Claude Code · ISTO.

Lesson 1 · What Claude knows

Why Claude Code

Why Claude Code: same Claude, set up for what we'll build.

See every fileThe project's documents, rules and scripts, open in VS Code right next to Claude.

Run scriptsSmall programs that do a job the same way every time. Claude runs them in the terminal and asks you first.

Share through GitHubIn week 3: one shared project, and changes are reviewed before they land.

Everything you learned in Cowork still applies: same Claude, same way of asking.

Your questionnaire

Your questionnaire: where the team is today.

How often you use AI

daily: 4

several times a week: 1

a few times a month: 1

How you start a task

a fresh prompt each time: 3

reused phrasing or a tool you built: 2

not using AI yet: 1

Today: write it once3 of the 5 who use AI every week start from a blank prompt each time. A file Claude reads at every start ends that.

People (of 6) spending 3+ hours a week on…

One person each: recurring reports by hand (→ Lessons 2–3), repeat judgment calls, scheduling, spreadsheet analysis.

6 of 8 answered.

Your questionnaire

What you want to build, and where Claude Code fits.

Lesson 1We are here

3 of 5 start from a blank promptWrite it once in CLAUDE.md; every session starts from it.Claude Code: files the whole team shares

Lesson 2

5 of 6 want a report3 of them on stock, sell-through or replenishment, from Sage and Shopify.Claude Code: one shared database with your definitions; calculations as scripts you can check and rerun

Lesson 3

2 of 6: a report that builds itself every MondayBuilt once, then it runs on its own.Claude Code: the job lives in the project, next to its data

3 of 6 tasks start when an email arrivesWhen X happens, Y happens.Claude Code: written once as a skill, run the same way every time

Lesson 4

Your own work, on Isto's dataA 4-minute demo of what you built.Claude Code: one shared project, owned by the team

Same Claude as Cowork. Where Claude Code has the edge for these goals: the work lives in one project the whole team shares, with the same definitions and the same numbers for everyone.

The building blocks

The six building blocks of this course.

Lesson 1We are here

VS Code tour

Where everything is: files, Claude, the terminal.

Memory and context

What Claude knows every time, and what must never happen.

Lesson 2

Tools and reach

How Claude acts: scripts, connectors, a shared database.

Plan, then build

Agree the plan, then build one phase at a time.

Lesson 3

Skills

A job handbook Claude follows every time.

Hooks and automations

When something happens, the next step moves on its own.

Sharing through review

One shared project on GitHub; every change is reviewed before it lands.Also in lesson 4

Lesson 4

Your 4-minute demo

Show the team what you built, on Isto's own data.

Your Agentic Map

What you built, where it lives, who owns it, what's next.

All six blocks together

Without them, you are Claude's memory and messenger: you paste the context in and carry the answer out. These blocks move that work into the project.

Your VS Code window

VS Code: four places you'll use every day.

VS Code with the practice project open: Explorer, Claude panel, a CLAUDE.md preview and the terminal

12234

1Explorer

The project's folders and files.

2Claude panel

Opens from the Claude icon in the sidebar. Where you talk to Claude Code about this project.

3Editor

Open any file to read it: a CLAUDE.md, a policy, a report.

4Terminal

Where commands run. Mostly Claude runs them for you.

Your turn · 3 min

Open VS Code. Find the Explorer, open a terminal (Terminal → New Terminal), then open the Claude panel.

Why VS Code

Why VS Code: you see the whole project, not just the chat.

1The whole project in view

Every folder and file, not only the ones Claude mentions. You learn how a project is organised.

2Files shown as pages

Documents shown as formatted pages; with free extensions, spreadsheets and PDFs too.

3Every change before it lands

What was there and what Claude proposes, side by side. Accept or reject each change.

4Not tied to one AI

The same files work with other AI assistants and models.

Seeing the whole project, every day, is how you learn to look after it as it grows.

Get the practice project

The practice project: a team handbook to open in VS Code.

Your turn · 4 min

Go to github.com/Adapt-to-AI-B2B/este-handbook
Green Code button → Download ZIP. No GitHub account needed.
Unzip it, move the folder into your projects folder, rename it to este-handbook.
In VS Code: File → Open Folder → este-handbook.

Este is a fictional clothing brand. The project holds documents only: how the company works, its policies, its words, some meeting notes.

Idea · Claude only knows what you give it

Claude's memory lives in files: the chat forgets, the files don't.

Your turn · 4 min

Open CLAUDE.md, then docs/returns.md: a document, exactly as the team wrote it.
Tell Claude your name and your role. Then start a new session and ask: "What's my name?"
In the new session, ask: "Read CLAUDE.md and tell me in two lines what this project is."

Like a new colleague who forgets every conversation overnight: what's written down is still there tomorrow. The second answer works every time, for anyone, because it comes from a file in the project.

Memory and context

CLAUDE.md files: what Claude reads when it starts.

1~/.claude/CLAUDE.md

You, in every project. Your own habits: language, how you like answers, who you are.

2este-handbook/CLAUDE.md

The team, in this project. How this project works, its rules, where everything is.

3este-handbook/docs/meetings/CLAUDE.md

One folder only. Extra instructions for meeting notes, added on top of 2.

Every session starts hereAt the start, Claude reads 1 and 2 (and any CLAUDE.md in the folders above the project). It adds 3 the moment it works in that folder.

Today: the project's filesYour own ~/.claude/CLAUDE.md comes back in week 2.

Memory and context

CLAUDE.md is the orientation manual + index for Claude.

The orientation manualWho we are, how to answer, the rules. Short: what Claude needs every time.

The book indexWhere everything else is. Claude opens a document only when the question needs it.

About youABOUT-ME.md who you are; pointed to from your own ~/.claude/CLAUDE.md.

RulesRULES.md or one file per topic in .claude/rules/ (next block).

MemoryMEMORY.md decisions and lessons worth keeping.

The project's documentssuppliers.md returns.md, words-we-use.md…

Common names, not required ones. What matters is that CLAUDE.md says where they are.

A good CLAUDE.md

A good CLAUDE.md is short, with a reason for every rule.

Short. Under 200 lines. A long file is followed less well.
Points, doesn't copy. "The returns policy is in docs/returns.md", not the policy pasted in.
Every rule has a "because". Without the reason, Claude has to invent one, and applies the rule in the wrong place.

The new colleague rereads the manual every morning: keep it short enough to read, and say why.

Memory and context

Definitions: add the word nobody wrote down.

Your turn · 8 min

Ask Claude: "What does 'weeks of cover' mean at Este?"
Ask Claude to add a definition to docs/words-we-use.md, with the reason we use it.
Start a new session and ask the same question again.

The meeting notes used a word the handbook never defined. Now the answer is the same for everyone, every time.

Memory and context

Rule files: one topic each.

1At the entrance

A file in .claude/rules/ is read at the start, like CLAUDE.md. Every trip passes it.

2On one road only

A "paths" line at the top: Claude sees it only when it works in that folder.

3In the wrong place

Wrong folder, a typo, two signs that disagree: no error. Nobody sees it.

A rule file is a road sign: one topic each, put where Claude will pass, and checked once to see that it's followed.

Memory and context

Rule files: write one, then test it.

Your turn · 8 min

Open .claude/rules/customer-replies.md. When does it apply?
Ask Claude to draft a reply in drafts/customers/ to Rui, who wants to return a sale T-shirt he bought online 20 days ago.
Create .claude/rules/delivery-dates.md: never promise when an out-of-stock item is back, because supplier lead times move.
New session: draft a reply to Ana, who asks when the Medium Green Men's T-shirt is back in stock.

Memory and context

Memory: you write the rules, Claude keeps notes.

Your turn · 4 min

Tell Claude: "Remember: I manage the Porto Foz store, and I like short answers in bullets."
Type /memory and open what Claude wrote.

Claude decides what's worth keeping. If the whole team needs it, put it in CLAUDE.md instead.

Memory and context

Where does each instruction go?

Together · 6 min

"Our returns window is 30 days."
"Always answer me in Portuguese."
"Sage is the source of truth for stock, because the stores sell through it."
"Replies to customers follow the brand voice."
"I like short answers in bullets."
"The password for the online shop admin."

Places: CLAUDE.md in the project · a rule file · a document in docs/ · ~/.claude/CLAUDE.md · auto memory · nowhere.

To keep

Three things from today.

The chat forgets. The files don't.
Write the reason, not just the rule.
A rule is a road sign: put it on the right road, then check Claude follows it.

Next week: plan and build on shared data. Before then: the short pack in the Academy.