Claude Skills in Magi: Turn your best marketer's filters into standardized content rules

Claude marketing skills now run inside Magi. Set a skill to run after every outline or draft, or call it from Copilot. How to use Claude Skills in Magi, plus the best starter skills.

Author image

Vaishnavi Balachandran

Published

Read Time

AI content editor

What are Claude Skills, and why marketers care

A Claude skill is a short instruction file that tells an AI agent how to do one job. Claude marketing skills are the same thing, written for marketing work: an SEO check, a voice pass, a claims review. Think of it as the brief you would give a new hire: what to check, in what order, to what standard. Anthropic released the format, also called Agent Skills, in October 2025 and made it an open standard in December. Roughly 40 tools now read the same file, including Claude, Cursor, Codex, and GitHub Copilot.

Marketers took to skills fast. Most of the coverage this summer says the same three things. Teams stopped pasting the same prompt into every chat and started handing the AI a job. The first skill almost everyone builds is a brand-voice skill, because it fixes the "every draft sounds different" problem. And the most common failure is publishing skill output without a fact check.

Our own marketing team lived this. We built a humanizer, an SEO grader, and a founder-voice skill in Claude before any of them were in Magi. They worked. They also lived in one person's Claude setup. Running one meant opening Claude, pasting the draft, running the skill, and pasting the result back into the editor. Every time.

That gap is why skills are now inside Magi. The draft, the brand context, and the check should be in one place.

Where skills fit in BrandOS. BrandOS holds your facts and voice. Templates hold the shape of each content type. Skills hold the procedures and checks that run on the work.

Two ways to run a skill: as a rule or as a tool

As a rule. Set a skill in Settings, scope it to content types, pick a stage, and turn on auto-run. Now it fires on every outline or draft of that type. Nobody has to remember it.

As a tool. Open any document, type "/" in Copilot, pick the skill, and it runs once on the doc in front of you.

Same skill. Two triggers. A humanizer on auto-run is a quality floor. A humanizer on slash is a button. Most teams want both: a few skills that always run, and a longer list you reach for when you need them.

How to use Claude Skills in Magi: setup

Go to Settings > Skills and click Create Skill. The form has eight parts.

  1. Name and description. Copilot shows the description in its skill menu, so write what the skill does, not what it is called.

  2. Operation type. Text Modification edits the draft. Analysis reads the draft and reports back without changing it. Image Generation creates or changes images.

  3. Instructions. Paste the skill body in markdown, or upload a .md or .txt file. If you already have a SKILL.md from Claude or Cursor, upload it as is.

  4. Execution stages. After Outline runs on the structure before drafting starts. After Draft runs on the finished text. Pick one or both.

  5. Content types. LinkedIn post, blog post, landing page, email, one-pager. A skill only runs on the types you tick.

  6. Active. Off means the skill exists but never runs anywhere.

  7. Auto-run. On means the skill runs at its stage without anyone calling it. Off means it only runs when someone picks it from Copilot.

  8. Execution order. The second tab. Drag skills into the order they should run within each stage. Order matters: run your SEO grader before the humanizer, or the humanizer will undo the keyword placement.

Anyone in the workspace can create or edit a skill.

How to use Claude Skills in Magi: run one from Copilot

Open an outline or a draft. In the Copilot box, type "/". A list of the skills scoped to this content type and stage appears. Pick one. A chip appears with an optional context box, so you can add a note like "focus on the intro" before you send.

The skill runs and applies its edits to the document. The thread shows "Running skill: name" while it works. From there you keep editing with Copilot as usual.

Copilot tells the difference between asking a skill to run and asking about it. "Humanize this" edits the draft. "What does the humanizer do?" gets you an answer and leaves the draft alone.

What a skill sees

In a chat window, the skill sees whatever you pasted. In Magi, it runs on an outline or draft that already carries the campaign and ICP. The voice, tone, keywords, and brand rules are there too. A claims-check skill can see what the brand is allowed to say. A humanizer can see the voice it is supposed to match. You did not paste any of that. It was already there.

Best Claude Skills for marketing teams to start with

Humanizer ships with every workspace. Here are seven more Claude marketing skills that cover the checks most B2B teams do by hand.

After Outline

  • Answer-first structure. Put a direct answer under every H2 and build an FAQ from the brief's questions. Good for AEO.

  • Argument check (Analysis). Flag any section that repeats a point already made, and any heading that names a topic instead of a takeaway.

After Draft

  • SEO grader. Keyword placement, meta title and description, internal link slots.

  • Claims and compliance check. Every number needs a source. Every product claim needs to be shipped. Add your banned words if you sell into a regulated market.

  • Readability pass. Short sentences, 7th-grade reading level, cut repeated paragraphs.

  • CTA check. One CTA per piece, matched to the funnel stage.

  • Repurposing. Blog to LinkedIn post. Blog to newsletter section. One-pager to email. Our own workspace runs two of these on every send.

Build one. Run it on three drafts. When it misses, fix the instructions, not the draft. The next run is better, and so is every run after it.

Bringing in skills from Claude, Cursor, or GitHub

If your team already has skills, you do not rewrite them. Create Skill, upload the .md, tick the content types and stage, save.

You will probably bring in more than you expect. On August 6, OpenAI launched Agent Plugins with Amazon, Microsoft, Cursor, and Vercel. It reads the same skill files. Skill files are now how instructions travel between tools, and Claude skills marketplaces already index close to two million of them. The June ecosystem report found marketing skills growing faster than coding skills.

OWASP published an Agentic Skills Top 10 this summer, and researchers showed skill-borne attacks at Black Hat in August. A skill pulled onto a laptop can rewrite anything the agent touches. A skill in your workspace runs only at a named stage, on a named content type, and shows up in the thread every time. Same file. Much smaller blast radius.


See it on your own draft

Want to watch a skill run on your own outline? Book a demo.

Already on Magi? Login and go to Settings > Skills.

Frequently asked questions

What are Claude Skills? A Claude skill is a markdown instruction file that tells an AI agent how to do one job. The format is an open standard, so one skill file runs in Claude, Cursor, Codex, Magi, and about 40 other tools.

How is a skill different from a prompt or a template? A prompt is one instruction for one moment. A template sets the shape of a content type. A skill is a saved procedure that runs on the outline or draft, either automatically at a stage or when you call it.

Do skills in Magi run on Claude? No. The skill format comes from Claude. Inside Magi, a skill runs on whichever model your pipeline uses.

Can I use skills from GitHub in Magi? Yes. Upload the .md file in Create Skill, set the stage and content types, and save.

Which content types support Claude skills in Magi? LinkedIn posts, blog posts, landing pages, emails, and one-pagers.

How do you use Claude Skills in Magi? Create the skill in Settings > Skills. Turn on auto-run so it runs after every outline or draft, or type / in Copilot to run it on one document.

Content Editor Demo

Book a demo

Content Editor Demo

Book a demo