Agent Skills

Are you ready to supercharge your AI agents? Imagine an AI that not only understands your requests but can also execute complex, multi-step tasks with precision. This is where Agent Skills come in, and they're set to revolutionize how we interact with AI!

1.About Agent Skills

Agent Skills, initially introduced by Anthropic in October 2022, an innovative way to share procedural knowledge with AI agents. At their core, these skills are simple yet powerful they're markdown files containing clear, specific instructions for your AI.

Think of them as mini-manuals that teach your AI exactly how to perform a task. Each skill comes with a YAML header that includes a name and description, giving your agent the necessary context before diving into action.

The brilliance of Agent Skills quickly caught on, leading to an open standard that has been embraced by major players in the AI world, including OpenAI Codex, Microsoft, GitHub, and Cursor. This widespread adoption means a more unified and portable way to enhance AI capabilities across different platforms.

2.Skills.sh - Your New Best Friend for Skill Management

Managing these powerful skills has never been easier, thanks to skills.sh, Vercel's new command-line interface (CLI) tool. This tool simplifies the process of installing and managing skills, making it accessible even for those new to AI agent development.

3.Getting Started

To install a skill, use the skills CLI:

npx skills add vercel-labs/agent-skills

This will install the skill and make it available to your AI agent.

Go to your project where you need a skill for one of the project i am using the anthropic skill for a simple website that is npx skills add anthropics/skills

Once the skill is installed you can choose the skill that you want in my case it is frontend-design and once it is choosed you can choose to which agent you want this skill

After choosing the agent you can see the skill repo is clone and it is added

If you open your project now you can see that skill. Example if you choose the claude agent you can see that skill like below

And now you can give your Claude agent the instructions that you want and it will consider the following instructions.

4.Creating your own skill

You can create your own skill and publish it to the vercel skills for that what you need to do is very simple that is

npx skills --help

you can see the list of options from the skills cli like below

Now we will create a new skill for code review using the below command

npx skills init skills code-review

which will create a skill like below

And if you open the project folder you can see a skill template like below

Now we will add a skill creator so that we will know the structure that we need to followup. Add the skill creator in the skill using the below command

npx skills add https://github.com/anthropics/skills --skill skill-creator

Once the skill-creator is installed you would see something like this in your project directory

Now you can update the instructions based on your need and can share the skill.

Will take one more example of downloading the react best practices that needs to be followed. Download the

npx skills add https://github.com/vercel-labs/agent-skills --skill vercel-react-best-practices

Once the skill is downloaded you can see 3 things inside the .claude directory inside the vercel-react-best-practices

• AGENTS.md which is mainly for agents and LLMs to follow when maintaining

• SKILL.md Guidelines for the skill

• rules directory - Here you can find the best practices along with examples like below

  

5.How to Validate a Skill

When building a skill, it's crucial to work iteratively on it until it reaches a state where you can fully delegate the task to Claude.

• Generate log files: Once you have validated that your skill works correctly, you should generate log files. This helps ensure that you don't accidentally use a skill that has not been properly validated.

• Teach Claude edge cases: You need to teach Claude or your AI agent the edge cases to ensure it handles them well, and then add these to the skill. This iterative process helps refine the skill until it consistently produces the desired results.

6.Sharing a skill via GitHub

Skills are shared via GitHub by housing them within a skills directory inside a GitHub repository.

Here's how it works:

• The entire set of skills for a project lives inside this skills directory.

• Each specific skill will have its own directory within this main skills folder.

• Inside each individual skill's directory, there is a skill.md file, which serves as the entry point for the AI agent. This file contains the skill's description and main instructions.

• Alongside the skill.md file, you can include other resources like scripts, additional markdown files, or templates that the skill might need.

• This structure allows skills to be easily cloned and installed by other users using tools like skills.sh.

This method is convenient because a skill can be located in the same repository as the tool it teaches the agent to use, making it easier to update the skill whenever the tool is updated

Voila we have learnt how to use a skill in our project to make things better as well as learned how to create our own skill.

You can find some great skills in the Vercel 🎉

Sky is the limit of what you can do with them.