Day 2: Give Claude Code Good Context

5-minute quick start

You will learn

Why a new session does not remember your last chat
What goes in a CLAUDE.md file, and what to leave out
How to write an ask with a file, a definition of done and a constraint

The one thing to remember

Write down what is always true in CLAUDE.md. Say what is true today in the prompt.

A working example

/init
# then trim to: commands, conventions, things to avoid

Short on time? That is enough to start. Jump to the full guide for the what, why and how of every step, in the terminal and in VS Code.

01Why context is everything

A new Claude Code session does not automatically carry over your previous conversation. It can read your files, but unless something tells it, it doesn't know your preferences, your team's rules or the command you use to run the tests. Continuity comes from what is written down: CLAUDE.md (and auto memory, if you have it on) is loaded at the start of every session. When it doesn't know, it guesses, and a confident guess in the wrong framework or style costs you a round of fixes.

Two habits remove most of the guessing: a CLAUDE.md file for what is always true about your project, and clear asks for what is true about this one task.

02CLAUDE.md: a briefing it reads every session

CLAUDE.md is a plain Markdown file in your project's root folder. Claude Code loads it at the start of every session, so you write the briefing once instead of repeating it. You can also keep a personal one at ~/.claude/CLAUDE.md for preferences that apply to all your projects.

The quickest start is to let Claude draft it. In a project folder, run:

>_PowerShell

—☐✕

PS C:\dev\my-project> claude

# inside Claude Code
> /init

It reads your project and writes a first CLAUDE.md. Then edit it. The draft is a starting point, and you know things it can't see.

03What belongs in it

Keep it short. A long file is a file that gets ignored, and every line uses up context. Include only what Claude can't work out by reading the code:

01Commands

How to install, run, test and lint. This is the single most useful section, because it lets Claude verify its own work.

02Conventions

The rules a new teammate would be told on day one: naming, where new files go, which library to use for what.

03Things to avoid

The mistakes you keep correcting. "Don't edit generated files." "Never commit directly to main."

A small example you can copy and adapt:

>_CLAUDE.md

—☐✕

# My Project
A web app for tracking orders (React + Node).

## Commands
- Install: pnpm install
- Test: pnpm test (run before finishing any task)
- Lint: pnpm lint

## Conventions
- TypeScript only, no new JavaScript files
- UI components live in src/ui; follow existing patterns
- API calls go through src/api/client.ts

## Avoid
- Don't edit anything in dist/ (generated)
- Don't add dependencies without asking

Rule of thumb: if Claude gets something wrong twice, add a line to CLAUDE.md. Treat it as a living file that grows from real mistakes, not a document you write once and forget.

04Clear asks: the other half

CLAUDE.md covers what's always true. Each request still needs to say what you want now. A good ask has three parts: where to look, what "done" means, and any constraint.

Vague"Make the search faster."

Clear"Search on the orders page takes 3 seconds. The query is in src/api/orders.ts. Find why it's slow and fix it without changing the response format. Run pnpm test when you're done."

You don't need long prompts. You need the right three sentences. Naming a file also saves Claude from searching the whole project, which keeps the session faster and focused.

05Try this today

InitRun /init in one of your projects.

TrimCut it down to commands, conventions and things to avoid.

AskGive your next task a file, a "done" and a constraint.

GrowAdd a line whenever Claude repeats a mistake.

Next up: giving it the right words, prompts that get the result you meant the first time.

It only knows
what you tell it.