Claude Code Deep Dive - Mad Skillz

Claude Code Deep Dive - Mad Skillz

One of the most iconic scenes in the Matrix is when Neo uploads Jiu Jitsu and Kung Fu into his brain. Claude Code skills are EXACTLY the same. Let’s learn about the why, what, and how of agent skills and even create our own skill so Claude Code can generate images for us.

“I know Kung-Fu.” ~ Neo

This is the fourth article in the CCDD (Claude Code Deep Dive) series. The previous articles are:

  1. Claude Code Deep Dive - Basics
  2. Claude Code Deep Dive - Slash Commands
  3. Claude Code Deep Dive - Total Recall

🧠 What are skills? 🧠

A skill is just a directory that contains a SKILL.md file plus optional scripts, templates, and docs. The SKILL.md file must have frontmatter with a name and a description:

---
name: awesome-skill
description: awesome skill for doing cool stuff!!!
---

There are a lot of optional frontmatter fields too. But, this is not a reference manual, so check them out yourself later!

Everything after the frontmatter is free-form Markdown. Think of it as a mini playbook for Claude to follow.

🧭 What’s the Big Deal about Skills? 🧭

So far that sounds a lot like the stuff you can put in CLAUDE.md. Right, except skills are a clean context-engineering solution. Instead of clattering CLAUDE.md with every workflow, you keep targeted knowledge in skills and let Claude Code load them on demand. You get focus because only the relevant instructions enter the context. You get reuse because a skill can live once and serve many projects. And you get automation because scripts and templates let Claude actually do the work instead of just describing it.

This approach is so useful that it became an open standard (https://agentskills.io) used by most of the industry. More on that later.

🧱 Where Do Skills Live? 🧱

Claude Code looks for skills in these locations:

  • Enterprise (managed settings.json)
  • Personal: ~/.claude/skills
  • Subdirectory: foo/.claude/skills
  • Project: .claude/skills
  • Plugin: <plugin>/skills

Nested skills behave like nested CLAUDE.md files: the closer the skill is to the file you’re working on, the more specific it is. If multiple skills share the same name, higher-priority locations win (enterprise > personal > subdirectory > project). Plugin skills use a plugin-name:skill-name namespace, so they don’t conflict with other levels. If a skill and a slash command share the same name, the skill wins.

πŸ”¦ How Are Skills Loaded? πŸ”¦

Only the frontmatter (name + description) is loaded into context at first. Claude sees a list of available skills and decides which one to load when it’s needed. When it chooses a skill, the full SKILL.md content is pulled in, and any referenced files like scripts or templates can be loaded as needed.

This is why the description matters. It’s the trigger.

πŸ› οΈ Create Your Own Skill πŸ› οΈ

Alright. Let’s create our own skill from scratch. Claude Code can analyze images, but it can’t generate images out of the box. This is a pretty useful capability, so I built a skill for that.

Here is the structure:

skills/generate-image/
β”œβ”€β”€ SKILL.md
└── scripts/
    └── generate-image.py

No need for too many explanations about how it works since the SKILL.md file does exactly that :-)

---
name: generate-image
description: Generate images using OpenAI's gpt-image-1.5 model
---

# Generate Image

Create images from text prompts using the `scripts/generate-image.py` script.

## Usage

./scripts/generate-image.py "prompt" [output_file]

## When to use

- User asks for an image, icon, logo, illustration
- Visual content would help explain something

The generate-image.py script does the actual work. Claude reads the skill, runs the script, which generates the image and saves it to the output file. That’s it.

You can check the existing skills available to you in Claude Code itself by typing /skills. It will show a list of skills as well as custom slash commands. And, yes you can run skills directly just like slash commands:

Let’s see it in action. I told Claude Code to generate an image of dogs playing poker (classic), and it figured out it needs to load the skill and then run the script, which executed perfectly.

I then asked Claude Code to open it, which it did (using the open command in the terminal via the built-in Bash tool)

img.png

That’s about it. You can ask Claude to help you write your own skills of course. It is very skilled at that (see what I did there?).

Let’s see where skills are in the overall AI agent community as of January 2026.

βš–οΈ Claude Code Skills vs Agent Skills βš–οΈ

Skills started at Anthropic, but the idea is now open and shared across the ecosystem. The emerging standard lives at agentskills.io and is supported by multiple AI coding agents. Claude Code uses the same core concept: a SKILL.md file with name/description frontmatter plus optional features. But, the Anthropic skills are a superset of agent skills, so if you use primarily Claude Code I recommend checking out the Anthropic skills docs (or this blog post).

If you stick to the lowest common denominator of Agent skills, then your skills will be portable across tools, which is exactly the point. One skill, many AI agents.

⏭️ What’s Next ⏭️

So far, we’ve covered the basics, slash commands, the memory system and now agent skills. The CCDD series will continue with more advanced topics:

  • MCP support
  • Subagents
  • Hooks
  • Plugins
  • Beyond the terminal
  • Managing multiple Claude Code sessions in parallel
  • Comparison with other AI coding agents

🏠 Take Home Points 🏠

  • Skills are on-demand knowledge modules, not permanent context
  • SKILL.md frontmatter (name/description) is the trigger
  • Keep skills small, focused, and reusable
  • Scripts and templates turn skills into real automation
  • The skills format is becoming a cross-agent standard

πŸ‡§πŸ‡¬ Π§Π°ΠΎ, приятСли! πŸ‡§πŸ‡¬