Understanding AI Skills: How To

We often think of AI like Claude as a brilliant conversationalist—a "know-it-all" friend you can chat with. But with the rise of Agent Skills (and resources like https://agentskills.io/home), we are moving from asking AI to talk about things, to teaching AI how to do things.

A Skill is essentially a standard operating procedure (SOP) for your AI. It transforms a generic Large Language Model (LLM) into a specialized agent capable of executing complex workflows reliably.

What Are Pre-Existing Skills?

Out of the box, Claude already possess "native" skills. These include:

• Coding: Writing and executing scripts.
• Web Browsing: Searching the internet for real-time data.

However, the real magic happens when you layer Custom Skills on top of these. This is where you define the specific business logic for your unique needs.

How SKILL.md Works

A skill is usually defined in a simple Markdown file (SKILL.md). It acts as a strict set of instructions. It generally consists of two parts:

1. The Metadata (YAML Frontmatter): Located at the top, this tells the AI the name of the skill and when to use it.
2. The Instruction Body: The specific steps, formulas, and formatting rules the AI must follow.

The Anatomy of a Custom Skill

Let's look at how to create a custom skill using a Marketing Campaign Analysis as our example.

To create this skill, you write a file that defines:

• Input Requirements: "I expect a CSV with columns for ad spend, clicks, and revenue."
• Logic & Formulas: "To calculate Net Profit, subtract $8 shipping per order and 35% product cost from the revenue."
• Benchmarks: "If the Click-Through Rate (CTR) is below 2.5% for Facebook, flag it."
• Output Format: "Always give me a table with status indicators like [OK] or [X]."

Putting It Into Practice: A Real-World Example

Imagine you are a Digital Marketer. You have a messy spreadsheet of last week’s ad performance across TikTok and Google.

Without a skill, you might paste the data to Claude and ask, "How did we do?" You’ll likely get a generic summary.

With the analyzing-marketing-campaign skill defined, here is what happens:

1. Trigger: You upload the CSV and ask, "Analyze our weekly performance."
2. Recognition: Claude reads the skill description (analyzing-marketing-campaign) and realizes this is the correct tool for the job.
3. Execution:

• It scans the file for errors (e.g., conversions without clicks).
• It calculates ROAS (Return on Ad Spend) and CPA (Cost Per Acquisition) using the math defined in your skill file, not generic math.
• It compares your TikTok ads against the specific benchmark (0.9% CVR) listed in the file.

4. Result: Instead of text, you receive a perfectly formatted Efficiency Analysis Table:

Interpretation: Google is driving profit, but TikTok is bleeding money due to high CPA. Recommended action: Pause TikTok and review creative.

Why This Matters

Skills turn "prompt engineering" into "software engineering." By writing a SKILL.md file, you stop hoping the AI guesses what you want and start guaranteeing it follows your business rules—every single time.

Whether you are analyzing marketing funnels, checking code for bugs, or writing SEO articles, defining a Skill ensures your AI partner works exactly the way you need it to.

Want to learn more about skills, no doubt to enroll in this course: Agent Skills with Anthropic