Week one is install, auth, a short CLAUDE.md, and one plan-mode change. Defer MCP sprawl, plugin catalogs, and custom agent frameworks until the basic loop works.

This is a path, not an encyclopedia. When you need depth, jump to the guides linked at the end.

Prerequisites

You do not need MCP, a custom slash library, or a perfect CLAUDE.md on day one.

Day 0 — Install and open a session

Official paths (verify against Claude Code docs before you ship this page — installers move):

Native install (recommended)

macOS / Linux / WSL:

curl -fsSL https://claude.ai/install.sh | bash

Windows PowerShell:

irm https://claude.ai/install.ps1 | iex

Homebrew: brew install --cask claude-code (does not auto-update — brew upgrade claude-code later)

Then:

cd your-project
claude

Log in when prompted. If ANTHROPIC_API_KEY is already set, approve the key instead.

Done when: you have an open Claude Code session in a real repo.

Day 1–2 — A CLAUDE.md that survives Monday

Write a short project CLAUDE.md (or run /init and cut it down). Keep only what is always true:

## Commands
npm run typecheck
npm run test

## Boundaries
- Preserve public APIs
- Never edit generated files
- Prefer www canonical URLs for public pages

## Definition of done
- Types pass
- Critical path verified

Rules that change weekly do not belong here. Put them in the ticket or the plan.

Done when: a new session reads the file and follows at least one boundary without you restating it.

Deep dive: The Complete CLAUDE.md Guide.

Day 3 — One plan-mode change

Pick a multi-file but small job (two to four files). Enter plan mode before edits:

Require the plan to list files, out of scope, and a verify command. Approve only when the file list looks right.

Field pattern we use: map → plan → build → verify. Example from a real hub change: three files (two locale pages + dictionaries), link targets forced to existing blog slugs before coding. See Claude Code Plan Mode.

Done when: one planned change shipped with a verify step you actually ran (for us: npm run typecheck + link check).

Day 4–5 — One slash command, one hook

Slash command (team-shared)

Create .claude/commands/map.md (or start with /plan / /ship from the same library):

---
description: Map the repo before changing anything.
argument-hint: [area or ticket]
---
Before writing code, map the working set for: $ARGUMENTS
# ... list entrypoints, files, blast radius, unknowns — no edits

Commit project commands so teammates get them on pull. Personal commands live in ~/.claude/commands/.

One hook only

Add a single PostToolUse or PreToolUse hook that earns its keep — e.g. typecheck before git commit, or protect .env. Do not install five hooks on day four.

Deep dive: Claude Code Hooks.

Done when: you ran the slash once and the hook fired once on purpose.

Day 6–7 — Review the loop, skip the rest

Ask yourself:

  1. Did CLAUDE.md save a repeated instruction?
  2. Did plan mode prevent a wrong file set?
  3. Did verify catch something before merge?

If yes, keep the loop. If a step felt empty, delete it — do not add MCP to fill the silence.

Skip list for week one

TemptationWhy wait
Every MCP server you saw on TwitterSecrets + permission surface; start with zero or one read-scoped server later
A 40-line CLAUDE.mdLong files rot; short ones get followed
Parallel subagents for a one-file fixCoordination cost > benefit
"Complete guide" bingeYou need one useful session, not coverage

When you do add MCP, keep secrets in env (e.g. ${GITHUB_PAT}), not in prompts. A sanitized shape looks like:

{
  "mcpServers": {
    "filesystem-docs": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-filesystem", "./docs"],
      "env": {}
    }
  }
}

Full MCP guide comes after this path is boring.

Checklist (print this)

Related on this site

Answer-ready block

Week one with Claude Code is install, auth, a short CLAUDE.md, and one plan-mode change with a real verify step. Defer MCP, large prompt libraries, and multi-agent setups until that loop is reliable.