A Beginner's Tour of Claude: From Web Chat to Plugins

How This Post Works

Claude can be as simple as opening a website and asking a question. Or as deep as orchestrating subagents that read your code, run tests, and open pull requests.

This guide goes from one to the other in twelve levels. Each level is short and concrete. Stop reading whenever you’ve got what you need. Come back when you want more.

The further you go, the more “developer-y” it gets. Levels 1 to 5 are for everyone. Levels 6 onward assume you’re either a developer or curious enough to want to be.

---

What Is Claude?

Claude is an AI assistant made by Anthropic. It’s a family of models. Opus is the most capable. Sonnet is balanced. Haiku is fast.

You don’t usually have to pick. The product picks for you based on what you’re doing.

The only thing worth knowing up front: Claude has a knowledge cutoff date. It doesn’t know what happened last week. If your question depends on current information, tell it what you know or let it search.

---

claude.ai (Sometimes Called Claude Chat)

The simplest way to use Claude is to go to claude.ai in your browser. You type. It answers. That’s the whole thing.

This is what most people mean when they say “I use Claude.” It’s a chat interface with conversation history, file uploads, and a few power features.

The most valuable thing to learn here is how to ask. There’s a simple formula:

Context + Goal + Constraints.

• Context: who you are, what you’re working on
• Goal: what you actually want
• Constraints: format, length, tone, anything that has to be there or can’t be there

Generic prompts get generic answers. Specific prompts get useful ones.

---

The Apps

Same Claude, but as a real app on your machine or phone. There are first-party apps for Mac, Windows, iOS, and Android.

Conversations sync between web, desktop, and mobile. You can start a thread in your browser at lunch and continue it from your phone on the bus.

Why use the app instead of the website?

• Keyboard shortcuts. A floating shortcut to summon Claude from anywhere is genuinely useful.
• Context capture. Apps can grab what’s on your screen or read selected text more easily.
• Offline-friendly drafts. Your queue persists if your connection drops mid-thought.

If you use Claude every day, install the app. If you use it occasionally, the website is fine.

---

Claude Cowork

A regular chat is ephemeral. You start a conversation, drop in a file, get an answer, and when you close the tab the file is effectively gone. Next time you have to upload it again.

Cowork gives Claude a real workspace, a durable place where files actually live. Claude can read from it, write back to it, and pick up exactly where you left off the next time you open it. The work persists. The context persists. You stop re-uploading the same documents.

A few practical things this unlocks:

• Continuity across sessions. The notes, drafts, and reference docs you put into the workspace are still there next week.
• Iterating on real artifacts. Claude can edit a document in place rather than handing you a new copy each turn.
• A single source of truth. Whatever lives in the workspace is what Claude sees. You’re not negotiating with copy-pasted snippets anymore.

You don’t need Cowork to get started with Claude. It earns its place once you have work that’s bigger than a single conversation, files you keep coming back to, or projects you want to live somewhere instead of disappear.

---

Claude Code

Here’s where it stops being a chat tool and starts being a coworker.

Claude Code is Claude with the ability to read your files, run commands, and edit code on your machine. It runs in two places:

• In your terminal as the claude command.
• In your IDE (VS Code, Cursor, JetBrains) as a built-in panel.

You point it at a project. You describe what you want. It plans, asks for permission on risky steps, and does the work.

Open a terminal, type claude, and you’ll see something like this:

The “approve plan” pause is intentional. You can always inspect what Claude is about to do and stop it before any file changes happen. This is the safety model: Claude has the capability, you keep the control.

Claude Code requires a Pro subscription. It’s worth it the moment you’re using Claude for real engineering work.

---

The Prompting Habits That Pay Off

If you only learn one thing about getting better with Claude, learn this: the more specific you are, the better the answer.

It sounds obvious. It’s harder than it sounds.

There’s a classic exercise: explain how to make a peanut butter and jelly sandwich. Sounds simple. Now imagine the listener has never seen bread, doesn’t know what a knife is, and will do exactly what you say with no interpretation. By the time you’re done, you’ve written a small instruction manual.

Talking to Claude is the same. Generic ask, generic answer. Specific ask, useful answer.

A few habits:

• Attach the document. Don’t describe what’s in it. Just upload it.
• Show, don’t tell. Paste the example, not your description of the example.
• Iterate. First answer is the draft. Push back. Refine.
• Tell it what not to do. “No headings, no bullet points, write like a human” is a real constraint.

Specific prompts produce specific, useful answers. The more detail you bring, the more value you get back.

---

Projects (claude.ai) and CLAUDE.md (Claude Code)

If you find yourself typing the same context every time (“I’m a teacher, K-5, here’s how I write lesson plans…”), you’ve discovered the need for standing context. Two ways to give Claude that context once and have it stick.

On claude.ai: Projects

A Project is a folder of related conversations that share:

• Custom instructions. A standing prompt that Claude reads before every conversation in that Project.
• Reference documents. Files Claude can pull from on every turn.

Create a Project for each major thing you work on. Add documents. Write instructions. Every chat in that Project starts with the context already loaded.

This single habit transforms how useful Claude feels.

In Claude Code: CLAUDE.md

In Claude Code, the same idea is a literal file called CLAUDE.md that lives in your project folder. Claude reads it automatically every session.

Layered, most-specific-wins:

Think of CLAUDE.md as the one-page “how we work here” doc you’d hand to a new assistant. Write it once. Claude reads it every morning.

Stuck on the blank page? Let Claude help.

The hardest part of a Project’s instructions or a CLAUDE.md is staring at the empty file. Good news: Claude is great at this exact task.

Open a chat and ask something like:

Help me write custom instructions for a Project I’m using to draft weekly status reports for my team.

Or, for Claude Code:

Help me write a CLAUDE.md for this repo. Ask me whatever you need to know about how we work.

Claude will ask follow-up questions about your role, your team, your conventions, and your preferences, then produce a draft you can paste in and refine. It’s the fastest way from “I don’t know where to start” to a working set of instructions.

---

Slash Commands

Slash commands are quick shortcuts you can call by typing /. They show up in both claude.ai and Claude Code.

Built-in ones include things like /help, /clear, and /compact. You can define your own for actions you find yourself doing repeatedly.

Custom commands are great for two-keystroke access to common actions. They’re cheap to define and they accumulate value over time.

---

Skills

A skill is a reusable capability with its own instructions, files, and (optionally) parameters. Where a slash command is “just a shortcut,” a skill is closer to a small program that Claude knows how to run.

Examples:

• A /lesson-plan skill that generates classroom material in your school’s exact format, pulling from your standards rubric.
• A /threat-report skill that ingests a news feed and produces a structured weekly threat intel summary.
• A /code-review skill that reads a diff and produces feedback in your team’s review template.

Every skill is invokable as a slash command. Not every slash command is a full skill. The difference is depth: skills carry their own context and can be elaborate.

If you find yourself doing the same complex task with the same shape every week, that’s a skill candidate.

---

Agents (Subagents)

A subagent is a specialized version of Claude that the main Claude can delegate to. They have their own focused system prompt and (often) their own set of tools.

Why use them?

• Parallelism. Multiple subagents can work on independent parts of a task at the same time.
• Context isolation. A research subagent can read 50 files without polluting your main conversation context.
• Specialization. A code-reviewer subagent can have a very different prompt than a documentation-writer subagent.

You don’t usually invoke subagents directly. The main Claude decides when delegation will help and spawns them.

This is where things start to feel less like “AI tool” and more like “AI team.”

---

Hooks

A hook is a script that runs automatically when something happens. Think before file edit, after tool call, or on session stop.

Hooks are how you encode rules that should always apply, without having to remind Claude every time.

Examples:

• A pre-commit hook that runs your linter and blocks the commit if it fails.
• A post-edit hook that auto-formats files Claude just modified.
• A pre-tool-use hook that double-checks risky commands before they run.

Hooks live in your settings file and run as shell commands. They’re how teams enforce “we always do X” without it being a memory-test for the model.

Hooks are an “advanced” feature in the sense that you usually don’t need them on day one. They become valuable once you have a workflow worth enforcing.

---

Plugins

A plugin is a bundle of skills, slash commands, hooks, and configuration packaged together as a single installable unit.

Why bundle them?

• Distribution. A team or community can ship a coherent set of capabilities as one install.
• Versioning. Plugins update together. Skills, commands, and hooks stay in sync.
• Discoverability. A plugin marketplace makes it easier to find capabilities than searching for individual skills.

Examples of what you might find as a plugin:

• A “PR review toolkit” that adds skills for code review, security review, and test analysis, plus matching slash commands.
• A “marketing automation” plugin that adds skills for blog drafting, social posting, and weekly performance reports.
• A “DevOps helper” plugin that adds skills and hooks for CI/CD workflows.

Plugins are a relatively new layer in the Claude ecosystem and they’re how the most powerful workflows tend to be shared.

---

What Order to Actually Try Things In

If this is all new, the order I’d recommend:

1. Open claude.ai, have a real conversation about something you actually need to figure out.
2. Create your first Project with custom instructions. Notice how the next chat feels different.
3. Iterate. Don’t accept the first answer. Push back. Get more specific. See what happens.
4. Install the desktop app if you’re on it daily.
5. If you’re a developer, install Claude Code and write a small CLAUDE.md for one project.
6. Once you’re comfortable, start writing slash commands for your repeat actions.
7. From there, skills, subagents, hooks, and plugins all become natural extensions.

You don’t need most of this to get value out of Claude. You can stop at level 2 and still get a huge amount done. The deeper levels are there when you’re ready.

If you want a tighter, faster intro to the same material, our getting started with Claude post is the short version. If you want to see real examples of using Claude for non-coding work, the putting agentic AI to work post walks through four concrete use cases.

And if you want to learn alongside other people doing the same thing, come to a Thursday class at the E.G. Fisher Library, or join the Discord. Most of what we know about Claude, we learned from each other.

See you out there.