Claude Code Guide

---

Transparency note: This guide reflects my personal experience after several months of daily Claude Code usage. I'm sharing what I've learned, not claiming expertise. The tool evolves constantly, and so does my understanding. Feedback welcome.

---

Privacy Notice: Claude Code sends your prompts, file contents, and MCP results to Anthropic servers.

• Default: 5 years retention (training enabled) | Opt-out: 30 days | Enterprise: 0
• Action: Disable training | Full privacy guide

---

Get Started in 60 Seconds

Option A: One-liner (no clone needed)

claude "Fetch and follow the onboarding instructions from: https://raw.githubusercontent.com/FlorianBruniaux/claude-code-ultimate-guide/main/tools/onboarding-prompt.md"

📸 Preview: Interactive Onboarding Flow

1. Choose your language	2. Select your goal
3. Pick your experience level	4. Set available time

Option B: From cloned repo

1. Copy the prompt from tools/onboarding-prompt.md
2. Run claude → Paste → Enter

Claude asks 2-4 questions then guides you through personalized content based on your profile.

Or browse directly: Cheat Sheet | Quick Start | Full Guide

---

5 Rules to Remember

1. Review diffs before accepting changes
2. /compact at 70% context usage (check with /status)
3. Be specific: WHAT + WHERE + HOW + VERIFY
4. Plan Mode first for risky/complex tasks (Shift+Tab×2)
5. Create CLAUDE.md for every project

---

Choose Your Path

I want to...	Go here
Learn step by step	Run the onboarding command above
Find something specific	Guide Navigation
Test my knowledge	Quiz
Fix a problem	Troubleshooting

---

Why This Guide?

Installation, agents, MCP servers, hooks, skills, and CI/CD integration—documented through several months of daily practice. A structured learning journey sharing what I've learned so far.

By Florian BRUNIAUX | Founding Engineer @Méthode Aristote | Connect on LinkedIn

The Problem: Most Claude Code resources are scattered blog posts or dense reference manuals. You're left piecing together workflows and discovering critical concepts too late.

This guide: A structured learning journey with ready-to-use examples. Your mileage will vary—the goal is to save you the exploration time I've already spent.

Traditional Docs	This Guide
Lists features	Teaches workflows
Reference lookup	Progressive learning
Theoretical concepts	Production-ready patterns
"Figure it out"	"Here's exactly how"

Reading time: The Quick Start takes ~15 minutes. Full guide is ~3 hours but most people read by section as needed.

If this guide saves you hours of trial-and-error, please star it.

Prerequisites

• Node.js 18+ (or use the shell installer on macOS/Linux)
• Anthropic API key — Get one here

---

⚡ Audit Your Setup

Already have Claude Code installed? Quickly scan your configuration:

curl -sL https://raw.githubusercontent.com/FlorianBruniaux/claude-code-ultimate-guide/main/examples/scripts/audit-scan.sh | bash

Instant checks: Config files, tech stack, extensions, security hooks, MCP servers, CLAUDE.md quality

Want deeper analysis? Use tools/audit-prompt.md for personalized recommendations (~3 min).

🤖 LLM Reference

Give any AI assistant instant Claude Code expertise (~2K tokens):

curl -sL https://raw.githubusercontent.com/FlorianBruniaux/claude-code-ultimate-guide/main/machine-readable/reference.yaml

Use cases: Paste output into ChatGPT/Claude/Gemini, add to system prompts, or reference in Claude Code with @machine-readable/reference.yaml

What's inside: Decision trees, command reference, context zones, MCP servers, agent templates, troubleshooting—optimized for machine consumption. Points to line numbers in the full guide for deep dives.

🔬 Deep Audit (Personalized Recommendations)

Get a comprehensive, context-aware audit that analyzes your project's README, CLAUDE.md files, and business domain to provide tailored recommendations:

🔒 Privacy: The audit downloads reference files from this repo, then analyzes YOUR local files with your Claude CLI. Your project files are sent only to your Anthropic API endpoint, not to this repository or any third party.

Quick Version (~10 sec):

curl -sL https://raw.githubusercontent.com/FlorianBruniaux/claude-code-ultimate-guide/main/examples/scripts/audit-scan.sh | bash -s -- --json | claude -p "Analyze this Claude Code setup. Give: 1) Health score 0-100 2) Top 3 quick wins 3) CLAUDE.md template for detected stack. Be concise."

Full Audit with Context (~30 sec, recommended):

# Claude Code Deep Audit - Context-Aware Version
# Downloads reference files, reads YOUR local files, analyzes with Claude
REF=$(curl -sL https://raw.githubusercontent.com/FlorianBruniaux/claude-code-ultimate-guide/main/machine-readable/reference.yaml)
SCAN=$(curl -sL https://raw.githubusercontent.com/FlorianBruniaux/claude-code-ultimate-guide/main/examples/scripts/audit-scan.sh | bash -s -- --json 2>/dev/null)
README_CONTENT=$(head -100 README.md 2>/dev/null || echo "No README.md found")
CLAUDE_MD=$(cat CLAUDE.md 2>/dev/null || echo "No CLAUDE.md found")
LOCAL_CLAUDE_MD=$(cat .claude/CLAUDE.md 2>/dev/null || echo "No .claude/CLAUDE.md found")

claude -p "Reference:
$REF

Scan results:
$SCAN

Project README (first 100 lines):
$README_CONTENT

Project CLAUDE.md:
$CLAUDE_MD

Local .claude/CLAUDE.md:
$LOCAL_CLAUDE_MD

IMPORTANT: Focus on INCREMENTAL improvements to existing setup. Don't suggest creating things that already exist. If CLAUDE.md exists, suggest specific improvements to it rather than a full rewrite.

Based on ALL this context, provide:
1. Stack Recap: runtime, framework, test runner, bundler, database, key integrations detected
2. Health Score (0-100) - be strict: penalize missing SSoT pattern if >100 lines without @refs
3. Findings table: Priority|Element|Status|Action (only gaps, not what exists)
4. Top 3 quick wins (<5 min) - MUST be specific to THIS project's domain (not generic advice)
5. If CLAUDE.md exists: list 3-5 specific improvements (not a full template). If missing: provide ~100 line template
6. Suggested agents/commands/hooks that DON'T duplicate existing ones - check extensions count first
7. Ideas to leverage Claude Code for this specific domain and detected integrations"

What you get:

• Stack recap: Runtime, framework, test runner, bundler, database, and key integrations auto-detected
• Strict health score (penalizes large CLAUDE.md without @refs)
• Incremental improvements: Specific fixes for YOUR setup, not generic advice
• Domain-aware suggestions (e.g., EdTech → session planning agents, E-commerce → inventory commands)
• Non-duplicate suggestions: Only recommends agents/comma