For developers who already use Claude Code and want to move from operating it to governing it, this course develops an engineering-agent mental model and practical skills for context management, layered CLAUDE.md and rules, Skills workflow design, Tools/MCP/Plugins, Hooks/Subagents/sandbox controls, Prompt Caching, verification loops, and long-term configuration maintenance.
Best for
Developers who already use Claude Code but are encountering context drift, ineffective rules, or tool misuse
Engineers who use Claude Code for real project changes, debugging, reviews, and automation
Technical leads designing CLAUDE.md, rules, Skills, Hooks, MCP, or Subagents for a team
Advanced users studying Prompt Caching, verification loops, and long-term Claude Code maintenance
Problems solved
Treating Claude Code like an ordinary chat tool, then losing control on complex engineering work
Allowing CLAUDE.md to grow while context becomes noisier and rules still fail to apply reliably
Confusing the boundaries among MCP, Tools, Skills, Hooks, Subagents, and Plugins
Expanding automation without matching controls for permissions, sandboxing, verification, and isolation
Allowing long sessions, model changes, compaction, and tool changes to destabilize cost, speed, and reliability
Outcomes
Explain Claude Code through an engineering-agent loop and identify the layer where a failure occurred
Choose MCP, Tools, Skills, Hooks, Subagents, or Plugins for the right scenario instead of adding tools blindly
Manage the context window with /context, /clear, /compact, HANDOFF.md, and Subagents
Place CLAUDE.md, rules, auto memory, Skills, and Hooks in distinct layers to reduce conflicts and context noise
Understand the main effects of Prompt Caching and define verifiable completion criteria for Claude Code tasks
Highlights
Starts from Claude Code's engineering runtime structure instead of a list of isolated tips
Progresses through context, rules, capability extensions, controls, and verification
Uses scenario judgments, counterexamples, boundaries, and practice-ready decision frameworks in every lesson
Uses the existing English Claude Code knowledge base for RAG support
Supports the move from advanced individual use to team-level Claude Code governance
All chapters
8 chapters
Lesson 1: Understand Claude Code as an Engineering Agent
Explain how Claude Code works through an engineering-agent loop of gathering context, taking action, and verifying results, then identify the layer where a failure most likely occurred.
Lesson 2: Distinguish MCP, Tools, Skills, Hooks, Subagents, and Plugins
Choose MCP, a Tool, a Skill, a Hook, a Subagent, or a Plugin for a real Claude Code scenario, then explain the choice and the risk of using the wrong mechanism.
Lesson 3: Context Engineering—Keep Useful Information Above the Noise
Identify sources of context pollution in Claude Code and choose among /context, /clear, /compact, HANDOFF.md, and Subagents to manage long sessions and noisy tasks.
Lesson 4: Decide What Belongs in CLAUDE.md, Rules, and Memory
Place long-term rules, path-specific rules, automatic memory, workflows, and mandatory checks in the right layers to reduce context noise and rule conflicts.
Lesson 5: Skills—Turn Experience into On-Demand Workflows
Decide whether a repeated task should become a Skill and draft a Skill with triggers, inputs, steps, outputs, stop conditions, and side-effect boundaries.
Lesson 6: Tools, MCP, and Plugins—Add Capability Without Confusing the Agent
Decide whether a Claude Code extension scenario needs a Tool, MCP, or a Plugin, then define clear boundaries for naming, output, errors, permissions, and distribution.
Lesson 7: Hooks, Subagents, and Sandboxing—Put Automation on Rails
Combine Hooks, permissions, sandboxing, Subagents, and worktrees for risky or noisy Claude Code tasks so automation risk and context pollution stay controlled.
Lesson 8: Prompt Caching, Verification Loops, and Long-Term Maintenance
Explain how Prompt Caching affects Claude Code cost, speed, and reliability, then define verifiable completion criteria and long-term configuration health checks.
