Documentation

Claude Agent Skills Guide

Agent Skills are modular capabilities that extend Claude's functionality. Through a standardized filesystem structure, they equip agents with domain-specific expertise.

Modular

Create once, reuse infinitely. No need to repeat prompts in every conversation.

Filesystem-based

Runs on a virtual filesystem. Manage AI skills just like you manage project code.

Progressive Disclosure

Loads files on demand, significantly saving Context Window usage.

How it Works: Progressive Disclosure

Click the cards below to see how Claude loads information in stages.

LEVEL 1

Metadata

Always Loaded

Loads only YAML frontmatter. Extremely low cost (~100 tokens).

name: pdf-tool
desc: Extract text...

LEVEL 2

Instructions

On Trigger

Reads SKILL.md when user intent matches description.

# PDF Processing
## Steps...

LEVEL 3

Execution

On Demand

Reads reference docs only when needed. Code logic stays out of context.

$ python run.py
> Output data

Skill File Structure

Interactive Viewer

Directory

Scripts

Tip

SKILL.md is the entry point. Split complex instructions into child files to avoid consuming too many tokens at once.

SKILL.MD

---
name: processing-pdfs
description: Extract text and tables from PDF files, fill forms.
---

# PDF Processing

## Quick Start
Use pdfplumber to extract text from PDFs:

```python
import pdfplumber
with pdfplumber.open("doc.pdf") as pdf:
    text = pdf.pages[0].extract_text()
```

## Features
- **Form Filling**: See [FORMS.md](FORMS.md)
- **Validation**: Run `scripts/validate.py`

Templates & Best Practices

Best Practices

Naming Convention

Use Gerunds (e.g., processing-data) to clearly state "what is happening".

Third-Person Description

Description is injected directly into System Prompt. Avoid "I help you...", use "Extracts data from..." instead.

---
name: your-skill-name
description: Brief description of actions (3rd person).
---

# Skill Name

## Instructions
[Step-by-step guidance]

## Examples
Input: ...
Output: ...