Extending Claude Code

Module 4 · Intermediate

Skills, subagents, hooks, MCP, and output styles — the extension layer on top of the agentic loop.

8 lessons. Work through them in order — each one builds on the last.

Ways to extend Claude CodeThe extension layer at a glance: skills, commands, subagents, hooks, MCP and output styles. What each one is for, and when to reach for which.

SkillsTurning a prompt you retype into a named, invocable skill. Frontmatter that controls invocability, arguments, and injected command output.

Commands vs skillsWhere single-file commands and skill folders overlap, which wins on a name conflict, and why new workflows should usually be skills.

SubagentsDelegating work to a subagent with its own context window and restricted tools. Configuring descriptions, tool access and persistent memory.

Hooks, part one: events and configurationThe hook events Claude Code fires, and how to register commands against them so a guarantee runs every time rather than when Claude remembers.

Hooks, part two: prompt hooks and securityPrompt-based hooks and the security model around them: what a hook can block, what it can inject, and how to validate untrusted input.

MCP serversConnecting external systems through MCP servers, at project or personal scope. Managing the context cost of the tools they expose.

Output styles and the status lineAdjusting how Claude works with output styles, and surfacing session state in the status line.