Agentic ai • March 18, 2026 • 5 min read

🤖 Claude Skills - Introduction & Overview 🚀

A **Skill** is an organized package of instructions, scripts, and resources that gives a Claude agent the ability to perform a specific task with expertise.

Agentic AI Claude Skills Claude Code AI Agents LLM Gen AI Ollama Local AI Models

What is a Skill?
Why Skills? Key Advantages
Two Types of Skills
Skill Structure
Progressive Loading
Using Skills in Claude Code
Skill Composability

1. What is a Skill?

A Skill is an organized package of instructions, scripts, and resources that gives a Claude agent the ability to perform a specific task with expertise.

Think of it like a playbook — when you want Claude to achieve a particular goal, you define a Skill so that Claude follows the precise instructions within it and delivers accurate, deterministic results.

💡 Simple Analogy: A generalist (base Claude) can do many things reasonably well. A Skill turns Claude into a specialist — it knows exactly what tools to use, what format to follow, and how to validate its output for that one job.

Skills are currently popular across AI Agent and LLM ecosystems because they bridge the gap between raw model capability and reliable, production-grade task execution. They are predefined executable patterns — and that predictability is their power.

2. Why Skills? Key Advantages

Advantage	Description
⚡ Token Optimization	Skills are loaded only when relevant, reducing unnecessary token consumption
📚 Easy to Learn	High-level abstractions — describe what to do, not how Claude reasons through it
🧩 Composability	Multiple Skills can be combined to achieve complex, multi-step goals
🚀 More Efficient	Pre-structured instructions mean less back-and-forth, fewer clarification loops
🎯 Deterministic & Accurate	Predefined executable scripts produce consistent, predictable results
🔒 Guardrails Built-In	Embed validation steps, output format rules, and error handling

3. Two Types of Skills

🏛 Anthropic Built-In Skills

Ready-to-use skills maintained and versioned by Anthropic:

Skill	Purpose
`xlsx`	Excel / spreadsheet generation
`docx`	Word document creation & editing
`pdf`	PDF generation, merging, OCR
`pptx`	PowerPoint deck creation
`frontend-design`	Production-grade UI components
`product-self-knowledge`	Accurate Anthropic product facts

🛠 User-Defined Custom Skills

Build anything your workflow needs:

Skill	Purpose
`stock-analyzer`	Technical + fundamental analysis
`security-monitor`	Threat detection & alerting
`agent-orchestrator`	Multi-agent task coordination
`email-processor`	Custom inbox workflows
`code-reviewer`	Opinionated code quality checks
`your-skill-here`	Anything your workflow needs!

Key difference: Anthropic-loaded Skills are maintained by Anthropic. Custom Skills are yours — you define them, own them, and they can be as domain-specific as needed.

4. Skill Structure

Every Skill lives in its own directory and revolves around a single key file: SKILL.md.

stock-analyzer/
├── SKILL.md              ← The brain · metadata + instructions (required)
├── helper-script.py      ← Shared utility functions
├── templates/            ← Output blueprints
│   ├── report.html       ← HTML report layout with 
│   └── email.txt         ← Email message template
└── resources/            ← Static reference data
    └── reference-data.json  ← Lookup tables, configs, benchmarks

What each file does

🐍 `helper-script.py` — Utility Script

Reusable Python functions that every other file in the skill depends on. Written once, called everywhere. No code generation at runtime.

# helper-script.py — shared utilities
def format_currency(value):
    return f"${value:,.2f}"

def pct_change(old, new):
    return f"{((new-old)/old)*100:+.1f}%"

📁 `templates/` — Output Blueprints

Pre-built output layouts with `` tokens. Claude populates them at runtime — guaranteeing consistent formatting every time.

templates/report.html

<h1> — Stock Report</h1>
<p>Price:   ·  RSI: </p>
<div id="chart"></div>

templates/email.txt

Subject:  Analysis — 

Hi ,
 closed at  ().
RSI:  · Signal: 

📁 `resources/reference-data.json` — Static Reference Data

Static facts Claude looks up on-demand instead of reasoning from scratch. Results are always consistent.

{
  "sector_pe_benchmarks": {
    "Technology": 28.5,
    "Healthcare": 22.1,
    "Financials": 14.3
  },
  "rsi_thresholds": { "overbought": 70, "oversold": 30 }
}

`SKILL.md` — The Core File

The SKILL.md file has two parts:

YAML frontmatter — lightweight metadata Claude reads first (always in memory)
Full instruction body — loaded only when the skill triggers

---
name: stock-analyzer               # max 64 characters
description: |                     # max 1024 characters
  Analyze stocks using technical indicators (RSI, MACD,
  Bollinger Bands) and generate reports with price
  projections. Use for swing trading and portfolio review.
  Triggers: "analyze AAPL", "stock report for MSFT",
  any ticker + time period combination.
---

# Full instructions below — loaded only when skill triggers
# Up to ~5,000 tokens of detailed guidance

Field	Limit	Purpose
`name`	64 chars	Unique identifier seen by Claude
`description`	1,024 chars	Determines when Claude triggers this skill
Full instructions	< 5k tokens	The detailed playbook
Linked files	On-demand	Scripts, templates, data

5. Progressive Loading

Claude never loads everything at once. Only what’s needed, when it’s needed.

Level 1: Metadata (name + description)
         → Always in memory · Zero cost

              ↓  skill triggers

Level 2: SKILL.md body
         → Full instructions load into context
         → User says "analyze AAPL" → description matches

              ↓  Claude needs detailed info

Level 3: Resources (indicators.md, reference-data.json)
         → Loaded only if required
         → e.g. Claude needs the RSI formula to explain output

💡 This keeps Claude’s context lean while providing deep expertise exactly when needed.

6. Using Skills in Claude Code

Claude Code is the CLI that lets you run Claude directly from your terminal — with native Skills support.

Step 1 — Install Claude Code

npm install -g @anthropic-ai/claude-code
claude --version
# Claude Code 1.x.x — ready

Step 2 — Place Your Skill Package

Drop your skill folder inside .claude/skills/ in your working directory:

claude-workspace/              ← open terminal from here
└── .claude/                   ← Claude Code config folder
    └── skills/                ← all skill packages live here
        └── stock-analyzer/    ← your skill package
            ├── SKILL.md       ← required
            ├── helper-script.py
            ├── templates/
            └── resources/

💡 .claude/ is to Claude Code what .git/ is to Git — the project-level config home.

Step 3 — Launch Claude Code & Run `/skills`

~/claude-workspace $ claude
✓ Claude Code started · project: claude-workspace

claude> /skills

📦 Available Skills (1 found in .claude/skills/)

  01  stock-analyzer   Comprehensive stock analysis using Yahoo Finance data

Type a query to trigger a skill automatically, or use /skill <name> to invoke directly

claude> Analyze the stock AAPL for last 6 months

⚡ Skill matched: stock-analyzer  (triggered by: "analyze the stock")
📥 Loading SKILL.md instructions...
🐍 Running helper-script.py → analyze_stock.py --symbol AAPL --period 6mo
📊 Generating report from templates/report.html...

✅ Analysis Complete — AAPL (6 months)
   Current Price: $195.71  |  6mo Return: +31.4%
   RSI: 58.4 (Neutral)     |  P/E: 32.4x
   Report saved: aapl_report_6mo.html

Bonus — Direct Skill Invocation

claude> /skill stock-analyzer
📦 Skill loaded. What would you like to analyze?

claude> Compare NVDA and AVGO over 1 year
⚡ Running analysis for 2 symbols: NVDA, AVGO...

Quick Reference

Command	What it does
`/skills`	List all skills in `.claude/skills/`
`/skill <name>`	Directly invoke a skill by name
Natural query	Claude auto-matches to the right skill
`.claude/skills/`	Place skill folder here to activate it

7. Skill Composability

One of the most powerful features — multiple Skills working together in a single request.

📊 xlsx  +  📄 docx  +  📈 stock-analyzer  =  🚀 Full Research Report

A single request — “Analyze NVDA and give me a Word report with an Excel data appendix” — triggers three Skills simultaneously. Each contributes its specialty, Claude orchestrates them seamlessly.

Key Takeaways

Skills = organized packages of instructions, scripts, and resources for specific tasks
Two types: Anthropic built-in (xlsx, docx, pdf, pptx) and user-defined custom Skills
SKILL.md is the core — YAML frontmatter for discovery, full instructions for execution
Progressive loading keeps token usage efficient — only load what you need
.claude/skills/ is where Claude Code discovers your skill packages
Skills are composable — combine multiple Skills for complex, multi-step workflows
Results are more deterministic because the execution path is pre-defined

🚀 Up Next: Hands-On Session — Build your own Skill from scratch

🤖 Claude Skills - Introduction & Overview 🚀

Table of Contents

1. What is a Skill?

2. Why Skills? Key Advantages

3. Two Types of Skills

🏛 Anthropic Built-In Skills

🛠 User-Defined Custom Skills

4. Skill Structure

What each file does

🐍 `helper-script.py` — Utility Script

📁 `templates/` — Output Blueprints

📁 `resources/reference-data.json` — Static Reference Data

`SKILL.md` — The Core File

5. Progressive Loading

6. Using Skills in Claude Code

Step 1 — Install Claude Code

Step 2 — Place Your Skill Package

Step 3 — Launch Claude Code & Run `/skills`

Bonus — Direct Skill Invocation

Quick Reference

7. Skill Composability

Key Takeaways

Color Themes

Table of Contents

1. What is a Skill?

2. Why Skills? Key Advantages

3. Two Types of Skills

🏛 Anthropic Built-In Skills

🛠 User-Defined Custom Skills

4. Skill Structure

What each file does

🐍 helper-script.py — Utility Script

📁 templates/ — Output Blueprints

📁 resources/reference-data.json — Static Reference Data

SKILL.md — The Core File

5. Progressive Loading

6. Using Skills in Claude Code

Step 1 — Install Claude Code

Step 2 — Place Your Skill Package

Step 3 — Launch Claude Code & Run /skills

Bonus — Direct Skill Invocation

Quick Reference

7. Skill Composability

Key Takeaways

Color Themes

🐍 `helper-script.py` — Utility Script

📁 `templates/` — Output Blueprints

📁 `resources/reference-data.json` — Static Reference Data

`SKILL.md` — The Core File

Step 3 — Launch Claude Code & Run `/skills`