Skip to content

Claude Code — Beginner to Pro

A sequenced curriculum: read the files in number order (00 → 57). Every file ends with a prev/next link, every stage ends with a milestone, and 45-capstone-projects.md has a hands-on project to prove each stage.

Works two ways:

  • Self-study — follow the stages, do the capstones
  • Teaching others — each stage below has 🎯 core ideas and a 🎬 live demo; see the Instructor Guide at the bottom

Every file ends with two built-in study aids:

  • 🧠 Quick Recall — a memory trick / mnemonic so the concept sticks (e.g. “Esc stops, Esc-Esc time-travels, Shift+Tab shifts trust”)
  • 👨‍🏫 Teach It — how to teach that topic: the one-liner to say, the demo to show, and the mistake learners will trip on

Sources: code.claude.com/docs · Anthropic courses (mapped in 44-anthropic-courses.md)


🎓 Course: Claude Code 101

#FileTopic
00OverviewWhat “agentic AI coding assistant” means, plain-language
01Installation & SetupInstall, login, config locations
02QuickstartYour first real session, step by step
03Core CLI UsageFlags, piping, one-shot mode
04Slash CommandsEvery built-in command
05Interactive Mode & ShortcutsKeyboard mastery
06How Claude Code WorksInternals: loop, tools, context

🎯 Core ideas: agent ≠ chatbot (it acts and verifies); the agentic loop (gather → act → verify); you approve every change by default; describe outcomes, not steps. 🎬 Live demo: in a real repo — “what does this project do?” → “fix this bug” → approve the diff → “commit it”. Nothing lands harder than a stranger’s bug fixed in 3 minutes. Milestone: install, navigate, prompt, explain the loop. → Capstone 1

#FileTopic
07Tools Deep DiveEvery tool with real examples
08Sessions & CheckpointingResume, name, fork, /rewind
09Context Window Management/context, /compact, caching
10Plan Mode & ThinkingPlan first, think harder, opusplan
11Real-World Workflows10 end-to-end workflows

🎯 Core ideas: the context window is finite working memory — manage it deliberately; /rewind makes mistakes cheap, so be ambitious; plan mode before big changes; /clear between tasks, /compact within them. 🎬 Live demo: let Claude make a mess on purpose, then double-Esc → restore code. The room relaxes visibly once they see undo exists. Milestone: you control how Claude works. → Capstone 2

Stage 3 — Configuration 🟡 (Intermediate)

Section titled “Stage 3 — Configuration 🟡 (Intermediate)”

🎓 Course: Claude Code in Action

#FileTopic
12Settings Systemsettings.json scopes & precedence
13Model Config & Env VarsAliases, effort, fast mode, env vars
14PermissionsRules reference
15Permissions — Real Examples5 real scenarios
16Sandboxing & SecurityOS-enforced boundaries, credential protection

🎯 Core ideas: four settings scopes (managed > local > project > user); permission rules decide what runs, the sandbox limits what running commands can touch — layers, not alternatives; match the model to the task (sonnet daily, opusplan for design, haiku for scripts). 🎬 Live demo: allow Bash(npm test:*) in project settings, re-run — the prompt disappears. Then /sandbox and show a network prompt fire on first domain use. Milestone: safe, prompt-free automation. → Capstone 3

#FileTopic
17MemoryCLAUDE.md & auto memory reference
18Memory — Real Examples3 real CLAUDE.md files
19Project StructureThe .claude directory layout

🎯 Core ideas: CLAUDE.md is standing instructions loaded every session — keep it under ~100 lines and it pays rent on every token; auto memory is Claude’s own notebook; mid-session CLAUDE.md edits apply only after /clear. 🎬 Live demo: ask for a commit → sloppy message. Add a commit convention to CLAUDE.md, /clear, ask again → perfect message. Cause and effect in 60 seconds.

Stage 5 — Extending: Skills, Hooks, MCP, Plugins 🟡

Section titled “Stage 5 — Extending: Skills, Hooks, MCP, Plugins 🟡”

🎓 Courses: Introduction to agent skills, Introduction to MCP

#FileTopic
20SkillsOverview
21Skills Step-by-StepBuild your first skill
22Skills — Real Examples10 complete skills
23HooksReference
24Hooks — Real Examples12 hook recipes
25MCP ServersReference
26MCP — Real ExamplesReal integrations
27PluginsPackage & share extensions

🎯 Core ideas: one mental model — skills = knowledge on demand (loaded when invoked), hooks = guaranteed automation (the harness runs them, not the model), MCP = new abilities (external tools/data), plugins = the shipping container for all three. 🎬 Live demo: build a /release-notes skill live from git log (5 min), then a PostToolUse format-on-edit hook — edit a file badly and watch it auto-format. Milestone: Claude Code customized to your team. → Capstone 4

Stage 6 — Agents & Parallelism 🟠 (Advanced)

Section titled “Stage 6 — Agents & Parallelism 🟠 (Advanced)”

🎓 Course: Introduction to subagents

#FileTopic
28Agents OverviewThe agent system
29SubagentsSpecialized workers
30Agent Project StructureDesigning the agent layer
31WorktreesParallel isolated sessions
32Agent TeamsCoordinated multi-session teams (experimental)

🎯 Core ideas: delegation is context protection — verbose work happens in the subagent’s window, only the summary returns; subagents report back, teammates talk to each other; parallel work needs file separation (worktrees or one-owner-per-area). 🎬 Live demo: run a test suite normally (watch context fill) vs. via a test-runner subagent (main context stays clean). /context before and after tells the story. Milestone: you delegate and parallelize. → Capstone 5

#FileTopic
33Headless & Scriptingclaude -p, piping, CI patterns
34Routines & SchedulingCloud automation, /schedule, /loop
35GitHub Actions@claude in CI
36Code Review/code-review, ultrareview, PR reviews

🎯 Core ideas: claude -p makes Claude a Unix tool — pipe anything in, get JSON out; scheduling ladder: /loop (in-session) → desktop tasks (your machine) → routines (cloud, machine off); layered review catches bugs where they’re cheapest. 🎬 Live demo: git diff main | claude -p "review for security issues" --output-format json | jq — one line, real review, parseable output. Milestone: Claude works while you don’t. → Capstone 6

Stage 8 — Surfaces: IDE, Desktop, Web, Chat 🟠

Section titled “Stage 8 — Surfaces: IDE, Desktop, Web, Chat 🟠”
#FileTopic
37VS Code ExtensionIDE integration
38Desktop, Web & Remote ControlTeleport, dispatch, work from anywhere
39Slack, Chrome & ChannelsChat-to-PR, browser automation, event push

🎯 Core ideas: one engine, many surfaces — CLAUDE.md/settings/MCP follow you everywhere; sessions move between devices (teleport in, remote-control out); meet work where it happens (bug report in Slack → PR back in Slack). 🎬 Live demo: start a task in the terminal → /remote-control → approve its next permission prompt from your phone, held up so the room can see.

🎓 Course: Building with the Claude API

#FileTopic
40Agent SDKBuild custom agents (Python/TS)
41Cost ManagementToken economics & optimization
42Large CodebasesMonorepos at scale
43Best Practices15 production practices

🎯 Core ideas: the pro bar is building agents for others, not just using one; cost is an engineering constraint you design for (caching, model mix, delegation); big codebases need scoped context (nested CLAUDE.md, targeted reads). 🎬 Live demo: 30-line Agent SDK script with one custom tool, run it, show the cost readout. Milestone: you build agents for others. → Capstone 7

Stage 10 — Practice & Certification 🏁

Section titled “Stage 10 — Practice & Certification 🏁”
#FileTopic
44Anthropic CoursesFull Skilljar catalog mapped to this path
45Capstone Projects7 real-case projects + scorecard

Stage 11 — Specialized & Reference 📚 (read as needed)

Section titled “Stage 11 — Specialized & Reference 📚 (read as needed)”

Complete coverage of every remaining topic — dip in when a feature becomes relevant.

#FileTopic
46JetBrains & Terminal SetupIntelliJ/PyCharm plugin, terminal tuning
47GitLab CI/CD@claude in GitLab pipelines
48Computer UseClaude controls your screen (macOS)
49Goal & Advisor/goal autonomy + stronger-model second opinions
50Agent View & WorkflowsBackground session manager + 100-agent orchestration
51Artifacts & Deep LinksPublish live pages; launch sessions from URLs
52Output Styles, Voice & UIPersonas (incl. Learning mode), dictation, status line
53Dev Containers & Sandbox EnvironmentsThe isolation ladder for autonomous runs
54Prompt Caching Deep DiveWhat invalidates the cache and why it matters
55TroubleshootingDiagnose install, session, MCP, and model issues
56Enterprise DeploymentProviders, managed settings, analytics, rollout
57GlossaryEvery term in plain language, cross-linked

🎯 Core ideas: the power-user trio is 49+50 (/goal + agent view + workflows = real autonomy); 54 (caching) quietly explains half of all “why is it slow/expensive” questions; 55 + 57 are the references you hand to anyone you onboard.


WeekFilesCourseDeliverable
100–06Claude Code 101Capstone 1 — tour-guide a strange repo
207–11Capstone 2 — plan, wreck, rewind, re-do
312–19Claude Code in ActionCapstone 3 — zero-prompt test loop
420–24Intro to agent skills2 skills + 1 hook shipped to a real repo
525–27Intro to MCPCapstone 4 — full onboarding kit + 1 MCP server
628–32Intro to subagentsCapstone 5 — review fleet, parallel worktrees
733–39MCP AdvancedCapstone 6 — hands-off pipeline (headless + routine + CI)
840–45 + skim 46–57Building with the Claude APICapstone 7 — ship a real Agent SDK agent

Stage 11 isn’t a week of its own — pull files in when they become relevant: 52 (Learning style) in week 1 if teaching, 49–50 in week 6–7 for autonomy, 53 before any unattended run, 54 the first time someone asks “why was that turn slow?”, 55/57 always open in a tab.


Instructor Guide — Teaching This Curriculum

Section titled “Instructor Guide — Teaching This Curriculum”

Your prep is already done: every file’s 🧠 Quick Recall block gives you the mnemonic to put on a slide, and its 👨‍🏫 Teach It block gives you the demo and the mistake to pre-empt. Before a session, read just those two blocks in each file you’re covering.

  1. Demo live, in a real repo — never slides. The product is the demo. Every stage above has a 🎬 suggestion; each takes under 5 minutes.
  2. Learners type from minute one. Watching an agent is boring; directing one is addictive. Pair up: one navigates (says what to ask), one drives (types). Swap.
  3. Capstones are the certification. A stage isn’t “taught” until the learner’s capstone runs. Use 45’s scorecard as the class tracker.
  • 60-min intro session = Stage 1: 00 talk-through (10 min) → live demo (10) → everyone installs + runs quickstart (25) → each person fixes one thing in their own repo (15)
  • Half-day workshop = Stages 1–3 + Capstone 1 as the closing exercise
  • Team onboarding (recommended) = the 8-week plan, one weekly 60-min session per row: 20 min recap + demo of the new stage, 40 min working on the deliverable together
  • Brown-bag single topic = any one file from Stage 11 (they’re self-contained)
  • Turn on the Learning output style for trainees (52): Claude explains decisions and leaves TODO(human) gaps — the tool teaches alongside you
  • Demo repo: pick something small with tests (a past internal project beats any toy repo)
  • Have /statusline show context % so learners see the window filling — makes Stage 2 concrete

Common beginner mistakes (address these proactively)

Section titled “Common beginner mistakes (address these proactively)”
MistakeFix to teach
Vague prompts (“fix the bug”)Outcome + location + proof: “fix the empty-form bug in ContactForm; tests must pass”
Pasting whole files into chatGive paths — Claude reads files itself, selectively
Never clearing context/clear between tasks is free; a 200-message session is not
Approving everything blindlyRead the diff for a week; then earn acceptEdits
Treating a wrong turn as failure/rewind or /branch — exploration is the workflow
One giant CLAUDE.md100-line budget; overflow goes to skills
Fear of breaking thingsShow checkpointing + git early; ambition follows safety

claude · /help · /clear · /model · Shift+Tab (plan mode) · Esc (interrupt) · Esc Esc (rewind) · /context · /resume · !cmd

Model the meta-skill: ask Claude Code itself, live — “how do I configure X?” It has a built-in docs agent. The lesson is that the tool answers questions about itself; 55 covers the rest.


  • “I just want to be productive this week”: 00–07, 11, 17–18, then capstones 1–2
  • “I’m setting this up for my team”: 12–19, 20–27, 36, 56, then capstone 4
  • “I want to automate everything”: 16, 29–36, 39, 49–50, 53, then capstone 6
  • “I’m building an AI product”: 40 + Building with the Claude API course + capstone 7
  • “Something’s broken”: 55 → 57 → claude doctor

Keep the capstone scorecard honest — a stage isn’t done until its capstone runs.

Written by Fenil Patel