The Five Free Marketing Skills for Claude Code I Use on Client Work

By Frederik Baun, conversion copywriter and founder of Words Sounds Visuals

I have written direct response copy since 2018, and every framework inside these five skills has shipped on paid client work.

What Are Marketing Skills for Claude Code?

A skill is a markdown file with one job. It might cover headline writing, an SEO audit, or how to structure a cold email sequence.

Open one and it reads like a briefing memo you would hand a new hire. A copywriting skill lists the formulas, the rules, and a checklist for what to cut.

Claude Code reads that file when a task matches its description. Then it follows the framework inside instead of inventing a format.

That is simpler than the name suggests, and it is close to the whole mechanism.

A skill folder can also bundle scripts for Claude to run. That is how Anthropic’s own document skills handle PDFs and spreadsheets.

Mine are markdown only, with nothing to execute.

I have been building this kind of system for client work for a while. I wrote up the architecture in how I use Claude Code for marketing.

What follows is the part I had never released, which is the five skills themselves.

Which Five Skills Did I Release?

All five are on GitHub at WordsSoundsVisuals/claude-code-marketing-skills, free, under a CC-BY-4.0 licence. No email, no signup, no gate of any kind.

They are the exact files I load when a client is paying me, copied straight out of my working system with the internal paths stripped.

The headline writer

This one returns exactly 16 headlines, one per formula, and it shows its working first.

Before the headlines appear, it emits a self-check block. That block names each headline’s anchor, counts its triggers, and scores it on the 4Us.

Those four are Useful, Urgent, Unique and Ultra-Specific. Anything under 12 out of 20 gets rewritten before it reaches you.

That scoring step is the whole point. Without it you get sixteen variations of the same bland line.

With it you get sixteen genuinely different angles. You can also see why each one scored what it scored.

The formulas themselves are the classic direct response set, and I broke them down separately in 16 headline formulas and landing page headline formulas.

The email subject line writer

This one returns seventeen subject and preview-text pairs, one per formula, capped at 60 characters each.

The rule that carries this skill is the anchor requirement. Every line has to name something the reader already cares about before they open.

“You” is not an anchor. “Your Gmail open rates” is an anchor.

If the skill cannot name the anchor in one or two words, the line fails and gets rewritten.

There is a second rule I care about more than any other in this file. The skill is banned from accusing the reader.

“What you are getting wrong about email” is rejected. “The Gmail update that tanked your open rates” is allowed.

The difference is who the bad actor is. Get that wrong and it costs you unsubscribes.

I wrote up all seventeen formulas in email subject line formulas.

The landing page writer

This covers the full opt-in page: headline, subheadline, value proposition, bullets, social proof, form, CTA button, microcopy and the privacy line.

What makes it useful is that it treats six page types as six different pages. A lead magnet page runs 300 to 800 words.

A free trial page runs 800 to 2,000 and needs risk indicators near the button. A waitlist page teases without revealing.

Splitting those six out stopped me reaching for one template on every offer.

It also carries the numbers I use to argue with clients. Every form field beyond the email address costs you conversions.

Cutting a four-field form to three is worth a real lift. Having that in the file means Claude pushes back when someone asks for a phone number.

The bullet point writer

Fascination bullets are the copy skill I see missing most often, and almost no AI tool writes them properly.

The engine here is the triple “so what?” chain. Take a feature, ask “so what?” three times, and the third answer is your bullet.

“Pre-built email templates” becomes “you send your first campaign this week” becomes “you stop losing momentum” becomes “you feel like someone who follows through.”

That third answer is what people buy.

Eight formulas sit on top of that, including the curiosity bullet, the warning bullet and the imagine bullet.

The offer architect

This is the biggest of the five and the one I would run first on anything new.

It walks 11 phases. Sophistication stage, awareness stage, the value equation, mechanism naming and three promise variants come first.

Then the value stack, the bonus stack, the guarantee, the objection register, price anchoring and a one-page summary.

The phases people skip are the ones that matter. The guarantee routing table picks between unconditional, conditional, anti-guarantee and performance-based.

Which one you get depends on your price and your buyer’s dominant fear. The objection register sorts objections into vehicle, internal or external.

I covered the thinking behind it in what makes an offer irresistible.

Why Do These Reject Output Instead of Describing Good Copy?

Here is the difference between these files and most of the skill libraries you can install.

Almost every copywriting skill I have read tells the model what good copy looks like. It lists formulas, explains features versus benefits, and gives a few examples.

The model reads that, agrees enthusiastically, and hands back something that reads like a template.

A model given only positive criteria will match the shape of the examples. It has nothing to fail against.

These five skills spend most of their length on rejection criteria instead. The headline writer has a hard floor of four triggers and a 12-out-of-20 score.

The subject line writer has an anchor test and a banned-construction list. Both emit a self-check block before the output.

That forces the model to score its own work in public before it hands anything over.

A reader emailed me in July after pointing a conversion skill at his own homepage. The audit came back clean and he shipped the rewrite that afternoon.

Two weeks later a customer told him the copy read like a template. The skill had compared his draft to best practice, not to how he talks to his customers.

Positive criteria alone will do that to you every time.

I wrote more about that failure mode in the words that give AI copy away and how to make AI writing sound human.

How Do You Install Them?

Two routes. The plugin route is the one to use.

Inside Claude Code, run these two commands:

/plugin marketplace add WordsSoundsVisuals/claude-code-marketing-skills
/plugin install copywriting-skills@wsv-marketing-skills

That is the whole install. All five skills land at once and Claude picks them up immediately.

If you would rather see the files first, clone the repository and copy the folders yourself:

git clone https://github.com/WordsSoundsVisuals/claude-code-marketing-skills.git
cp -r claude-code-marketing-skills/plugins/copywriting-skills/skills/* ~/.claude/skills/

Dropping them in ~/.claude/skills/ makes them available in every project. Dropping them in .claude/skills/ keeps them local to that project.

Restart Claude Code and they are live.

How Do You Use a Skill Once It Is Installed?

You do not call these by name, which surprises people.

Claude reads the description line at the top of each skill and loads the right one when your request matches it. So you ask for the work, not for the skill:

• “Write me headlines for a landing page selling bookkeeping to dental practices”
• “I need subject lines for tomorrow’s email about the pricing change”
• “Build the offer for a $2,000 done-for-you LinkedIn service”

Each skill lists the inputs it wants at the top of its file. The headline writer asks for your audience, your format and any proof you have.

The subject line writer asks for the topic, the list temperature, and whether you want opens, clicks or replies.

Skip those inputs and you get generic output. That is true of every skill from every author, and it is true of every human copywriter you have ever briefed badly.

One habit worth building is giving the skill your real numbers. “A 37% lift for a dental client in six weeks” beats “good results.”

The skills are built to use proof wherever it fits. They have nothing to work with if you withhold it.

What Does the Output Look Like?

Here is the shape of a headline writer run, trimmed:

SELF-CHECK
Headline 1 (F1, How To): anchor=[cold email] | triggers=[specificity,
  benefit, mechanism, direct address] | count=4 | 4Us=U:4/Ur:3/Un:4/US:5 (=16/20)
Headline 2 (F2, Number+Things): anchor=[reply rate] | triggers=[number,
  specificity, benefit, audience] | count=4 | 4Us=U:4/Ur:3/Un:3/US:5 (=15/20)
END-SELF-CHECK

Formula 1: How To
Type: Benefit
Headline: How to Book 12 Dental Practices a Month Without Cold Calling

The self-check block is the part worth reading. It tells you which headline is strongest and why, so you are picking from scored options rather than guessing.

The subject line writer does the same thing in a different shape, naming each line’s anchor and counting its triggers before the seventeen pairs appear.

Three of these also run as free browser tools, with no install and no signup.

They are the headline writer, the email subject line writer and the offer architect.

The frameworks and the rules are identical. You just do not need a terminal.

Which Skill Should You Run First?

Run the offer architect first whenever the offer is new or has never sold.

Every other skill in this set assumes the offer is settled. The landing page writer assumes there is something worth opting in for.

The headline writer assumes there is a promise worth making. Neither can rescue an offer nobody wants.

A lot of copy work turns out to be offer work in disguise.

If the offer is proven and you are rewriting an existing asset, start with the headline writer. The headline carries most of a page’s conversion performance, so it is where the leverage sits.

Then the order runs offer, page structure, headline, bullets, and finally the subject lines for the email driving traffic to it.

Each hands off to the next. The landing page skill names the other two it expects to call.

How Do These Compare to the Big Skill Libraries?

There are much larger libraries than mine, and for some jobs they are the better pick.

The honest read is that these are different products. If you want one skill for every marketing function, install a big library.

If you want copy that does not read like a template, the depth per file is what gets you there. You cannot have both in one download.

There is a practical reason to keep your installed set small, too. Every skill loads its description into context, so fifty of them makes it harder for Claude to pick.

I would rather run five files I trust than fifty I have not read.

Are Claude Skills Safe to Install?

It depends on the skill, and the assumption most people start from is wrong.

Plenty of skills are markdown only, and installing one of those is closer to pasting a prompt than to running a program.

But the spec lets a skill folder bundle scripts alongside the markdown. That is how Anthropic’s own document skills handle PDFs and spreadsheets.

So “it is just a text file” is not a safe default assumption.

The second risk is subtler and applies even to markdown-only skills. A skill can tell Claude to run shell commands, fetch URLs or read files.

The model already has those tools. A malicious skill needs no code of its own to do damage.

So read the file before you install it. All five of mine are plain markdown you can read in ten minutes.

None of them runs a shell command or fetches a URL. A few tell Claude to open a reference file sitting in the same folder, and you can read that yourself first.

Apply the same test to any skill from a stranger, including the big libraries.

What a Skill Cannot Do for You

Installing a skill answers what to do. It does not decide who checks the draft before a client reads it.

I split my own system into three layers rather than one file per task.

Directives say what a task requires. Orchestration decides what runs and in what order.

Deterministic code then does the work the same way every time.

The five skills in this post are row one. My blog’s own publish gate is row three.

It blocks any post carrying a banned pattern or an unsourced statistic, and it has caught real violations in my drafts before they went live.

Feed Claude a skill file with no layer above it and it improvises the rest. It decides which draft to trust, when a claim needs a source, and whether the tone fits.

Those are calls that need to be right every time rather than most of the time.

The two pieces I have kept back are my editor pass and my compliance checker. They are the layers that turn a decent draft into something I will put my name on.

I wrote about how the checking layer works in how to fact-check AI content.

Should You Write Your Own Skills?

Yes, and it is a smaller job than it looks.

A skill is a folder with a SKILL.md file in it. The frontmatter needs two things, a name and a description.

The body is whatever instructions you would give a competent freelancer.

The description matters more than anything else in the file. It is the only part Claude reads when deciding whether to load the skill.

So name the trigger phrases someone would type. Write “use whenever headlines, headline options or an H1 are requested” rather than “helps with headlines.”

Two rules I learned the hard way. Anthropic’s own best-practice guidance is to keep a SKILL.md under 500 lines and under 5,000 tokens, and that ceiling is real.

Split the overflow into a reference.md the main file points at. And write the rejection criteria, not just the framework.

Start by taking a task you have briefed three times this month and writing the brief once, properly. That is a skill.

What This Means for Your Marketing

Marketing skills for Claude Code are worth installing. They save you writing the same briefing document every time a task repeats.

The good ones encode judgement you would otherwise have to supply yourself.

They are also one layer of three. The part that decides whether a draft is safe to send sits above the skill file, in whoever or whatever checks the work.

You do not need my stack to start. Take the five files and run one against a real task this week.

Watch what it gets right and what it misses. That miss is where your own checking layer starts.

If you would rather have the system than the files, book a call and we can talk about building it around your business.

If you just want the files, take them. There is no email gate on this and there is not going to be one.