🟢 Beginner–Intermediate ⚙️ Type: AI Agent Skill / OSINT Research 💸 Free & Open Source (MIT) ⭐ 38,000+ GitHub Stars
What is last30days-skill?
last30days-skill is a massively popular, open-source AI agent extension created by developer mvanhorn. It transforms your local AI coding assistant (like Claude Code, Gemini CLI, or Cursor) into an incredibly powerful, real-time research engine.
Traditional AI web searches rely on standard SEO-driven engines, which often return outdated blog posts or heavily curated corporate documentation. last30days flips this model upside down. When you give it a topic, it bypasses generic search engines and queries Reddit, X (Twitter), YouTube, Hacker News, Polymarket, GitHub, and TikTok simultaneously.
It utilizes a two-phase search architecture to extract raw community discussions, code repositories, and prediction market odds from the last 30 days. It then deduplicates the data, ranks it by real human engagement (upvotes, views, and actual betting money), and synthesizes a highly grounded, actionable briefing complete with real citations.
Who is it for?
- Developers and Prompt Engineers trying to discover the most effective prompting patterns or library workarounds being discussed by the community right now, rather than reading 6-month-old documentation.
- Sales Teams and Founders doing deep-dive meeting prep. You can run the skill on a prospective client to instantly pull their recent X posts, YouTube interviews, and public product issues.
- Analysts and Investors tracking emerging trends, sentiment shifts, or using the built-in Polymarket integration to evaluate real-time financial odds on global events.
- Content Creators looking to see what specific angles, memes, or questions are currently generating the highest engagement across social platforms.
What makes it special?
- Engagement-Weighted Scoring — The AI doesn’t just read the internet; it scores it. A Reddit thread with 15,000 upvotes carries far more weight in its final summary than a corporate press release with zero engagement.
- Financial Intelligence (Polymarket) — It natively integrates prediction markets. If you research an unfolding news event or tech release, it automatically pulls live betting odds to show you what the market actually believes will happen.
- Multi-Harness Compatibility — It isn’t locked to one AI platform. It acts as a universal “Agent Skill” that works flawlessly inside Claude Code, Gemini CLI, Cursor, OpenAI Codex, OpenClaw, and over 50 other agent hosts.
- HTML Brief Generation — By appending
--emit=htmlto your prompt, the skill will instantly generate a beautiful, dark-mode, print-friendly HTML dashboard of its findings that you can share with your team. - Comparative Mode — Instead of researching a single tool, you can ask it to compare frameworks (e.g., “OpenClaw vs Hermes”). It will run parallel discovery, pull live GitHub star counts, and output a side-by-side comparative analysis.
Requirements before you start
To run this skill locally, you need an AI agent to act as the “host” and a modern Python environment for the underlying scraping engine:
- Python 3.12 or 3.13 — The background scripts require modern Python. Older versions (like 3.10) will trigger a preflight failure block.
- Node.js (npm/npx) — Required to install the skill package across your system globally.
- An AI Agent Host — You need a CLI agent like Anthropic’s Claude Code (recommended), Gemini CLI, or OpenClaw installed on your machine.
- API Keys (Optional but Recommended) — Standard sources (Reddit, Hacker News, Polymarket, GitHub) work out-of-the-box for free. Fetching deep X/Twitter or specialized video data requires adding your own scraper API keys to the configuration.
Step-by-step installation
There are multiple ways to install the skill depending on your daily AI driver. Here are the three most common methods:
Method 1 — Claude Code Marketplace (Recommended)
If you use Anthropic’s Claude Code, you can install it directly from the terminal using the plugin marketplace, which ensures you always get automatic updates:
/plugin marketplace add mvanhorn/last30days-skill
/plugin install last30days
Method 2 — Global Install via NPX (For Cursor, Gemini CLI, etc.)
If you use multiple agents and want the skill available everywhere on your machine, use the standard Agent Skills package installer:
npx skills add mvanhorn/last30days-skill -g
Method 3 — Claude.ai Web Interface
If you do not want to use a terminal and strictly use the Claude web app in your browser:
- Go to the GitHub Releases page and download the
last30days.skillfile. - Log into
claude.ai. - Navigate to Settings > Capabilities > Skills.
- Click the + button and upload the file. (Ensure that code execution and file creation are enabled for your Claude web account).
Step 4 — Run Your First Research Brief
Open your AI agent and type the slash command followed by your topic:
/last30days "Claude Sonnet 4.6 community reception" --emit=html
The AI will trigger the background Python scripts, scrape the internet, and drop a comprehensive briefing directly into your chat (and save an HTML copy to your Documents folder)!
Common errors and fixes
| Error | What it means | How to fix it |
|---|---|---|
ERROR: last30days v3 requires Python 3.12+ | The skill’s preflight check detected that your system’s default Python version is too old. | Install Python 3.12 or 3.13. The skill script will automatically search your PATH for python3.12 or python3.13 and use it. |
Unknown command: /last30days (in Claude Code) | You are using a newer version of Claude Code (v2.1.105+) with an older, cached version of the skill plugin that had a path escape bug. | Run /plugin uninstall last30days, then reinstall it using the marketplace command to pull the latest V3 patch. |
| X (Twitter) or specific video sources return zero results | The free, unauthenticated fallback scrapers are being rate-limited by the social platforms. | The skill will natively fall back to free sources (like Reddit and HN). To unlock premium walled-garden scraping, you must add a SCRAPECREATORS_API_KEY to the ~/.config/last30days/.env file. |
Free vs Paid comparison
| Feature | last30days-skill (Local Extension) | Perplexity Pro (Commercial API) |
|---|---|---|
| Cost | $0 (Uses your existing Agent’s API tokens) | $20 per month |
| Search Focus | 🟢 Social Engagement, Sentiment, & Forums | 🟡 Broad Web & Academic Papers |
| Terminal Integration | ✅ Runs seamlessly inside your coding environment | ❌ Must switch to a web browser |
| Privacy | ✅ Research memory saved to local SQLite | ❌ Search history stored on cloud servers |
Bottom line: If you are already working inside the terminal with an AI coding assistant, installing the
last30days-skillis an absolute necessity. It brings the power of a dedicated OSINT research team directly into your development workflow for free. While commercial tools like Perplexity are better for general knowledge queries, nothing beats last30days for discovering what the tech community is actually arguing about right now.
Alternatives — 3 similar tools
1. Perplexity AI
The consumer standard for AI web searching. While it lacks the deep, raw engagement-weighted scraping of social platforms (like pulling exact Polymarket odds or YouTube comments), it is significantly faster and doesn’t require terminal installation or API keys to use.
2. Exa (Formerly Metaphor)
An API-first search engine built explicitly for Large Language Models. If you are building your own custom Python agents from scratch and need them to search the web based on meaning rather than keywords, the Exa API is a phenomenal alternative to relying on local scraping scripts.
🔗 exa.ai
3. SearchGPT (OpenAI)
Integrated directly into ChatGPT, this provides real-time web browsing capabilities. It is fantastic for general news and documentation lookups, but its strict safety guardrails and anti-scraping compliance often prevent it from deeply analyzing raw, unfiltered Reddit threads or Twitter debates the way last30days can.
🚀 Want more free AI tools like this?
We find, test, and write setup guides for the best free and open-source AI tools — so you don’t have to dig through GitHub yourself.Browse Free AI Tools at globalaiforce.com/shop →
📸 Follow us for daily AI tool tips and tutorials: instagram.com/globalaiforce