A copywriter does not need fifteen Skills. They need four, each one earning its trigger, with one of them refusing to let slop words through. Here is which four, what goes in each, and a worked SKILL.md for the refusal-rules one.
Most posts about Claude Skills for copywriting are lists of forty. That is the wrong instinct. A copywriter has a small number of things that are true on every job, every time, and those are the things worth wrapping in a Skill. Everything else is a prompt you run once. Build four narrow Skills that load on their own, and Claude stops drifting off-voice and stops reaching for "delve" the moment you are in a hurry.
This is the copywriting cut of the wider Claude Skills for marketing guide. If you have never built a Skill, read the 20-minute build walkthrough first, then come back. The four below assume you know the SKILL.md shape: a few lines of YAML, then instructions.
Three things define a good Skill: the trigger (the description that tells Claude when to load it), what the Skill knows (the reference it reaches for), and what it refuses (the lines it will not cross). A copywriting Skill that has no refuse-condition is just a prompt with extra steps. The refusal is the whole point.
Not forty. Four. Each one does a single job, loads on its own, and earns the slot it takes up in Claude's context.
Trigger: any external-facing writing or rewriting task. Knows: the voice guide for the brand, loaded from a reference file in the folder. Refuses: to invent voice attributes that are not in the guide, and to "elevate" or "punch up" the voice into something it is not.
This is the one most copywriters build first, and it pairs with a real deliverable: the voice guide itself. The Skill is plumbing. The guide is the work. If your voice guide produces the same flat output for every brand, the problem is upstream of the Skill, and why most brand voice prompts produce identical outputs covers the three structural moves that fix it.
Trigger: any generated or rewritten copy, on every job. Knows: the banned-word list and the rules about invented claims. Refuses: to ship a single tell-word. This is the Skill that bans delve, leverage, and best-in-class before they ever reach a draft.
This is the most underbuilt Skill in copywriting and the one with the highest payback. It is the worked example below. Every copywriter has a mental list of words that make their skin crawl, and that list is forgotten the moment a deadline is close. A Skill never forgets.
Trigger: any request for headlines, subject lines, hooks, or ad copy variants. Knows: how many variants to produce, which angles to cover (benefit, curiosity, objection, specificity), and the brand's headline rules. Refuses: to hand back three variants that are the same sentence reworded.
The default failure mode of headline generation is ten options that are all the same idea wearing different hats. A good headline Skill forces spread: one benefit-led, one curiosity-led, one objection-led, one stupidly specific. The refuse-condition is "no two variants may share the same angle." That single line is what makes the output useful instead of filler.
Trigger: any draft that needs a final pass before it ships. Knows: the brand's do/don't pairs, pulled from real samples. Refuses: to pass a draft that hits a "don't" without flagging it and offering the "do" version.
Voice guides describe the voice. Do/don't pairs enforce it. "Say it plainly, do not hedge." "Lead with the concrete, do not open on abstraction." This Skill turns those pairs into a checklist Claude runs against every draft, citing the pair it caught. It is the difference between a voice you describe and a voice you ship.
The refusal-rules Skill is the best one to build first, because it is small, it is the same across most of your work, and the payback is immediate. The trigger is broad on purpose: you want it loading on every piece of copy. The knows is a flat list. The refuses is the entire point.
Here is the YAML frontmatter. The description names the trigger words explicitly, so Claude loads it on any writing task.
---
name: refusal-rules
description: Use this Skill whenever the user is writing,
drafting, rewriting, editing, or critiquing any copy
(emails, ads, headlines, web copy, social, blog drafts).
Triggers on "write," "draft," "rewrite," "edit," "polish,"
"make this better," or any request to produce or revise
marketing copy. Enforces the banned-word list and the
no-invented-claims rules on every output, before the draft
is shown to the user.
---
Below the YAML, the instructions. Notice the shape: a short knows section, then a long refuses section. For this Skill, the refuses section is the deliverable.
# Refusal Rules ## What this Skill knows This Skill enforces the copy standards that apply to every piece of writing, regardless of brand or format. It runs as a final filter on any draft before it reaches the user. ## What this Skill refuses Never ship a draft containing these words or phrases: - delve, leverage, best-in-class, seamless, robust - tapestry, landscape (as a metaphor), realm - elevate, unlock, harness, navigate the complexities - "in today's fast-paced world", "game-changer" - "resonates with", "drives engagement", "at scale" # If a banned word is the only accurate one, flag it # and ask, do not silently substitute a worse word. Also refuse: - Invented statistics, percentages, or study citations - Fake testimonials or named-customer quotes - Claims of "leading", "#1", or "award-winning" unless the user has supplied the proof - Three "variants" that are one sentence reworded When a draft would hit any of these, rewrite around it silently for the word list, and flag the claim issues to the user with the line that triggered them.
That is a complete, shippable Skill. Save it as SKILL.md in a folder called refusal-rules, package it, and install it. From then on, every copy task you run loads it without you asking. The banned-word list is yours to extend. Run a draft through the Slop Sniffer first to find the tell-words your own writing leans on, then add them to the list. The Skill turns a one-time check into a standing rule.
A Skill that only says "write well" is a horoscope. A Skill that says "never let the word leverage through, and flag any statistic the user did not supply" is a contract. Claude can act on the second one. Specificity in the refusal is what separates a Skill that changes your output from a Skill that decorates your folder.
Four Skills is a ceiling, not a target. If you only build one, build the refusal-rules Skill, because it is the one that stops the most embarrassing output and it is the same across nearly every job you take. Add the brand-voice Skill next. The headline and do/don't Skills are worth it once you feel the first two earning their keep.
And the same rule applies that applies to every Skill: it is only as good as the file it points at. A refusal-rules Skill with a lazy banned-word list is a lazy Skill. Spend the twenty minutes building the list from your own writing, not from a generic "AI tell-words" article. The list that bans your tics is worth ten lists that ban someone else's.
If you would rather start from a working template than a blank file, the four free Claude Skills on the Hub are the same four free prompts packaged as installable Skills, with the YAML and refuse-conditions already wired. Voice Extractor, Competitive Teardown, SERP-Informed Brief, Performance Readout. CC BY 4.0, drag-and-drop, no email gate. Open one, read the SKILL.md, and you have the shape for your own refusal-rules Skill.
Get the 4 free Skills