Use These 17 Claude Plugins, It Will Make You 10x Better.

CChase AI
Computing/SoftwareSmall Business/StartupsInternet Technology

Transcript

00:00:00While ClaudeCode out of the box is one of the strongest AI tools on the planet, we can make it
00:00:04way better by simply adding plugins, skills, and CLIs to our stack. The problem is there are
00:00:11hundreds of plugins and skills and things you can add to ClaudeCode. So how do you know what is
00:00:15actually worth your time, especially if you're just getting started? Well, in this video, I'm
00:00:20going to help you out by going over 17 different plugins that I actually use and think are worth
00:00:24your time. We're going to break it down into three different buckets, data, design, and
00:00:28productivity. And for each one of these, I'm going to go over what it is, how it works, and why you
00:00:33should care. We got a lot to cover, so let's hop in. One of the things you hear over and over again
00:00:39when it comes to design is the idea that AI lacks taste. Well, what if we had a skill that gave it
00:00:45taste? And that's exactly what the taste skill does. This is an open source GitHub repo that's all about
00:00:51defeating the AI slot monster. Now, the taste skill GitHub actually includes a number of sub-skills,
00:00:56things like image to code skill, the redesign skill, the output skill. And all of this is meant
00:01:01to give you better front end design, make your websites look better. This works with every single
00:01:05agent out there, not just ClaudeCode. So if you're using AI to create landing pages or anything that
00:01:10is design oriented, you're going to want to check out the skill. Here's an example of one of their
00:01:14websites. We can see we have some sort of like scrolling animations. And in general, this is definitely a
00:01:19different vibe than what you normally see from something ClaudeCode would produce. Here's another one
00:01:23that's a bit more on the traditional SaaS side. But again, if ClaudeCode essentially one shot
00:01:28something like this for you, I think most of us would be happy. So if you're looking to go beyond
00:01:32the default Anthropik front end design skill, this is something I would take a look at. Now,
00:01:36number two on the list is impeccable. And this is another open source front end design skill that is
00:01:41meant to get rid of AI slot and improve on what Anthropik already gives us. Now you're going to hear a
00:01:46lot more about impeccable because just a couple of days ago, they became a built in layer for everybody
00:01:51who uses the GitHub Copilot app. So GitHub themselves saw this, thought they did such a good job and was like,
00:01:56Hey, let's actually make this native to our AI tool. Now, just like the taste skill, there are layers to this
00:02:01impeccable skill. It is one single skill, but it includes 23 different commands, everything from
00:02:07documenting what it's done to critiquing what you've already created, polishing it, making it bolder, quieter, et cetera,
00:02:13et cetera. There's honestly so many commands that it's hard to know what this actually does. Now, the easiest way to see all
00:02:17these commands in action is on the impeccable website, which is impeccable.style. So if I scroll down, I can
00:02:22see all the commands here on the left. I get a quick explanation of what it does. And here on the right, I can
00:02:28actually see a difference between what cloud code would create and what impeccable would give us. So this is
00:02:33the distill command where we have this sort of mini dashboard and we want to just kind of strip it down to
00:02:38its bare parts. Well, distill would give us all that same information, but in a cleaner profile. Another cool thing
00:02:43about impeccable is the browser editor. So you can actually bring up your website on your computer. This
00:02:49is in beta. And instead of having to edit these things inside the terminal, you can actually go onto
00:02:53the website, select different things and say, Hey, I want to do X, Y, and Z with impeccable to change
00:02:57this. And you can see the changes live instead of having to do everything through code gives you sort of a
00:03:01visual element to all this. And I think this live editor is what really separates from something like the
00:03:06pace skill. So if you really want to get hands-on with your design work, I highly suggest you give it a shot.
00:03:11Now, moving on to our third tool on the list, we have awesome design.md. This is all about taking
00:03:17websites that already exist and using them as templates for what we are going to build. This
00:03:22is based on the design.md principle from Google stitch. Now, if you've never used Google stitch
00:03:27before, I highly suggest you check it out. Totally free tool for front end design work. But what we're
00:03:31talking about here is this design.markdown file. So it's over here on the right. And this is what
00:03:37Google came up with in terms of creating the perfect prompt to have AI create strong front end design.
00:03:42So awesome design.md uses this as a template and applies it to websites that already exist. Now,
00:03:50what do I mean by that? Well, you see right here a collection of a bunch of different websites. So
00:03:54let's say we like how Airtable does their website. If I click on Airtable,
00:03:59I then get a breakdown of their entire site structure, colors, surfaces, text, typography,
00:04:06spacing, buttons, everything that I need to use their website as a building block for my own. So
00:04:12we're not cloning their website, but we're sort of cloning their design language and applying it to
00:04:18our own. And they have this for a ton of websites and they have it broken down by use case. So fintech
00:04:23and crypto, designing creative tools, productivity, SaaS, on and on and on and on. So this is perfect
00:04:29if you're someone who is really just looking to take what's already working and apply it to your
00:04:34website. Now, before we jump into the next tool, a quick word from today's sponsor, me. So I just
00:04:40came out with a Cloud Code Masterclass and it is the number one way to go from zero to AI dev,
00:04:44especially if you don't come from a technical background. I focus on real use cases. It's
00:04:48updated every single week. So if you're a little confused and you don't know where to begin with all
00:04:53this Cloud Code stuff, this is the best place to start. You can find a link to it in the pinned
00:04:58comment. It's inside of Chase AI+. I hope to see you there. Now, our first tool in the productivity
00:05:04bucket is going to be Ponytail. This is one of the fastest growing AI repos in the world right now,
00:05:09and for good reason, because it allows Cloud Code to be way more efficient. It makes it faster,
00:05:14it makes it cheaper, and it allows it to write less code while maintaining the same outputs.
00:05:19So they're claiming Cloud Code will be just as good, but faster and cheaper. Now, we can look at some of
00:05:24the numbers right here. And so with Ponytail versus the baseline, we're looking at 50% less code written,
00:05:3022% less tokens used, 20% cheaper cost, and then time-wise, it's 27% faster. Well, how does this
00:05:37actually work? Well, it essentially asks Cloud Code a few questions before it actually writes any code.
00:05:42First, it asks, does this need to exist? Then, is this already in the code base? Does a standard
00:05:47library do this? Does a native platform feature actually have this? Is it an installed dependency?
00:05:52Is it one line? And only then, after it goes to those questions, does it write any code? And it says
00:05:57it's going to write the minimum that works. The whole idea here is that Cloud Code tends to be rather
00:06:01verbose, and it likes to build things from scratch, even if they've already exists. So it's pretty much
00:06:06asking, hey, does this feature that you want to build on your own, is it already out there? Is it
00:06:11already in the code base? Is there already a library for that? If so, just use that. Now,
00:06:15a note on these numbers, these are all with Haiku. In the video I did, we did the same test,
00:06:19the same benchmarks using Opus, and the numbers were even more drastic. So I highly suggest you check
00:06:24this out if you're someone who's looking for a quick, easy way to save on tokens. Now, this next one
00:06:29is one of my favorites, and that is notebooklm-pi. This allows us to connect Cloud Code to
00:06:34notebooklm. Now, if you've never used notebooklm before, it's one of Google's best products. It's
00:06:39completely free. You essentially throw any sort of resources you want at it, whether that's PDFs,
00:06:44documents, YouTube videos, anything, or have it go find these documents, and then you can chat with
00:06:49AI about those things. On top of that, you can ask it to create deliverables for you. I'm talking like
00:06:54slide decks, images, infographics, videos, whatever you want. Again, completely free.
00:06:58Now, there's not an API for notebooklm, but this CLI essentially gets around that and allows us
00:07:04to do everything you would normally do inside of notebooklm, but through the terminal, which is
00:07:08great if you're trying to create some sort of workflow that brings in notebooklm. And again,
00:07:12you're sort of offloading all this work to the Google servers for free. On top of that,
00:07:16the CLI actually gives you more tools than the web app itself. We can do batch downloads,
00:07:21quiz and flashcard export. We can save chat to nodes, on and on and on. So this is,
00:07:25you know, actually more powerful than being here as you normally would.
00:07:30Now, the GitHub gives you a bunch of use cases, but I tend to use it the most
00:07:33in combination with YouTube. So it's able to get links to YouTube videos. It gets the transcript
00:07:37and it answers any questions about it. And again, doing that through the terminal
00:07:40is awesome. Next up, we have Playwright CLI. I think this is one of the most powerful ones
00:07:44we're going to go over today. This is all about browser automation. So if you've ever done something
00:07:48or you want to create an automation where you're like, ah, there's not an API that exists,
00:07:51but I want Cloud Code to almost act like a human and go on the website and click on things
00:07:55and fill out certain forms. Well, the Playwright CLI can do that. And beyond browser automations,
00:08:00this is also great for front end design. I actually thought about putting it in the design
00:08:04bucket. So imagine you're creating some sort of website and you have some sort of form people
00:08:08need to fill out. Well, instead of going through all the edge cases yourself manually, you can spin
00:08:12up Playwright CLI. It will open up a ton of browsers, test it all for you in just a couple
00:08:17minutes, and you're completely hands off. Now, don't confuse this with the Playwright MCP.
00:08:21The Playwright CLI is much more effective than the MCP and it uses way less tokens.
00:08:26Next up, we got the Codex plugin for Cloud Code. This is an official plugin from OpenAI,
00:08:30and this allows you to connect Codex and the GPT models to Cloud Code. Now, this is perfect for
00:08:36things like code review or adversarial review, because we all know that Cloud Code loves the
00:08:41code it writes itself. And sometimes we want a second set of eyes to take a look at what it did.
00:08:45Well, this plugin allows you to do that. It even has commands like Codex Rescue, where you can offload
00:08:50entire features to Codex. So you can have Cloud Code work on one aspect of your application and then
00:08:57have Codex work on another simultaneously, and you can sort of get the best of both worlds.
00:09:01Next on the list is GWS, the Google Workspace CLI. This is not an official Google product,
00:09:06but it was created by a Google developer and it got so popular that it actually got the guy fired.
00:09:11Now, this is great if you're inside the Google Workspace all the time and you want more functionality
00:09:15than what the Google Connector gives you. Because if you've used the Google Connector, you know that
00:09:19you can't do things like send emails. But what if you want to offload that ability to AI? Well,
00:09:24this allows you to do that. And not only gives you that functionality, it includes 40 plus skills to
00:09:29do all sorts of things. So if you're doing things like weekly digest, you want a standup report,
00:09:34a meeting prep, an email to task. There's a bunch of like preloaded workflows all ready to go
00:09:39inside of here. Now, the installation and setup for this tool can be a little more complicated.
00:09:44But if you really want to add power to your Google stack with Cloud Code, you can't go wrong with
00:09:49this. Now, here's one everyone should already have at this point. And if not, it should be like the
00:09:53first thing you install. And that is the GitHub CLI. If you are creating anything with Cloud Code,
00:09:58at some point, you are going to want to push it to GitHub. And the CLI makes it extremely easy to do so.
00:10:04Now, the last one in the productivity bucket is the skill creator skill. And this is an
00:10:07official anthropic skill. And this does more than just create new skills. It allows us to modify and improve
00:10:12existing skills and measure skill performance. Skills are awesome. But sometimes we don't even know if
00:10:19this skill is something that needs to exist, or if we're trying to improve a skill that already exists,
00:10:24how do you know it's actually making it better? Well, the skill creator skill allows us to
00:10:28automatically do A-B tests with the first iteration of our skill and the so-called improved version,
00:10:33and also test things like how does this work with the skill versus without the skill. So you can get
00:10:38some sort of objective proof if the skill you're trying to work on actually makes sense. And since
00:10:45skills are one of the most powerful things we have inside of Cloud Code, I would argue that the skill
00:10:49creator skill is the most important skill out there. Now, to install this is really simple. Once you're
00:10:54inside of Cloud Code, just do forward slash plugin. And then go ahead and search for the skill creator
00:11:00skill. Once you do that, go ahead and install it. And if you shift over to installed, you'll see it
00:11:06right here, skill creator. So now let's talk about data, whether that's databases, research, anything
00:11:12about grabbing information and storing it and things like memory. And the first one I want to talk about
00:11:17is last 30 days. At one point, this was the number one repo on GitHub. And this is all about doing
00:11:23research with Cloud Code and research that goes well beyond a simple web search. And instead of a simple
00:11:28web search, this goes really deep on a number of specific sources. We're talking Reddit, Twitter,
00:11:34YouTube, TikTok, Reels, Hacker News, Polymarket, on and on and on and on and on. So you get a really deep
00:11:40look at a particular subject and what people are talking about when it comes to that subject on these
00:11:45different platforms. This is perfect for anybody who's doing any sort of like daily briefing type things
00:11:51inside of Cloud Code or creating any sort of deliverables that rely on getting legit data. And this is also
00:11:58a great alternative to just throwing slash deep research at everything and burning a bunch of tokens with
00:12:03dynamic workflows. And speaking of research, let's talk about the Firecrawl CLI. Now Firecrawl is one of the best
00:12:08tools out there for scraping the web, especially when we're trying to scrape web pages that have a lot of bot
00:12:13protections. Now Firecrawl has a paid version and an open source version. The paid version gives you the most power in
00:12:21terms of being able to get on websites that have these sort of like bot protections due to their own
00:12:26proprietary model. But we can get a lot of the same functionality with their open source version. So there is an open
00:12:32source version of Firecrawl. This is their GitHub. And so this is perfect if you just want one tool to do specific web
00:12:39scraping, you don't need last 30 days, you're not trying to hit a billion sources at once. I just want one tool for the job.
00:12:44And Firecrawl is awesome for that and definitely a step above the standard Cloud Code web search. And that's because it can do a
00:12:49little more than just scrape, it can actually interact with pages, it can discover all the URLs on a website, it can crawl all the URLs on the
00:12:57website. So you have a lot of settings you can essentially tweak if you're very specific about what you want to grab. Next up, we have auto research. And this comes from
00:13:06Karpathy. This is essentially machine learning in a box. What this allows us to do is take some sort of application that we are
00:13:14trying to improve, we throw auto research at it, and then it's going to automatically run loop after loop after loop, trying to
00:13:21improve whatever we've decided we want to improve. For example, let's say we had some sort of Python application that we wanted to
00:13:27run faster. That's a specific success criteria that's related to time, right? It took one second, it took 0.99 seconds. So we
00:13:36would throw that into auto research, and it's going to run experiment after experiment until it gets that
00:13:41runtime as low as possible. And you can see an example of something like that. In this case, it ran 83
00:13:46experiments. After every experiment, it tried something new and it has a log showing here's what we tried,
00:13:52here's what worked, here's what didn't. And after 83 experiments, it had 15 improvements. And this was all
00:13:58done automatically. Again, this is sort of like a machine learning type thing in a box that we can add the
00:14:03cloud code. The cool thing about this project is it's also pretty lightweight. But remember, we can't use
00:14:08this for everything. We need to be working on some sort of application or some sort of task that has
00:14:13very specific success criteria that is objective. Again, think of time and numbers. Next up, we have
00:14:19the Supabase CLI. And I love Supabase because it has a generous free tier. And it covers a lot of our
00:14:25bases when it comes to creating applications. So imagine you're creating a website that has a form
00:14:29submission that includes somebody's emails. You're trying to grab emails. Well, where are those emails
00:14:33going to live? Well, they need to be in a database. Supabase can handle that. It can create the databases
00:14:38from cloud code. I just say, hey, use a Supabase CLI, create a database that makes sense. And it will do
00:14:43all of it automatically with natural language. It also handles things like authentication. So do you need
00:14:49some sort of login for your website? Well, Supabase can do that as well. So it allows us to do a lot
00:14:53from one place. And again, some of this stuff can get complicated if you've never done it before. But
00:14:58using cloud code and giving it this sort of Supabase tool to use allows it to be very simple. You just
00:15:05use natural language and it will guide you through the whole process. And Supabase is also something
00:15:09you can run locally if you want to do that as well. Next up, we have Obsidian. And Obsidian is one of
00:15:13the easiest ways to improve cloud code's memory. Obsidian allows us to designate folders on our computer
00:15:19as vaults. And these vaults essentially become a treasure trove of information that we put inside
00:15:23of there. And when we open up cloud code inside these vaults, it now essentially becomes connected
00:15:29through this knowledge graph of different documents that we've added. And if we set it up correctly,
00:15:35and I have tons of content on how to do so, this gives cloud code a map of all of our documentation,
00:15:41which means it can answer questions about it, but answer questions about it very effectively and
00:15:45efficiently. Now, there are also skills that we can add to cloud code that help it use Obsidian in the
00:15:50best way. If you just go to GitHub and you search up Obsidian skills, this simple GitHub repo was created
00:15:56by the founder of Obsidian and essentially teaches cloud code the best practices for how to integrate
00:16:01itself into your Obsidian stack. Now, speaking of memory, let's talk about RAG or retrieval augmented
00:16:07generation. This is definitely a step up from something like Obsidian because this isn't like a
00:16:12quote unquote fake knowledge graph. This is the real deal when we're talking about knowledge graphs and
00:16:16embeddings. And light rag is one of my favorite tools for creating these sort of knowledge graph
00:16:21RAG systems. It's lightweight, it's fast, and it's a great introduction to more complicated RAG systems.
00:16:27And again, you can essentially run everything through cloud code. You can do all your queries through
00:16:32cloud code, and it's very easy to connect these two things together. And again, I have content
00:16:36on this exact setup. And once you get comfortable with things like light RAG, you can move into even
00:16:41more advanced stuff like RAG anything, which moves us away from simple PDF and text documents and allows
00:16:46us to include things like images and graphs and charts, things that are usually harder for traditional RAG
00:16:53systems to handle or even systems like Obsidian. And last but not least, we have the Stripe CLI. This is
00:16:58great if you're creating any sort of applications that you want to actually make money from. If you're doing
00:17:02any sort of transactions, you're going to want the Stripe CLI because it makes handling Stripe way
00:17:08easier. The Stripe UI can be kind of a pain in the butt to go through. And again, anything that
00:17:13allows us to control our application and edit it through the terminal and through natural language
00:17:17and cloud code is a huge benefit to you. So that's where I'm going to end this for today. As always,
00:17:23let me know in the comments what you thought. Make sure to check out Chase AI Plus if you want
00:17:26to get your hands on my cloud code masterclass, and I'll see you around.

Key Takeaway

Enhancing ClaudeCode with specific CLI tools, plugins, and custom skills like Ponytail and Impeccable significantly increases development speed, lowers token costs, and improves design quality.

Highlights

  • Integrating specialized plugins into ClaudeCode enhances performance by automating browser tasks, improving design aesthetics, and optimizing token usage.

  • The Ponytail tool reduces code verbosity by 50% and decreases token usage by 22% by verifying existing dependencies and libraries before writing new code.

  • The 'Skill Creator' skill enables objective A/B testing of custom skills to verify if iterations actually improve outcomes within ClaudeCode.

  • The Impeccable tool provides a live visual browser editor for front-end design, allowing users to modify website elements directly rather than through terminal commands.

  • Auto Research automates improvement loops by running iterative experiments on applications based on specific, objective success criteria like execution time.

  • The 'Last 30 Days' skill enables deep research across Reddit, Twitter, YouTube, and other platforms as an alternative to token-heavy dynamic searches.

Timeline

Front-End Design Enhancements

  • The Taste skill offers sub-skills like image-to-code and redesign to improve front-end aesthetics.
  • Impeccable integrates directly into GitHub Copilot and features a live visual browser editor for real-time design adjustments.
  • Awesome design.md utilizes Google’s design principles to clone the design language of established websites for new projects.

These tools address common limitations in AI-generated design by providing specific skills to improve visual output. Impeccable separates itself from other tools by offering a visual editor that removes the need to perform all edits within the terminal. Awesome design.md provides a systematic way to deconstruct the layout, typography, and spacing of existing successful websites to inform new designs.

Productivity and Workflow Automation

  • Ponytail optimizes ClaudeCode performance by reducing redundant code and cutting token costs by 20%.
  • The notebooklm-pi CLI allows terminal-based interaction with Google's NotebookLM for processing PDFs, videos, and documents.
  • Playwright CLI facilitates browser automation and testing without relying on specific APIs.
  • The Skill Creator skill allows for automated A/B testing of custom skills to measure performance improvements.

These tools focus on efficiency and offloading complex tasks. Ponytail enforces a check-list of existing libraries and features before allowing the AI to write new code. The notebooklm-pi CLI creates a pipeline for analyzing large sets of resources for free. The Skill Creator stands as a vital tool for verifying if additions to the ClaudeCode stack actually deliver measurable results.

Research, Data, and Memory Systems

  • Last 30 Days conducts deep research across multiple social media and news platforms.
  • Auto Research uses machine learning loops to iterate on specific success criteria like code execution time.
  • Obsidian and Light RAG improve memory and context retrieval for AI models by using local vaults and knowledge graphs.
  • The Stripe CLI simplifies transaction management for applications developed within the terminal.

These tools expand the data-handling capabilities of ClaudeCode. Auto Research automates experimental cycles, such as optimizing a Python application, until a specific metric is met. Memory management is addressed through Obsidian, which connects folders as knowledge graphs, and Light RAG, which provides a more sophisticated retrieval system for complex information.

Community Posts

View all posts