Claude Code Personal OS: 10-Minute Quick Start Setup

Stop prompting. Start operating.

A Personal OS is the minimum viable structure that turns Claude Code from a chatbot into an agentic workspace. This guide gets you there in ten minutes.

For the full deep-dive — architecture, auto-dreaming, security hardening, parallel sessions — see Build Your AI Chief of Staff: A Complete Claude Code Personal OS Guide.

Minute 1–2: Your CLAUDE.md

Create ~/.claude/CLAUDE.md with exactly three things:

# My Operating Principles

## Identity
You are my engineering partner. You work on my machine, with my stack,
under my constraints. Think out loud before acting.

## Non-Negotiables
- Never edit .env files without explicit confirmation
- Never delete files without asking first
- Never push to main without review
- Prefer simple over clever

## Stack Defaults
- Runtime: Node.js 18+, Python 3.10+
- Database: SQLite before PostgreSQL
- Format: Prettier with single quotes
- Tests: Run before marking anything done

That’s it. 27 lines. Claude reads this every session. The specificity — “SQLite before PostgreSQL” instead of “choose the right tool” — is what changes actual output.

Minute 3–4: Install MCP Servers

MCP servers are how Claude talks to the real world. Start with two that require zero configuration:

claude mcp add fetch
claude mcp add brave-search

• fetch — reads any web page. No API key needed.
• brave-search — real web search from within sessions. Free tier available.

Verify they’re active with /mcp inside a Claude session.

Minute 5–6: Write One Skill

Skills are reusable workflows. Pick something you do every day. I started with /review:

Create ~/.claude/skills/review.md:

---
name: review
description: Pre-commit code review checklist
---

## Review Protocol
Before I commit, check:

1. **Security** — hardcoded credentials, SQL injection, missing validation
2. **Types** — `tsc --noEmit` passes
3. **Tests** — `npm test` passes
4. **Error handling** — every non-trivial operation has try/catch or error return
5. **Edge cases** — empty states, null inputs, rate limits

Report what passes and what fails. Do not proceed if any check fails.

Now type /review in any session. It takes 90 seconds and catches real bugs.

Minute 7: Add Project Context

Every project gets its own ./CLAUDE.md committed to git:

# Project: [Name]

## Architecture
- Monorepo with apps/ and packages/ directories
- Express API in apps/api, React frontend in apps/web
- SQLite via better-sqlite3 for data

## Conventions
- Feature branches off main
- Conventional commits (feat:, fix:, chore:)
- Tests alongside source files (*.test.ts)

This file travels with the repo. Every team member and every CI pipeline gets the same context.

Minute 8: The Self-Improvement Habit

Claude proposes a new rule, adds it with a date stamp. Your OS learns from every correction. After a month, your CLAUDE.md is a remarkably accurate model of how you work.

Minute 9–10: First Session

mkdir my-first-os && cd my-first-os
claude

Then run your first command:

Read my CLAUDE.md. Scan this directory. What do you understand about how I work?

If Claude accurately describes your preferences and constraints — you’re operational.

What You Have After 10 Minutes

This is the minimum viable Personal OS. From here, every session adds capability.

The full system — auto-dreaming, agent personas, security sandboxing, parallel sessions — takes weeks to build and tune. Start here. Add one component per week. Let it compound.

See the complete build guide: Build Your AI Chief of Staff — covers architecture, 8 setup steps, 14 real-world use cases, and the 90-day fluency roadmap.