Course Overview: What you will learn in 15 chapters

Web mental model and MCP (Chapters 7–8)

• Describe the three layers of a modern web app: frontend, backend, AI backend and reason about where latency, failures, and data live
• Install Playwright MCP and extend Claude Code beyond its training data so it can browse live URLs, read real documentation, and research current sources
• Articulate what MCP is, how MCP servers register tools, and why an agent's usefulness depends on the tools it is connected to

Sub-agents and pipelines (Chapters 9–11)

• Distinguish Skills from sub-agents: when each runs in the main session vs. a fresh isolated context
• Build a research-agent that Claude delegates to automatically based on its description field
• Add a prd-reviewer and a read-only code-reviewer that return structured, actionable audits (risks rated HIGH/MED/LOW, edge cases, scope flags, line-level code findings)
• Chain three agents in one prompt to go from market question to reviewed PRD in roughly five minutes with no manual handoffs

Evaluation and deployment (Chapters 12–13)

• Build a ground truth table of ten real inputs with expected outputs and scoring criteria
• Run /skill-evaluator to move a Skill from "it seemed to work" to "it passes 8/10 on real cases" and read failures to find structural prompt issues
• Deploy a Claude-powered app to Vercel, set ANTHROPIC_API_KEY correctly in the deployment environment, and share a live URL with stakeholders for real feedback

Reflection and next steps (Chapters 14–15)

• Audit your own toolkit including CLAUDE.md, Skills, agents, evals, deployed app, and choose the next real problem to apply it to
• Identify which workflows in your role are worth encoding as Skills or sub-agents based on frequency and isolation needs
• Adopt a sustainable habit of using Claude Code on one small, real project per week instead of treating it as a one-off experiment

What you will have built by Chapter 15

• A populated CLAUDE.md that orients every future session
• A .claude/skills/ directory with /prd-generator, /skill-evaluator, and /deploy-checklist
• A .claude/agents/ directory with research-agent, prd-reviewer, and code-reviewer
• A reviewed PRD for a real feature idea, with risks flagged
• An evaluated Skill scoring 8/10 or better against a ground truth table
• A deployed application with a live URL that real users can try

How to use this course

Each lesson is short with 5 to 10 minutes to read and 15 to 30 minutes to complete the action. Do them in order. Skipping the CLAUDE.md, Skill, or agent setup lessons leaves later lessons running against an empty project. The setup is what makes everything after it feel useful.

When you finish, take the course assessment in the sidebar. It will show you which ideas stuck, which deserve a second pass, and which lessons are worth revisiting before you start applying any of this at work.

→ Browse the course repo → Ready to go deeper? Enroll on Maven (code: claudecode)