Claude Code Ecosystem
How do MCP, skills, plugins, hooks, and subagents work together?
Combine tools, rules, and reusable capabilities.
📌 Learning goals
- Explain what CLAUDE.md, Skill, MCP, Hook, Plugin, and Subagent each do.
- Choose the smallest useful component instead of making simple work complex just to look sophisticated.
- Create a shareable, inspectable, safe-by-default Claude Code project configuration.
- Know when Claude Code alone is enough, and when a worktree or the Claude Agent SDK is warranted.
Entry conditions
Be comfortable driving Claude Code (or any CLI agent) with a throwaway practice repo; read the quickstart before starting.
🧭 Lessons on this site
Read in the suggested order; checkboxes share the same browser progress as the /learn track pages.- 01What Is Claude Code: The Reference Template for Terminal Agents
The flow of one real session explains what Claude Code is: a coding agent that lives in your terminal, reading files, editing files, running commands and iterating until the task is done. With installation methods, the CLAUDE.md memory mechanism, form factors beyond the terminal, the billing model, and the three kinds of people who should not start with it.
9 min - 02Installing Claude Code: A Getting-Started Path for the Terminal Agent
Anthropic's official terminal coding agent Claude Code, from system requirements, four install methods (native install script, Homebrew, WinGet, npm), the verification command, and browser sign-in, to permission modes and your first change. Every step spells out the output you should see and the real error messages listed in the official docs.
19 min - 03A Deep Dive into Claude Code
From installation and configuration to advanced techniques, fully master the Claude Code CLI tool, actually powered by DeepSeek V4 / ModelStudio, to boost everyday development efficiency.
5 min - 04MCP (Model Context Protocol) Ecosystem Explained: The Three Core Primitives of Tool, Resource, and Prompt
A complete guide to Anthropic's MCP protocol: the design principles behind Tools, Resources, and Prompt Templates, practical configuration, and Claude Code integration.
10 min - 05The Role and Configuration of the MCP Protocol in a Three-Layer Agent Framework
The respective roles and configuration of the Model Context Protocol in Hermes Agent (not supported), OpenClaw (orchestration layer), and Claude Code (execution layer).
5 min - 06Complete Guide to the Slash Command System
From trigger points to real-world use: an in-depth look at the design philosophy behind 30+ slash commands, the reasons they were created, practical scenarios, and how to use this system to greatly boost work efficiency.
14 min - 07A Complete Tutorial for Matt Pocock Skills: The 115K Star AI Coding Engineering Discipline System, a Full Walkthrough of 15 Skills from Installation to Practice
mattpocock/skills: 35 Agent Skills released by TypeScript legend Matt Pocock from his own .claude directory, with 115K+ Stars, 2.7M+ installs, and 60K newsletter subscribers. This is not vibe coding; it is an AI coding discipline system condensing decades of software engineering experience.
24 min - 08In-Depth Analysis of Addy Osmani's agent-skills Open Source Project
A complete introduction to Google Chrome Engineering Director Addy Osmani's agent-skills project: the design philosophy behind 21 AI-assisted development skills, real-world applications, and how it inspired the upgrade of our slash command system.
10 min - 09Build Complete Infrastructure for Your AI Agent in Half an Hour: The Complete Agentic Infrastructure Ten-Piece Guide
From zero to fully running: ten prompts to establish a gate pair, vector memory, skill curation, and scheduled inspection. Solves the skill-skipping problem caused by LLM confidence bias, so your Agent never loses its memory again.
22 min
📚 Required reading
- 1.Claude Code Quickstart⭐⭐⭐⭐⭐Install and open your first session.
- 2.Extend Claude Code（官方零件總表）⭐⭐⭐⭐⭐One official table separating CLAUDE.md, Skill, MCP, Hook, Plugin, and Subagent.
- 3.How Claude remembers your project⭐⭐⭐⭐The boundary between CLAUDE.md, rules, and auto memory.
🎯 Curated resources
| Resource | Who it's for | Priority | Why |
|---|---|---|---|
Official docs Claude Code — Skills | Before Exercise 2 | ⭐⭐⭐⭐⭐ | The old .claude/commands/ still works; new teaching uses SKILL.md. |
Official docs Claude Code — Hooks | Before Exercise 3 | ⭐⭐⭐⭐⭐ | Events, inputs/outputs, and blocking rules. |
Official docs MCP Specification | When checking the protocol | ⭐⭐⭐⭐ | Check the dated version when consulting the spec. |
Official templates anthropics/skills | Want official skill templates | ⭐⭐⭐⭐⭐ | Track current Claude Code skill support. |
Official templates anthropics/claude-plugins-official | Study plugin structure | ⭐⭐⭐⭐⭐ | The first template for official plugins and marketplace structure. |
Community skills mattpocock/skills | Study short, work-oriented skills | ⭐⭐⭐⭐ | Watch how the community writes short, work-oriented skills. |
Community skills obra/superpowers | Learn TDD and planning skills | ⭐⭐⭐⭐ | See how TDD, debugging, and planning skills combine. |
Subagents wshobson/agents | Study agent-definition division of labor | ⭐⭐⭐⭐⭐ | Study naming and division of labor across many agent definitions; start with a few files and do not copy permissions. |
Subagents anthropics/claude-cookbooks | Read official notebooks | ⭐⭐⭐⭐⭐ | Read the official tool-use and orchestration notebooks. |
Agent SDK anthropics/claude-agent-sdk-python | Embedding an agent in your own program | ⭐⭐⭐⭐⭐ | Python SDK client, message types, and options; not the full Claude Code runtime source. |
MCP servers modelcontextprotocol/servers | Connecting a first MCP in Exercise 4 | ⭐⭐⭐⭐ | Officially maintained reference servers; audit permissions before connecting. |
Chinese walkthrough ZhangHanDong/harness-engineering-from-cc-to-ai-coding | Want a Chinese walkthrough | ⭐⭐⭐⭐ | A Chinese harness reading; verify facts against official docs. |
🛠 Hands-on practice (upstream)
Full exercises & starter codeSummaries from the upstream curriculum; full code, cost, and latency estimates live upstream.
- Exercise 1: a minimal project rulebook — a short CLAUDE.md with only Purpose / Do not / Verify / Deliver (under 200 lines).
- Exercise 2: turn a repeated flow into a Skill — a readme-check SKILL.md with disable-model-invocation to control timing.
- Exercise 3: add a logging-only Hook — every time Claude tries to write a file, append the event and tool name to a jsonl log.
- Exercises 4–5: connect MCP and package a plugin — pick one minimal connector from the official servers, then box your skills/hooks into a plugin.
- Main project: a safe Claude Code practice repo — add one component per exercise, and only move on after the previous one succeeds.
✅ Self-check
- Can you explain CLAUDE.md, Skill, MCP, Hook, Plugin, and Subagent in one sentence each?
- Have you finished at least the first three exercises without handing an unfamiliar script or high-privilege token to the agent?
- Can you say what different problems subagents and worktrees each solve?
- Can you say which category Claude Code, OpenRouter, OpenCode/Pi, and Ollama each belong to?
- Can you tell whether you need the CLI or truly need the Agent SDK?
Adapted from awesome-agentic-ai-zh (MIT, by Wenyu Chiou) v2026.09.23; links checked 2026-08-27. Stars mark learning priority (⭐⭐⭐⭐⭐ = you will get stuck without it), not popularity. MIT License · Curriculum structure last updated 2026-10-03. Content is still being filled in; lessons marked “in progress” are not live yet.