Lessons
Twelve levels. Each lesson names the Claude Code version it was checked against, and every claim about how the tool behaves links to the official documentation.
Level 1
- 1.1What agentic coding actually changes (and what it doesn't)9 min
- 1.2Install, authenticate, and prove it works12 min
- 1.3Anatomy of a session: prompt, tool call, permission prompt, status line11 min
- 1.4Your first real task: ask a question about code you did not write13 min
- simFirst Contact20 turns · 200,000 tokens
- quizAssessment8 questions · pass 80%
Level 2
- 2.1The four kinds of context, and which one you left out12 min
- 2.2Rich input — references, images, URLs and pipes13 min
- 2.3Point at the pattern, not just the outcome12 min
- 2.4Describing a bug so it can be found11 min
- 2.5When you are the missing context, get interviewed12 min
- quizAssessment10 questions · pass 80%
Level 3
- 3.1What a 200k window is actually spent on14 min
- 3.2Reading the budget and spending it — /context, /compact, /clear, /btw16 min
- 3.3One session, one task13 min
- 3.4Course-correcting early — Esc, rewind, and "undo that"13 min
- 3.5Resuming — --continue, --resume, and naming things14 min
- simThe Long Afternoon18 turns · 200,000 tokens
- quizAssessment12 questions · pass 85%
Level 4
- 4.1The loop, and why exploration is not the optional part14 min
- 4.2Plan mode: the four ways in, and what it actually blocks16 min
- 4.3Reading a plan critically — and editing it before you approve16 min
- 4.4When not to plan: the one-sentence-diff test13 min
- 4.5Commit hygiene under an agent: described, reviewed, revertable16 min
- simTwo Roads14 turns · 200,000 tokens
- quizAssessment10 questions · pass 80%
Level 5
- 5.1Why a verifiable loop beats a better prompt14 min
- 5.2The escalation ladder — prompt, /goal, Stop hook, second opinion20 min
- 5.3Verifying work that has no test — screenshots, runs, logs16 min
- 5.4Handing over the test first — the failing test on purpose15 min
- 5.5The trust-then-verify gap, and the green that isn't15 min
- simGreen on Red16 turns · 200,000 tokens
- quizAssessment10 questions · pass 80%
Level 6
- 6.1Where instructions come from — scope, load order, and @ imports15 min
- 6.2What belongs in CLAUDE.md — and the far longer list of what does not15 min
- 6.3.claude/rules/ and paths: — instructions that load when they are relevant15 min
- 6.4Auto memory — what Claude writes about you, and where it lives14 min
- 6.5The four commands: /init, /memory, /import, and trimming with /doctor12 min
- 6.6The over-specified CLAUDE.md: how a long file makes results worse14 min
- quizAssessment12 questions · pass 80%
Level 7
- 7.1Skills, and what happened to custom commands14 min
- 7.2Frontmatter that matters: description, invocation, tools16 min
- 7.3Arguments: $ARGUMENTS, positions, names and ${CLAUDE_*}15 min
- 7.4Dynamic injection with !`cmd` — and when to refuse it16 min
- 7.5Progressive disclosure: what stays in the window15 min
- 7.6Resolution and precedence: which /deploy actually runs14 min
- quizAssessment10 questions · pass 80%
Level 8
- 8.1The one thing subagents are for14 min
- 8.2What a subagent inherits, and what it does not16 min
- 8.3Fork or subagent: full context or clean slate13 min
- 8.4Authoring an agent file18 min
- 8.5Writer, reviewer, and the findings that are not real15 min
- 8.6Parallel sessions without collisions16 min
- simThree at Once16 turns · 200,000 tokens
- quizAssessment10 questions · pass 80%
Level 9
- 9.1The hook lifecycle — picking the event that actually fires13 min
- 9.2Exit codes and JSON — what actually blocks14 min
- 9.3Handler types and timeouts — command, http, mcp_tool, prompt, agent13 min
- 9.4Hook footguns — the gate that is open and the gate you cannot get past14 min
- 9.5MCP — transports, scopes and OAuth15 min
- 9.6Prompt injection through MCP — the tool result is not your friend17 min
- 9.7Non-interactive automation — claude -p, stream-json and CI14 min
- quizAssessment14 questions · pass 85%
Level 10
- 10.1The Six Modes and What Each One Actually Permits12 min
- 10.2Rule Syntax, Precedence, and Why Deny Wins14 min
- 10.3Auto mode: what the classifier reviews and where it hands back13 min
- 10.4The Bypass Reflex13 min
- 10.5Secrets, .env, and what leaves your machine12 min
- 10.6Untrusted content is data, never instructions13 min
- quizAssessment12 questions · pass 85%
Level 11
- 11.1Scoping a repo too big for the window20 min
- 11.2Code intelligence instead of brute-force reading12 min
- 11.3Asking a codebase questions as a workflow14 min
- 11.4Git under an agent: what undo actually covers18 min
- 11.5Three review workflows, and what each one is worth20 min
- 11.6Refactoring safely: pin the behaviour first16 min
- simThe Monorepo16 turns · 200,000 tokens
- quizAssessment12 questions · pass 80%
Level 12
- 12.1Shared conventions: what goes in the repo, what stays personal18 min
- 12.2Distributing skills, agents and hooks: plugins and managed settings20 min
- 12.3Rollout that works: one workflow, one team, one measurable outcome16 min
- 12.4Measuring honestly: accept rate, PR attribution windows, and the metrics that reward the wrong behaviour20 min
- 12.5Security review for a team rollout: managed policy, allowManagedHooksOnly, allowlisted hook URLs20 min
- 12.6Keeping current when the tool ships weekly14 min