Creating Agent Skills FAQs

Question 1

How does this skill help optimize the context window?

Accepted Answer

It provides guidance on a Three-Tier Architecture: Tier 1 (Metadata) for discovery, Tier 2 (Full Content) for instructions, and Tier 3 (References) for deep data. This ensures Claude only loads the information it needs, preventing context bloat.

Question 2

When should I use the Creating Agent Skills capability?

Accepted Answer

Use this skill when you are designing new SKILL.md files, organizing directory structures for complex tools, or trying to improve the discovery and reliability of your custom Claude Code extensions.

Question 3

What exactly is a Claude Code Agent Skill?

Accepted Answer

An Agent Skill is a modular capability that extends Claude's functionality. Unlike simple documentation, skills are formalized 'playbooks' or procedural workflows stored in structured folders that guide Claude through complex, multi-step tasks reliably.

Question 4

Why is the {baseDir} variable important for skill development?

Accepted Answer

{baseDir} is critical for portability. It allows you to reference bundled scripts and assets using relative paths that resolve correctly regardless of where the skill is installed, preventing broken paths across different user environments.

Question 5

What is the difference between instruction-based and code-based skills?

Accepted Answer

Instruction-based skills rely primarily on text guidance within a SKILL.md file. Code-based skills leverage executable scripts (Python, Bash) where only the script's output consumes context tokens, making them highly efficient for data processing.

Creating Agent Skills

Creating Agent Skills

Key Features

Use Cases

Key Features

Use Cases