Build day · pick one · ship it

Build day. Pick one thing and ship it.

A whole day of building, end to end. Set up your tools once, then choose one playbook that matches what you actually want - a brand & social kit, an SEO website, or a SaaS, portal or web app - and go deep. You're not meant to do them all. One thing, finished and live, beats a stack of half-done ones.

Every prompt here is a starting point, not a magic word. Read what the agent does, then steer it in your own words - that is the skill. Facilitators are walking the room; if you're stuck for more than five minutes, put your hand up.

Build day

Set up, then pick your build

Everyone starts with Step 0 · Setup. Then choose one playbook - click its tab and it opens on its own. You're only meant to do one, so pick what you'd actually use on Monday.

Step 0

Set up once - everyone

The whole build day runs on the same two tools: VS Code to see the files you're making, and Claude Code to build them. Ten minutes now, then you never touch setup again.

Before you build · ~10 min · everyone

Install VS Code, Node.js & Claude Code

VS Code is your window onto the files; Node.js is the engine most builds run on; Claude Code is the agent you drive from the terminal. Get all three once.

Setup~10 minMac / Windows

Get your tools installed

A code editor so you can see your work, Node.js so it can run, and the Claude Code CLI so you can build by talking.

  1. Install VS Code from code.visualstudio.com - this is where you'll see every file Claude creates.
  2. Install Node.js 18 or newer (LTS) from nodejs.org.
  3. Install the Claude Code CLI in a terminal:
Install the Claude Code CLI
npm install -g @anthropic-ai/claude-code

Then open a project folder in the terminal, launch Claude Code, and log in the first time it asks:

Start Claude Code
claude

The first run opens a browser link to sign in with your Claude account. You can also type /login inside Claude Code. New to VS Code? File → Open Folder… points it at any project folder, and Terminal → New Terminal opens a terminal right inside it.

Troubleshooting
  • command not found: claude - close and reopen the terminal so it picks up the new command.
  • Prefer a one-line installer? On macOS/Linux: curl -fsSL https://claude.ai/install.sh | bash
  • There's a step-by-step PDF in your workshop folder: Claude_Code_Setup_Guide_Recreated.pdf.
  • Only doing the Claude Design playbook? You still want VS Code so you can open and tweak the design files - but that one runs mostly in the browser, so don't let setup block you.
Playbook 1 · easiest

Claude Design - Social & Brand

Import a ready-made design system, drop in your brand, and walk away with on-brand social posts and a mini brand kit. Mostly in the browser.

  • Import the AI Champions Club design system
  • Apply your logo, colours & fonts
  • Generate a batch of on-brand social posts
Start the design build →
Playbook 2 · builder

SEO Website

Start from a complete, production-style website template, run its built-in kickoff skill to make it yours, then deploy it free so every git push republishes it.

  • Ready-made React static-site template
  • Kickoff skill interviews you or imports your current site
  • Deploy free on GitHub + Cloudflare Pages
Start the website build →
Playbook 3 · deepest

Build a SaaS / Portal / Web App

Build a working full-stack app (TaskFlow) one small step at a time - database and all - then push it to GitHub and deploy it live.

  • Next.js · TypeScript · Prisma · SQLite
  • The engineer's loop: Ask → Review → Run → Check
  • Ship it on GitHub + Railway
Start the app build →
Playbook 4 · your own agent

AI Agent/Automation

Define your own AI agent in plain English - its role, the tools it can touch, the knowledge it reads, and hard guardrails - and a live builder writes the prompt for you. Then build it, dry-run it, and set it loose safely.

  • Describe role, tools, knowledge & guardrails
  • A live builder writes your agent prompt
  • Build with agent-creator, dry-run & review
Build your own agent →
Playbook 5 · marketing bench

Agentic Marketing

Set up a reusable AI Marketing Bench for your company: brand foundations, editable creative, Google and Meta ad campaigns, and matching organic social with Buffer scheduling.

  • Interview your business and configure the bench
  • Design a paid Google or Meta ad campaign
  • Add matching organic posts, then hand off safely
Build your marketing bench →
Playbook 1

Claude Design - Social Media & Brand

Most useful if you produce brand or social content. Import a complete example design system into Claude Design, iterate it into your brand, then generate on-brand campaigns from it - and hand those straight to the marketing engine.

Playbook 1 · ~45-60 min · outcome: your own design system + a branded campaign

Your brand, as a reusable design system

By the end you'll have a design system wearing your brand - live in Claude Design - plus a ready-to-post social campaign generated from it, ready to hand to the marketing engine.

1.110 minClaude Desktop · Link Code

Create the design system in the desktop app

  1. Open the Claude Desktop app.
  2. Go to Design in the bottom-left.
  3. Across the three tabs, go to Design SystemsCreate Design SystemCreate Here.
  4. Type in your company name - e.g. Mike's Mowing.
  5. Link all your company brand assets, logos, colors, fonts, pdf files anything you can find.
  6. Click Continue to Generation.
Checkpoint: Claude imports the your designs and generates your new design system - tokens, components and templates - under your company name.
1.215 minMake it yours

Iterate it into your brand

Import first, then iterate - turn the example brand into yours.

Open the Design Systems tab, double-click your system to open it, then iterate with Claude until it's on your brand.

Make it your brand
Apply my brand to this design system - use my logo, colours and fonts, and keep the structure and templates. Here's my brand: [attach your assets if you haven't already / name your colours / name your fonts]. Update every asset so it's on-brand. If anything's missing, ask me before guessing.

No brand files to hand? Say "pull my colours, fonts and logo from [your-website-url] and use those."

Checkpoint: the imported system, now unmistakably yours - the moment you see your logo on the templates, you're there.
1.310 minAdd the templates

Add the supplied social posts to your brand

Bring every social-post template from the provided brand design into your own design system, restyled for your brand.

ai-champions-club.zip · download the supplied design
  1. Download ai-champions-club.zip using the download link above.
  2. Unzip it and note where the unzipped AI Champions Club design folder is saved.
  3. Open your branded Design System in Claude Design.
  4. Click the + button.
  5. Select Link local code.
  6. Choose and link the unzipped AI Champions Club design folder.
  7. Use this prompt to add the complete set to your design system:
Add all supplied social posts to your brand design
Add all of the social posts from the linked AI Champions Club brand design to a new page in my brand design system. Include every supplied social-post format and layout, but restyle each one using my logo, colours, fonts, imagery and brand voice. Keep the original dimensions and make every template editable and reusable. Do not leave any of the supplied social-post templates out.
Checkpoint: your Design System now has a page containing all of the supplied social posts, rebuilt in your brand.
1.415 minShip it

Generate a campaign & hand it off

With your design system selected, generate an on-brand social campaign - then feed it to the marketing engine.

  1. Come out of the design system and go back to the Claude Design home.
  2. Select your design system from the drop-down.
  3. Now ask it to create a campaign - a social media campaign is a good first one:
Create a social media campaign from your design system
Can we please create a new marketing campaign for the socials for my business? The target market is [X]. The campaign I want to run is [Y]. Use my design system so everything is on-brand.
➡️ Hand it off: once you've created a campaign or any asset, download it and drop it into the Agentic Marketing engine (Playbook 5) to turn it into a full multi-channel push.
Checkpoint: your own design system live in Claude Design, plus a downloaded, on-brand campaign ready for the marketing engine.
💡 Pro tip: springboards.ai is handy for brand ideas only - a slogan, a campaign angle, a tagline. It's not a replacement for Claude Design; use it to spark an idea, then build the actual assets here.
Playbook 2

SEO Website

Start from a working, production-style website codebase - routing, static generation, a markdown blog, SEO, forms, redirects and quality checks already built - instead of asking an AI to recreate all of that every time. Run its built-in kickoff skill to make it yours, then deploy it free so every git push redeploys.

Playbook 2 · 60-120 min · outcome: a live, SEO-ready website

Business-grade website with strong SEO

The download is already a runnable website. Its built-in $website-kickoff skill interviews you, can inventory your current public site, and replaces the starter business - then you build, verify, and deploy it free on Cloudflare Pages.

⚠️ Good to know: this is a static site - no server-side code. It's perfect for marketing pages, services, locations and a blog, but it can't run dynamic forms, logins, payments or a database on its own. Contact forms need a form service (or a developer to wire up a backend) before they'll actually send.
ai-website-kickoff.zip · complete website template
2.15 minDownload & open

Open your ready-made project

Get a private working copy onto your computer, then open it in VS Code or Claude Desktop.

Option A · VS Code and the terminal

  1. Download ai-website-kickoff.zip using the chip above.
  2. Unzip it somewhere you own, such as Documents/Projects. Do not work inside the ZIP.
  3. Open the unzipped template-project folder in VS Code: File → Open Folder…
  4. Open Terminal → New Terminal. Check that the prompt is inside template-project.
  5. Type claude and press Enter to start Claude Code in that folder.
Option B · Claude Desktop
Open the Claude Desktop app, choose the Code tab, then open or select the unzipped template-project folder. Claude will work directly inside that folder, so you do not need to launch it from the VS Code terminal.
Paste this into Claude
Set up this website project and prove the starter works. Install the required packages, run the production build, fix any setup errors, then start the local development site for me. Give me the local URL to open. I should see the generic “Your Business” website. Explain what you do in plain English and ask before making any major change.

Open the local URL Claude gives you. You should see an intentionally generic “Your Business” website.

Command examples Claude may run
npm install
npm run build
npm run dev

To stop the development site manually, use Ctrl+C.

Troubleshooting
  • npm not found → install the current Node.js LTS release, then reopen the terminal.
  • Wrong folder → the terminal must contain package.json, src, and START-HERE.md.
  • Port already in use → accept Vite's next available port or stop the other dev server.
2.220-40 minBuilt-in skill

Run the business kickoff

Give the agent facts once. It creates the brief, route plan, content gaps and implementation from the systems already in the project.

Start Claude Code or Codex from the project folder, then use one of these prompts.

Starting without an existing website
Use $website-kickoff to configure this website for my business. Interview me a few questions at a time. Suggest sensible defaults when I do not know, but never invent proof or business facts. Show me the business brief and route plan before you change the starter site, then continue until the production build passes.
Starting from your existing public website
Use $website-kickoff to configure this website for my business. My existing website is https://YOUR-DOMAIN-HERE.com. Inventory its public pages and reuse the useful facts, URLs and content structure. Confirm anything stale or uncertain, interview me only about gaps, and show me the business brief and route plan before implementation. Then continue until the production build and placeholder scan pass.
Important: the crawler reads public pages only. You still decide what is true, current and approved. Never let the agent invent testimonials, results, customer logos, addresses, awards or credentials.

The first network crawl may ask for permission. Approve it only for the business website you own or are authorised to migrate.

2.320 minReview like the owner

Review the brief before the build

Catch wrong assumptions while they are still a short plan, not spread across twenty pages.

  1. Read kickoff/business-brief.md. Correct positioning, customers, offers, locations, tone and CTA.
  2. Read kickoff/route-plan.md. Remove weak pages; add important services, industries, solutions and locations.
  3. Read kickoff/content-gaps.md. Decide which missing proof/assets can be supplied now.
  4. Tell the agent what to change, then explicitly approve implementation.
Useful steering prompt
The brief is close, but change [specific fact]. Remove [page] because we cannot make it useful yet. Add [offer/audience] and use [tone]. Keep [existing URL] because it already has traffic. Update the three kickoff files, show me the revised route plan, then implement it.
2.420-40 minBuild & verify

Run it, inspect it, and finish the gaps

Leave with a site that builds to real static HTML and a precise list of owner-supplied items—not a confident-looking unfinished demo.

Final verification prompt
Run the full website-kickoff verification now. Build every route to static HTML, check titles, descriptions, H1s, canonicals, sitemap membership, internal links, noindex pages, mobile navigation, forms and remaining starter markers. Fix every code or content-structure failure you can. Then give me the local preview command, deployment settings, route list, environment variables, publishing instructions and only the factual assets or approvals that still require me.

Preview the finished site locally:

Paste this into Claude
Build the finished website and start a local production preview for me. Fix any build errors you find, then give me the preview URL to open.
Command examples Claude may run
npm run build
npm run preview
  • Open the home page, one service, one industry/solution, one case study, one blog post and the contact form.
  • Resize to mobile width and use the menu by keyboard.
  • Check that all demo labels and “Your Business” text are gone.
  • Do not publish until form destination, domain, legal details, proof, analytics consent and missing imagery are resolved.
Checkpoint: the project builds, the content is your business, every route works directly, the sitemap uses your domain, utility pages are noindexed, the form has a real destination, and kickoff/content-gaps.md contains no launch-blocking item.
2.5Ship it · 15 minGitHub + Cloudflare

Deploy it for free - GitHub + Cloudflare Pages

Static sites host free on Cloudflare Pages, and every git push redeploys automatically.

A · Put the code on GitHub. Log into the GitHub CLI first: ! gh auth login (in your terminal), then:

Let Claude Code do the git work
Initialise a git repo here, add a sensible .gitignore for a Node/Vite project, commit everything, then create a new private GitHub repo called bobs-website and push to it using the gh CLI.
B · Connect Cloudflare Pages · C · The magic loop

B - dash.cloudflare.com → Workers & Pages → Create → Pages → Connect to Git → choose your repo. Build config:

  • Framework preset: None
  • Build command: npm run build
  • Build output directory: dist

Save and Deploy → ~1–2 min → a live URL like https://bobs-website.pages.dev.

C · The magic loop - change something with Claude Code ("add a contact page"), then say "commit and push." Cloudflare redeploys on its own. That's continuous deployment, for free.

Cloudflare build fails? Open the build log. Build command must be npm run build, output dist. Node-version error → add env var NODE_VERSION = 20. Paste the log to Claude and it'll fix it.
Final checkpoint: your site is live on a public URL, and pushing a change redeploys it automatically - total hosting cost: $0.
🔁 Next step - keep iterating. You're not done, you're live. Ask Claude Code for changes ("add a testimonials section", "write three more blog posts", "tune the homepage headline"), then "commit and push" - Cloudflare redeploys each time. Growing the site is just more of this loop.
Playbook 3

Build a SaaS / Portal / Web App

The deepest build: a working full-stack app with a real database. You don't need to understand every line - the goal is the engineer's loop: Ask → Review → Approve → Run → Check → Improve. Go one small step at a time.

Playbook 3 · ~90 min · outcome: a deployed app

Build TaskFlow, then ship it

Use a beginner-friendly task tracker as the example, or adapt the same process to your own SaaS, portal or web app: build real features, save data in a database, then push it to GitHub and deploy it live.

Part A · Build it locally

Choose your build

3.1Next.js · Prisma · SQLite

Build it one small step at a time

Not sure what to build? Follow TaskFlow as a complete sample app. Already have an idea? Switch to the “My Own App” instructions.

Choose this if you are unsure what to build. TaskFlow is a complete sample app that teaches the whole process with clear, ready-to-paste prompts.
Step 1 - paste this into Claude
Set up and run a new Next.js app for me. Do the setup yourself rather than only telling me which commands to type.

First, check whether Node.js and npm are installed and suitable for the latest Next.js. If either is missing or outdated, install the current Node.js LTS release and anything else required, or guide me through the installation if you need my permission. Confirm both tools work before continuing.

Create a new empty folder called `taskflow`, switch into that folder, then actually run this command there:
`npx create-next-app@latest . --typescript --tailwind --eslint --app --src-dir --import-alias "@/*"`

The `.` means create the app directly in the current `taskflow` folder. Do not create another project subfolder inside it.

Once it finishes:
- install any remaining dependencies the project needs
- start the development server with `npm run dev`
- fix any setup errors
- give me the local URL to open and confirm the starter page loads

Explain what you do in plain English. Ask me before anything needs administrator access or changes software outside this project. Do not build the TaskFlow features yet.
Command examples Claude may run
mkdir taskflow
cd taskflow
npx create-next-app@latest . --typescript --tailwind --eslint --app --src-dir --import-alias "@/*"
npm install
npm run dev
Step 2 - ask Claude to understand it & plan (don't edit yet)
Inspect this new Next.js project and explain the important files in simple language.
I'm not a software engineer, so keep all explanations plain and beginner-friendly -
and remember that about me as we work.
Then propose a short plan for building a beginner-friendly full-stack task tracker called TaskFlow.

The app should use:
- Next.js App Router
- TypeScript
- Tailwind CSS
- Prisma with SQLite for local database storage
- Server Actions or API routes, whichever is simpler for this project

Do not edit files yet. First explain the plan and wait for me to approve it.
Show steps 3–9 (UI → database → features → polish → memory)
Step 3 - first UI with sample tasks
Build the first version of the TaskFlow home page using temporary sample tasks only.

Requirements:
- Replace the default Next.js starter page.
- Show the title "TaskFlow" and a short subtitle.
- Show an input box and an Add button (they don't need to save yet).
- Show 3 sample tasks with a completed state.
- Show filter buttons for All, Active, and Completed.
- Use clean Tailwind CSS styling and keep it beginner-friendly.

After editing, tell me which files changed and how to run the app.
Step 4 - add the database
Add local database storage for tasks using Prisma and SQLite.

Requirements:
- Install and configure Prisma if not already installed.
- Create a Task model with id, title, completed, createdAt, and updatedAt.
- Add a seed step with a few example tasks.
- Keep the database local. After each step, explain in plain English what you just did and why - I'm not a developer.
- Do not remove the UI from the previous step.
Step 5 - connect UI to real tasks
Connect the TaskFlow page to the real SQLite database.
- Load tasks from the database and display them.
- Add a new task on submit; prevent empty tasks; show a validation message.
- Use Server Actions or API routes consistently.
- Tell me how to test adding a task in the browser.
Step 6 - complete, delete, filter
Add the main task management features to TaskFlow.
- Mark a task completed or active; delete a task.
- Make the All, Active, and Completed filters work.
- Show a progress summary like "2 of 5 tasks completed".
- Give me a manual browser checklist to verify the features.
Step 7 - polish · Step 8 - review · Step 9 - memory
Polish the TaskFlow app: spacing, colours, layout, empty states, small loading states. Keep it clean, not flashy. Do not add new major features.

Then: Review this project as if I am new to coding - what it does, key files, how the frontend talks to the backend, how tasks are saved, and how to run it tomorrow. List any bugs before fixing.

Finally, run /init to create CLAUDE.md, then add a rule: keep this project simple and beginner-friendly. Prefer small, readable changes over clever code.
Done when: npm run dev runs, you can add / complete / delete / filter tasks, they survive refresh, the design is clean, and there's a CLAUDE.md.
3.2Iterate

Keep building with AI until it's yours

TaskFlow - or your own idea - is rarely done in one pass. Keep the loop going: add features, fix what's off, refine the look, one small step at a time, until it does what you actually want.

Steer it, one change at a time
Let's keep improving the app. Next I'd like: [describe one change in plain words].
Remember I'm not a software engineer - explain what you're doing simply.
Make one small change at a time, tell me which files changed and how to check it in the browser,
and stop to ask me before anything risky.
Good things to ask for: a new feature, a bug fix ("X breaks when I..."), a design tweak, or "review what we have and suggest the three most valuable next steps." When it does what you want, move on to shipping it below.

Part B · Ship it

Deploy - GitHub & Railway

Back up your build online, then put it on the internet with a public link.

3.3GitHub

Push your project to GitHub

Create an empty private repo named after your app (don't add README/.gitignore/license), sign in with gh auth login, then let Claude do the git work.

Paste into Claude Code
I have created an empty private GitHub repository at:
<PASTE YOUR REPOSITORY URL HERE>

Please:
- Initialise git in this project if it is not already.
- Make sure a sensible .gitignore is in place (node_modules, .env, the local SQLite file).
- Create a first commit.
- Add that GitHub repository as the remote named "origin".
- Push the code to the main branch.

After each step, explain in plain English what you just did and why - assume I'm not a developer. Stop and ask me before anything needs my GitHub login.
3.4Railway

Deploy to Railway

Sign up at railway.com (sign in with GitHub), install the Railway CLI, run railway login, then ask Claude to set up Postgres and deploy.

First, give it a real database. Your SQLite database is just a local file - it won't survive on a host - so we move to Postgres, which Railway hosts for you, before deploying.

Step 1 - set up Postgres (paste into Claude Code)
I have installed the Railway CLI and run `railway login`.

Please set up a Postgres database for this app on Railway:
- Create a new Railway project and add a Postgres database to it.
- Switch this app's Prisma datasource from SQLite to Postgres.
- Wire up the DATABASE_URL from Railway, run the migrations, and keep my existing tables and seed data.
After each step, explain in plain English what you just did and why - assume I'm not a developer. Stop and ask me before anything needs a login or could cost money.
Step 2 - deploy (paste into Claude Code)
Now deploy this Next.js app to Railway using the Railway CLI:
- Deploy it into the same Railway project as the Postgres database.
- Configure the build and start commands correctly for a Next.js app.
- Make sure the app connects to the Railway Postgres database via DATABASE_URL.
- Give me the public URL once it is live.
After each step, explain in plain English what you just did and why - assume I'm not a developer. Stop and ask me before anything needs a login or could cost money.
Why Postgres first? A local SQLite file doesn't persist on a host - a deploy can wipe it. A Railway-hosted Postgres database lives separately from the app, so your data survives between deploys.
Final checkpoint: a working SaaS app you built step by step, backed up on GitHub and live on a public Railway URL, with its data in a real Postgres database.
3.5Learn · 5 minGoing to production

From your laptop to production - the grown-up version

What you just did deploys straight from your machine. That's perfect for a demo - but not how real products go live.

⚠️ Never deploy to production from a local device. 'Friends on let friends do this' Not yours, not a friend's. Whatever's on someone's laptop might be half-finished, untested, or full of debug changes. Production should only ever deploy from a dedicated production branch, automatically.

Here's the approach - the shape of how you'd do it properly. Want to try it now? Claude can walk you through each part:

  1. Create a production branch in GitHub (your main/day-to-day work stays separate from what's live).
  2. Ask Claude to set up the Railway production environment and a GitHub Action that auto-deploys to it - the single prompt below does both.
  3. Push your changes up to a normal branch and check they're good.
  4. When you're happy, have Claude promote it into the prod branch - the Action deploys it for you.
Set up production + auto-deploy (paste into Claude Code)
Set up a proper production deploy for this app on Railway, then wire up automatic deploys from GitHub.
- Create a separate Railway production environment, kept apart from my day-to-day setup.
- Add a GitHub Action that automatically deploys to Railway production whenever I push to the prod branch.
- Commit and push the workflow and any config to my GitHub repo.
After each step, explain in plain English what you just did and why - I'm not a developer. Stop and ask me before anything needs a login or could cost money.
The point: deploys are triggered by the branch - reviewed, repeatable and automatic - not by whoever happens to have the CLI logged in.
3.6Ship it · 15 minSecurity

Check the security before real users

The final step. Before anyone real touches it, have Claude hunt for security holes and lock down who can do what.

Ask Claude to security-review it
Review this codebase for security holes before real users touch it.
- Find common vulnerabilities: injection, exposed secrets or API keys, insecure dependencies, and missing input validation.
- Set up proper authorization: a logged-in user must only be able to see and change their OWN data, and must not be able to call API endpoints or perform actions they aren't allowed to. No endpoint should be usable just because someone knows the URL.
- Review the API against the OWASP API Security Top 10 and tell me exactly where it falls short.
List what you find by severity, fix what you can safely fix, and flag anything that needs my decision or a real developer.
⚠️ Until this passes - and ideally a real developer has looked too - don't collect real customer data, take payments, or share it beyond a trusted test group.
Done when: Claude has reviewed the code and the API against OWASP, authorization is enforced so users can only do what they should, and you have a clear list of anything left for a developer.
Playbook 4

AI Agent/Automation

Design your own AI agent - an assistant that owns a recurring job. Describe what you want in plain English and the builder below writes the prompt for you. The golden rule stays the same: it drafts, you approve - a human gate on anything that matters.

Playbook 4 · ~45-60 min · outcome: your own working agent

Define your agent, then build it

Every reliable agent comes down to four things: a clear role & scope, the tools it can touch, the knowledge it reads, and hard guardrails. Fill those in below and the builder assembles a ready-to-paste prompt - then the agent-creator skill turns it into a working agent.

agent-creator.skill · the builder
4.115 minDefine it

Describe your agent - the builder writes the prompt

Fill in the four boxes. As you type, a complete agent prompt builds itself - copy it straight into Claude Code.

How knowledge works: every agent lives in its own folder - that's its home base, holding its SKILL.md, a CLAUDE.md with its process, and any examples. It can also read knowledge from other directories you point it at - a brand-voice file, a policies folder, past examples. In the Knowledge box, list both: its own folder, plus any outside paths it should pull context from.

Who is it and what one job does it own? Where does its job start and stop?

Which connectors, skills or apps may it touch? Note read-only vs allowed to act.

Its own folder, plus any other directories or files it should read from.

The hard rules. Be specific about anything it should never touch or decide.

Your agent prompt - copy it into Claude Code
4.215 minagent-creator

Build it with the agent-creator skill

Paste your prompt into Claude Code. It already asks Claude to plan first and flag what you've missed - read that plan and fill the gaps before you approve.

Your prompt tells Claude to use the agent-creator skill, plan the agent first, and tell you what you've missed before building. So the flow is: paste it, read the plan, patch the gaps, then approve. Claude builds a SKILL.md and a CLAUDE.md in the agent's own folder, kept as local files.

Examples are the strongest context. If the agent produces something in your voice or to a standard - replies, summaries, posts - drop one or two great past examples into its folder and point the Knowledge at them. One good example beats a paragraph of description.
4.3Ship it · 20 minDry-run · review · schedule

Dry-run it, review like a manager, then schedule

Run it without touching anything real, hold it to every guardrail, then set it loose on a schedule.

Do a dry run first - always. Ask the agent to show exactly what it would do without doing it: which items it found, what it would draft, what it would touch. Check its inputs and outputs against your role and guardrails. Send it back when it slips: "You didn't flag the overdue invoice - your job says you must. Redo." Iterate, iterate, iterate - the first pass is never 100%. Only once the dry runs are clean, set it up as a scheduled task.

Dry-run it (paste into Claude Code)
Before doing anything for real, do a dry run of this agent.
Show me exactly what you WOULD do, step by step, without sending, changing or deleting anything:
- what you read, and the items you found
- what you'd draft or produce for each
- anything you'd touch that I should know about
Then wait for me to review it against the role and guardrails before we ever run it for real.
Schedule it & let it improve itself
Put it on a schedule
Now set this up as a scheduled task (e.g. every hour, or each morning). On each run, produce a short summary of what you did and anything waiting on me - and never take a guarded action without my approval.
Let it improve itself
Look back over your last few runs. Where did you get it wrong or miss something? Update your CLAUDE.md and skills so the next run is better - and tell me exactly what you changed and why.
  • It tried to act instead of draft → reinforce the guardrail ("NEVER send - only draft").
  • Its output isn't right → give it one great example. Examples are the strongest context.
  • It's inconsistent → make the must-dos explicit checks, not suggestions.
Checkpoint: an agent you defined, built with agent-creator, proven with clean dry runs, and running on a schedule - that never takes a guarded action without you.
Playbook 5

Agentic Marketing

A reusable, company-owned marketing workspace. Configure your business and brand once, then use Claude to design Google or Meta ad campaigns, editable visual assets, and matching Buffer-ready organic posts - with human approval before anything goes live.

Playbook 5 · ~75-95 min · outcome: your own reusable marketing bench

Set up once, then run paid ads and organic social

The download is a complete, company-owned marketing workspace. It already contains setup, creative-production, Google Ads, Meta Ads and Buffer skills, plus templates and fictional examples to adapt. You configure it once, then lead with a paid campaign and add matching organic social.

ai-marketing-bench.zip · your reusable marketing bench
Quick start
  1. Download and unzip ai-marketing-bench.zip somewhere you own - one copy per company. Do not work inside the ZIP.
  2. Open the unzipped AI-Marketing-Bench folder in Claude Code.
  3. Read START-HERE.md and PLAYBOOK.md - they explain the workspace and the safe operating rhythm.
  4. Run the setup below (5.1) to make the bench describe your real business.
  5. Then work through the campaigns in 5.2-5.4.
5.120 minSet up

Make the bench yours

Interview the business once and turn the answers into a practical foundation Claude can reuse for every campaign.

Run this inside your copy of AI-Marketing-Bench
Use the setup-marketing-bench skill and interview me in short rounds.
Help me define the company, offers, audience, brand voice, visual direction,
channels, capacity, budget, KPI definitions and approval rules.
Update the foundation files, then run the readiness checker.
Label assumptions and leave TODOs for anything we cannot answer yet.
Checkpoint: the foundation files describe the real business, and Claude can explain the first small campaign it recommends and why.
5.225 minPaid · creative

Create your first paid ad campaign (Google or Meta)

Design a real Google or Meta campaign from your foundations - brief, targeting and creative - kept paused and unspent.

Build the paid campaign
Create a paid ad campaign for [offer]. Read the foundation files first.
Decide with me whether Google or Meta fits this offer and audience better.
Copy the matching campaign template under 03-paid/, then use the
creative-production skill to adapt ad-mockups.html and ad-mockups-new.html
to our brand. Complete the brief, creative, targeting, budget, launch
checklist and stop conditions. Give me the ad copy, headlines and
exact-size creative. Keep every status as Draft and the campaign paused -
do not launch, activate or spend anything.
Checkpoint: a complete Google or Meta campaign folder - brief, targeting, budget, creative and stop conditions - that all reflect the same offer, audience and brand, and stays a paused draft.
5.320 minOrganic · creative

Add a matching organic social campaign

Reuse the same offer and brand to produce editable organic posts you can repeat.

Build the organic campaign
Now create a small organic social campaign for the same offer.
Use the post template and the creative-production skill. Adapt the
editable social-card templates to our brand. Give me platform-specific
copy, square and story/reel designs, alt text, UTMs, an asset manifest
and a review checklist. Keep every status as Draft.
Checkpoint: editable source templates, exact-size exports and channel copy that match the paid campaign's offer and brand - and remain drafts.
5.415-30 minApprove · hand off

Validate and practise a safe hand-off

Check both campaigns against the playbook and rehearse going live - without publishing or spending money.

Validate without going live
Review both campaigns against the playbook and foundation files.
For the paid campaign, confirm the Google or Meta brief, targeting, budget,
tracking and stop conditions, and show the exact account and actions that
would create it paused. For organic, prepare the Buffer queue and run
validation in dry-run mode. Show me every account, content, date/time,
timezone, budget and action that would happen next.
Do not publish, launch, activate or spend anything.
⚠️ Human gate: drafting and dry runs are safe. Creating or activating ads, and publishing to Buffer, each require a separate, explicit approval after reviewing the exact account, channel, content, timing and budget.
Final checkpoint: a reusable company bench, one paid and one organic campaign reviewed together, and validated next steps that have not gone live.

Wrap-up

What to take home

🚀

Something real, shipped

You leave with one finished thing - a branded kit, a live website, or a deployed app - not a stack of half-built ones.

🔁

The loop

Ask → Review → Approve → Run → Check → Improve. The same loop works for any build, at any size.

📁

Your project folder

It's all just files in a folder. Yours to keep, reopen in VS Code, and keep building on Monday.

After the training

Where to go from here with Team 400

From self-paced learning to a trained AI engineer embedded in your team.

See what's next →