56,000 stars for the idea: an agent is more important than yet another search model
Sh0ny
1 августа 20262 min read
In short
The last30days-skill project has topped all GitHub trends not because it searches better than Google, but because it connects isolated platforms through a single agent. I’ll explain why the architectural approach here is more interesting than the features.
The GitHub repository currently has 56,000 stars—and it doesn’t train models or build a new search engine. last30days-skill is a skill for AI agents that simultaneously scrapes Reddit, X, YouTube, Hacker News, Polymarket, and the web in parallel, and then synthesizes all of this into a single brief. The stars here aren’t for hype, but for an architectural idea that’s becoming increasingly relevant: no single model has access to all platforms at the same time. Google doesn’t touch Reddit comments. ChatGPT has a deal with Reddit but can’t see X or TikTok. Gemini knows YouTube but not Reddit. Claude has no native access to any of them. Each platform is a walled garden with its own API, tokens, and authentication. The last30days-skill solves this not through a deal with the platform, but through an agent to which you provide your keys and browser sessions. It then searches all sources at once, ranks the results by actual engagement—upvotes, likes, and funds on Polymarket—and provides a summary. The practical value is obvious: instead of Googling someone before a meeting, you get a picture of what they’ve actually been doing over the past month—tweets, podcasts, GitHub commits, and Reddit discussions. None of these pieces are available individually on Google. It’s a fair trade-off: only Reddit, HN, Polymarket, and GitHub work with zero configuration. To connect X, YouTube, TikTok, and arXiv, you need to go through the setup wizard—it takes 30 seconds, but it’s still a manual process. You’re dependent on your own keys and sessions, and if a platform changes its API or blocks access, the agent loses its data source. This isn’t a “search killer,” but proof that value today lies not in yet another large model, but in who’s better at bridging isolated data. An agent that accesses a dozen walled gardens simultaneously delivers what no single model can on its own. The question that remains open is: how long will platforms tolerate agents that pull their data through user sessions before they close this loophole as well?