Claude Code, shaped to your project
The Intermediate course. Turn the instructions you keep repeating into skills, split project memory so it loads when it applies, pick the model and effort on purpose, move the guardrails into the repository with settings and the sandbox, enforce rules with hooks, delegate to subagents, connect tools over MCP, and package the lot as plugins. Eight lessons and a capstone. About 41 minutes of reading, and every exercise lands in your own repo — shaping it is the point of the level.
The level at a glance
Beginner made one change safe, in one session, by one developer — and left the project’s rules where CLAUDE.md could only suggest them. This level moves them into the repository: skills and memory that commit with the code, settings scopes that decide who a rule affects, a sandbox the operating system enforces, hooks that hold in every permission mode. The one-developer setup, in Wesam’s phrase — and the reason we teach it: suggestion does not scale to Monday.
You are ready if you can do what the Beginner exit asks: install and start a session, choose permission modes and approve a plan, take one change from request to commit, keep a session on track, and leave the project a CLAUDE.md a fresh session follows.
One platform note: from this level on, the lessons are written for macOS, Linux and WSL 2 — the sandbox needs one of them, and the hook scripts are bash. The exercises assume a repository you can experiment in; the capstone finishes the one you have been shaping.
The lessons
In order, each one building on the last; the capstone uses all eight.
Turn a repeated workflow into a skill
Every session, you type the same instructions again. Lesson 1 turns one of them into a file — and the repetition stops.
Organize project memory, and see what loaded
A two-hundred-line CLAUDE.md is a tax every session pays in full. Imports organize it; scoped rules make most of it conditional — and there is a way to see exactly what loaded.
Pick the model and the effort
Two knobs, often turned in the wrong order. The model decides how capable the answer can be; the effort decides how much work goes into it. This lesson is mostly about not touching either one.
Permissions, settings scopes, and the sandbox
Beginner kept the guardrails personal — you, answering prompts. This lesson moves them into the repository: rules everyone gets, and a sandbox the operating system enforces whether Claude asks or not.
Enforce a rule with a hook
CLAUDE.md is advice a model can miss. A hook is your own command that Claude Code runs at a fixed point in every session — and one exit code turns it from observation into enforcement.
Delegate to a custom subagent
Some work should not happen in your conversation at all. A subagent is a second Claude with its own context window, its own instructions, and only the tools you list — it reports back a summary.
Connect a tool with MCP
MCP lets Claude use tools that belong to another program — a browser, a database, an API. One server, added twice: once for yourself, once for the whole project, committed and pinned.
Install and manage plugins
Skills, hooks, subagents, MCP servers — this whole level, packaged so it installs as one unit. Plugins are the level’s parts bin, and the last lesson before the capstone puts it together.
No walkthrough this time. A teammate starts Monday and gets only what is in git — the brief says what the team needs; you choose how to build every part of it.
The Intermediate exit
The capstone checks what git can see; you tick the rest, honestly. Every line below is the entry test for installment 3 — and the reason a teammate’s sessions work like yours from their first clone.
Sources and attribution
This installment is our adaptation of Wesam Abousaid’s open curriculum Claude Code: Everything You Need to Know (v2026.10), used under its MIT license. The lesson spine, the exercise pattern and the capstone design are his; the phrasing, the opinions and the field notes from daily agent delivery are ours. Facts are stated as the official Claude Code documentation states them at the verified version above.
We are not affiliated with or endorsed by Anthropic. If you want the automated per-lesson checks and the purpose-built practice repo, use the upstream template.