The last30days Skill: How to Fix Claude Code's Research Problem
The last30days Skill: How to Fix Claude Code's Research Problem
Claude Code has a research problem, and a single open-source skill called last30days fixes it. It sits in the gap between shallow web search and heavyweight deep research — scraping Reddit, X, YouTube, TikTok, Instagram, Hacker News, GitHub, and more to pull actual user sentiment instead of whatever ranked well on Google. It's the #1 trending GitHub repo of the day with over 55,000 stars, and it installs in a single line. I've been running it for days, and here's exactly what it does and how to set it up.
What Is the last30days Skill?
The last30days skill is an open-source Claude Code skill that scrapes the major social platforms, reads the comments, and synthesizes real user sentiment into a research brief. Instead of summarizing the top-ranking articles like a normal web search, it goes to the grassroots level — the places where people actually say what they think.
Here's the problem it solves. Whenever you ask Claude Code to research something, you're stuck choosing between two bad options:
- Web search — basically a glorified Google search. Fast, but surface-level. You get a summary of whatever articles rank well on SEO.
- Deep research — dynamic workflows that spin up hundreds of sub-agents, take 5, 10, 15, even 20 minutes, and can burn through millions of tokens.
The last30days skill is the middle ground. It casts a wide net across platforms, digs into the comment sections, and hands you back user sentiment — not headlines. For anybody who does research inside Claude Code, this is the tool you've been missing.
Which Platforms Does last30days Scrape?
The skill pulls from nearly every major platform: Reddit, Hacker News, Polymarket, GitHub, YouTube, TikTok, Instagram, LinkedIn, Pinterest, Bluesky, and X/Twitter. That breadth is the whole point — it's collecting signal from places a standard web search never touches.
The value-add is the sources. These are the conversations you won't see with out-of-the-box Claude Code. When you install the skill, most of these platforms are set up for you automatically — no API key, nothing to pay for. A few require dependencies, and Claude Code walks you through those when you run it.
Only a couple of sources cost money, and the repo itself breaks down exactly which ones and what they run. Two to know:
- X/Twitter requires an xAI API key. It's the most expensive source, but "expensive" is relative — across every run I did, it averaged about 10 cents.
- TikTok and Instagram reels require a Scrape Creators API key. But when you install the skill, it sets you up with a subsidized account that includes several thousand free calls. Run it every single day and you likely won't pay Scrape Creators anything for about 6 months.
How Does the last30days Skill Actually Work?
You invoke the skill inside Claude Code with /last30days (or just plain natural language), hand it a topic, and it fans out across every relevant platform in parallel before synthesizing the results into a ranked brief. There's no back-and-forth interrogation — you run it and it gets to work.
The flow looks like this:
- You give it a topic — something like "What are people saying about Claude Opus 5?"
- It sharpens your prompt. Whatever rough prompt you hand it, the skill rewrites it into something better before searching.
- It scopes intelligently. If you ask about a person or product, it already knows which subreddits are relevant and which X handles matter. You don't have to spell it out.
- It searches in parallel. A deterministic Python script hits all the platforms at once, so the whole thing moves quickly.
- It reads the comments. This is the part that matters — it's not just scraping post titles. It goes into the comment sections to see whether people agree with the original poster or whether there's a fight happening underneath.
- It ranks by cross-platform repetition. If the same take shows up on Reddit AND X AND YouTube, it gets ranked way higher. One stray upvoted Reddit comment won't hijack the whole report. The logic is simple: if it's showing up everywhere, there's probably something to it.
You can also scope it down. Don't want the full sweep? Tell it to only check Reddit, or only Twitter, or only YouTube. And if you're not sure how to use a flag, just ask — Claude Code downloaded the skill, so it knows the skill's own best practices and can teach you.
last30days vs. Claude Web Search: What's the Difference?
The clearest way to see the value is a side-by-side: the same question run through last30days versus a plain Claude web search. I asked both, "What are people saying about Claude Opus 5?"
The web search result wasn't wrong. It just gave me a summary of the major articles you'd find on Google. It never went a layer deeper into what people were actually saying on X or Reddit. It never hit the grassroots.
The last30days report cast a far wider net, and the sources made it obvious. For that one question, it pulled from:
- 22 Reddit threads
- 13 X posts
- 14 YouTube videos
- 26 TikToks
- 13 Instagram reels
- 16 Hacker News stories
- 25 GitHub items
- 1 Polymarket entry
That's a mountain of signal. And because it read the comments, it surfaced things a headline scan never would: the dominant framing that "Opus 5 is half the cost of Fable 5," the backlash living in the comment sections rather than the posts themselves, and the general sentiment from professional reviewers. On top of the brief, it dumps everything — every transcript, every comment — into a JSON file so you can sift through the raw data yourself.
That's the difference between reading the headline of a Reddit post and actually seeing whether the top comment is agreeing or tearing it apart.
What Does a last30days Report Look Like?
When I ran it on the query "what are people saying about the last30days skill," it took about 5 minutes and came back with a structured brief. Here's what a report gives you:
- A one-liner summary up top — in my run, it opened with "this repo is growing faster than the hype posts can keep up with it," plus live star numbers.
- The general through-line — the single narrative tying the research together (for last30days, that it "fixes generic AI research").
- Key patterns it spotted across the research.
- A per-platform breakdown — five threads from Reddit, 22 from X, and so on down the list.
- Pointers to the raw results — the raw markdown by default, and the full JSON if you ask for it.
For contrast, I ran the same prompt again but explicitly told it not to use the skill — pure web search. The report wasn't wrong, but it was shallow. Even when it described what people praised, it stayed generic, because it wasn't reading comments or transcripts.
When Should You Use last30days Instead of Web Search?
Use last30days when you need real user sentiment; stick with web search for quick, day-to-day questions. This is the honest part most tool reviews skip: you should NOT use last30days for everything. It's a bigger operation than a quick lookup, and firing it off for every random question inside Claude Code is overkill.
Think of it as three tiers:
- Web search — surface-level, fast, fine for most day-to-day questions.
- last30days — the in-between. Real sentiment from real people, in about 5 minutes.
- Deep research — the dynamic-workflow, thousands-of-sub-agents heavy artillery.
For pulling genuine user sentiment, I'd argue last30days might even beat deep research. It's aimed squarely at what people are actually saying, and it gets there faster and cheaper.
How Do You Install the last30days Skill?
Installation is a single line of code inside Claude Code — the command lives in the GitHub repo, and running it auto-configures every source that doesn't need an external API key. That's genuinely the whole setup for most people.
The repo goes deeper than just Claude Code, too. It shows you how to wire the skill into other agents like Codex and Cursor, and even how to run it on the Claude.ai web app. The only thing that might trip you up is connecting the specific paid sources — and even there, the fix is easy: throw the repo link at Claude Code and let it walk you through the exact source you're setting up.
The one source that will actually cost you money is X/Twitter, and again — roughly 10 cents a run.
Frequently Asked Questions
Is the last30days skill free?
Mostly, yes. Most platforms are set up automatically with no API key and no cost. X/Twitter requires an xAI API key and runs about 10 cents per search. TikTok and Instagram use Scrape Creators, but the install gives you a subsidized account with several thousand free calls — enough to run daily for around 6 months without paying.
How is last30days different from Claude's deep research?
Deep research spins up hundreds or thousands of sub-agents, can take 15-20 minutes, and may burn millions of tokens. last30days is a lighter middle ground that finishes in about 5 minutes and focuses specifically on scraping platforms for user sentiment. For sentiment-driven questions, it's often the better and cheaper choice.
What platforms can last30days scrape?
Reddit, Hacker News, Polymarket, GitHub, YouTube, TikTok, Instagram, LinkedIn, Pinterest, Bluesky, and X/Twitter. The repo documents which sources need dependencies or API keys and what each one costs.
Do I have to use last30days for every research task?
No, and you shouldn't. For quick, everyday questions inside Claude Code, plain web search is fine. Save last30days for when you specifically need to know what real people are saying across platforms, not just what ranked on Google.
Can I limit which platforms it searches?
Yes. By default it uses every source it thinks is relevant, but you can scope it down — tell it to only check Reddit, only X, or only YouTube. If you're unsure how, just ask Claude Code, since it already knows the installed skill's options.
If you want to go deeper into building research workflows with Claude Code, join the free Chase AI community for templates, prompts, and live breakdowns. And if you're serious about building with AI, check out the paid community, Chase AI+, for hands-on guidance on how to make money with AI.
