Top 10 Claude Code Skills, Plugins & CLIs (April 2026)
Top 10 Claude Code Skills, Plugins & CLIs (April 2026)
If I was starting with Claude Code from scratch right now, these are the ten skills, plugins, and CLIs I'd install before writing a single line of code. I've spent the last year testing tools across client projects, my own agency work, and a growing Skool community, and this is the short list that actually earns its spot. A few of these will be new to you even if you've been around Claude Code for a while.
The ecosystem is moving fast. New plugins, skills, and command-line tools hit GitHub every single day, and most of them are noise. The ten below are the ones I come back to on real projects.
What Is the Codex Plugin for Claude Code and Why Use It?
The Codex plugin for Claude Code lets you run OpenAI's Codex as a second set of eyes on code that Opus 4.6 or Sonnet 4.6 just wrote. That matters because large language models grade their own work kindly. Opus doesn't look at a pull request it just generated and say "my code sucks."
Adversarial review is the killer feature. If you don't come from a technical background, you can't always tell whether what Claude Code produced is structurally sound. Telling Claude Code to review its own output is basically useless. Pointing a different model family at the codebase, with a prompt designed to find problems, surfaces the stuff Anthropic's models happily miss.
Installation is two commands. Search for "Codex plugin Claude Code" on GitHub, copy the marketplace-add command into Claude Code, run /reload plugins, then run /codex setup. You need an OpenAI account — a $20 ChatGPT Go subscription is more than enough for heavy use.
Once it's installed, run /codex adversarial review inside any project. It also ships with a standard (less aggressive) review command, plus codex rescue, which hands control to Codex if you want Claude Code to stop trying and let the other model finish a specific feature.
How Does Obsidian Work With Claude Code?
Obsidian is free markdown-note software at obsidian.md, and it turns any folder on your computer into an instant knowledge base that Claude Code can read and write to. You install Obsidian, point it at a folder (mine is literally called "the vault"), and open Claude Code inside that folder. Every markdown file Claude Code creates becomes part of your Obsidian graph automatically.
The reason this matters: Obsidian becomes a lightweight RAG system without the overhead of running a real RAG stack. Folders become topics. Internal links become your retrieval layer. For anything up to a few thousand markdown files, it's all you need.
To get the most out of this setup, install the Obsidian skills from the Obsidian CEO's GitHub repo. It's a small set of skills that teach Claude Code how to navigate and maintain a vault — sub-indexes, wiki-style cross-links, sensible folder conventions. I walked through an end-to-end build of this system in a recent video, modeled on Andrej Karpathy's viral tweet about LLM-friendly note structures.
If you're using Claude Code for any kind of personal-assistant project or a knowledge base that's going to grow over time, this combination is the first thing I'd set up.
What Is Auto Research and How Does It Improve Claude Code Output?
Auto Research is a machine learning algorithm in a box that automatically runs experiments on any program or skill you're trying to improve. You install it with a few lines, point Claude Code at it, and say "use Auto Research to optimize this." It runs a batch of ML experiments, throws away the changes that make things worse, and commits the ones that improve performance.
This is the most hands-off optimization workflow I've found. You don't babysit it. You don't tune hyperparameters. It just runs until the program gets measurably better.
Use it when you've got a skill, prompt, or program where you know the output could be sharper but you can't pin down the right changes manually. Auto Research does the search for you.
What Is awesome-design.md and Why Does Claude Code Need It?
awesome-design.md is a GitHub repo of design markdown files extracted from high-quality production websites like Claude, ElevenLabs, and Cohere, which you hand to Claude Code as front-end design templates. It hit 38,000 stars in its first week, and it fixes Claude Code's biggest weakness: front-end design is still rough even with the official front-end-design skill.
The inspiration is Google Stitch, an AI design tool that generates a design.md file for every website it builds — a very detailed, well-structured prompt covering typography, color, spacing, button behavior, and layout patterns.
awesome-design.md ports that idea to a community repo. You browse the catalog, pick a markdown file that matches the aesthetic you want (Notion-style, Claude-style, ElevenLabs-style), and feed it to Claude Code as your design baseline. Buttons, colors, fonts, spacing — all structured as a prompt template instead of a vibe.
Install it with the single command on the repo's README. If you're shipping anything with a public-facing front end, this is the fastest way to stop shipping AI slop.
What Is Firecrawl CLI and When Should You Use It?
Firecrawl CLI is a command-line web scraper designed for sites with anti-bot protection, and it returns structured, LLM-friendly output by default. Claude Code's built-in web search runs into walls on heavily protected sites. Firecrawl doesn't.
Two things make this worth installing. First, it punches through bot detection that would block a naive fetch. Second, it returns data in a clean format optimized for LLM ingestion — so Claude Code can actually use the results without extra parsing.
The main Firecrawl product is a paid API, but there's also an open-source version. The open-source version doesn't include the proprietary engine that beats the tougher bot walls, but for most scraping work, it's plenty.
Every CLI should come with a matching skill — the skill teaches Claude Code how to use the CLI correctly. Install both. That's the standard pattern for all the CLI tools in this list.
Is the Playwright CLI Better Than the Playwright MCP?
The Playwright CLI is the latest and cheapest way to give Claude Code real browser automation, and it's significantly better than the Playwright MCP. Ignore anyone still recommending the MCP — the CLI is faster, cheaper, and more reliable.
Here's the technical reason. The Claude in Chrome extension and similar screenshot-based browser tools work by taking pictures of the page and passing them to the model. Screenshots are slow and expensive. Playwright doesn't work that way. It reads the page's accessibility tree — actual structured page data — which is dramatically faster and uses fewer tokens.
With Playwright CLI installed, you can tell Claude Code something like "spin up a few Chrome instances and test the form submission on my landing page" and it will actually do it. Multiple tabs, real interactions, real assertions — controlled from inside your Claude Code session.
If you're building web apps, doing QA, or automating anything browser-based, this is non-negotiable.
Why Use the NotebookLM CLI With Claude Code?
The NotebookLM CLI connects Claude Code to Google's NotebookLM web app, which normally has no API, so you can run research and analysis against Google's servers instead of burning your own tokens. For anyone struggling with Claude Code usage limits, this is a cheat code.
NotebookLM is Google's research app — you feed it sources (YouTube videos, PDFs, URLs), and it builds a queryable knowledge base. The CLI gives Claude Code programmatic access to every feature, plus a few the web UI doesn't expose: batch downloads, slide revision, full-text access, programmatic sharing.
The tokens savings are real. When Claude Code uses NotebookLM to analyze sources, Google's infrastructure does the heavy lift. Claude Code just consumes the output. On research-heavy projects, this can cut your usage by a meaningful percentage.
Install the CLI with a couple of lines. Better yet, paste the GitHub URL into Claude Code and say "install this for me" — Claude Code is smart enough to read the README and run the right commands.
What Is the Skill Creator Skill and Why Is It the Most Important One?
The Skill Creator skill is Claude Code's official plugin for building and benchmarking custom skills, and it's the only way to quantifiably measure whether a skill actually improves your output. Skills are arguably Claude Code's most powerful native feature, and before this existed, there was no real way to A/B test one.
The big unlock is performance measurement. The Skill Creator skill will take your new skill, run it against a benchmark set, and give you concrete numbers on whether it's beating the baseline. It also tests incremental improvements. Edit a skill, rerun the benchmark, see whether your change moved the needle or not.
Install it through the marketplace: run /plugin, search for "skill creator skill," install, reload plugins. From that point on, every custom skill you build or modify should go through it.
If you're ever going to write your own skill — and you should — don't skip this one.
What Is LightRAG and When Should You Use It Over Obsidian?
LightRAG is an open-source graph-RAG system that kicks in when your knowledge base outgrows Obsidian — typically several thousand documents or a client project that needs real retrieval infrastructure. Obsidian is incredible for personal-scale knowledge, but there's a point where a proper RAG system is cheaper and faster.
LightRAG is the one I reach for because it's lightweight and free. The alternatives (graph-RAG from Microsoft, for example) get expensive fast. LightRAG gives you graph-structured retrieval without the enterprise price tag.
Use it for:
- Client projects that need their own knowledge base
- Corpuses with thousands of documents
- Any scenario where Obsidian's folder-and-link approach starts creaking
Personal knowledge vault under 1,000 markdown files? Stick with Obsidian. Cross that threshold, or need graph traversal that Obsidian can't cleanly do, and LightRAG earns its spot.
How Does GWS Connect Claude Code to Google Workspace?
GWS is a CLI tool built by Google engineers (not an official Google product) that gives Claude Code full access to your Gmail, Google Docs, Drive, and Calendar. If you're trying to use Claude Code as a personal assistant, you need this.
Before GWS, every attempt to connect Claude Code to Google Workspace was hacky and brittle. GWS is the first setup that feels stable — because the people writing it work at Google and know the APIs from the inside. It isn't the official Google stamp, but it's one step below that.
The one downside is setup. You have to spin up a Google Cloud project and enable a handful of APIs, which is more work than most of the tools on this list. There's a video linked from the repo walking through it.
The payoff is the skill library. GWS ships with a long list of workflow-level skills — rescheduling meetings, organizing Drive folders, creating recurring events, drafting replies — so you don't have to stitch the primitives together yourself. Don't load them all at once; point Claude Code at the repo and ask which ones are actually relevant to your day-to-day.
This is the closest thing to Anthropic's "Open Claude" personal-assistant vision available today.
FAQ
Should you use the Playwright MCP or the Playwright CLI?
Use the CLI. It's roughly 90% more token-efficient, faster, and easier for Claude Code to use correctly. MCPs have significantly more overhead, and the Playwright CLI is the canonical example of why the ecosystem is shifting away from MCPs for tools that have a terminal-native option.
Do you need to pay for anything to use these tools?
Most are free. The Codex plugin requires a ChatGPT account ($20/month Go plan is plenty). Firecrawl's managed API is paid but has an open-source version. NotebookLM is free. GWS, LightRAG, Auto Research, Playwright CLI, and the Skill Creator skill are all free. Only Firecrawl's premium engine and Codex need a paid tier.
What's the easiest way to install a Claude Code plugin or CLI?
Paste the GitHub URL into Claude Code and ask it to install the tool. Claude Code reads the README and runs the right commands. Works for almost every CLI, MCP, and plugin in this list.
Which tool should I install first if I'm brand new to Claude Code?
Start with Obsidian plus the Obsidian skills for a knowledge base, then the Skill Creator skill so you can build and benchmark your own workflows. Those two alone change how you work. Everything else layers on top once you know what you actually need.
How do I avoid installing too many CLIs and MCPs at once?
Less is more. The trap is installing every tool you hear about and ending up with a bloated environment full of skills and servers Claude Code doesn't know when to invoke. Pick tools based on real use cases you're hitting, not hypothetical ones you might hit later.
If you want to go deeper into building real workflows with Claude Code, 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.
