Wiki.Quizz.vn

This page is read only. You can view the source, but not change it. Ask your administrator if you think this is wrong.
====== Claude Code Mastery Roadmap ======

===== Overview =====

  * Level 0 → Basic Prompting
  * Level 1 → Structured Prompt Engineering
  * Level 2 → Rules & Skills Layer
  * Level 3 → MCP & Tool Use
  * Level 4 → Agent Architecture
  * Level 5 → RAG for Large Codebases
  * Level 6 → CI/CD Automation
  * Level 7 → Multi-Agent Systems
  * Level 8 → Production AI Engineering Platform

----

===== Level 0 — Basic Prompting =====

**Goal:** Use Claude effectively in daily coding tasks.

**Learn:**
  * Role prompting
  * Clear constraints
  * Output formatting
  * Step-by-step reasoning
  * Diff generation

**Keywords:**
  * Prompt engineering
  * Zero-shot vs Few-shot
  * Chain-of-thought
  * Structured output
  * Markdown diff

----

===== Level 1 — Structured Prompt Engineering =====

**Goal:** Make prompts predictable and reusable.

**Learn:**
  * Prompt templates
  * Role / Task / Constraints / Output format structure
  * Self-critique loops
  * Persona switching
  * JSON output enforcement

**Keywords:**
  * Meta-prompt
  * Output schema
  * Deterministic formatting
  * Guardrails
  * Prompt injection awareness

----

===== Level 2 — Rules & Skills Layer =====

**Goal:** Create reusable AI capabilities.

**Learn:**
  * Global system rules
  * Skill abstraction
  * Prompt libraries
  * Code-style enforcement
  * Domain-specific DSL prompts

**Keywords:**
  * System prompt
  * Skill templates
  * Reusable instruction blocks
  * Behavioral constraints
  * AI coding standards

----

===== Level 3 — MCP & Tool Use =====

**Goal:** Let Claude interact with your codebase.

**Learn:**
  * Model Context Protocol (MCP)
  * Tool calling
  * Filesystem access
  * Git integration
  * Context window management

**Keywords:**
  * MCP
  * Tool invocation
  * Function calling
  * Agent tool loop
  * Context control

----

===== Level 4 — Agent Architecture =====

**Goal:** Build Claude as a reasoning agent.

**Learn:**
  * Observe → Plan → Act → Reflect loop
  * Multi-step reasoning
  * File search orchestration
  * Error recovery loops
  * Iterative refinement

**Keywords:**
  * ReAct pattern
  * Agent loop
  * Reflection prompting
  * Autonomous task execution
  * Task decomposition

----

===== Level 5 — RAG for Large Codebases =====

**Goal:** Scale Claude to large repositories.

**Learn:**
  * Embeddings
  * Vector databases
  * Chunking strategies
  * Retrieval ranking
  * Code summarization

**Keywords:**
  * Retrieval-Augmented Generation (RAG)
  * Vector DB (FAISS, Pinecone, Weaviate)
  * Similarity search
  * Context compression
  * Hybrid search

----

===== Level 6 — CI/CD & Automation =====

**Goal:** Integrate Claude into development lifecycle.

**Learn:**
  * Git hooks
  * PR review bots
  * GitHub Actions
  * Structured diff comments
  * AI quality gates

**Keywords:**
  * AI code review
  * Pre-commit hooks
  * Pull request summarizer
  * CI automation
  * Static analysis augmentation

----

===== Level 7 — Multi-Agent Systems =====

**Goal:** Multiple AI roles collaborating.

**Example Agents:**
  * Architect Agent
  * Refactor Agent
  * Test Generator Agent
  * Security Agent
  * Performance Agent

**Keywords:**
  * Multi-agent orchestration
  * Task planner agent
  * Tool router
  * Hierarchical agents
  * Cooperative AI

----

===== Level 8 — Production AI Engineering Platform =====

**Goal:** Build internal AI coding infrastructure.

**Learn:**
  * Token budgeting
  * Cost optimization
  * Logging & observability
  * Output evaluation metrics
  * Human-in-the-loop systems

**Keywords:**
  * AI observability
  * Token efficiency
  * Acceptance rate metrics
  * Prompt versioning
  * Model fallback strategy

----

===== Advanced Topics =====

  * AST-based code modification
  * Tree-sitter parsing
  * Semantic diff generation
  * LLM hallucination detection
  * Spec-driven development
  * AI security hardening
  * Sandboxed execution

----

===== Recommended Learning Timeline =====

**Month 1**
  * Level 0–1
  * Structured prompts + JSON output

**Month 2**
  * Level 2
  * Git hook integration

**Month 3**
  * Level 3–4
  * Agent loop + MCP

**Month 4**
  * Level 5
  * RAG for your repository

**Month 5**
  * Level 6+
  * CI automation + Metrics

----

===== Final Mindset =====

Claude Code is not just AI that writes code.

It becomes a programmable reasoning engine integrated into your engineering workflow.

Serious power begins at Level 4 and above.