I sat down with matt van horn (Matt Van Horn) and watched him turn claude code into a real-time research engine with his /last30days claude code skill he "fixes" claude code in 30 seconds. this skill pulls what’s actually working right now from x, reddit, and the web, then... feeds that context straight into your prompts so you stop building off stale advice. we went from trending rap songs → cold email frameworks → researching clawdbot → planning and building a competitor live, with almost zero hand-written code. pretty nifty little claude code skill share this with a friend / full ep available on The Startup Ideas Podcast (SIP) 🧃 where i will give you ideas/tools/tutorials to make your dreams a reality i will not hold back any alpha and this claude skill is alpha forsure you can install this claude code skill in 30 seconds dream big my friendsshow more

Thanks for having me. Super fun jamming on examples and ideas. @slashlast30days and install here

@slashlast30days 🙏

@mvanhorn can you make this a paid skill @mvanhorn so instead of me having to pay for X API and set my API keys I just run 'npx skills add --skill last30days' and then everytime I call /last30days I just pay $0.01 via x402? Happy to help you set it up

@mvanhorn The real unlock is making the workflow deterministic: extract → score → cite → ship. Otherwise it’s just vibes. Curious how he handles dedupe + rate limits across sources?

@mvanhorn @mvanhorn is forever the goat due to June. Still makes me the perfect bacon slices.

@mvanhorn Bro, you are always dropping bangers after bangers

@mvanhorn Appreciate you Full ep over here

@mvanhorn Thanks, off to youtube then

@mvanhorn Most people are still playing with prompts. This is building with real time leverage.

skills that inject live context are the unlock most people are sleeping on. the base model knows everything up to its training cutoff, but the gap between training data and right now is where all the alpha lives. this pattern of feed the agent what's working today instead of what worked last year is going to become standard.

"Real-time research engine" is a search API wrapper with extra steps. Pulling trending content from X and Reddit isn't novel. It's scraping with AI summarization. The "stale advice" framing assumes trends from the last 30 days are more valuable than accumulated knowledge. Sometimes true. Often not. Building a competitor to Clawdbot live with "almost zero hand-written code" sounds impressive until you ask if it actually works in production. Demos always work. Maintenance is where things break. The skill itself—feeding recent context into prompts—is useful for certain tasks. Tracking what's buzzing right now. Catching shifts fast. But calling it alpha is overselling. It's automation anyone technical could build in an afternoon. Installing in 30 seconds just means importing someone else's config. Not building capability yourself. "Dream big" while using tools that lower the barrier to execution is good advice. Just don't confuse speed with substance. Most people will install this, use it once, realize it doesn't fundamentally change their workflow, and move on. The ones who get value are already doing research systematically. This just makes them faster. Everyone else gets a shiny new toy that doesn't solve the real problem: knowing what questions to ask in the first place.

@mvanhorn Real-time feeds from the last 30 days turn Claude into a powerful hype detector, but for shipping durable indie products, the mainstream rush to zero-handwritten-code ignores that truly reliable architectures usually take longer than 30 days to prove themselves in production

@mvanhorn how do you filter out prompt injections from scraping this data?

@mvanhorn gm Greg ow man things move so fast

@mvanhorn developers who build custom skills for their tools get 10x more done. thats it

@mvanhorn 😮‍💨🚀 thanks Greg & Matt!

The real shift is context awareness. Claude pulling from X, Reddit, and live web means prompts can reference real-world signals (trending complaints, competitor moves, breaking news) instead of static docs. We're seeing this collapse research → ideation → build cycles from weeks to days.

@mvanhorn So apparently there’s levels to this shit @_jklnz

@mvanhorn Mind doing a stream showing a workflow vibe coding a web app monetized for mobile as well? I haven’t seen a real video on that through the whole process UI / UX -> backend

@mvanhorn Billions are being wiped from software stocks as we speak… The development of these tools are quite literally leading to market disruption. What does the future hold??? Are we going to continue to be given access to these technologies? This is an arms race😭

@mvanhorn That journey from music to cold email frameworks is inspiring! Cold emailing can unlock a world of possibilities. It's all about smart outreach and building connections. Any reason you are not using Mystrika, yet?

@mvanhorn This is the kind of tool that actually turns research into action fast

We built something similar for our auto parts business.. Claude Code scans our competitor pricing daily, tracks Amazon BSR rankings, and feeds real-time market data into our buying decisions. 30+ containers a year from Asia and knowing what's trending RIGHT NOW vs last quarter is the difference between a full warehouse and a profitable one. The skill ecosystem around Claude is going to be massive.

@mvanhorn This is the missing piece most people skip. LLMs are only as good as the context you feed them. Fresh real-world data beats training data every time. Been building similar context pipelines and the quality gap is night and day. Claude Code skills are genuinely underrated.

@mvanhorn Impressive to see such rapid innovation in real time research tools, agility in technology is crucial for staying relevant.

@mvanhorn This demonstrates the power of AI skills as dynamic knowledge engines.

@mvanhorn You see the new Claude code /insights command that gives you info on how you can use it better? So legit

@mvanhorn Clawdbot is legitimately my second brain at this point. Wild seeing it on the podcast.

@mvanhorn a real-time research engine pointed at clawdbot, what could possibly go wrong 😅 for real though, that's a killer skill. so much of the web is just stale advice.

@mvanhorn How does a "fixed" language model like Claude Code compare to a custom-built solution for a specific use case, like the finance tracking app I'm building?

@mvanhorn Old advice slowed me down. Live data changed decisions instantly.

@mvanhorn real-time research fed directly into Claude Code this is actually genius

@mvanhorn Matt didn’t just debug Claude; he gave it WiFi for reality. Now it learns from the internet faster than most people find their car keys.

@mvanhorn @mvanhorn great interview! I am new to this how did you find great skills like compound engineering and super powers? I ask because I tried and my results suck.

@mvanhorn that sounds like a total game changer for anyone who uses claude code regularly, can't wait to check out the full episode on startup ideas pod

@mvanhorn claude code skills that pull live community data are underrated. the best prompts come from what's working *right now* not what worked 6 months ago. love this pattern.

@mvanhorn last30days is really good. was hesistant to setup an xai api key (been paying for so many that why lol).. im happy i did it!

@mvanhorn thirty seconds to fix it feels mildly illegal in the best way

@mvanhorn Full ep over here