Agent Skills with Anthropic

Taught by Elie Schoppik, Anthropic's Head of Technical Education, this 2 hour 19 minute course is the guided tour of Agent Skills — the open folder-based format in which a SKILL.md file, plus optional scripts and reference documents, gives an agent on-demand expertise. The sequence is deliberate. Two 'Why Use Skills' lessons establish the problem: repeating the same instructions across conversations, and paying context tokens for capabilities you rarely invoke. A dedicated lesson then draws the boundaries between skills, tools, MCP servers and subagents, which is the distinction most teams get wrong first and the reason many end up building an MCP server where a Markdown file would do. From there it is hands-on: an 18-minute walkthrough of Anthropic's pre-built skills for Excel, PowerPoint and skill creation; a 16-minute lesson on authoring custom skills covering the YAML frontmatter, the name and description fields Claude matches incoming requests against, and how progressive disclosure keeps bundled reference files out of the context window until something reads them; then three substantial deployment lessons — 17 minutes on the Claude API, where skills run inside the code-execution container behind a beta header, 24 minutes on Claude Code's filesystem layout under personal and project skills directories, and 20 minutes on the Claude Agent SDK. The payoff is portability: because the format is an open standard, one authored skill runs unchanged across Claude.ai, Claude Code, the API and the SDK. It is labelled Beginner and lists no prerequisites, though the API and SDK lessons assume you can read Python.

Topic

Skills

Level

Beginner

Format

Course

Cost

Freemium

Duration

2h 19m, self-paced — 10 video lessons plus a graded quiz

Provider

DeepLearning.AI (with Anthropic)

Hands-on

Yes — code/exercises

Certificate

Available

• ✓Structure a skill folder with SKILL.md, YAML frontmatter, reference files and executable scripts
• ✓Write a description field that reliably triggers the skill on the right requests and no others
• ✓Apply progressive disclosure so bundled reference material costs nothing until it is read
• ✓Decide correctly between a skill, a tool definition, an MCP server and a subagent
• ✓Use Anthropic's pre-built Excel, PowerPoint and skill-creation skills as composable building blocks
• ✓Deploy the same skill across Claude.ai, Claude Code, the Claude API and the Claude Agent SDK
• ✓Build custom skills for code generation, code review, data analysis and research workflows
• ✓Combine skills with MCP servers and subagents into a multi-step agentic system

• •Taught by Anthropic's own Head of Technical Education, so the guidance matches how the feature was designed to be used
• •The skills-vs-tools-vs-MCP-vs-subagents lesson answers the single most common architecture question about this format
• •Over an hour of the runtime is deployment across four different surfaces, not conceptual repetition
• •Uses the open Agent Skills standard, so what you learn transfers to any skills-compatible agent, not only Claude
• •Free to watch in full; only the graded quiz and completion badge require a DeepLearning.AI PRO membership

This page was written from 2 sources, 1 on domains other than deeplearning.ai.

1. 1.deeplearning.ai — agent skills with anthropicvendor
2. 2.platform.claude.com — overview