5 Claude Code skills for beginners: Humanizer, Skill Creator, Grill Me, Caveman, Marketing Skills

If you don't have Claude Code yet, do the first two steps under “Before you start” below (Node.js and Claude Code). Then make a folder for your marketing work, start Claude in it, and paste the prompt. It installs all five, runs a first test that sends nothing anywhere, and hands you a cheat sheet.

This is the manual way: every step and every command. Claude follows it too.

The five don't plug into each other, but they line up nicely on one real job, like a new service page:

Words you'll see

• Skill: a folder with a SKILL.md file: instructions that teach Claude one job.
• Slash command: typing / and a skill's name in Claude Code (like /humanizer) to run that skill on purpose. Most skills also kick in on their own when what you ask fits them.
• Terminal: the window where you type commands (Terminal on a Mac, PowerShell on Windows).
• Node.js and npx: Node.js runs JavaScript programs on your computer; npx comes with it and downloads and runs a tool in one go.
• Skills CLI: the free command-line tool that installs, lists, updates and removes skills. You run it as npx skills.
• Global (-g) vs project: global means every folder on your computer (~/.claude/skills; the ~ is your home folder). Without -g, the skill goes in .claude/skills inside the folder you're in, and only works there.
• Tokens and limits: tokens are the chunks of text Claude reads and writes. Longer answers use more of them, and your plan's usage limit runs out faster.

Made by blader, MIT license. It's built on Wikipedia's Signs of AI writing, the field guide Wikipedia's editors use to catch undisclosed AI text. It checks for 26 patterns, strongest first: “it's not just X, it's Y”, one-line closers like “Let that sink in.”, deep-sounding sayings, staged run-ups, stock AI words like “testament” and “delve”, and dashes everywhere.

Two things I like about it. It doesn't make anything up: if a sentence needs a name, a number or a date it doesn't have, it asks you instead of inventing one. And it's upfront that it edits for readers, not to sneak past AI detectors; its README says detectors still flag most of its output.

Made by Anthropic (the company behind Claude), Apache 2.0 license. Its job, in its own words: create new skills, improve existing ones, and test how well they work. You tell it the job, it drafts the SKILL.md, writes a few test prompts, runs them, shows you the results and rewrites the skill from your feedback.

The testing part runs small Python scripts and takes a while. If you just want the skill written, say so: its own instructions tell it to skip the formal tests when you'd rather just work through it together.

Made by Matt Pocock, MIT license. He calls it one of his most popular skills, and it's for non-code plans too. You tell it what you want to do, and it interviews you in rounds: numbered questions, each with its own recommended answer (so “yes” takes the recommendation), until every decision in your plan is settled. It won't act on the plan until you confirm you're on the same page.

One thing most install videos miss: grill-me is a two-line skill that hands off to a second skill called grilling, which holds the actual questioning. You need both, which is why the command names both.

Made by Julius Brussee, Apache 2.0 license. The joke in the video is “me read file, me fix bug”. The real skill is less cartoon than that: its rules say answer first, cut the greeting and the recap, short sentences, and no “me think” theater. It never drops words like not, never or only, keeps numbers and commands exactly as they are, and switches back to full sentences for security warnings and anything you can't undo. Once on, it stays on for the whole session until you say “stop caveman” or “normal mode”.

Example (made up). Normal: “Sure! I'd be happy to help. The reason your contact form isn't sending emails is most likely that the email setting was left blank when the form was set up.” Caveman: “Form email not sending. Email setting blank. Fill it in, then send a test.”

Made by Corey Haines, MIT license. On 2026-10-31 the repo had 53 skill folders: copywriting, SEO audits, ads and ad creative, emails, landing pages (cro), pricing, referrals, schema, cold email and more. The one to run first is product-marketing: it interviews you about your business, customers and how you talk, and saves it as .agents/product-marketing.md in your folder. Every other skill reads that file first, so you stop explaining your business every time.

Fifty-three skills is a lot to load everywhere, so I'd put them in one marketing folder only (no -g). Claude Code loads a short description of every skill into each conversation, and when there are too many it starts cutting descriptions, which makes it worse at picking the right one.

Then start claude in ~/marketing and type /skills: you should see all five. Claude Code picks up new skills without a restart. The one exception: if ~/.claude/skills didn't exist when your session started, type /reload-skills once.

This runs on your Claude plan with nothing posted, sent or published. Paste it into Claude Code:

Good: it should flag the “more than just a roof” and “don't just install roofs” lines, the “testament” and “passionate about elevating” filler, the “peace of mind, quality, and excellence” list of three and the “Let that sink in.” closer, and the rewrite is shorter and plain. It may ask you for a real fact to put in place of the fluff (how long you've been in business, what you actually install). That's it refusing to make one up. Bad: Claude rewrites it without listing what it changed. Then the skill isn't loading: type /skills and check humanizer is there.

Then try /grill-me on something small you're actually planning. It costs you a few minutes of answering questions, and it's the quickest way to see what it does.

Skills installed with the Skills CLI don't update themselves. Check now and then:

The Skills CLI sends anonymous usage data by default. Set either of these to 1 to switch it off:

The real cost is your plan's usage limit. Every skill runs on it, and a long Grill Me session or a Skill Creator test run uses a good chunk. That's where Caveman earns its spot. And watch one thing: the skills are free, but a skill can point Claude at a paid tool (an SEO data service, an ad account). If one asks for a key or an account, stop and read why before you go on.