Agenda

What is Claude Code?
Getting to know concepts of Claude Code

Live Demo
Watch Claude Code build a REST API in real time

Skills, Agents & MCP
Extending Claude Code with custom skills, sub-agents, and integrations

Workflows
How you can use it, and tips for using it as a team

Pizza!

Hands-On
Try it out for yourself

What is Claude Code?

An agentic coding tool that lives in your terminal

• Reads, writes and runs your code autonomously
• Understands your entire project through context
• Uses tools — file system, terminal, browser, sub-agents
• You describe what, Claude figures out how

Context

The context window is Claude's working memory — every message, file and tool result fills it up

• Fills up → oldest information gets lost
• Clear between tasks → /clear gives a fresh start
• Sub-agents → get their own context, keeping yours clean
• CLAUDE.md → re-read on every new session, always "remembered"

CLAUDE.md

Your project's instruction manual — Claude reads it at the start of every session

# Project: Task API
Tech: Node.js, Express, PostgreSQL
Style: ESLint + Prettier, camelCase
Test: `npm test` (Jest)
Build: `npm run build`

• Generate with /init, refine as your project grows
• Better context = better output

Agents & Skills

Skill — a set of instructions Claude follows before doing a task
Agent — an autonomous worker that executes tasks

Skills

A skill is a markdown file with instructions Claude reads before working

• Project skills — live in your repo, shared with the team
• User skills — personal, apply across all your projects
• Built-in skills — come with Claude Code (PDF, Excel, etc.)

What to put in a skill

• Step-by-step workflows: "when deploying, always run tests first"
• Templates: "PR descriptions should follow this format"
• Conventions: "use factory pattern for test data"
• Guardrails: "never delete migration files"

Agents

Agents are autonomous Claude instances that work independently

Sub-agents

• Spawned by Claude during a task
• Own context window — keeps your main session clean
• Example: "research the best auth library" runs in isolation

Agent memory

• Agents can read/write to files to persist knowledge across sessions
• Useful for long-running projects — learnings don't get lost on /clear

Agent teams

A Team Lead breaks down your task and delegates to Teammates in parallel

• Each teammate is a full Claude Code session with its own context
• They coordinate via a shared task list and messaging system
• Teammates can challenge each other's work
• Monitor with Shift+Up/Down, Ctrl+T, and Enter

MCP — Model Context Protocol

An open protocol that connects Claude to external tools and services

• GitHub — read issues, create PRs, review code
• Slack — search messages, post updates
• Databases — query schemas and data directly
• Custom — wrap any API your team uses

Think of it as giving an LLM hands to reach into your toolchain

 
 
But there are things to consider: context

Prototyping

Validate ideas in minutes instead of days

• "I want to automate this repetitive task" → just throw together a little program
• "Will this API design actually work?" → build it and find out
• "Can we parse this file format?" → throw together a proof of concept
• "What would this architecture look like?" → scaffold it, poke at it, throw it away

The code doesn't need to be perfect — it needs to answer a question

Workflows

How to actually use this day-to-day

• Plan → clear → execute — let Claude plan first, then /clear and work with a fresh context
• Write good prompts — be specific, give examples, state what you don't want (and don't overthink!)
• Review everything — Claude is fast, but makes mistakes. Read the diffs, review the code.
• Git branching — let Claude work on a branch, review before merging (almost like normal work )
• Team conventions — agree on CLAUDE.md standards, shared skills, PR review rules