From Chatting to Doing: Implementing Real-World AI Skills for Claude

alirezarezvani/claude-skills

Here is a breakdown of why this is a game-changer and how you can get started.

In the world of Agentic AI, we want models that don't just "talk" about code but actually do things. This repository provides a structured way to give Claude specific "skills"—pre-defined functions or subagents that handle repetitive or complex tasks.

Autonomy
Instead of you copying and pasting output, Claude can use these skills to interact with your file system or external APIs.

Consistency
By using standardized "Commands" and "Subagents," you ensure the AI follows a specific protocol every time it performs a task like documentation or testing.

Efficiency
It reduces the "context window" fatigue. Instead of explaining how to do a task every time, you just give Claude the skill to execute it.

To use these skills, you generally need to be using Claude Code (the CLI tool from Anthropic) or a framework that supports MCP (Model Context Protocol).

Most of these skills are designed to be integrated into your local environment. You can typically clone the repository and link the scripts to your execution path

git clone https://github.com/alirezarezvani/claude-skills.git
cd claude-skills

You'll need to tell your Claude instance where these skills live. If you are using the claude CLI, you might add these to your configuration file (often found in ~/.claude/config.json or similar, depending on your setup).

Let’s look at what these "skills" actually look like in practice. One common pattern in this repo is the use of Subagents—specialized instructions that handle one specific part of a project.

Imagine you want Claude to automatically generate technical documentation for a new module.

The Skill Definition (Simplified Logic)

How you would trigger it in the terminal

# Using a custom command defined in the skills repo
claude "Run the doc-gen skill on ./src/auth-module.ts"

What happens under the hood (Sample Code)
The skill might involve a script that tells the subagent

Read the file content.

Identify exported functions.

Write a Markdown file using a specific template.

// A conceptual look at how a skill might be structured
export const documentationSkill = {
  name: "generate_docs",
  description: "Reads a file and creates a README.md snippet",
  execute: async (filePath: string) => {
    // Logic to parse the file and return a structured prompt to Claude
    const code = await readFile(filePath);
    return `Analyze this code and generate JSDoc comments: ${code}`;
  }
};

This repository is a great starting point for building your own AI-driven development environment. It moves you away from "Chatting with AI" and toward "Collaborating with an AI Agent."