A complete worked build: the brand voice extractor Skill, with the YAML frontmatter, the SKILL.md, the install, and the mistake everyone makes on their first one.
Claude Skills are the most underused feature in the product for marketers. Everyone has heard of Projects. Almost nobody is writing Skills. Which is strange, because a Skill is the easier of the two to ship: one markdown file, a few lines of YAML at the top, drag into Claude Desktop, done.
This post walks through building one specific Skill from start to finish. The Brand Voice Skill: a tiny Skill that loads automatically whenever you ask Claude to write something in your voice, so you do not have to keep pasting your voice guide into every conversation. Twenty minutes to build. You can ship this tonight.
A Skill is a folder containing a SKILL.md file (the instructions) and optional supporting files. The folder gets zipped and dragged into Claude. From then on, Claude loads the Skill automatically whenever it decides the Skill is relevant to what you are asking.
Two things make this powerful for marketers. First: Skills work across every Project and every conversation, so a Brand Voice Skill is loaded whether you are drafting an email or briefing an agency. Second: only the Skills that are relevant load, so you can have ten of them installed without bloating the context for unrelated questions.
The hardest part of a good Skill is the description. Claude uses the description to decide when to load the Skill. If the description is too vague, the Skill loads on every conversation and bloats the context. If it is too narrow, it never loads when you need it. For the Brand Voice Skill, the trigger is: any task that involves writing or rewriting external-facing content for the brand.
The description should name the trigger explicitly. Bad description: "Helps with brand voice." Good description: "Use this Skill whenever the user is writing, drafting, rewriting, or critiquing any external-facing content (emails, social posts, ads, web copy, blog drafts, scripts) where it needs to sound like the brand. Triggers on words like 'write,' 'draft,' 'rewrite,' 'in our voice,' 'on-brand.'"
The SKILL.md file starts with a YAML block that names the Skill and tells Claude how to recognise when to load it. Keep it tight. Name in kebab-case, description in operator voice.
---
name: brand-voice-extractor
description: Use this Skill whenever the user is writing, drafting,
rewriting, or critiquing external-facing content (emails, social
posts, ads, web copy, blog drafts, scripts) where it needs to sound
like the brand. Triggers on words like "write," "draft," "rewrite,"
"in our voice," "on-brand," or any request to produce or revise
copy for the company. Loads the brand voice guide and applies the
three structural moves (paired contrast, pattern-level rhythm,
do/don't pairs from real samples) to every output.
---
Below the YAML, write the actual instructions Claude will follow when the Skill loads. Three sections work for most marketing Skills: what the Skill knows, what it does, what it refuses.
# Brand Voice Extractor ## What this Skill knows This Skill loads the brand voice guide for [Brand Name]. The full guide is in voice-guide.md (in this Skill's folder). The three structural moves the brand voice is built on: 1. Paired contrast: every attribute named with what it is NOT (e.g. "confident, not arrogant. Specific, not technical. Plain, not flat.") 2. Pattern-level rhythm: short sentences. Concrete examples. No hedging. One italic emphasis per heading at most. 3. Do/don't pairs from real samples: see voice-guide.md for the 6 pairs that define on-brand vs off-brand. ## What this Skill does When loaded, apply the voice to every piece of output: - Use the rhythm patterns from voice-guide.md - Cite the do/don't pair if the user's draft hits a don't - Suggest revisions that move drafts toward the do side - If the user has not specified a register, default to the brand's standard register from voice-guide.md ## What this Skill refuses - Do not invent voice attributes that are not in voice-guide.md - Do not "elevate" or "punch up" the voice. The voice is the voice. Stay in it. - Do not use the words: delve, leverage, tapestry, landscape (as a metaphor), resonates with, drives engagement, navigate the complexities. These are explicitly off-brand. - If the user asks for output that contradicts the voice (e.g. "make it more corporate"), ask before you do it.
Save the file as SKILL.md inside a folder called brand-voice-extractor. Put your actual voice guide (the markdown file you extracted using the Voice & Tone Extractor) in the same folder, named voice-guide.md. Zip the folder. Drag the zip into Claude Desktop's Skills panel. Restart Claude.
Test it by opening any conversation (any Project, no Project, does not matter) and typing: "Write me a 100-word intro for a new product feature." If the Skill is working, the output should sound like your brand. The Skill activation will be visible in the conversation: Claude will note that the Brand Voice Skill loaded.
Trying to make one Skill do everything. The first Skill someone builds is almost always called "Marketing Skill" and has fifteen things in it: voice, briefs, readouts, social, email, the works. It loads on every conversation. It bloats the context. It produces worse output than a vanilla prompt would.
The right move is to build three to five narrow Skills, each one earning its trigger. Brand Voice. Refusal Rules. Output Format. Each one does one thing well. The smallest unit that earns its slot in Claude's context is the unit you want. The decision tree on Skill vs Project vs prompt covers when something belongs in a Skill in the first place.
A Skill is only as good as the file it points at. The Brand Voice Skill loads voice-guide.md, but voice-guide.md is the actual deliverable. If the voice guide is half-baked, the Skill is half-baked. Extract the voice properly first (the free Voice & Tone Extractor is built for this), then wrap it in the Skill.
Same applies to every Skill you build. The Skill is plumbing. The reference file is the work.
If you do not want to build your own from scratch, the four free Claude Skills on the Hub are the same four free prompts packaged as installable Skills, with the YAML and instructions already wired. Voice Extractor, Competitive Teardown, SERP-Informed Brief, Performance Readout. CC BY 4.0. Same template you would write yourself, ready to install.
Get the 4 free Skills