AILoadoutAILoadout
AI Guide

What are AI Skills?

AI Skills (Agent Skills) are “specialized know-how and workflows” packaged into a file (SKILL.md) that AI picks up automatically when it hits a matching task — like a skill for making PDFs, reviewing code, or analyzing Excel. In short, they teach the AI “how to do one job well” and keep it on hand, so it loads and uses it on its own without you re-explaining every time.

Updated July 2026·~7 min read·Level: Beginner

What AI Skills actually are

AI Skills — fully known as Agent Skills — come from Anthropic's Claude Agent Skills. The idea is to write specialized know-how and workflows into a file called SKILL.md that the AI picks up on its own when it hits a task that matches — for example a PDF skill, a code-review skill, or an Excel-analysis skill.

Many people ask how this differs from “just a prompt.” The difference is that a prompt is a one-off command you type yourself, while a Skill is know-how installed ahead of time that the AI chooses to load only when needed (progressive disclosure) — normally it only reads each skill's name and short description, then pulls the full details in when a matching task appears. On top of that, a Skill can bring along supporting files and scripts, not just plain text.

How AI Skills work

The heart of a Skill is the SKILL.md file, split into two main parts:

  • The header (metadata) — the skill's name and a short description of “when to use it,” which tells the AI what kind of task should trigger this skill.
  • The body (instructions) — the steps, workflow, examples, and possibly references to supporting files/scripts bundled in the same folder.

Roughly how it goes in practice:

  1. You install the skill with the AI (drop the folder with SKILL.md in the designated place — a one-time setup)
  2. The AI reads only each skill's name and description first, not the full body
  3. When you give a task that matches a skill, the AI pulls in that skill's full body to read and follow
  4. If the skill includes scripts or supporting files, the AI can call them as the skill instructs
💡 Key point: Progressive disclosure means the AI doesn't have to carry every skill's details at all times — it grabs only the skill that fits the current task, so it can hold hundreds of skills without filling up the context, and works more accurately.

What AI Skills let AI do

AI Skills add specialized know-how and workflows, helping AI handle tasks that need specific steps or standards — for example:

  • Create documents — produce PDF, Word, PowerPoint files in a set format
  • Review code — check code against a team's checklist and standards
  • Analyze data — open Excel/CSV files to summarize, calculate, and chart
  • Follow specific procedures — e.g. how a team deploys or writes commits
  • Use specialized tools via scripts bundled with the skill

AI Skills add specialized know-how, while MCP changes what AI can access and AI Rules set the standards it must follow.

Real-world AI Skills examples

Popular skills people actually use:

  • PDF / documents — create and edit PDF, Word, PowerPoint files with a professional format
  • Excel / spreadsheets — read, edit, calculate, and chart from table files
  • Code Review — check code against standards, find bugs and things to improve
  • Brand / Design — format work to match brand guidelines
  • Skill Creator — a skill that helps you write new skills

See curated AI Skills, with badges for which are official / which AIs they support, in the AI Skills library.

Does it work with Claude, Cursor and Codex

Yes — AI Skills started with Claude, but SKILL.md is a plain text file, so it's flexible and adaptable elsewhere:

  • Claude (Claude.ai, Claude Code) — full Agent Skills support; the originator of the idea
  • Cursor — can use similar files as specialized instructions for the AI in a project
  • OpenAI Codex and others like Windsurf, Cline — use project-attached instruction/guide files in a comparable way
Because SKILL.md is plain text, a single skill can often be adapted across tools — write it once, build on it in many places.

How to start using AI Skills

Getting started with AI Skills takes 3 steps:

  1. Pick a skill that fits the job — want AI to make PDFs? Choose the PDF skill. Open the AI Skills library and browse by task.
  2. Install the skill into your AI tool — place the folder containing SKILL.md where the tool expects it, following each tool's guide.
  3. Try a task that matches the skill — give a task that needs it; if set up right, the AI picks up the skill on its own.

Don't want to wire them one by one? Pick a Loadout with AI Skills — bundled by task, ready to use.

Are AI Skills safe

A Skill is a set of instructions, possibly with scripts, that the AI follows — so use only skills you trust, since they affect how the AI actually works. Keep these in mind:

  • Install only trustworthy skills — pick official or well-reviewed open-source ones, and read the SKILL.md first
  • Be careful with skills that bundle scripts — make sure you know what the script does, don't run it blindly
  • Read the “when to use” description — a skill with a clear description gets triggered at the right moment and won't fire on the wrong task
✅ We curate only official / well-reviewed open-source AI Skills, with full source and license attribution, so you can choose with confidence.

AI Skills vs MCP vs Rule

These three are often confused, but they do different jobs:

DifferenceAI SkillsMCPRule
What it changesSpecialized know-how / workflowWhat AI can access / act onThe standards / rules it must follow
Like a…Manual / built-in skillPlug for toolsHouse rules
When it runsWhen a matching task appearsWhen AI needs data/toolsAll the time, every task

Read on: What is MCP · What are AI Rules · What are AI Agents

Loadouts with AI Skills

In practice you rarely use one Skill alone — you combine several with MCP/Rules/Prompts to finish one job. A bundle assembled like that is called a Loadout.

For example, a web-dev Loadout might include a code-review Skill + a document Skill, an MCP for GitHub, and a coding-standards Rule in one set — ready to use without assembling it yourself.

Next steps