The Claude Bible

Don't want to wait?

Install Claude Code in 3 minutes.

Here's the setup that takes most of the book's Part 2. Run these now and you're live. The book picks up from here with what to build first.

Step 1 · Install

Mac/Linux: open Terminal and paste:

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

Windows: open PowerShell (not Git Bash, that one breaks) and paste:

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

Step 2 · Authenticate

Type claude in the terminal. It'll open a browser tab and prompt you to sign in with your Claude Pro ($20/mo) or Max account. After auth, you're back in the terminal, blinking cursor, ready.

Step 3 · Create a CLAUDE.md

Drop a file called CLAUDE.md at the root of your project. This is the single biggest upgrade. Even a 10-line version is enough to start.

# Project: Sales Ops Tools

## Stack
- Node 20, TypeScript, Postgres

## Conventions
- All scripts in /scripts, use npm run for entry
- Never touch /vendor or /.archive

## Test commands
- npm test
- npm run typecheck

One pro tip the book opens withPress Shift+Tab twice to enter Plan Mode. Claude reads files, answers questions, sketches a plan, but won't touch anything until you confirm. Use this for everything for your first week. The other modes are for when you trust the setup.

What's inside

Six parts, no fluff.

Part 1

The Claude Model Family

→ Opus 4.7, Sonnet 4.6, and Haiku 4.5 — when to use which
→ How context windows, output limits, and pricing actually compare
→ The 'start with Sonnet' rule and when to break it
→ Common Claude vocab that confuses people (and what each term really means)

Part 2

Claude Code Setup and First Project

→ Install on Mac, Linux, and Windows (with the Windows gotchas)
→ CLAUDE.md — the single biggest upgrade you can make in an hour
→ Plan mode, auto-accept mode, and the slash commands worth memorizing
→ Your first paste-ready project: an end-of-week report from messy data

Part 3

Skills, Hooks, MCP, and Sub-agents

→ Skills = what to do. MCP = how to do it. Hooks = when to enforce it. Sub-agents = who does it.
→ Writing your first skill (the 5-line minimum that actually works)
→ Wiring an MCP server in under 5 minutes
→ Pre-commit and post-tool hooks that save you from yourself
→ When to spin up a sub-agent vs. running the work inline

Part 4

The Anthropic SDK

→ Tool use, batch API, files API, and citations
→ Prompt caching: how to drop your bill 50-90% without losing quality
→ Extended thinking, memory, and compaction
→ Migrating between model versions when Anthropic ships a new one

Part 5

Capstone Labs

→ Build a sales follow-up agent that runs against your CRM
→ Build a meeting-notes → SOP pipeline
→ Build a multi-source weekly report generator
→ Each lab includes the full prompt, the file structure, and the gotchas I hit

Part 6

Production and Safety

→ Prompt injection patterns to test against
→ Cost controls for teams
→ Logging, tracing, and what to do when Claude is wrong
→ The 'safe defaults' settings every team should ship with

10 questions, real answers

The questions everyone asks me about Claude.

These are the actual top friction points I see from operators, builders, and founders trying to use Claude in their business. Each answer is the version I'd give in a 1-on-1 call. The book goes deeper on every one.

What's the difference between Claude and Claude Code?+

Claude is the AI model family (Opus, Sonnet, Haiku) — you can talk to it in a chat window at claude.ai. Claude Code is a separate tool: it runs in your terminal, reads files on your computer, writes and edits code, and runs commands. Same underlying model, but Claude Code has hands. You use claude.ai when you want a chat. You use Claude Code when you want something built or changed on your machine.

How do I install Claude Code on Windows?+

Open PowerShell (not Git Bash — Git Bash will throw a 'Raw mode is not supported' error) and run: irm https://claude.ai/install.ps1 | iex. Restart your terminal, then type 'claude' to launch. If the command isn't found after install, add the install path to your User PATH in System Properties → Environment Variables. On Mac or Linux, run: curl -fsSL https://claude.ai/install.sh | sh. You need a Claude Pro ($20/mo) or Max plan to authenticate.

Do I need to be a developer to use Claude Code?+

No, but it helps to be comfortable in a terminal. Claude Code talks to you in plain English. You describe what you want, it writes the code, it asks before changing files. The buyers who get the most out of it are operators (marketing managers, sales ops, finance analysts) who already have technical problems they're paying engineers to solve. The book has 3 capstone labs aimed at non-engineers.

Which model should I pick — Opus, Sonnet, or Haiku?+

Default to Sonnet. It hits 90% of use cases at a fraction of Opus cost. Use Opus only for the hard stuff: long-horizon planning, complex multi-step reasoning, and tasks where one wrong step ruins the output. Use Haiku for high-volume structured tasks: JSON extraction, classification, ticket triage. The Bible has a decision tree on page 31 you can screenshot.

What's CLAUDE.md and do I need one?+

Yes, this is the single biggest improvement most people miss. CLAUDE.md is a file you drop in your project root that tells Claude Code how your codebase works: your stack, your conventions, your test commands, files to never touch. Claude Code reads it automatically on every session. It's the difference between Claude getting up to speed in 30 seconds vs. asking you the same setup questions 50 times.

What's the difference between Skills, Hooks, MCP, and Sub-agents?+

Skills are markdown files that tell Claude what to do (e.g., 'when the user asks for a PR review, do these 5 things'). MCP servers are external tools Claude can call (your database, Slack, a custom API). Hooks are scripts that fire automatically at lifecycle events (e.g., 'always run the linter before committing'). Sub-agents are isolated Claude instances that handle heavy work in their own context window. Part 3 of the book has 1-page rulebooks for each.

How much does Claude Code cost?+

Subscription path: Claude Pro $20/mo gets you light usage, Max $100 covers most professional work, Max $200 is unrestricted. API path: pay per token (Sonnet around $3-15 per million tokens depending on cache). Real-world: most operators run on Max $100 and stay well under the 5-hour token caps. The book has a cost-control chapter with prompt caching tactics that drop bills 50-90%.

Can Claude actually run parts of my business?+

Yes, with guardrails. Realistic wins for week 1: meeting notes → action items, multi-source weekly reports, ticket triage, lead enrichment, content repurposing. Not realistic: handing it a vague goal and walking away. Claude is a junior dev that's read everything but has no memory of your specific business. The book's capstone labs are the patterns that survive contact with reality.

Why does Claude sometimes refuse to do what I ask?+

Two main reasons. One: it interpreted your request as potentially harmful (legal advice, medical advice, anything sketchy). Fix: clarify intent and context — 'this is for my own personal use,' 'this is sample data,' etc. Two: it's uncertain and being cautious — sometimes false refusals happen on benign requests. Fix: rephrase, add context about who you are and why you need it, or switch to a stronger model. Don't argue with the refusal, just give it more context.

What's the safest way to use Claude with company data?+

Three rules. (1) Never paste production credentials, customer PII, or unredacted financials into claude.ai — use the API or Claude Code with explicit data-handling rules. (2) Use Claude Code's permission system to restrict file access (Read-only mode for sensitive folders). (3) Run a sub-agent for anything touching sensitive data so its output is isolated. The Bible's Production and Safety section walks through Anthropic's enterprise data policies.