Claude Code Hands-On: CLAUDE.md, Hooks, Skills, Subagents

What you'll learn

• ✓Run a Claude Code session the safe way: ask, act, review, verify
• ✓Choose permission modes and write allow and deny rules for a repository
• ✓Write a CLAUDE.md that actually changes behavior, and know what belongs in rules, memory or hooks instead
• ✓Build hooks that format code after every edit, block edits to secrets and add context to every prompt
• ✓Write skills with arguments, pre-approved tools and scripts, and fix a skill that never fires
• ✓Write subagents with restricted tools, run them in parallel, and know when not to delegate
• ✓Recover from a bad change, manage context and cost, and audit what your repository ships to the team

About this course

Claude Code is a coding agent that lives in your terminal. It reads your repository, edits files, runs commands and tests, and asks before it does anything you have not allowed. This course teaches you to drive it deliberately, in a small FastAPI project that you can follow along with, one feature at a time.

Section 1 is the working loop: ask before you order, review every edit and command, resume and compact sessions, choose a permission mode on purpose, read what the tool lines hide, write prompts with a definition of done, use plan mode, and pick a model for the difficulty of the task. Section 2 is CLAUDE.md: what belongs in it, project versus personal versus path-scoped rules, how auto memory differs, and the lines that change nothing or do harm. Section 3 adds hooks, the only mechanism that is enforced rather than suggested: a formatter after every write, a guard that blocks edits to secrets, the session events, and a hook that gives the agent context it cannot read. Section 4 is skills: arguments, pre-approved tools, auto-invocation, forked context, why a skill is never selected, and skills that run scripts. Section 5 is subagents: the built-in Explore and Plan agents, a code reviewer you write yourself, parallel and inline agents, when not to use one, and /subtask. Section 6 is the capstone, a real SQLite migration reviewed and committed, followed by recovery, context and cost, and background work. Section 7 is working for real: several repositories, tests as the specification, a large refactor in reviewable steps, shell mode, auditing what a teammate actually gets from your repository, and a full day at speed.

Every terminal lecture is a real session, recorded as it happened, including the moments where Claude asks for permission. Each lecture starts from a tagged snapshot of the demo project, so you can watch any lecture on its own.

**What this course is not.** It does not cover the desktop app, the IDE extensions or the web UI, and it does not teach Python or FastAPI. The follow-up course, *Claude Code in Production*, covers MCP servers, headless mode, CI and the Agent SDK.

Curriculum9 sections · 40 lectures · 1h 58m

Read alongside the course