Transcript
00:00:00Cloud Code has a research problem, and this open source GitHub repo solves it.
00:00:05It's called Last 30 Days, and it helps you with the dilemma you face
00:00:08whenever you try to have Cloud Code search for something.
00:00:10Either we rely on web search, which basically is a glorified Google search,
00:00:14or we have to use something like deep research, which takes 5, 10, 15, 20 minutes,
00:00:19hundreds of sub-agents, and potentially millions of tokens.
00:00:23Instead, the Last 30 Days skill gives us a perfect middle ground
00:00:26where we can scrape a ton of different platforms
00:00:28and figure out actual user sentiment,
00:00:31not just the headlines and not just the articles that rank really well on SEO.
00:00:35This is the perfect tool for anybody who does research with Cloud Code,
00:00:38so you're going to want to stick around for this one.
00:00:40So the Last 30 Days skill is an open source GitHub repo that's got 55,000 stars.
00:00:44It's been the number one GitHub repo of the day,
00:00:47and essentially what it does is it allows Cloud Code
00:00:49to scrape all the major social media platforms.
00:00:52I'm talking things like Reddit, Hacker News, Polymark, GitHub, YouTube, TikTok, Instagram,
00:00:57all of them. It goes out, finds the information you need to know.
00:01:00Again, we're getting a lot of user sentiment here that you wouldn't get with a normal web search,
00:01:04and then it synthesizes it all and gives you a report.
00:01:07And that report looks something like this.
00:01:08So what I asked Cloud Code was,
00:01:10what are people saying about Cloud Opus 5?
00:01:12On the left, we have the report using the last 30-day skill,
00:01:15and on the right, we have the Cloud Plus web search.
00:01:17Now, first, looking at the Cloud Plus web search, you know, sort of output, what do we get?
00:01:22Well, we essentially get a summary of the major articles you would find on Google
00:01:27if you said, what are people saying about Opus 5?
00:01:30So it's not necessarily wrong, but it's not going a layer deeper
00:01:33and seeing what people are actually saying on things like Twitter or Reddit
00:01:37and that sort of thing, right?
00:01:38It isn't getting to the quote-unquote grassroots level.
00:01:41Meanwhile, if we look at the last 30 days report,
00:01:43you can see that we cast a much wider net,
00:01:45and that's super obvious when we take a look at the actual sources.
00:01:49So for this report, you looked at 22 Reddit threads.
00:01:52It had 13 X posts, 14 YouTube videos, 26 TikToks, 13 Instagram reels,
00:01:5716 stories from Hacker News, 25 items from GitHub, anything from Polymarket.
00:02:02So that's a lot of information.
00:02:04And so what it's able to tell us is stuff that's, again,
00:02:07very nitty-gritty in terms of what people are saying.
00:02:10So it's talking about how it did on Instagram
00:02:12and how it did on TikTok in the short-form space.
00:02:14It's talking about, hey, here's sort of like the framing that's been going on,
00:02:18mainly like Opus 5 is like half the cost of Fable 5.
00:02:21Talks about some of the backlash that lives in the comments, not just the posts.
00:02:24So it's one thing to, again, read sort of the headlines of these posts on like Reddit.
00:02:27It's another thing to see, okay, like what do the comments say?
00:02:30You know, are they agreeing with what the original post has said?
00:02:33Or is there like a ton of fights going on?
00:02:35And lastly, it tells us in terms of professional review,
00:02:37what the general sentiment is.
00:02:48So you put that all together and you get much deeper insight
00:02:51with this last 30-day skill.
00:02:57Now diving a little deeper on this, with the last 30-day skill,
00:03:00this report is just a small sort of synthesis of everything it's gathered.
00:03:08Behind the report you see, it also creates a markdown file.
00:03:11That's just a slightly larger summary.
00:03:13So you can kind of see that in its raw format right here,
00:03:16where it talks a little bit more about what it found on each platform.
00:03:23And beyond that,
00:03:23it collects everything and puts it into a JSON file,
00:03:27like all the transcripts, all the comments, all those things.
00:03:30And so like, that's what we see right here.
00:03:32And there's a ton of stuff, right?
00:03:34So just way more data for you to sift through.
00:03:37Now, before we jump into more details on this skill,
00:03:39a quick word from today's sponsor, me.
00:03:42So inside of Chase AI+,
00:03:44I have just released the Cloud Code Masterclass.
00:03:46And it is the number one way to go from zero to AI dev,
00:03:49especially if you don't come from a technical background.
00:03:52Focus on real use cases.
00:03:53This is constantly updated.
00:03:55So if you're someone who wants to level up their AI game,
00:03:58but doesn't really know where to start,
00:04:00I highly suggest you check us out.
00:04:02There's a link to it in the pinned comment.
00:04:04Hope to see you there.
00:04:05Now, when you see all that,
00:04:06it becomes very obvious what the value add is here
00:04:08with the 30-day skill.
00:04:09And that is the sources.
00:04:10These are the sources you won't necessarily be able to see
00:04:12with out-of-the-box Cloud Code.
00:04:14And it's pretty expansive.
00:04:15You know, like we have the big ones like Reddit
00:04:17and YouTube and Hacker News and all that.
00:04:19But you also can do LinkedIn, Pinterest, Blue Sky.
00:04:22You really have everything.
00:04:23Now, some of these are really easy to set up.
00:04:25And when you install this skill,
00:04:27which, again, is one single line.
00:04:28I'll show it in a second.
00:04:29It will just automatically be set up for you.
00:04:31There's no like API key required.
00:04:32We don't have to pay for anything.
00:04:34Other things require certain dependencies.
00:04:36And again, Cloud Code will walk you through that
00:04:38when you run this skill.
00:04:39And lastly, there are certain platforms
00:04:42that just require an API key,
00:04:43like things like X or Twitter.
00:04:44Yeah, you will need an X AI API key
00:04:49to actually scrape Twitter.
00:04:50That being said, it's not particularly expensive.
00:04:53Like Twitter is really the most expensive thing here.
00:04:55And for every single run I did,
00:04:56it was usually on average about 10 cents.
00:04:59Now, the other thing you want to pay attention to
00:05:01is the ones like TikTok and Instagram Reels.
00:05:04It requires an API key from scrape creators.
00:05:08However, when you run this and install this skill,
00:05:11what it's going to do
00:05:12is it's going to set you up
00:05:14with like essentially a subsidized account
00:05:16that gives you several thousand free calls.
00:05:20You can essentially run this for free
00:05:21without ever paying scrape creators
00:05:22for like six months
00:05:23if you use this every single day.
00:05:25So this is actually like very easy to set up
00:05:28and you get access to a ton.
00:05:30And inside the repo itself,
00:05:31it breaks that down to
00:05:31which sources require what and what they cost.
00:05:35Now, before we go into the install and the demo,
00:05:36let's talk about how it works.
00:05:38So basically what you're going to do
00:05:40is you are just going to invoke the skill
00:05:42inside of Cloud Code.
00:05:43So you're going to do slash last 30 day skill
00:05:45or just say use the 30 day skill.
00:05:46And you're going to give it a topic.
00:05:48Just like I showed you in the example
00:05:49where I said,
00:05:50what are people saying about Cloud Opus 5?
00:05:52It's then going to take your crappy prompt,
00:05:55whatever that happens to be.
00:05:56It's actually going to make it a little bit better.
00:05:58And then it's going to spread out
00:06:00and start hitting all these different platforms.
00:06:02Like it says here, it's smart about it.
00:06:03So if you said something like,
00:06:05hey, I want you to search for things about Kanye West,
00:06:07it knows which sort of subreddits
00:06:09it should look at
00:06:10and which sort of Twitter handles are relevant.
00:06:12It then searches all of them in parallel.
00:06:14So this is actually a pretty quick process.
00:06:16And then it goes in depth.
00:06:17Like I said, we aren't just looking at titles of posts.
00:06:20We are looking at the comments.
00:06:21From there, it's going to rank the information
00:06:22and synthesize it into a brief.
00:06:24So if it's seen the same things over and over again
00:06:27on different platforms,
00:06:27it's going to rank that way higher.
00:06:29There's not just going to be like one Reddit comment
00:06:31that kind of like messes it all up.
00:06:32And it's telling you report like,
00:06:34this is the big deal.
00:06:35This was the comment that got upvotes.
00:06:36It's like, okay, hey, if I saw this on Reddit
00:06:38and I saw this on Twitter,
00:06:38I saw this on YouTube,
00:06:39there's probably something here.
00:06:41Now for install, very easy inside of Cloud Code.
00:06:43It is a single line of code.
00:06:44This will be in the GitHub repo
00:06:46and I'll have that link in the description.
00:06:48Furthermore, this repo goes pretty in depth
00:06:49in terms of other sort of agents.
00:06:51So if you're using anything else
00:06:53like Codex and Cursor and all that,
00:06:54very simple to do.
00:06:56And also even shows you how to do it
00:06:57on the claw.ai web app.
00:06:59So now that you understand what it does,
00:07:00how it works and how to install it,
00:07:02let's just demo it real quick.
00:07:03All right, so to use this skill is very simple.
00:07:05Like any skill, we can invoke it
00:07:06by simply doing forward slash last 30 days
00:07:09or we can use natural language.
00:07:11From there, we are just going to give it our prompt.
00:07:13So our prompt is do some research
00:07:15on what people are saying
00:07:16about the last 30 days skill.
00:07:18Now, you saw all those different sources
00:07:21we had available to us.
00:07:22If I don't give it any like flags
00:07:24or additional information,
00:07:25it's going to use all of them
00:07:26if it thinks it's relevant.
00:07:28You also have the ability
00:07:29to kind of like scope down.
00:07:31So I could say,
00:07:31hey, I want to know about last 30 days,
00:07:33but only check Reddit
00:07:34or only check Twitter
00:07:35or only check YouTube.
00:07:36So it's pretty flexible in that regard.
00:07:38And remember, if you don't know
00:07:40like what your options are
00:07:42in terms of how to use this skill,
00:07:44well, clawed code knows.
00:07:45The fact that it downloaded the skill,
00:07:46it knows about the skill.
00:07:47It can sort of teach you the best practices
00:07:49if what you want to do isn't covered,
00:07:50you know, in this video.
00:07:52So when I go ahead and run it,
00:07:53it's not really going to ask me
00:07:54any questions or anything.
00:07:55There's no back and forth.
00:07:56It's just going to invoke the skill
00:07:57and get to work
00:07:58and start running a bunch of agents
00:08:00in parallel
00:08:00to hit all those different platforms.
00:08:02So to actually execute the search,
00:08:03it runs a deterministic Python script
00:08:06to go ahead and scrape everything.
00:08:07So after five minutes,
00:08:08this is what it came back with.
00:08:09It gives us sort of that one-liner summary
00:08:11at the beginning
00:08:11that this repo is growing faster
00:08:13than the hype post can keep up with it.
00:08:14And it gives us sort of the live numbers.
00:08:16It talks about sort of like
00:08:17the general through line for it all,
00:08:19which is that it fixes generic AI research.
00:08:22And it sort of shows me
00:08:23like where other people are talking about this
00:08:25in terms of YouTube and Twitter.
00:08:26It talks about the key patterns.
00:08:27It's off from the research
00:08:28and then it also breaks it all down by platform.
00:08:31So five threads for Reddit,
00:08:3222 from Twitter, et cetera, et cetera.
00:08:35And then it shows us
00:08:36where we can find the raw results.
00:08:38So just the raw markdown.
00:08:39And if you want the raw JSON,
00:08:41you can ask for that as well.
00:08:43For contrast,
00:08:43let's see what we get back
00:08:44when we give it essentially
00:08:45the same exact prompt,
00:08:47but we tell it explicitly,
00:08:48do not use the 30-day skill for this.
00:08:50And here's the report it gets us
00:08:51with the pure web search.
00:08:52And again, it's not wrong.
00:08:54It's just much more shallow.
00:08:55And we don't really get things
00:08:56like user sentiment.
00:08:57Even when it talks about
00:08:58what people praise,
00:08:59it's rather generic
00:09:00because it's not looking at comments
00:09:01and it's not looking at transcripts
00:09:03or anything like that,
00:09:04which means using web search
00:09:06for like your day-to-day
00:09:07on like random questions
00:09:08inside of Cloud Code isn't wrong.
00:09:10We don't have to use
00:09:12last 30 days for everything.
00:09:13In fact, that would suggest you don't
00:09:15because it is kind of like a big deal.
00:09:17But I think what this really buys us
00:09:19is something in between web search,
00:09:22which is surface level
00:09:23and doing, hey,
00:09:24let's do slash deep research,
00:09:25dynamic workflows,
00:09:27you know, 3,000 subagents, right?
00:09:29This is that in between.
00:09:30And especially,
00:09:31and I would argue
00:09:31it might even be better
00:09:32than deep research
00:09:32when it comes to user sentiment.
00:09:35So like you saw there,
00:09:36really simple to use.
00:09:38The install is one line.
00:09:39The only thing that will possibly trip you up
00:09:41is when you're trying to connect
00:09:43specific sources.
00:09:44But again, Cloud Code,
00:09:46just throw out this repo link
00:09:47and it will walk you through
00:09:48the specific ones you need.
00:09:49And really,
00:09:50the only one that's going to cost you money
00:09:52is going to be the X in Twitter one.
00:09:54And for reference,
00:09:55every time I've used this,
00:09:55like I said before,
00:09:56about 10 cents,
00:09:58give or take.
00:09:59So as always,
00:09:59let me know what you guys thought
00:10:00of this video.
00:10:01Make sure to check out Chase AI Plus
00:10:03if you want to get your hands
00:10:04on my Cloud Code masterclass.
00:10:05There's a link to that
00:10:05in the pinned comment.
00:10:07And I'll see you around.