17 Claude Code Plugins, Skills & CLIs Worth Installing
17 Claude Code Plugins, Skills & CLIs Worth Installing
Claude Code is one of the strongest AI tools on the planet out of the box — but the right plugins, skills, and CLIs make it dramatically better. The problem is there are hundreds of them, and most "top tools" lists are padded with stuff nobody actually runs. These are the 17 I use myself, broken into three buckets: data, design, and productivity. For each one you'll get what it is, how it works, and why you should care.
Let's get into it.
What Are the Best Claude Code Design Plugins?
The most common complaint about AI-generated front-ends is that AI lacks taste. Everything comes out looking like the same purple-gradient SaaS template. These three tools exist to fix exactly that.
Taste Skill
The Taste skill is an open-source GitHub repo built around one goal: defeating the AI slop monster. It's not a single trick — it bundles several sub-skills, including an image-to-code skill, a redesign skill, and an output skill, all aimed at making your front-end design look genuinely good.
A key detail: it works with every agent, not just Claude Code. If you're using AI to build landing pages or anything design-oriented, this is the first thing to reach for beyond Anthropic's default front-end design skill. The example sites in the repo — scroll animations, real visual identity — look nothing like the default output Claude Code produces on its own.
Impeccable
Impeccable is another open-source front-end design skill aimed at killing AI slop, and it's about to get a lot more attention: GitHub made it a built-in layer inside the GitHub Copilot app. When GitHub itself decides your skill is good enough to ship natively, that tells you something.
It's a single skill, but it packs 23 different commands — documenting what it built, critiquing what you've already made, polishing, making a design bolder or quieter, and so on. The clearest way to see it work is on impeccable.style, which shows side-by-side comparisons of default Claude Code output versus Impeccable output. The distill command, for example, takes a cluttered mini-dashboard and strips it down to a cleaner profile with the same information.
The real differentiator is the browser editor (currently in beta). Instead of editing everything through the terminal, you pull your site up in the browser, click on elements directly, and tell Impeccable what to change — and you see it live. If you want to get hands-on with design, this is the one.
awesome-design.md
awesome-design.md takes existing websites and turns them into templates for what you're building. It's based on the design.md principle from Google Stitch — Google's free front-end design tool, and the prompt structure Google engineered to get AI to produce strong design.
Here's the useful part: it maps that principle onto real sites. Pick a site you like — say Airtable — and you get a full breakdown of its structure, colors, surfaces, text, typography, spacing, and buttons. You're not cloning the site; you're cloning its design language and applying it to your own project. Everything is organized by use case — fintech and crypto, design and creative tools, productivity SaaS — so you can grab what's already working in your category.
What Are the Best Claude Code Productivity Plugins?
This bucket is about making Claude Code faster, cheaper, and connected to the tools you already live in.
Ponytail
Ponytail is one of the fastest-growing AI repos in the world right now, and for good reason: it makes Claude Code write less code while keeping the same output. The claimed numbers, measured against the baseline, are hard to argue with — 50% less code written, 22% fewer tokens used, 20% cheaper, and 27% faster.
How? Before Claude Code writes anything, Ponytail forces it through a checklist: 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? Is it one line? Only after clearing those questions does it write the minimum that works. Claude Code tends to be verbose and loves building from scratch even when the thing already exists — Ponytail is the guardrail against that. Note: those numbers were measured on Haiku. On Opus, the savings were even more drastic.
NotebookLM-Pi
NotebookLM-Pi connects Claude Code to Google's NotebookLM — one of Google's best and completely free products. You throw resources at NotebookLM (PDFs, documents, YouTube videos) and chat with AI about them, or have it generate slide decks, images, infographics, and videos.
NotebookLM has no public API, and this CLI gets around that — it lets you do everything through the terminal, which is exactly what you need to wire NotebookLM into a workflow. It actually exposes more tools than the web app: batch downloads, quiz and flashcard export, saving chat to notes. The combination I lean on most is NotebookLM plus YouTube — feed it a video link, it grabs the transcript, and answers questions about it. And you're offloading all the compute to Google's servers for free.
Playwright CLI
The Playwright CLI might be the most powerful tool on this list. It's browser automation: when there's no API and you want Claude Code to act like a human — visiting a site, clicking things, filling out forms — this handles it.
It doubles as a design tool too. Building a site with a form? Instead of manually clicking through every edge case, spin up the Playwright CLI and it opens a stack of browsers, tests everything in a couple of minutes, hands-off. One important distinction: do not confuse this with the Playwright MCP. The CLI is far more effective and uses way fewer tokens.
Codex Plugin
The Codex plugin is an official plugin from OpenAI that connects Codex and the GPT models to Claude Code. It's built for code review and adversarial review — because Claude Code loves the code it wrote itself, and sometimes you need a second set of eyes.
It goes further than review. Commands like Codex rescue let you offload entire features to Codex, so Claude Code can work on one part of your app while Codex works on another simultaneously. Best of both models at once.
GWS (Google Workspace CLI)
GWS is the Google Workspace CLI — not an official Google product, but built by a Google developer, and it got so popular it reportedly got the guy fired. If you live inside Google Workspace and want more than the standard Google connector gives you — the connector can't send emails, for instance — this fills the gap.
It ships with 40+ skills: weekly digests, staying on top of reports, meeting prep, email-to-task, and a pile of other pre-built workflows. Setup is a little more involved than most tools here, but for adding real power to your Google stack, it's hard to beat.
GitHub CLI
The GitHub CLI is the one everyone should already have — and if you don't, install it first. Anything you build with Claude Code eventually needs to get pushed to GitHub, and the CLI makes that dead simple. There's not much to debate here; it's table stakes.
Skill Creator Skill
The Skill Creator skill is an official Anthropic skill, and it does more than create new skills — it modifies existing ones, improves them, and measures skill performance. That last part matters most. It can run A/B tests between the first version of a skill and a supposedly improved version, and test performance with the skill versus without it. That gives you objective proof a change actually helped instead of guessing.
Since skills are one of the most powerful things in Claude Code, the Skill Creator skill is arguably the most important skill you can install. To get it, run /plugin inside Claude Code, search for "skill creator," and install — it'll show up under your installed plugins.
What Are the Best Claude Code Data Plugins?
This bucket covers everything about grabbing information, storing it, and giving Claude Code memory — databases, research, RAG, and payments.
last-30-days
last-30-days was at one point the number one repo on GitHub. It's research that goes well beyond a simple web search, pulling deep from specific sources — Reddit, Twitter, YouTube, TikTok, Reels, Hacker News, Polymarket, and more. You get a genuine read on what people are actually saying about a subject across platforms.
It's ideal for daily briefings or any deliverable that depends on legit data, and it's a smart alternative to throwing /deep research at everything and burning tokens — its dynamic workflows are far more targeted.
Firecrawl CLI
The Firecrawl CLI is one of the best tools for scraping the web, especially pages with heavy bot protection. Firecrawl has a paid version (its proprietary model gets past the toughest bot defenses) and an open-source version that covers most of the same ground.
Reach for this when you want one focused tool for scraping — you don't need last-30-days hitting a thousand sources, you just want the job done. It does more than scrape, too: it can interact with pages, discover every URL on a site, and crawl them all, with plenty of settings to tune when you're specific about what you want.
auto-research
auto-research, from Andrej Karpathy, is essentially machine learning in a box. You point it at something you want to improve, and it runs loop after loop, trying new experiments until it hits your success criteria.
The catch — and it's important — is that it only works when you have an objective, measurable success criterion, usually time or a number. Say you have a Python app you want to run faster: it took 1 second, now it's 0.99. auto-research runs experiment after experiment to drive that runtime down. In one example it ran 83 experiments and logged what it tried, what worked, and what didn't — landing 15 improvements, fully automatically. It's lightweight, but don't throw it at fuzzy goals; it needs something objective to optimize against.
Supabase CLI
The Supabase CLI earns its spot on the generous free tier alone, and it covers a lot of ground for building real applications. Say you've got a site with a form that collects emails — those emails need a database, and Supabase handles it. Tell Claude Code, "use the Supabase CLI, create a database that makes sense," and it builds the whole thing in natural language.
It also handles authentication — logins for your site — from the same place. Database work can get complicated if you've never done it, but handing Claude Code the Supabase tool keeps it conversational. And you can run Supabase locally if you'd rather.
Obsidian
Obsidian is one of the easiest ways to improve Claude Code's memory. You designate folders on your computer as vaults, fill them with information, and when you open Claude Code inside a vault it connects through the knowledge graph of your documents. Set up correctly, it gives Claude Code a map of all your documentation so it can answer questions about it quickly and accurately.
There are also skills that teach Claude Code to use Obsidian well — search GitHub for "Obsidian skills" and you'll find a repo from the founder of Obsidian himself that teaches Claude Code the best practices for integrating into your Obsidian stack.
Light RAG
For real memory, there's RAG — retrieval augmented generation — a step up from Obsidian's knowledge graph into genuine embeddings and knowledge graphs. Light RAG is my favorite entry point: lightweight, fast, and a great introduction to more complex RAG systems. You run all your queries through Claude Code, and connecting the two is easy.
Once you're comfortable, you can move up to RAG-Anything, which goes beyond simple PDFs and text to include images, graphs, and charts — the stuff traditional RAG systems (and tools like Obsidian) usually struggle with.
Stripe CLI
The Stripe CLI is for anyone building applications they actually want to make money from. If you're handling transactions, this makes working with Stripe far easier — the Stripe UI can be a pain to navigate, and controlling your payment setup through the terminal with natural language in Claude Code is a real upgrade.
Which Claude Code Plugins Should You Install First?
If you're just starting, install the GitHub CLI immediately — you'll need it the moment you build anything. From there, pick by what you're doing: Ponytail to cut tokens and cost on every project, the Skill Creator skill to build and measure your own skills, and one design tool (Taste or Impeccable) if you're shipping front-ends. Add data and research tools like last-30-days, Firecrawl, or Supabase as your projects demand them.
The point isn't to install all 17 at once. It's to know which one solves the problem in front of you.
Frequently Asked Questions
What's the difference between a Claude Code plugin, skill, and CLI?
A skill is a packaged set of instructions and sub-skills Claude Code loads to perform a task; a plugin bundles skills and commands you install through Claude Code's /plugin menu; a CLI is a standalone command-line tool Claude Code can call. In practice they overlap — many tools on this list are technically CLIs that Claude Code drives, while others like Skill Creator and Impeccable are true skills.
Do these plugins only work with Claude Code?
No. Several — the Taste skill most notably — are designed to work with any AI agent, not just Claude Code. Others like the Skill Creator skill and the Codex plugin are tied more directly to the Claude Code ecosystem. Check each repo, but a lot of this stack is portable.
How do I actually install a Claude Code skill or plugin?
For official plugins and skills, run /plugin inside Claude Code, search for the tool by name, and install it — it'll appear under your installed plugins. For open-source repos and CLIs, follow the setup instructions in each project's GitHub. Some, like GWS, have more involved setup than others.
Which plugin saves the most on tokens and cost?
Ponytail is the clearest win for cost: measured against the baseline it cut code written by 50%, tokens by 22%, cost by 20%, and time by 27% — and those numbers were on Haiku, with even bigger gains on Opus. The Playwright CLI is also worth noting, since it uses far fewer tokens than the Playwright MCP for the same browser automation.
Is there a plugin that improves Claude Code's memory?
Yes — start with Obsidian for a lightweight knowledge-graph memory over your own documents, then step up to Light RAG for real retrieval-augmented generation with embeddings, and RAG-Anything when you need to include images, graphs, and charts.
If you want to go deeper on building your Claude Code stack, join the free Chase AI community for templates, prompts, and live breakdowns. And if you're serious about building with AI, check out the paid community, Chase AI+, for hands-on guidance on how to make money with AI.
