# maitriser-claude-code-formation-pedagogique-deep-research-2026-02

## Veille

Complete Claude Code training: 12 pedagogical modules on agentic coding - Deep Research

## Titre Article

Maîtriser Claude Code — Détail complet des 12 modules et ~60 leçons

## Date

2026-02

## URL

https://github.com/dgirard/fiches-veille/blob/main/docs/deep%20research/202602/Ma%C3%AEtriser%20Claude%20Code.md

## Keywords

Claude Code, agentic coding, pedagogical training, Socratic pedagogy, CLAUDE.md, MCP, Model Context Protocol, slash commands, permissions, sandboxing, CI/CD, headless mode, subagents, skills, hooks, context window, Git, Explore-Plan-Code workflow, Haiku, Sonnet, Opus, FSRS flashcards

## Authors

Deep Research Veille Interne

## Ton

**Profile**: Structured pedagogical curriculum, didactic and interactive register, progressive technical level (beginner to advanced)

**Description**: Complete training document designed according to a 5-step Socratic method (hook, guided exploration, dialogue, active practice, FSRS flashcard). The tone alternates between that of a supportive tutor who poses provocative questions to spark reflection and that of a technical expert who delivers precise knowledge. Analogies are pervasive (architect/bricks, an Airbus A380 to fetch bread, a desk with 200,000 sticky notes) to anchor abstract concepts. The target audience is the professional developer seeking to move from traditional coding to agentic orchestration.

## Pense-betes

- **Pedagogical structure**: 12 modules (A1-A12), ~60 lessons, 60 FSRS flashcards, estimated duration 3h-5h
- **Track A** covers Claude Code end to end: concepts → installation → conversations → commands → memory → security → context → Git → MCP → CI/CD → customization → best practices
- **5 steps per lesson**: Socratic hook (30s) → Guided exploration (90s) → Socratic dialogue (60s) → Active practice (60s) → Synthesis + Flashcard (30s)
- **Fundamental concepts**:
- Agentic coding = the developer describes the intent, the agent explores/plans/codes/tests
- Claude Code = terminal-first agent (not a chatbot), can read/write files, execute bash, manage Git
- Three models: Haiku (fast/economical), Sonnet (balanced/default), Opus (intelligent/complex)
- **Anthropic-recommended workflow**: Explore → Plan → Code (Plan mode via Shift+Tab)
- **4-level memory system**: Enterprise Policy > Project CLAUDE.md > Project Rules (.claude/rules/) > User CLAUDE.md
- **CLAUDE.md**: Git-versioned file providing the project's permanent context, created with /init, kept concise to avoid "fading memory"
- **Context management**: 200k-token window, /compact (summarize) vs /clear (clear), /cost and /context for monitoring
- **Multi-layer security**: 4 permission modes (Normal, Auto-accept edits, Plan, Bypass), OS sandbox (Seatbelt macOS, bubblewrap Linux), PreToolUse hooks, allow/deny in settings.json
- **MCP**: open standard for connecting Claude Code to external tools (GitHub, Brave Search, Context7, Sentry, Playwright)
- **Headless mode**: -p flag for non-interactive use, CI/CD integration via GitHub Actions, text/json/stream-json output formats
- **Skills vs slash commands**: commands = triggered manually (/command), skills = automatically activated by Claude when the description matches
- **Hooks**: deterministic scripts on 11+ events (PreToolUse, PostToolUse, Stop, Notification...), exit code 2 = block
- **5 professional patterns**: Explore→Plan→Code→Test→Review, Context priming, Iterative refinement, Model surfing, Session hygiene
- **Recommended paywall**: Modules A1-A4 free (basics), A5-A12 Pro (expertise)

## RésuméDe400mots

This document constitutes a complete pedagogical curriculum for mastering Claude Code, Anthropic's agentic CLI tool. Organized into 12 modules (A1 to A12) totaling approximately 60 lessons, it follows a rigorous progression from foundational concepts to advanced best practices, with an estimated duration of 3 to 5 hours.

**Foundations (A1)**: Agentic coding is a paradigm shift in which the developer moves from line-by-line writing to orchestrating AI agents. Claude Code is distinguished by its terminal-first approach and provides access to three models (Haiku, Sonnet, Opus) suited to different levels of complexity.

**Getting started (A2-A3)**: Installation is done with a single command (curl or npm). The conversational workflow relies on specific, contextualized prompts. Four golden rules of agentic prompting: be specific, provide context, state constraints, one objective at a time. The Explore → Plan → Code workflow is recommended by Anthropic for complex tasks.

**Commands and memory (A4-A5)**: The essential slash commands (/init, /clear, /compact, /cost, /model, /rewind) structure daily work. The 4-level hierarchical memory system (Enterprise Policy, project CLAUDE.md, Project Rules, User CLAUDE.md) allows context to be shared and customized. Auto Memory allows Claude to write its own notes between sessions.

**Security and context (A6-A7)**: Four permission modes (Normal, Auto-accept, Plan, Bypass) offer a control-autonomy spectrum. The OS sandbox (Seatbelt on macOS, bubblewrap on Linux) isolates commands at the kernel level. The 200,000-token context window requires active management via /compact and /clear.

**Integrations (A8-A10)**: Claude Code excels at Git management (intelligent commits, conflict resolution, automated PRs). MCP (Model Context Protocol) connects Claude Code to external tools (GitHub, Brave Search, databases). Headless mode (-p flag) enables CI/CD integration via GitHub Actions with budget control.

**Advanced customization (A11-A12)**: Custom slash commands (.claude/commands/), skills (automatically detected SKILL.md files), specialized subagents, and deterministic hooks (11+ events) enable extensive automation. The five professional patterns (Explore→Plan→Code→Test→Review, Context priming, Iterative refinement, Model surfing, Session hygiene) form the foundation of expert mastery.

Each lesson integrates a provocative Socratic hook, a guided exploration, a tutor-learner dialogue, a hands-on exercise, and an FSRS flashcard for spaced repetition.

## GrapheDeConnaissance

- Anthropic —a_créé→ Claude Code (TECHNOLOGIE, 0.99)
- Claude Code —est_instance_de→ agent terminal-first (CONCEPT, 0.99)
- Claude Code —utilise→ Haiku 4.5 (TECHNOLOGIE, 0.97)
- Claude Code —utilise→ Sonnet 4.5 (TECHNOLOGIE, 0.97)
- Claude Code —utilise→ Opus 4.5 (TECHNOLOGIE, 0.97)
- Claude Code —mesure→ 1 milliard $ revenus annualisés (MESURE, 0.95)
- Anthropic —recommande→ workflow Explore-Plan-Code (METHODOLOGIE, 0.98)
- CLAUDE.md —fait_partie_de→ système de mémoire hiérarchique (CONCEPT, 0.97)
- Enterprise Policy —fait_partie_de→ système de mémoire hiérarchique (CONCEPT, 0.96)
- MCP —permet→ connexion à outils externes (CONCEPT, 0.95)
- mode headless —permet→ intégration CI/CD (METHODOLOGIE, 0.94)
- pédagogie socratique —s_applique_à→ Claude Code (CONCEPT, 0.97)
- hooks —s_applique_à→ événements PreToolUse et PostToolUse (CONCEPT, 0.93)
- FSRS —permet→ mémorisation espacée des flashcards (CONCEPT, 0.92)

---
Canonical: https://www.thekb.eu/en/fiches/maitriser-claude-code-formation-pedagogique-deep-research-2026-02/