Transcription
Cloud code seems to be changing by the day, and as someone who's literally trying to stay on top of this space to 10x myself as a cloud code user, I get overwhelmed. We even have legends like Andre Carpathy saying that even he feels behind when it comes to keeping track of everything that's coming out and shipping.
So naturally, I built something to fix this. A system that shows exactly what changed with Claude Code, analyzes what matters, and even can send you a voice note to your email to read all the updates to you. So instead of getting lost in the sauce and lost in the myriad of posts, you can know exactly what changed, why it matters, and most importantly, how you can use it to level yourself up.
So in this video, I'm going to walk you through exactly how the system works and give you everything that you need to build it on your end. Let's get into it.
Now, if you're not familiar, Anthropic has a GitHub repository where they maintain all the updates they make to Cloud Code. So this is called the change log, and you can see this is the last latest version right here. It goes through different slash commands they've added, bug fixes, overall enhancements. But if you're just looking at bullets, it's hard to differentiate what's a nice to know and what could fundamentally change the way that you build using cloud code.
Now, theoretically, you could just subscribe to this repo and every single time a change comes out, you could be notified by GitHub, take a look at the bullets, take the bullets, throw it into some form of workflow to break it down for you, and then sift through the TL;DR to see what matters to you. But there must be a better way. Right? Right?
Which is why I built this. Now, on the face of it, looks very similar to what I just showed you. Maybe just a little bit prettier. But trust me, this is way more potent than meets the eye. Outside of the marginally more aesthetic bullets here, where you can see very quickly what is a new feature versus a fix, you can also do the following, where you can click play, and it will outline each bullet out to you in a custom voice from Gemini's audio API.
So, if I go on 2X here, I can have my AI assistant friend Sharon break this down for me. So, if I click on play, you'll hear the following.
>> Version 2.0.74 changes include new feature added LSP language server protocol tool for code intelligence features like go to definition, find references, and hover documentation. New feature added back/terminal. >>
You get the idea. But wait, there's much more. You can use this little chat assistant. And this chat assistant always looks at the latest thing that has been fetched from the GitHub repo. And you can chat with one version or multiple versions. So I can click on 2.074 and say what changed. And then it should be able to go through all the notes and break it down to me in plain English. And if I want I could click on multiple versions. There you go. Here's the breakdown. And if I want I could say something like this. Okay. Thanks for the breakdown. What is LSP? Can you break it down for me like I'm in grade eight or nine? I don't know how this would fundamentally change that I the way I would build with cloud code. Break it down for me so you can actually clarify what matters and how it matters to you. And you'll see here if we zoom in just a tad it says LSP stands for language server protocol. Think of it as a universal translator for programming languages. Before imagine you're trying to learn 10 different languages and every time you switch to a new notebook the way you write and look up words changes. So this seems like it's very developer oriented. It breaks it down.
Now, on top of that, we can click on multiple versions to actually ask a question about the progression of changes. So, from all of these different changes and change logs, what do you think are the top three things that would really impact the way that I would build my kind of Claude code ecosystem? Right now, I'm trying to build my second brain cloud code. So, just walk through the core things that might matter to me the most. So, now you can really zero in on what matters. And if we send this over, the best part of this is every single time there's a fundamental change in the change log, it should autosync here. So this second sidebar will always be up to date. It now comes back with a grounded and relevant answer tailored to our exact use case. So it says for building a second brain and robust ecosystem within cloud code, these are the three most impactful updates. And rightfully so, it came up with this, the Claude in Chrome beta, which was a great update, which if you're not familiar, allows Claude to take control of a Chrome browser through the Claude extension. So, you have better browser access and the ability to do headless browsing, scraping, etc.
Now, if you've been a watcher of this channel for months, you know I'm not going to stop there. So, we have yet another tab called the What Matters tab. And this tab gives you the TL;DR of the most recent update that you can again click this speaker icon and have it transcribe for you. Or you can go down to the bullets and you can see what the critical changes are, any major features, important fixes, and if you need some form of accountability to make sure you test out these features at the very bottom here on the action items feature, you can see it tells you you should go and run the terminal setup if you're using the following languages. Adjust muscle memory for thinking toggle moving from tab to alt. So, it gives you all the key things you know behavior-wise to use this tool to the best potential possible.
And of course, if this light mode is blinding you, then you can always click on this dark mode and go to your settings tab and turn it on to be dark mode by default. And beyond that, you can also enable cron jobs. Now, if you don't know what a cron job is, it's essentially a time-based scheduler. So, I think it's derived from the Greek god Kronos, which is also related to the concept of time. And what you could do is email notification sending. So what you can do is essentially set the email notifications so it can send once a day, once a week, whenever it fits for you. And so long as this cron job is running, every single time it detects that there's a change in the change log, it will send you an email like this where if you're on the go, if you're at the gym, if you're a general nerd like me, where I even want to learn about cloud code while walking my dog, it will send you this TL;DR with the features and naturally this little voice note that you can listen to wherever you are. So this leaves you with little to no excuse to be a Claude code power user.
And the way I set this app up is it uses something called SQLite, which is a free database you can run locally. So you don't need to pay 10 bucks to Supabase or have something on cloud. This can all run on your local computer. So this is the TL;DR of the app. And now I'm going to walk you through exactly how I built it.
So now I'm going to walk through this excel sheet, which breaks down how this app is built in easy to understand images that even if you're not technical, you should be able to follow along. Now, at the root of it, we only have a few services being used to bring this all together. One, naturally, we have cloud code. Two, we're using the Gemini API for inference on the chatbot and for the auto transcription. And then lastly, we have a service called Resend, which essentially allows you to easily send yourself an email and ping yourself from something like cloud code.
Now, typically another alternative you could do is set up some form of webhook on Zapier where you ping that server, then you take that ping and then you send it to a Gmail node where you log in and then go through that whole wraparound. But with something like Resend, it has a very generous free tier. So, as long as you're using this for personal use, you shouldn't have to exceed it. You can do the exact same process without having to add overhead of an additional workflow, webhook management, etc. And all you really need to do with Resend is just give the API key to cloud code and it will take care of the rest.
Now getting more in the nitty-gritty, all that's happening behind the scenes is there is a GitHub URL. If we go back to the app for a second, if you look at this little link here, you'll see this is the exact URL we're referencing, which if we click back, you'll see that this is the raw user content from the change log. So what's happening is once a day when this runs, this goes and grabs the raw content and then pushes it through the rest of the system. Once we fetch it, it parses it. Basically breaks it down into very easy to understand and easy to process text. We send that to Gemini 3 Flash, which is very quick, very cheap, and very smart for that level of inference. And then we push it to our SQLite. This is our database, our free database that lives on our system. And anytime it is updating, it goes and updates the UI which is on React. It's just a framework. And then there's a change log view like I showed you. There's the what matters view which has to run to update for all the new TL;DR, the new insights, etc. And then we have the chat which only works when you actually invoke it and send a message. And then naturally we have the audio which does text to speech. And then we have this email Resend. So it's not actually very complicated. It just thinking through what this would look like to be as simple as possible but still powerful is the hardest part.
And if you need yet another visual, we have this raw change log that I just showed you. Gemini Flash applies some system prompts to it to put it in this format. So you have a TL;DR, you have the updates, you have the features, you have the fixes, and then all of these get mapped to these different cards here. And these are the cards that you saw on the front end.
Now to make the app efficient enough so that if I ever invoke this little transcription function that we don't have to re-transcribe it every single time, we are caching any audio that we've generated. And in terms of the text to speech, all that's happening is if we click on this button here, and we've never clicked it before. So I'll give you an example. So if I click on this bad boy, and then we click on this speaker icon right now, you'll see it's invoking the API. It's going to come back with an audio that will probably autoplay randomly. Once it is invoked and we've generated the audio, we are caching it within our local browser. So you don't have to regenerate it every single time, which is why you can see this one. I can keep playing on demand. And to make it play on your browser, all that's happening is we're taking the Gemini text to speech. It's getting formatted into something called Base64. If you've ever worked with Zapier and multimedia files, you'd have had to encounter the concept of Base64 before. We then convert it to a wave file, which just allows you to easily play it from your browser. And that's pretty much it from a nutshell.
So, with this understanding and the prompt playbook I'm going to walk you through and give you in the second link in the description below, you should be able to recreate my system in no time.
Now, to come up with the initial prompt that I gave Claude Code to build this beautiful app, all I did was actually go to Perplexity and enable the labs feature. I really like it cuz it's quick. It does really proper research. It knows to look at the most recent sources. And I just said, "I have an idea to ask Cloud Code to build an app for the Anthropic change log so I can keep track of all the newest features." That's pretty much the TL;DR. Then I gave it a link to the change log. I told it I want to be able to use Gemini speech generation. So I just gave Perplexity everything it needs to dive deeper. And then it created this change log app.markdown file. This is essentially a full breakdown of what the app entails. I take this MD file along with a few other files I asked for. So I asked for a markdown file to feed cloud code on how to use the Gemini 3 API along with the text-to-speech API. So despite you seeing all these different folders and files, the core ones I started with to build this were just these three right here. And this is the change log app that has all the specifications that Perplexity using Claude under the hood came up with. And it was very thoughtful about the way it generated it.
Now, I rebuilt this app, by the way, two or three times because I would get halfway. I realized I built it inefficiently. So, what you will have in the second link in the description below is the latest and the most comprehensive change log markdown file. And the best part of this file is you can reuse a lot of the structure to build all kinds of apps without having to spend days. I built this in total, even after redoing it two, three times, in 6 hours. And it has basically everything from where it should look to what the general flow of the app should be on the back end, the tech stack that we need to use, the fact that we need to create an environment file. This is where you store all of your passwords, your API keys, etc. So, we need like we said Gemini, and we need Resend. Those are the main ones we need. And then obviously the email that needs to be notified. That's what Resend takes as an input. So, once you get those API keys, I throw them in. And this is the rest of the guide that you can read through after you watch this video.
Gemini 3 markdown file. Nothing fancy here. I just took from Gemini itself. I just said Gemini 3 Flash API. And then I just went to the developer guide. And my favorite thing to do is I click on this little dropdown. I click on view as markdown. I take this bad boy. I copy it into a brand new file. Just create a brand new file called Gemini whatever.markdown. And that's why I gave it all the documentation it needs to understand how to use this latest API. One thing I'm going to double-click on is if you just tell it "use Gemini 3 Flash" and don't tell it to go research it or remind it that it doesn't know it as a part of its training, it will use Gemini 2.0 Flash. Reason being these models are still not updated to be literally as of this month or as of this quarter. I think Claude is still back in late 2024 or early 2025.
And the audio understanding markdown file, exact same thing. It just breaks down all the different voices it can pick from so it can enable that for me because one thing I did not show you is if I go back to this UI, I can choose from the other voices from the API as well. So I wanted the ability to not have to hardcode that myself and just give it all the documentation it needs to understand it.
Once I had these three core files, all I had to do was do slash initialize and then I would allow it to read these three files, create the Claude.md file, and then I would start prompting from there. But instead of dragging you through the mud and walking you through my really crappy versions of prompts from the terminal itself, I created this prompt playbook for you. So it has the real prompts from this build along with what the better prompt should have been. So you're going to see my lazy prompt here that still ended up culminating in this app. This was "Build me a change log tracking app that fetches Claude Code's change log, analyzes it with Gemini, generates audio, and sends email notifications." And then I tagged like Lily did, an @ right here. So I do @claude change log if it pops up. I'm basically tagging that file saying go and implement everything that's already in there. So I don't have to reprompt the whole idea. And I have a refined version of this prompt that you could use. And then the second part was "Allow me to monitor multiple change log sources, not just Cloud Code." So I wanted to be able to add different URLs over time.
Now, at some point when it came to the UI, it looked very simplistic and I just said, "Make this look more like Anthropic's design." And I screenshotted different parts of Claude on the front end. A better version of that prompt would have actually told it, "Use these light backgrounds with these hex codes, these dark backgrounds, these primary accents, etc." And then in terms of the UI, make the favicon more vibrant. Initially, it was just like a normal global icon that looks very cookie-cutter. I just wanted to give it some more personality since I'll have it on and running in the background. And then this was the easy way to do that.
One thing that Claude came up with was adding a tag for sentiment analysis. So it would basically go back to here in the app. If I go back to this what matters, it would say this is a positive update. Very rarely would an update to the change log be a negative one. Every progress is good progress end of the day. So that was irrelevant and I removed that from there. And then one thing I noticed is that I couldn't go back in time. So if it generated a TL;DR and I refreshed it, it would go away. Meaning if we had a brand new update, it would just disappear. I wanted some form of traceability. So we have a SQL database that lives on our computer where I could go back to the old TL;DR if I wanted to.
And just to finish off this section, just for some prompting principles, instead of saying things like "use Google's AI," you want to be specific again and say "use Gemini 3 Flash for analysis with JSON mode." And when you want to cache something on the front end, you want to be specific. So if you want to be able to generate audio and have that audio persist in your browser and on your local computer, you'd say "cache AI analysis in my SQLite database by version hash. Check cache before each API call." And if you don't know what any of those words mean, by the way, you could just go and tell another chat in Claude, open a brand new chat and tell it, "Hey, I'm noticing that this app is not doing X. I want it to do Y." Write me a prompt that I can give to Claude code to allow me to do so.
Now, on top of this prompting playbook, I'll also give you the Claude Code Change Log Tracker, which goes through the stack once again, the architecture, the tech stack that I used, the environment setup, API integrations, everything you need to understand this. Beyond that, all you'd have to do is go to Resend, log in. Once you log in, you'll be able to click on your API keys right here in this section, get the API key, tell it exactly where the email should go to, and it should be able to take care of the rest from there. And Resend reliably keeps a whole log of every email that's been sent. So, you can see the status, the subject, when it was last sent, and that's pretty much it.
So, I'll make everything I just showed you available in the second link in the description below, so you can build this for yourself and hopefully be one step closer to mastering cloud code that much better. And if you want to skip having to go through the process and go through the prompts and just do a one-click clone of my repo along with tons of other systems that I'm building for my exclusive community, then check out the first link in the description below. One of my main goals for the community this year is to deliver as many valuable systems that my members can take off the shelves and get ROI from within minutes. So, if that interests you, then maybe I'll see you inside. For the rest of you, I'll see you in the next.