Claude Code Deep Dive: Zero to Hero

What This Course Is

This is not a getting-started tutorial. If you have already used Claude Code to scaffold a project, fix a bug, or generate some boilerplate, you are past the starting line. This course takes you behind the curtain — into the architecture, protocols, and design decisions that make Claude Code work the way it does.

You will learn the six-step agentic loop that drives every Claude Code session, how context is assembled from a four-level memory hierarchy, how tools are dispatched and permissions are enforced, how hooks let you inject custom logic at lifecycle boundaries, and how MCP servers, multi-agent orchestration, and the SDK extend Claude Code into a programmable platform. Each concept is grounded in real internals — not abstractions.

The target audience is experienced developers who have used Claude Code casually but want to understand why it behaves the way it does, how to control that behavior precisely, and what becomes possible once you understand the machinery underneath.

What You Will Build

Across all ten modules and the capstone, you will build a Personal Learning Path Manager — a full-stack application for curating, tagging, and organizing learning assets (articles, videos, repos, courses) into structured learning paths.

The tech stack:

  • Python 3.13 with uv for dependency management

  • FastAPI for the REST API

  • Pydantic V2 for data validation and structured outputs

  • PostgreSQL for persistence

  • React for the frontend

Each module adds a meaningful layer to this project. Module 1 scaffolds the project. Module 3 builds the database schema and CRUD API. Module 8 adds an MCP server for RSS feed ingestion. Module 9 builds the React frontend with parallel agents. By the time you reach the capstone, you have a fully functional application — and a deep architectural understanding of the tool that helped you build it.

How Each Module Works

Every module follows the same three-part rhythm:

How It Works

An architecture deep dive into the relevant Claude Code subsystem. Expect mermaid diagrams, internal data flows, and design rationale. You will understand the mechanism before you use it.

See It In Action

Introspection exercises that let you observe the concept in a live Claude Code session. You will inspect ~/.claude/ files, analyze transcripts from the claude-code-transcripts repository, and run targeted experiments to confirm your mental model.

Build With It

Apply the concept to the Personal Learning Path Manager. Each module’s build section advances the project by one meaningful increment, so the architecture knowledge and the working application grow in lockstep.

Module Map

Module Title Architecture Concept Build Milestone

1

The Agentic Loop

Six-step cycle: input, context, reasoning, permissions, tools, loop

Scaffold project with uv + FastAPI

2

Context & Memory

4-level CLAUDE.md hierarchy, @includes, granular rules

Configure project memory and settings

3

Tools & Tool Dispatch

Built-in tools, checkPermissions, tool selection

Postgres schema, Pydantic models, CRUD API

4

Permissions & Trust

6 permission modes, rules, bash patterns

Project permission rules

5

Hooks & Lifecycle

6 events, 4 hook types, exit codes

Auto-format, test, lint hooks

6

Skills & Workflows

SKILL.md, namespacing, arguments, paths

Custom project skills

7

Extending Claude Code

Superpowers, marketplaces, ecosystem

Asset tagging via guided workflow

8

MCP Servers

Protocol, transports, config, safety

Custom FastMCP server for RSS ingestion

9

Multi-Agent Patterns

Sub-agents, isolation, worktrees

React frontend via parallel agents

10

SDK & Integration

Subprocess model, messages, sessions

CI automation script

C

Capstone

Full architecture review

Complete learning asset manager

Getting Started

Before starting, verify your setup on the Prerequisites page.

Source Material

This course draws on the following references: