Agent Skills with Anthropic

Course Overview

The Agent Skills course from Anthropic (via DeepLearning.AI) teaches how to extend AI agent capabilities using the Agent Skills open standard. Spanning 7 lessons, the course covers everything from understanding why skills matter to building production applications with the Python Agent SDK. Skills provide procedural knowledge loaded on demand, making agents more capable without bloating their system prompts.

Course Details

What You'll Learn

This course covers 7 essential lessons:

1. Why Agent Skills? — Understand the paradigm shift from specialized agents to general-purpose agents + domain-specific skills
2. Skills vs Tools, MCP, and Subagents — Learn how skills complement MCP connections and differ from built-in tools and subagents
3. Exploring Pre-Built Skills — Work with Anthropic's skill repository and compose skills into end-to-end workflows
4. Creating Custom Skills — Author your own skills with the standard directory structure: SKILL.md + assets, references, and scripts
5. Skills with Claude API — Embed skills in custom applications with single-turn, multi-turn, and streaming patterns
6. Skills with Claude Code — Leverage advanced frontmatter fields, user-invocable skills, and subagent integration
7. Skills with Claude Agent SDK — Build Python research agents with subagents, MCP integration, and cost control strategies

Course Structure

L1: Why Agent Skills?

The course opens with a worked example — an analyzing-marketing-campaign skill — demonstrating the gap between ad-hoc prompting and packaged skills. You'll see firsthand how uploading a skill transforms agent behavior, enabling it to apply budget reallocation rules automatically.

The second part introduces the paradigm shift: moving from specialized agents (Research, Coding, Finance) with narrow focus, to general-purpose agents that gain domain expertise through skills loaded on demand.

Key characteristics of Agent Skills:

• Portable — Works across Claude Code, Claude.ai, Agent SDK, and Claude API
• Composable — Chain skills into complex multi-step workflows
• Progressive Disclosure — Metadata → Instructions → Resources, loaded only as needed

L2: Skills vs Tools, MCP, and Subagents

A critical lesson that draws clear distinctions:

L3: Exploring Pre-Built Skills

Work with Anthropic's skill repository including pptx and skill-creator. Learn Skill Composition — combining skills into end-to-end workflows with three stages:

1. Data Retrieval — Gather information from sources
2. Domain Analysis — Apply specialized knowledge
3. Output Generation — Produce final deliverables (e.g., Marketing skill + Brand Guidelines skill + pptx skill)

L4: Creating Custom Skills

The skill authoring workflow:

1. Define purpose and trigger conditions
2. Write SKILL.md with YAML frontmatter + instructions
3. Add supporting files (assets/, references/, scripts/)
4. Test and iterate

Standard directory structure:

skill-name/
├── SKILL.md           # Required
├── assets/            # Optional: templates, images
├── references/        # Optional: documentation
└── scripts/           # Optional: executables

L5: Skills with Claude API

Learn to embed skills in custom applications by passing them alongside prompts in API requests. Covers application patterns including single-turn, multi-turn, and streaming — no Claude subscription required (API key only).

L6: Skills with Claude Code

Deep dive into Claude Code's advanced skill features:

• Frontmatter fields: allowed-tools, model, user-invocable, context, agent
• User-invocable skills: Manual invocation via /command
• Skills + Subagents: Two approaches — inject skills into subagent definition, or use context: fork for isolated execution

L7: Skills with Claude Agent SDK

Build Python-based research agents with subagents, integrating MCP for external services (e.g., Notion export). Includes practical cost control strategies — choosing between haiku (~$3.43) and sonnet (~$6.35) subagents based on task complexity.

Key Takeaways

By the end of this course, you'll understand how Agent Skills transform the way AI agents work:

• Skills are the bridge between raw LLM capabilities and domain expertise
• The open standard ensures your skills work everywhere — no vendor lock-in
• Composable skills enable complex workflows without complex code
• Skills work across the entire Anthropic ecosystem: Claude Code, Claude.ai, API, and SDK

Start Learning

Ready to extend your AI agents with skills? Enroll in this free course on DeepLearning.AI: