What they are, why they matter, where they live, how to build one, and the Skills worth setting up for each marketing discipline. Written by an operator who runs them weekly, not a vendor selling you on them.
Claude Skills are the most underused feature in the product for marketers, and the fastest-moving. Six months ago almost nobody was writing them. Now there are GitHub repos with dozens, directories full of them, and a genre of "70 free Claude skills" posts appearing every week. Most of those posts are link-dumps with no judgment. This one is the opposite: the complete, honest guide to what Skills actually are, when they earn their place, and how to use them without bloating your setup.
If you read one thing about Claude Skills, read this. If you want the discipline-specific detail (Skills for SEO, email, social, and the rest), the cards further down link to a focused guide for each.
A Skill is a small folder containing a SKILL.md file (the instructions) and optional supporting files. The folder gets packaged and made available to Claude. From then on, Claude loads the Skill automatically whenever it decides the Skill is relevant to what you are asking.
That last part is the whole point, and the thing most explainers skip. A Skill is not a prompt you paste. It is not a setting you toggle. It is a capability that Claude reaches for on its own, based on the description you give it. You write the description well, and the Skill shows up exactly when it should. You write it badly, and the Skill either never loads or loads on every conversation and clutters everything.
The mental model that works: a Skill is something Claude needs to know about how you work, not what you are working on this week. Your brand voice is a Skill. Your refusal rules are a Skill. The format your CMO expects every brief in is a Skill. These are true across every conversation, so they live in a Skill that travels with you everywhere.
A Skill is a reusable instruction pack that Claude loads on its own when relevant, so you stop pasting the same context into every conversation and start treating it as infrastructure.
Most marketers using Claude are paying a tax they cannot see. They paste their brand voice guidelines into a conversation, do some work, close the tab, and tomorrow they paste the same guidelines again. Every single time. The voice drifts because they paraphrase it slightly differently each day. The refusal rules ("do not invent statistics, do not use the word delve") get forgotten the moment they are in a hurry. The output format is reinvented per task.
A Skill kills that tax in one move. Build the Brand Voice Skill once, and every conversation, in every Project, inherits your voice without you lifting a finger. Build the Refusal Rules Skill once, and Claude stops reaching for "leverage" and "best-in-class" whether you remembered to tell it to or not. The work gets more consistent precisely because the consistency stops depending on your memory.
There is a second reason, and it is the one that compounds. Marketing teams are where prompt quality goes to die, because a good prompt in one person's head never reaches the rest of the team. A Skill is shareable. Build the team's voice and standards into Skills, and the intern produces on-brand work on day one, not month three. That is the difference between a tool one person uses well and infrastructure a whole team depends on.
This is the question that confuses everyone, and the answer is simpler than the confusion suggests. The three containers map to one question: how often is this thing true?
The trap is using one container for everything. Some marketers live in plain prompts forever, paying the paste-the-voice-again tax daily. Others build one giant Project called "Marketing" and stuff everything in, then wonder why Claude keeps citing a campaign from two quarters ago. The fix is matching the work to the right container, and Skills are the layer most people are missing entirely. The full decision tree post goes deeper with a worked week-in-the-life.
This is where the practical questions start, and where most guides get vague. There are two places Skills can live, and they behave differently.
Claude Code (the CLI tool). Skills live in a folder, typically ~/.claude/skills/, with each Skill in its own subfolder. Claude Code detects them automatically. This is the most flexible setup: Skills are just files, you can version them in git, share them as repos, and edit them in any text editor. This is what the "70 free Claude Code skills" repos you find online are built for.
Claude Desktop and Claude.ai. Skills are added through the app interface, packaged as a folder you drag in. Same SKILL.md format, same automatic loading, but managed through the app rather than a folder you edit directly.
For a marketer, the honest guidance: if you live in Claude Desktop or the web app (most marketers do), use the app's Skill management. If you are comfortable in the terminal or already use Claude Code, the folder approach gives you version control and easy sharing. Either way the SKILL.md you write is the same. The decision tree and the build walkthrough below work for both.
Every Skill is one markdown file. It starts with a YAML block (the metadata that tells Claude when to load the Skill), then the instructions. Here is the shape, using a Brand Voice Skill as the example.
--- name: brand-voice description: Use this Skill whenever the user is writing, drafting, rewriting, or critiquing external-facing content (emails, social posts, ads, web copy, blog drafts) where it needs to sound like the brand. Triggers on "write," "draft," "rewrite," "in our voice," "on-brand." --- # Brand Voice ## What this Skill knows The brand voice guide is in voice-guide.md (this folder). The three structural moves the voice is built on: 1. Paired contrast: every attribute named with its opposite 2. Pattern-level rhythm: short sentences, concrete examples 3. Do/don't pairs from real samples ## What this Skill refuses - Do not use: delve, leverage, tapestry, best-in-class - Do not invent statistics or testimonials - Do not "elevate" the voice. Stay in it.
The single most important line is the description. Claude reads it to decide when to load the Skill. Too vague ("helps with voice") and the Skill loads on every conversation and bloats the context. Too narrow and it never loads when you need it. Name the trigger words explicitly, as above. That YAML frontmatter is doing more work than the entire rest of the file.
Twenty minutes, four steps. The fastest first Skill is a Brand Voice Skill, because it pairs with a prompt you can run for free and it pays back the first time you brief out any copy.
voice-guide.md, package the folder, install it, restart Claude, and test with "write me a 100-word intro for a new feature."The full build walkthrough has the complete SKILL.md, the install steps, and the one mistake everyone makes on their first one (trying to make a single Skill do everything). If you do not want to build from scratch, the four free Skills are the same template, pre-built and ready to drag in.
Three to five narrow Skills, each earning its trigger, beats one bloated Skill every time. The question is which ones. That depends on the work you actually do. Below is a focused guide for each marketing discipline: which Skills are worth building, what goes in each, and the prompts they pair with.
1. One Skill to rule them all. The first Skill people build is called "Marketing" and has fifteen things in it. It loads on every conversation, bloats the context, and produces worse output than a vanilla prompt. Build narrow Skills, each earning its trigger.
2. A vague description. "Helps with marketing" tells Claude nothing about when to load. The description is the most important line in the file. Name the trigger words.
3. A half-baked reference file. A Brand Voice Skill that points at a weak voice guide makes every output weakly on-brand. The Skill is plumbing. The reference file is the actual work. Get the voice guide right first.
4. Treating Skills as prompts. A Skill is not something you paste. If you find yourself manually invoking it, the description is wrong. Fix the trigger so it loads on its own.
5. Never refreshing them. Voice drifts. Standards change. A Skill you built in January and forgot is a Skill quietly enforcing last year's standards. Review the important ones quarterly, the same cadence as the Project rebuild.
If this is more theory than you wanted, here is the shortcut. The four free Skills on the Hub are the same four free prompts (Voice Extractor, Competitive Teardown, SERP-Informed Brief, Performance Readout) packaged as installable Skills, with the YAML and instructions already wired. Drag them in, no email, no signup, CC BY 4.0. They are also the best way to learn the format: open one, read the SKILL.md, and you have a working template for your own.
Voice Extractor, Competitive Teardown, SERP-Informed Brief, Performance Readout. Installable Skills, ready to drag into Claude Desktop or Claude Code. CC BY 4.0. No email gate.
Open the Skills page