15 Claude Skills for Marketing Workflows

AI gets you a first draft in thirty seconds. Then the work starts.

You still have to check the angle, fix the CTA, add the proof, cut the filler, answer the objection and decide whether it is good enough to publish. That is where most AI-assisted marketing work dies, and it does not die at the blank page. It dies in the gap between "there is a draft" and "this is finished".

Call it the last twenty percent, which is a figure of speech rather than a measurement, because nobody has measured it and this post is not going to pretend otherwise.

I built fifteen Claude Skills for that gap. They are free, MIT licensed, and they work on the drafts already sitting in your folder.

Why "improve this landing page" does not work

Because it never says what finished means.

A broad instruction gets you a broad answer. The model produces something plausible, you read it, something is off, you ask again, and the second version is different rather than better. Nothing in the exchange defines the standard, so nothing can be checked against it.

Weak, because nothing ends it:

Improve this landing page.

Strong, because a stranger can tell when it is finished:

Keep revising until the hero states one outcome, every claim on the page
has a source or a visible gap marker, the CTA matches the offer, and the
remaining issues are ones only the owner can decide.

The second one is not a longer prompt. It is a different kind of instruction: it carries a definition of done.

That definition is what makes another pass possible. Without one, a revision has nothing to aim at.

The three things every skill in this pack carries

Review criteria. What done means for this artifact, written down before any rewriting starts. A list a stranger could check, not a feeling. The criteria come from the skill, not from you, because encoding that judgment is most of the value.

A verification pass. A separate read against those criteria, run after the writing pass rather than during it. The model is worse at checking work while it is still attached to producing it, so the two jobs are split on purpose.

A stop condition. The explicit sentence that ends the work. Either every criterion passes, or the remaining gap needs a human decision, and the skill says which. Without this you get one pass. With it you get work that keeps going until it meets a standard you set.

The 15 skills

Three jobs, five skills each. Every one of them starts from something you already have: a draft, a URL, a recording, or numbers you paste in.

Producing something new

• Creative Brief Writer turns a rough idea for a campaign, a piece of content, a design or a video into a brief a freelancer, designer, writer or editor can start on without a round of questions first. It forces you to state what done looks like, what is out of scope, and which decisions are already made. The questions it cannot answer go into an open decisions table with an owner, rather than quietly missing.
• Lead Magnet Polish takes a draft asset and works it until the title, promise, structure, examples, usage guide and CTA are all there and all honest.
• Recording to Assets takes one webinar, podcast or customer session and works out which assets it can honestly support. The useful half of its output is the list of assets it cannot: a ninety minute recording rarely contains three blog posts.
• Customer Words to Copy mines the language customers actually use, from calls, reviews and support tickets, and rewrites your copy to use their words instead of your internal vocabulary.
• LinkedIn Asset Post turns a finished asset into a post that earns the comment, with hook variants and the DM delivery copy.

Fixing something that underperforms

• SEO Article Refresh rewrites a published article until intent match, claim freshness, internal links and CTA relevance all pass, with every unsupported claim marked rather than smoothed over.
• Landing Page Rewrite names the top conversion issues, rewrites the hero and the CTA, makes the proof gaps explicit and gives you a before and after rationale a stakeholder can approve.
• Offer Sharpener works the promise, cuts the padding, names the missing proof and answers the main objection. Run this before the page rewrite: most pages that will not convert are fine as pages and weak as offers.
• Newsletter Issue Rewrite works a draft until it earns exactly one click. A high open rate with no clicks is a failure of the body, not the subject line.
• Resource to Demo Email writes the one email that bridges from a downloaded resource to a real next step, with a shorter variant and subject lines.

Deciding what to do next

• Experiment Decision Memo turns a finished test into a memo that ends in a decision, honest about the difference between a signal and a result, with kill and continue criteria written down.
• Weekly Report Decisions takes a report that currently ends in numbers and rewrites it until it ends in exactly three decisions, each with its evidence and an owner.
• Paid Ads Readout turns pasted campaign numbers into something a client or a boss can read: what the data says, what it does not, the missing data, and the next action with an approval gate.
• Ad Angle Generator takes your offer and produces practical angles with the pain behind each, the proof each one would need, its risk, and a ranked shortlist to test.
• Competitor Message Map turns pasted competitor copy into the claims the whole category repeats, the gaps nobody is filling, and the headline territories still open. It wants a real sample, five competitors with several claims each, and below that it gives you the claim inventory and refuses to call anything a category script, because a gap in three pages is a gap in your clipboard.

Install

git clone https://github.com/mardab96/marketing-workflows-claude-skills.git
mkdir -p ~/.claude/skills
cp -r marketing-workflows-claude-skills/*-* ~/.claude/skills/

Start a new Claude Code session. The skills usually activate on their own when what you are doing matches, and you can always name one directly.

How to get more out of them

Two habits change the output more than anything else.

• Say what you already know is wrong. "The hero is fine, the proof section is the problem" saves an entire pass, because the skill stops rediscovering what you had already worked out.
• Answer the missing-data questions instead of ignoring them. Every skill marks what it could not verify from your input. Those marks are the honest part, and filling one in is usually faster than another rewrite.

They also chain. A brief into a lead magnet into a LinkedIn post is the whole path from idea to published asset. Customer language before any writing skill changes the vocabulary of everything downstream.

What they will not do

They do not publish, send, post or spend anything. They do not log into your accounts, because none of them have access to one and all of them work from what you paste.

They do not invent statistics, quotes or results your inputs do not support. Where a claim cannot be verified it gets marked rather than smoothed over, which is occasionally annoying and always the right behaviour.

Everything that leaves your machine is your decision, made after you have read the draft.