Course Outline | Claude Code Training for Marketers

The intro video and the map of the training - what you need, what the first week looks like, where you type. Watch this first.

0.1
Start hereClaude Code is an agent that runs on your computer and does the work. Anything you can do on your computer, it can do too - directed in plain English.

Install Claude Code, say your first job out loud and finish it - all in the first hour, nothing typed but the install.

1.1
Where you typeThe plain text window that's already on your computer - what it is and how to open it.
1.2
Install Claude Code on MacThe only thing you'll ever type to install. Paste 1 command, sign in, done.
1.3
Install Claude Code on WindowsThe same 1 command, in PowerShell. Nothing else by hand.
1.4
Talk to ClaudeTalk to it like a colleague, out loud, not typed - the habit that decides how much you get out of it.
1.5
Set up your status line30 seconds of setup that shows which model you're on and how full the context window is, every session.
1.6
Set up your Claude Code folderOne folder, inside Dropbox or Google Drive so it backs up as you go. Get this wrong and you'll regret it 2 months in.
1.7
Approve each step firstThe 3-option prompt Claude shows before it acts - and why manual approval wins in the first weeks.
1.8
Your first real jobPick the admin job that's been sitting on your list for weeks. Don't build anything - just finish it.
1.9
Files in, files outDrag a file in, paste a screenshot, or type @ - everything Claude does runs on files.
1.10
Plan first, then runFor anything bigger than a one-liner, put Claude in plan mode and let it ask you questions first.
1.11
When it goes wrongSomething breaks in week 1. The fix is nearly always 1 of 4 moves, starting with a screenshot.

Connect Claude to the platforms you work in every day and ask them the questions you'd ask a team member.

2.1
What an MCP isThe bridge between Claude Code and Google Ads, Meta, Klaviyo, GA4, Shopify and dozens more.
2.2
Install your first MCPOne account links Claude to your paid media platforms in 2 web pages and a few clicks. Nothing typed.
2.3
MCP, CLI or API keyThe 3 ways Claude reaches a platform - and how it picks for you most of the time.
2.4
Your .env fileEvery key you'll ever give Claude lives in 1 plain text file. Claude reads it; you never paste a key into the prompt.
2.5
Connect every account, then never stop askingConnect every account you run, including the channels that never worked - then ask the platforms questions daily.
2.6
MCPs per projectEach project gets its own connectors - fewer tokens, no cross-over between clients.
2.7
Read vs writeRead is safe. Write is real. Keep every connector read-only until you've earned it.
2.8
The MCPs we useThe full connector stack we run at Webprofits, grouped into 3 tiers by setup effort.
2.9
The one-prompt unlock18 months of campaigns turned into a performance-based style guide - the highest-leverage prompt once a platform is connected.

Claude forgets in 2 different ways. This module fixes both - a briefing file, then a database that becomes its long-term memory.

3.1
Why Claude forgetsBetween sessions it starts blank; within a session its working memory fills up. Both have a fix.
3.2
What CLAUDE.md isThe briefing document Claude reads at the start of every session - 1 global, 1 per project.
3.3
Write your global CLAUDE.mdThe 5 things Claude should know about you in every session. Claude drafts it by asking you questions.
3.4
Write a project CLAUDE.mdThe client, the channels, the account IDs, the brand, what matters this quarter - per project folder.
3.5
Local memory firstA plain text handoff note gets you through week 1, before any database.
3.6
Supabase - the big unlockA free database that becomes Claude's long-term memory - every session, decision and rule, across every project.
3.7
Install /startup and /wrapupThe 2 commands that load and save your memory every session. 1 zip, 1 prompt, done.
3.8
Your first full loopEvery working session from now on has the same shape - and this is the first run of it end to end.
3.9
Keep it safe and keep it cleanBackups on, the memory audited for duplicates and stale rules.
3.10
Upgrade to the full memory systemReplaces the simple loop: the system we run at Webprofits tracks every workstream, not 1 summary a session.

The terminal, files, sessions, the context window, tokens and permissions - explained after you've used them for a week.

4.1
What the terminal isA 50-year-old text window that talks directly to your computer - and why Claude Code lives in it.
4.2
Terminal appsTerminal, PowerShell, VS Code, Cursor and the paid apps all run the same Claude Code. The question every live session asks, answered.
4.3
Files and folders are the workspaceClaude stands in the folder you started it in - that's how you control what it can touch.
4.4
SessionsOpen with a goal, work in short instructions, close by taking stock. Several tabs means several sessions.
4.5
The context windowClaude's working memory for 1 session, and the single most important mechanic in Claude Code.
4.6
Tokens and modelsRoughly 3 words for every 4 tokens, resent with every message - and the 4 things that load before you type.
4.7
PermissionsManual, accept edits, auto - what each mode asks about and when each is safe.
4.8
The 3 layers of memoryThis session, CLAUDE.md, Supabase - what lives where and for how long.
4.9
Change how Claude Code worksThe settings screen most members never open - how Claude replies, which model it starts on, how much it asks.

The habits that turn Claude Code from a tool you try into the way you work.

5.1
The session loopThe same 6 beats every session - and the 2 things that happen between them.
5.2
Most of the work is feedbackYou don't type and you don't write. You question and you correct.
5.3
Push backClaude's first answer often isn't its best, and 'I can't do that' is usually wrong.
5.4
Context before skillsBuild the context first, then the style guide from it, then execute. Skills come last.
5.5
Not everything needs to be builtFinish the task, or build something that does it every time - choosing is the skill.
5.6
Shipping is all that mattersTime in Claude Code means nothing until something goes live. Ship 1 external thing a day.
5.7
What to build firstStart with the workflow closest to the money and fix it end to end.
5.8
Work across sessionsEvery close leaves specific pending items; every open reads the arc, not just the last session.

A skill is a repeatable workflow packaged into 1 command. Build them by describing them, then make them reliable.

6.1
What a skill isA 20-step process packaged into 1 slash command - where your methodology gets encoded.
6.2
Using existing skillsFinding, installing, customising and disabling skills from the ecosystem - and which are worth installing.
6.3
Building your first skillDescribe it. Claude interviews you, writes it and saves it. Nothing to install.
6.4
Keep your skills leanEvery installed skill loads before you type a word. Skills are not free.
6.5
Thinking + process skills7 ready-to-install skills that change how Claude reasons, not just what it produces.
6.6
Make skills deterministic5 simple controls so a skill gives the same quality every run, not a different shape each time.
6.7
Templates for repeatable deliverablesGive the skill a template to fill instead of letting it design the layout from scratch.
6.8
Test skills with evalsAn automatic test that runs the skill and tells you pass or fail.
6.9
Keep skill token costs lowA well-built skill costs a fraction of a sloppy one - reuse, read only what's needed, check instead of regenerate.
6.10
Simplify skills as models improveStrip the instructions that were there to compensate for last year's model.

Claude running on a schedule without you at the keyboard - when to use one, how to build it, how to know it ran.

7.1
What a Routine isSame skills, same connectors, same CLAUDE.md - except nobody's watching.
7.2
When to use a Routine4 qualification questions so you only automate jobs worth automating.
7.3
Build your first RoutineName it, schedule it, write the prompt, verify the output - every step.
7.4
Context + secretsA remote Routine starts with nothing. Where it gets its context and its keys.
7.5
Monitor + debug runsRoutines fail silently unless you build in run history, output checks and failure notifications.
7.6
Routines + Skills + MCPsThey compose, not compete - the decision framework for which to reach for.
7.7
Build Routines that email you - AgentMail.toWire a Routine to its own inbox so the result of every run lands in your email.

Turn a report or a page into a link you can send, with version history, from 1 prompt at the end of a session.

8.1
GitHub vs DropboxThe question every intake asks - and why you need both.
8.2
What GitHub isCloud storage for files with version history and a free publishing system.
8.3
Connect GitHub1 prompt, 1 browser click. You won't type a GitHub command.
8.4
Project repos + Pages repoEach project folder becomes its own private repo when it needs one; 1 public repo publishes.
8.5
Commits, pushes, deploysLocal file to live URL. 1 prompt at the end of every session.
8.6
Encryption for sensitive pagesPages are public by default - password protection that genuinely scrambles the page.
8.7
Keeping repos cleanThe 10-minute monthly clean-up: archive, remove, review.

When a task needs a plan, how to write one, and how to organise your folders so nothing crosses over.

9.1
Plan vs promptThe 5-question framework for knowing which a task needs.
9.2
Brainstorming with ClaudeGet clear on goal, constraints and unknowns before you write a plan.
9.3
Writing a planPhases, steps, dependencies, time and risks - a plan that's a tool, not a document.
9.4
Context files + @file syntaxBrief, plan, spec and notes as separate files, referenced surgically so Claude loads only what it needs.
9.5
Executing in chunks1 phase at a time across sessions, with a clean open and close each time.
9.6
The 4 structuresAgency, brand, solo, channel-first - and 3 questions to pick the right one.
9.7
Agency structureEach client in its own folder with its own memory, connectors and credentials.
9.8
Brand structure1 brand across many channels - organise by discipline, not by client.
9.9
When to reorganiseThe signs a tidy is overdue, and how to do it without breaking everything.

What Claude Code actually costs to run all day, and the levers that cut it without a quality hit.

10.1
Trimming overheadWhat loads before you type is the most leveraged number to cut - every token saved comes off every message.
10.2
The 50% ruleWhat to do when the context window hits half, and why compacting isn't the answer.
10.3
3 levers that cut tokensThe model you run, how much Claude says back, how hard it thinks.
10.4
See where your tokens goClaude Code tracks your own usage - the 3 built-in views that show what's eating your plan.
10.5
What a day costsThe plan covers the day to day; what a failed experimental build day actually runs to.
10.6
Build on the best, run on the workhorseThink on the most capable model, execute on the cheaper one pinned inside the skill.

The rules that stop most of the damage, then the 4 layers that keep keys and client data out of the conversation.

11.1
The never-go-live rule1 line in your global CLAUDE.md: nothing publishes, launches, pauses, sends or deletes without your say-so.
11.2
What never goes in the promptNever type or paste a key, a password or customer data into the prompt - the general pattern.
11.3
Prompt injectionMalicious content in a file or webpage trying to hijack Claude mid-session - rare, but know what it looks like.
11.4
Security - the mapWhere the first 3 lessons leave you, and what the rest of the module adds.
11.5
Assume it could leakAn API key is a password for a program. How long chats are kept, and what that means for what you paste.
11.6
The 4 protection layersA deny list, an automatic guard for secret files, scripts that hold the key, keys in a password manager.
11.7
Make a leak survivableMake each key worth as little as possible if it ever gets out - decided ahead of time.
11.8
Hidden instructions + risky add-onsInstructions hidden in something Claude reads, or in a tool it uses - how to spot both.
11.9
Run a security auditThe skill that checks all 5 layers for you, on any project.

Claude builds fast, invents numbers and can't judge fit for a client. You are the quality layer.

12.1
You are the quality layerWhat Claude can't verify about its own work, and what that leaves for you.
12.2
The data-integrity skillEvery number in a deliverable traces to a real source or it doesn't go in.
12.3
QA checklists per deliverableReport, audit, landing page, email, proposal, ad - each has its own failure modes.
12.4
Using Claude to QA itselfThe 5 checks Claude runs on its own output when asked the right way.
12.5
Never-let-Claude-do-X10 categories of action too risky to execute without a written plan and human sign-off.
12.6
Look at the platformAfter anything goes live, open the platform and look. 'I did it through the terminal' isn't an excuse.

A team runs on shared context, not shared folders - memory through Supabase, work through GitHub, standards in 1 file.

13.1
The team share modelEach person their own seat, folders and memory; shared databases for the clients.
13.2
Share through SupabaseThe database is the share layer - a company-wide one for clients, a personal one each.
13.3
Share through GitHubA team organisation, not personal accounts, so the work stays with the company.
13.4
Shared rules + standardsThe team CLAUDE.md - spelling, brand, formatting and safety set once so every output matches.
13.5
Share your skills1 shared skills repo: clone to set up, pull to stay current, push to share.
13.6
Onboard your teamYou first, then 1 champion, then the one they recruit - on 1 real problem, almost always reporting.
13.7
Manage cost + accessKeep the spend cap on. Zero usage is a signal, not a saving.

The references you come back to, the fixes for when connections fail, and every skill from the training in 1 download.

14.1
When connections failAuth expiry, outages, rate limits, wrong account, missing tools - the predictable patterns and their fixes.
14.2
GlossaryEvery term used across the training, defined once.
14.3
Cheat sheetEvery command, prompt and habit from the training on 1 page.
14.4
Starter skill packAll 12 skills from the training bundled in 1 download.
14.5
The Webprofits stackThe full inventory of what we run day to day, on top of the connectors.
14.6
Claude Code in a browserWhen IT won't let you install it, a full computer in the cloud through a browser tab.

Claude Code Training
for Marketers

The training, module by module.

Quick start

Foundations

Build

Scale

Run your marketing
through Claude Code.

Claude Code Training for Marketers