1 / 11
← → Space · N for notes
Speaker Notes
Claude Code in Practice — Course Overview

Ship with
Claude Code

Skills · Sub Agents · Hooks · Multi-Agent Systems
Explained for product managers who want to build.

Claude Code Skills Sub Agents Hooks Multi-Agent
The Foundation

Claude Code is an AI pair programmer
that lives in your terminal

Not a chatbot. Not a plugin. A full agentic CLI that reads files, runs commands, and ships real things — with your approval at every step.

📂
Reads & edits your whole repo
Understands context across files, git history, and error logs — not just the snippet you paste in.
⚡
Runs commands & tests
Executes shell commands, runs your test suite, and fixes failures — all inside one conversation.
🚀
Ships with your approval
Every file write and command runs only after you confirm it — you stay in control of what deploys.
Why This Matters

PMs can now ship in hours what used
to take engineering weeks

Before Claude Code
  • Write a ticket and wait for sprint
  • Back-and-forth on requirements
  • Engineering builds, PM reviews
  • Bugs discovered after deploy
  • Iteration takes days or weeks
After Claude Code
  • Describe what you want in plain English
  • Claude reads your codebase & builds it
  • You review & approve each change
  • Errors fixed in the same session
  • Ship the same day you have the idea
01
Concept 01 — Skills

Skills are reusable AI workflows triggered by a slash command

Define a workflow once in a markdown file. Run it forever with /skill-name.

How they work
Each skill is a SKILL.md file in .claude/skills/ that describes the goal, steps, and output format. Claude executes those instructions exactly.
Why they matter
Encode your team's best practices once. Every PM runs the same high-quality workflow — no prompt engineering needed each time.
Example trigger
Type /prd-generator and Claude gathers requirements, writes a full PRD, and asks clarifying questions — in one command.
Skills — Examples

One command covers your entire PM workflow

prd-generator
Turn a feature idea into a structured PRD with acceptance criteria
competitor-research
Research 3–5 competitors with positioning, pricing & gaps
test-writer
Write unit tests for any function — happy path, edge cases, errors
frontend-slides
Generate animation-rich HTML presentations from any topic
deploy-checklist
Pre-deploy checklist before pushing to Vercel or Fly.io
stakeholder-update
Write concise weekly updates from raw notes in seconds
02
Concept 02 — Sub Agents

Sub Agents let Claude
delegate work
to specialized AI workers

Define specialist agents in .claude/agents/. The orchestrator spawns them, they execute, it synthesizes.

Specialization
Each sub agent has a specific role: one reads docs, one writes code, one runs tests. Better results than a generalist doing everything.
Isolation
Sub agents get their own context window — they don't pollute the orchestrator's session with long intermediate results.
Tool scoping
Each agent can be given only the tools it needs. A research agent reads web pages. A code agent edits files. No scope creep.
03
Concept 03 — Hooks

Hooks fire shell commands automatically
when Claude takes specific actions

Configure in .claude/settings.json. Automate quality checks, logging, and notifications around every Claude action.

PreToolUse
Fires before Claude uses any tool. Can block the action. Use to enforce policy — e.g. prevent editing production files.
PostToolUse
Fires after every tool call completes. Use to auto-run linters, formatters, or log audit trails after each file edit.
Stop
Fires when Claude finishes a task. Use to send a Slack ping, play a sound, or trigger a downstream process automatically.
Notification
Fires when Claude pauses for approval. Use to alert you via email or push notification so you don't miss a waiting session.
04
Concept 04 — Multi-Agent Systems

Multi-Agent orchestrates fleets of AI workers in parallel

One orchestrator. Many specialists. Tasks that would take hours serially complete in minutes in parallel.

Parallelism
Multiple agents run simultaneously — a research agent, a writing agent, and a review agent all work at once while the orchestrator coordinates.
Scale beyond context
Large tasks are split across agents so no single context window is overloaded. Each agent handles a bounded sub-problem.
Real-world example
Competitor research agent + pricing analyst + positioning writer → finished competitive analysis in one command, not one week.
The Full Picture

Skills + Sub Agents + Hooks + Multi-Agent
stack into production-grade AI workflows

Multi-Agent
Orchestrator coordinates multiple specialized agents in parallel. Handles tasks too large for a single context window.
Sub Agents
Specialist Claude instances — each with a defined role, scoped tools, and isolated context. Spawned by the orchestrator.
Skills
Reusable workflows that encode PM best practices. Used as building blocks by agents or triggered directly with /commands.
Hooks
Shell commands that fire on tool events — auto-formatting, logging, notifications, and policy enforcement across every layer above.
Real-World Output

Real PMs are shipping tools, dashboards
& agents — without engineers

Competitor Research Agent
Multi-agent system that researches 5 competitors, extracts pricing and positioning, and delivers a formatted report.
Built in 2 hours
AI Meeting Summarizer
Full-stack web app — paste a transcript, get action items, decisions, and risks. Deployed and shareable via URL.
Built in 1 session
Video Content Pipeline
Automated Remotion pipeline that turns a 1-line prompt into a narrated Instagram Reel explainer video.
Built in 1 day
PRD + Story Generator
Skill chain that goes from feature idea → PRD → user stories → acceptance criteria → JIRA-ready tickets.
5 min per feature
Getting Started

Start shipping with Claude Code
this week

1
Install Claude Code & run your first command
Install via npm install -g @anthropic-ai/claude-code. Open your project. Type claude. Ask it to explain your codebase.
2
Run a skill on a real PM problem you have today
Pick any skill from this course repo — /competitor-research, /prd-generator, /stakeholder-update. Use it on an actual problem.
3
Build one multi-agent workflow for your team
Define an orchestrator and two specialist sub agents. Pick a recurring research or documentation task. Automate the whole thing.
maven.com/boring-bot/claude-code-in-practice