookies or files.
Who should try it — and who should skip
Reach for /last30days if you're a developer, founder, or salesperson who wants one command to pull together what people are currently saying about a person, company, or topic across Reddit, X, YouTube, and similar sources instead of checking each site by hand. Skip it if a plain Google search already answers your question, you don't want to manage API keys or browser cookie access for the paid sources, or your AI host isn't in the README's supported list.
Related repositories
Want a second opinion on last30days-skill?
Ask an AI that can read this page — one click and you get its take on last30days-skill.