Before You Start
One tool installs the first four
Every skill below (except the one you build) installs through the same little command-line tool: npx skills. It's the package manager for Claude skills, and it works across Claude Code and 70+ other agents.
You don't install it separately. npx runs it on the fly. As long as you have Node on your machine (most people already do), you paste one line per skill and it's done. The -g on the end installs it globally, so it's there in every project.
Run these in your terminal, or just paste the command to Claude Code and let it run it for you. Either works.
find-skills
by Vercel · vercel-labs/skills
Start here, because this one finds the rest. find-skills searches the entire open skill ecosystem for a skill that already solves your problem, then hands you the install command.
You describe what you're trying to do in plain English. It checks the leaderboard at skills.sh, searches GitHub, ranks what it finds by install count and stars, and tells you exactly what to install. No more digging through random repos hoping something exists. It almost always does.
Install it
npx skills add vercel-labs/skills --skill find-skills -g
Try this
I need to turn a folder of messy PDFs into clean summaries. Is there an existing Claude skill for this? Find it, check the install count, and give me the command to install it.
mcp-builder
official, by Anthropic · anthropics/skills
An MCP is the wire that connects Claude to an outside tool. Some tools ship an official one. Most don't. mcp-builder builds you one for anything that has an API.
Point it at a tool you want Claude to control, hand it the API docs, and it researches the protocol, writes the server (TypeScript or Python), tests it, and wires it in. This is how you get Claude talking to the niche app your business actually runs on, the one nobody made a connector for.
Install it
npx skills add anthropics/skills --skill mcp-builder -g
Try this
Build me an MCP server that connects Claude to [the tool you want]. Here are the API docs: [paste the docs link]. Use TypeScript, then test it and show me how to use it.
stop-slop
by Hardik Pandya · hardikpandya/stop-slop
The best skill I've found for killing the telltale signs that something was written by AI. No more "here's the thing," no robotic rhythm, no em dash every other sentence.
It rewrites the slop out and scores your draft so you know when it actually reads like a human wrote it. If you publish anything written with AI, run it through this first. Do us all a favor.
Install it
npx skills add hardikpandya/stop-slop -g
Try this
Run stop-slop on this draft. Strip every AI tell, then give me the human score: [paste your text]
last30days
by Marcus Van Horn · mvanhorn/last30days-skill
The best research skill out there. Give it any topic or person and it searches Reddit, X, YouTube, even GitHub all at once, then hands you one clean brief on what people actually said in the last 30 days.
No single AI tool you already pay for can do this. ChatGPT can't read TikTok transcripts. Perplexity can't pull live Reddit threads ranked by upvotes. This stitches it all into one answer. I run it before writing almost any piece of content.
Install it
npx skills add mvanhorn/last30days-skill -g
Try this
/last30days [topic or person]. Give me what real people actually said in the last 30 days, ranked by engagement.
inbox-triage
custom · you build this one
This one you don't download. You build it, and it takes about two minutes. It's the skill I reach for most.
It reads your inbox from the last 24 hours, drafts replies to anything that needs one, tells you what's actually worth your attention, and clears out the junk. Every morning, in one command. The reason you build this one yourself instead of installing it: it has to plug into your email, in your voice, with your rules about what's safe to delete. The prompt below builds exactly that.
You'll need one thing first: Claude connected to your email.
Use the Gmail connector in the Claude desktop app, or the free Google Workspace CLI if you're in Claude Code. The build prompt asks which one you have and adapts to it, so don't overthink this.
Build it (copy-paste prompt)
You are going to build me a reusable Claude skill called "inbox-triage." Build it as a skill file I can run any time by saying "triage my inbox." What the skill does when I run it: 1. Pull my inbox messages from the last 24 hours. 2. Read the body of each one. Never classify on the subject line alone. 3. Sort every message into exactly one of three buckets: - RESPOND: a real person who needs a reply from me. A direct question, request, or a thread I started. - WORTH ATTENTION: not urgent, but I'd want to know about it. Surface it, don't act on it. - JUNK: promos, newsletters I don't read, stale automated notifications, cold outreach I never engaged with. 4. Show me two lists: everything in RESPOND (with a one-line "why" and a 2-line snippet each), and everything in JUNK (as a compact table with a one-line reason per row). Tell me the count of WORTH ATTENTION items and list them briefly. 5. For the RESPOND list, draft a reply to each one in my voice and save them as drafts. Never send anything automatically. 6. Wait for my explicit "yes" before touching the JUNK pile. When I approve, move those to trash (reversible), never permanent-delete. Hard rules to bake into the skill: - Never delete or send without my explicit approval. - When unsure which bucket, default to WORTH ATTENTION. Never guess something into JUNK. - Always protect: anything mentioning invoice, contract, legal, refund, password, or verification, plus anything from a real person I've already replied to. First, ask me which email tool I want this to use (the Gmail connector, the Google Workspace CLI, or whatever I have connected). Then write the skill file, save it, and show me how to run it.
Quick Reference
All four install commands
Paste all four at once
npx skills add vercel-labs/skills --skill find-skills -g npx skills add anthropics/skills --skill mcp-builder -g npx skills add hardikpandya/stop-slop -g npx skills add mvanhorn/last30days-skill -g
Number five, inbox-triage, you build with the prompt above. That's the full stack.
Work with Me
Want me on your problem for an hour?
Book a 1:1 call and we build the fix live — the workflow you want automated, the tool you can’t crack, whatever’s stuck. Direct, hands-on, no pitch waiting at the end.