Claude Code (referred to in the video as "Cloud Code") is already one of the strongest AI tools available out of the box, but Chase AI argues you can make it dramatically better by adding plugins, skills, and CLIs to your stack. The problem is there are hundreds of options, so this video narrows the field to 17 tools Chase actually uses, organized into three buckets: design, productivity, and data. For each one he covers what it is, how it works, and why you should care.
A common complaint is that "AI lacks taste." The taste skill is an open-source GitHub repo built around "defeating the AI slop monster." It bundles several sub-skills, including an image-to-code skill, a redesign skill, and an output skill, all meant to give you better front-end design and make websites look better. It works with every agent, not just Claude Code, so it suits anyone using AI to build landing pages or anything design-oriented. Chase frames it as going beyond Anthropic's default front-end design skill.
Another open-source front-end design skill aimed at killing AI slop and improving on what Anthropic gives you. It recently became a built-in layer for everyone using the GitHub Copilot app, signaling its quality. It is a single skill but ships 23 different commands, covering documenting work, critiquing what you created, polishing it, and making it "bolder" or "quieter." The impeccable.style website shows every command with before/after comparisons (for example, the distil command strips a busy mini-dashboard down to a cleaner profile with the same information). A standout feature is the browser editor (in beta): you can pull up your site, click elements visually, request changes, and see them live instead of editing only through the terminal. This live visual editing is what most separates it from the taste skill.
This is about using websites that already exist as templates for what you build, based on the design.md principle from Google Stitch (a free front-end design tool). Google's design.markdown file is essentially the "perfect prompt" for strong front-end design. awesome-design.md applies that template to real sites: pick a site you like (Chase's example is "error table"), and you get a full breakdown of its structure, colors, surfaces, text, typography, spacing, and buttons, so you can use it as a building block. You are not cloning the site, you are cloning its design language. Entries are organized by use case (fintech and crypto, design and creative tools, productivity, SaaS, and more).
One of the fastest-growing AI repos right now. It makes Claude Code more efficient: faster, cheaper, and able to write less code while keeping the same outputs. Reported numbers versus baseline: 50% less code written, 22% fewer tokens, 20% cheaper cost, and 27% faster (more drastic with Opus). It works by forcing Claude to answer a series of questions before writing any code: Does this need to exist? Is it already in the codebase? Does a standard library do this? Is there a native platform feature? Is it an installed dependency one line away? Only then does it write the minimum that works, countering Claude's tendency to be verbose and rebuild things from scratch.
One of Chase's favorites. It connects Claude Code to NotebookLM, one of Google's best free products, where you load resources (PDFs, documents, YouTube videos) and chat with AI about them or generate deliverables like slide decks, images, infographics, and videos. NotebookLM has no API, so this CLI works around that, letting you do everything through the terminal, which is great for building workflows. The CLI actually exposes more tools than the web app: batch downloads, quiz and flashcard export, and saving chat to notes. Chase uses it most with YouTube, pulling transcripts and answering questions about videos, all while offloading work to Google's servers for free.
One of the most powerful tools covered. It handles browser automation: when no API exists, Claude Code can act like a human, visit sites, click things, and fill out forms. It is also excellent for front-end design (Chase nearly put it in the design bucket): spin it up on a site with a form and it opens many browsers, tests every edge case in minutes, completely hands-off. Importantly, do not confuse it with the Playwright MCP, the CLI is more effective and uses far fewer tokens.
An official plugin from OpenAI that connects Codex and the GPT models to Claude Code. It is ideal for code review or adversarial review, since Claude Code "loves the code it writes itself" and sometimes you want a second set of eyes. It includes commands like codex rescue for offloading entire features to Codex, so Claude Code and Codex can work on different parts of your app simultaneously, giving you the best of both worlds.
Not an official Google product, but built by a Google developer and so popular the developer was hired. It is great for heavy Google Workspace users who want more than the standard Google connector, which can't do things like send emails. GWS adds that ability plus 40+ skills with preloaded workflows (weekly digest, standup prep, meeting prep, email-to-task). Setup is more complicated, but it adds serious power to a Google-centric Claude Code stack.
The one everyone should already have, and if not, it should be the first thing installed. If you build anything with Claude Code, eventually you will want to push it to GitHub, and the GitHub CLI makes that extremely easy.
An official Anthropic skill that does more than create new skills, it also modifies and improves existing skills and measures skill performance. It can run A/B tests between an original skill and an "improved" version, and test performance with versus without the skill, giving objective proof that a change actually helps. Chase argues that because skills are one of the most powerful things in Claude Code, the skill-creator skill may be the most important skill of all. Install it from inside Claude Code via /plugin, then search for and install "skill creator."
Once the number-one repo on GitHub, this is about research that goes well beyond a simple web search. It digs deep across specific sources, including Reddit, Twitter, YouTube, TikTok, Reels, Hacker News, and Polymarket, to surface what people are actually saying about a subject. It is perfect for daily-briefing workflows or deliverables that need legitimate data, and is a great alternative to throwing /deep-research at everything and burning tokens.
One of the best tools for scraping the web, especially pages with heavy bot protection. Firecrawl has a paid version (its proprietary model best handles bot protection) and an open-source version that delivers much of the same functionality. Use it when you want one focused tool for web scraping and don't need last-30-days hitting many sources at once. Beyond scraping, it can interact with pages, discover all the URLs on a site, and crawl them, with lots of settings to target exactly what you want, a clear step above standard Claude Code web search.
From Andrej Karpathy, described as "machine learning in a box." You point it at an application you want to improve, and it runs loop after loop automatically to improve a specific, objective success criterion. Example: a Python app you want to run faster (1 second to 0.99 seconds), it keeps running experiments until runtime is as low as possible. In one example it ran 83 experiments, logged what was tried and what worked, and produced 15 improvements, all automatically. It is lightweight, but only works when the task has very specific, objective success criteria, think time and numbers.
Chase loves Supabase for its generous free tier and broad coverage. If you build a site with a form that collects emails, those emails need a database, and Supabase handles it. You can create databases from Claude Code with natural language ("use the Supabase CLI, create a database that makes sense"), and it also handles authentication (logins). It does a lot from one place, and can run locally too. Natural language plus Claude Code makes otherwise complicated setup simple.
One of the easiest ways to improve Claude Code's memory. Obsidian lets you designate folders as vaults that become a treasure trove of information. Open Claude Code inside a vault and it connects through the knowledge graph of your documents. Set up correctly, this gives Claude Code a map of all your documentation so it can answer questions effectively and efficiently. There are also Obsidian skills on GitHub (created by the founder of Obsidian) that teach Claude Code best practices for integrating into your Obsidian stack.
A step up from Obsidian into real retrieval-augmented generation with genuine knowledge graphs and embeddings (not a "fake" knowledge graph). my-rag is one of Chase's favorite tools for building knowledge-graph RAG systems: lightweight, fast, and a great introduction to more complex RAG. You run queries through Claude Code and connect the two easily. Once comfortable, you can move to RAG-Anything, which goes beyond PDFs and text to include images, graphs, and charts, things traditional RAG and Obsidian struggle with.
Great for any application you want to make money from. If you handle transactions, the Stripe CLI makes working with Stripe far easier than the sometimes-painful Stripe UI. As with the other CLIs, controlling and editing your application through the terminal and natural language inside Claude Code is a major benefit.