Skip to main content
A skill is a set of instructions an AI agent loads when a request matches it. Each one turns a document in this knowledge base into something that gets done the same way every time: the same intake questions, the same limits, the same output format. Skills are portable. The format is a plain folder of markdown, so the same skill runs on any agent that supports them. The documents come first. A skill reads the workbooks you have already filled in, applies the rules the framework sets, and hands back finished work. Nothing here replaces a brief or a template, and no skill invents a fact your documents do not contain.

Skills and the MCP

Skills and the Snow Doughnut MCP are two ways of running the same framework. A skill is a folder you install on your own agent; the MCP is a connected server that exposes the whole framework as operation commands, with your setup remembered between sessions. The published skills below work on their own, with or without the MCP.

How a skill gets picked up

Every skill carries a description listing the requests it covers. The agent reads that description against what you asked for and loads the skill when the two match. You do not have to name it. Both of these reach the same skill:
  • “Write me 15 PMax headlines from this brief.”
  • “/google-ad-copy”
Ask in your own words and the right skill loads. Name it directly, which most agents accept as a slash command, when you want to be certain.

What a skill is made of

A skill that can check its own work carries a script for it. The Google ad copy skill validates every character count before it renders anything, which is why its output never exceeds a limit.

Skills call other skills

A skill rarely runs alone. Ad copy pulls the copy template, the brief, the brand voice, and a hook list before it writes a word; a report pulls the flat file, the KPI definitions, and the report template. Each skill’s runbook carries a Pulls row naming everything it loads, so an agent gathers the full set in one pass and you never assemble it by hand.

The runbook

Every skill page ends with a runbook: the run card an agent executes from. It holds the trigger, the intake questions in order, what the skill pulls, the rules it enforces, and the output contract. The rest of the page explains the skill to a person. The runbook is the same skill compressed for the agent, run under how the MCP talks: pick-list questions in a multiple-choice dialog, up to four at a time, options from the framework’s lists, and nothing asked twice.

Published skills

Installing a skill

Skills live in a folder per skill, named after the skill, with SKILL.md at the top level. That layout is the same everywhere, so one folder works across every agent you use. Where the folder goes depends on the agent:
  • Agents that read from disk, such as command-line tools and editor extensions, take a skills directory. One in your home folder makes the skill available everywhere; one inside a project scopes it to that project.
  • Agents you use in a browser or desktop app take an upload under settings. The skill then applies to every conversation on your account.
Check your agent’s own documentation for the exact path or upload screen, since both differ by product. Reload the agent after adding a skill so it reads the new description, and use whatever command it offers for listing skills to confirm the skill is loaded.
Type: hub
Domains: mcp
Provides: skill-index
Uses: conversation-contract, skill:ad-creation, skill:deai, skill:google-ad-copy, skill:linkedin-ad-copy, skill:meta-ad-copy, skill:tiktok-ad-copy, skill:video-editing, skill:yapper-framework
Triggers: “what can you do” · “which skill handles X”Output: Route request to skill by description match

Get the complete Marketing OS

Subscribe to Snow Doughnut, our free weekly newsletter, and get the Notion workspace where you do the work: tasks, a marketing calendar, name generators, briefs, and more.

Have us run it for you

We implement this system, use it to help you grow and scale, then hand it all off to you when you’re ready.
Last modified on November 2, 2026