Wahrwelt Notes

quickstart-claudecode-pdf

4 min read

Claude Code Quickstart From The Slides

Fast path

If you only care about getting productive quickly, do this:

  1. Read Best Practices.
  2. Understand the Feature Overview Map.
  3. Set up CLAUDE.md.
  4. Set up Permissions.
  5. Give Claude useful CLI tools.
  6. Add Hooks and Skills once your workflow repeats.
  7. Use Subagents and MCP when the work gets bigger.
  8. Leave Plugins and Agent Teams for later.

1. Best Practices

TL;DR

  • Claude Code works best as an agent, not just a chatbot.
  • The practical setup order is: context, permissions, tools, verification.
  • The highest-value habits are: let it verify its own work, give it rich context, and configure the environment instead of re-explaining things every session.
  • The most important workflow hint is: let Claude explore first, then plan, then code.

2. Feature Overview Map

TL;DR

  • CLAUDE.md, Permissions, and Hooks are the control layer.
  • Skills, Subagents, MCP, and Plugins are the extension layer.
  • Start with the control layer first.
  • Add extension features only when your basic setup is already working.

3. CLAUDE.md

TL;DR

  • CLAUDE.md is the main place to put persistent instructions.
  • Use it for repo conventions, architecture rules, common commands, testing expectations, and domain context.
  • Project and team rules belong here before you reach for more advanced features.
  • Child-directory CLAUDE.md files are useful when only part of the repo needs special instructions.
  • If you are just starting, this is the single highest-leverage thing to set up well.

4. Permissions

TL;DR

  • Permissions decide whether Claude can execute, has to ask, or gets blocked.
  • The order that matters is Deny > Ask > Allow.
  • This is how you make Claude useful without letting it run wild.
  • Set permissions early so normal safe actions are smooth and risky actions still stop.

5. Hooks

TL;DR

  • Hooks are for deterministic automation around lifecycle events.
  • Use hooks when you want something to happen every time, not just when Claude remembers.
  • Good early uses are formatting, validation, quality gates, protected-file checks, and notifications.
  • CLAUDE.md is advice; hooks are enforcement.

6. Skills

TL;DR

  • Skills are reusable workflows loaded only when needed.
  • They are better than stuffing everything into CLAUDE.md.
  • Use skills for repeated tasks like PR review, deploy steps, release flow, debugging playbooks, or onboarding.
  • If you find yourself pasting the same instructions over and over, that should probably become a skill.

7. Subagents

TL;DR

  • Subagents are for splitting work into isolated specialist tasks.
  • Use them when one agent doing everything starts getting messy.
  • Good use cases are code review, research, frontend/backend splits, and bounded investigations.
  • This is usually the first advanced feature worth using after CLAUDE.md, permissions, hooks, and skills are already solid.

8. MCP

TL;DR

  • MCP is the standard way to connect Claude Code to external services.
  • Use it when Claude needs structured access to tools, resources, or prompts from outside the repo.
  • This matters once your workflow depends on systems like GitHub, databases, Slack, Notion, Sentry, or internal APIs.
  • If you do not yet need that, skip it for now.

9. CLI Tool Integration

TL;DR

  • CLI tools are often the fastest and simplest external integration.
  • If you already use gh, aws, gcloud, or sentry-cli, teach Claude those before building heavier integrations.
  • A very practical pattern is to tell Claude to inspect foo-cli --help.
  • For many people, good CLI access matters before MCP does.

10. Plugins Later

TL;DR

  • Plugins bundle multiple Claude Code features into something version-controlled and shareable.
  • This is useful for teams, not for day-one setup.
  • Care about plugins when local config starts turning into team infrastructure.

11. Agent Teams Later

TL;DR

  • Agent Teams are for coordinating multiple Claude Code instances.
  • This is explicitly experimental in the deck.
  • Most people should ignore this until they already know why one main agent plus subagents is not enough.

12. Subagents vs Agent Teams

TL;DR

  • Use Subagents for one-to-many delegation inside one main session.
  • Use Agent Teams for broader coordination across multiple parallel agents.
  • For a beginner/intermediate setup, Subagents are the thing to learn; Agent Teams are optional.

What To Actually Set Up First

If you want the shortest useful setup checklist:

  1. Write a real CLAUDE.md.
  2. Configure permissions so safe commands flow and risky ones stop.
  3. Make your normal tools available.
  4. Give Claude a way to verify success with tests, screenshots, or expected output.
  5. Add hooks for repeatable checks.
  6. Turn repeated workflows into skills.
  7. Use subagents when task complexity starts to justify them.

Ignore For Now

If you are still getting comfortable with Claude Code, you can safely postpone:

  • Plugins
  • Agent Teams
  • deep MCP setup you do not already need

Files

  • Note: derekszen/quickstart-claudecode-pdf.md
  • Images: derekszen/assets/quickstart-claudecode-pdf/

Graph View