The Claude Code handbook you actually finish. Bilingual (EN/FR), free. Hooks, sub-agents, skills, plan-first dev, audit loop. For developers and knowledge workers. Le manuel Claude Code que tu finis vraiment. Bilingue (EN/FR), gratuit.
Latest version: V2 (May 2026), 52 pages, 14 chapters in 5 parts.
| English | Français | |
|---|---|---|
| Read online | themiryon.github.io/claude-code-handbook/en/ | themiryon.github.io/claude-code-handbook/fr/ |
| EPUB (V2) | en/claude-code-handbook-v2.epub |
fr/le-code-du-claudeur-v2.epub |
- Full chapter on skills: the third primitive of Claude Code, alongside hooks and sub-agents (Chapter 06)
- Plan-first development with a
plan-reviewersub-agent acting as staff engineer (Chapter 07) - The audit loop: capture lessons in Git-versioned files instead of local-only memory (Chapter 08)
- Parallel worktrees for running multiple Claude sessions side by side (Chapter 10)
- The memory matrix: clear rules for the 4 memory locations (Chapter 11)
- Two anonymized case studies: a 6-month solo developer + a 3-month knowledge worker (Annex H)
- Glossary for newcomers (Annex F)
- Reading paths and difficulty badges so you can read the book in the order that fits you
- Stricter print design: "1 concept = 1 page" where possible, no orphan headings
See CHANGELOG.md for the full change log.
The guide covers 14 chapters in 5 parts. If you want to jump straight in, Annex G has a "30-minute quick start" checklist.
Or skip ahead and just copy the templates below.
All the configuration files described in the guide, ready to drop into any project. Adapt paths, project names, and stack to your context.
templates/
├── CLAUDE.md ← the project brain (root of your repo)
├── .gitignore.sample ← lines to add to your .gitignore
├── .agents/
│ └── skills/
│ └── conventional-commits/ ← example skill (SKILL.md + references/), NEW in V2
└── .claude/
├── settings.json ← permissions + hooks
├── COMMANDS.md ← optional: index of your slash commands
├── PATTERNS.md ← optional: copy-paste recipes
├── hooks/
│ ├── pre-tool-guard.sh ← defensive: block rm -rf, .env writes, --force pushes
│ ├── post-edit-format.sh ← auto-format after Write/Edit
│ ├── session-start.sh ← project recap at session open
│ ├── activity-log.sh ← zero-token activity log
│ ├── coach-suggest.sh ← Coach mode (Stop event suggestions)
│ └── extract-lesson.sh ← audit-loop hook, NEW in V2
├── agents/
│ ├── code-auditor.md ← read-only code quality audit
│ ├── security-auditor.md ← OWASP / GDPR / secrets audit
│ └── plan-reviewer.md ← staff-engineer plan review, NEW in V2
├── commands/
│ ├── ship.md ← verify + commit + push in one line
│ ├── audit-quick.md ← code-auditor + security-auditor in parallel
│ ├── standup.md ← morning recap
│ ├── coach.md ← "what's the next step?"
│ ├── coach-mute.md ← turn off Coach suggestions
│ ├── coach-on.md ← turn back on
│ ├── new-feature.md ← plan-first pipeline, NEW in V2
│ └── extract-lesson.md ← capture a lesson to CLAUDE.md, NEW in V2
└── agent-memory/
└── README.md ← cross-session memory pattern
From your project root:
# 1. Copy the templates
cp -r /path/to/this/repo/templates/CLAUDE.md ./
cp -r /path/to/this/repo/templates/.claude ./
cp -r /path/to/this/repo/templates/.agents ./ # NEW in V2: skills
# 2. Append .gitignore additions
cat /path/to/this/repo/templates/.gitignore.sample >> .gitignore
# 3. Make hooks executable
chmod +x .claude/hooks/*.sh
# 4. Customize CLAUDE.md with your project specifics
$EDITOR CLAUDE.mdWindows users: run these commands from git-bash, WSL, or any POSIX shell. The native PowerShell terminal doesn't have
cp -r,cat >>, orchmod; the install steps assume a Unix-style shell. (Once installed, Claude Code itself runs fine on Windows.)
Then launch Claude Code from your project, it picks up everything automatically.
After running the templates and reading the guide:
- Permanent project context: Claude loads
CLAUDE.mdat every session, you stop re-explaining your stack - Defensive hooks: destructive commands (
rm -rfon dangerous paths,--forcepushes,.envwrites) blocked before execution - Auto-format: files formatted automatically after each edit, zero friction
- Session recaps: git state shown when you open a session, no need to type
git status - Specialized sub-agents: security, code, and plan-review audits in their own context
- Slash commands:
/ship,/audit-quick,/standup,/new-feature,/extract-lesson, etc. - Coach mode: suggests the right command at the right moment, without polluting the conversation
- Versioned domain knowledge via the skill system (NEW in V2)
- Plan-first pipeline with second-opinion review (NEW in V2)
- Lesson-capture loop that survives reclones (NEW in V2)
- Solo devs who code with Claude daily
- Teams who want consistent behavior across machines
- Knowledge workers (researchers, writers, analysts), V2 extends explicitly beyond devs
- Beginners moving past "vibe coding" toward a reproducible workflow
The guide and templates apply to all 6 Claude Code surfaces:
- Terminal CLI
- VS Code extension (also works in Cursor)
- JetBrains plugin
- Desktop app (macOS / Windows)
- Web (
claude.ai/code) - iOS app
For the Claude.ai chat product (not Claude Code), only the CLAUDE.md concept transfers via "Projects".
See BUILD.md for the Pandoc-based EPUB pipeline.
Found a factual error, a typo, or a setup that fails in your environment? Open an issue or a PR. This is a living document.
MIT. See LICENSE. Use, share, modify, sell; just keep the copyright notice.
Built by @TheMiryon. Inspired by the Claude Code crash course by Mayank and Anthropic's official documentation.