Claude Code

A few habits that make a real difference when working with Claude Code day to day.

1. Run from root

When you first launch Claude Code, run it from your project's root directory. Claude works best when it can see your whole project structure.

claude

2. Run /init immediately

Creates a CLAUDE.md file. Think of it as Claude's long-term memory — a persistent context file that gets loaded with every session. /init is just the starting point. Review it, refine it, make it yours.

3. CLAUDE.md is hierarchical

Claude reads memory from multiple levels, and the most specific one wins.

Level	Location
Global	`~/.claude/CLAUDE.md`
Project	`./CLAUDE.md`
Subdirectory	`./subdir/CLAUDE.md`

You can also pick which memory to use with the /memory command:

1. Project memory   →   ./CLAUDE.md
2. User memory      →   ~/.claude/CLAUDE.md

4. Keep CLAUDE.md concise

It gets prepended to every prompt. A bloated file eats your context window and degrades performance. Focus on what matters — cut everything else.

5. Structure: What, Domain, Validation

A good CLAUDE.md covers three things:

What — your tech stack (languages, frameworks, tools)
Domain Context — what each part of the codebase does
Validation Steps — how to verify correctness (tests, linters, type checkers)

Keep each section short and specific. Claude doesn't need prose — it needs signal.