Topics
Every answer, in one place.
One page per subject. Each gathers the definitions, the capabilities, the "X vs Y" comparisons and the long reads for a single question — so you stop opening five tabs to understand one thing.
Core
What you need on day one.
What is Claude Code and how do I actually start using it? What Claude Code is (and how to start) Claude Code is an AI assistant that works on your actual files instead of just talking about them. You download the Desktop app, open a folder, and ask in plain English. It proposes changes, and you review each one and accept or reject it. 20 entries Do I need to use a terminal, or can I just use the app? The app vs the command line You can just use the app. Claude Code Desktop and the command-line version are the same engine behind two different interfaces. Desktop is an ordinary window you point and click. The terminal is for automation, scripting, and the Linux distributions Desktop's beta doesn't cover. 9 entries What do I actually need to know about how my computer stores things? Files, folders and paths Files are named blobs of data that do nothing until an app reads them. Folders nest inside folders to make a tree. A path is a file's address through that tree. That's the whole model, and it's enough to work with Claude on real files. 4 entries What is an LLM and how does it actually work? How AI actually works A large language model is software that learned patterns from an enormous amount of text and now predicts what should come next, one small chunk at a time. Everything it does — answering, writing, summarising, coding — comes out of that single ability. 9 entries What's the difference between a chatbot and an AI agent? What an AI agent is A chatbot answers one message and then waits for you. An agent works toward a goal in a loop — it acts, looks at what happened, decides the next step, and repeats until the job is done. Claude Code is an agent. 6 entries How do I write a prompt that gets a good answer? Asking for what you actually want Write it like a brief, not a search query. Say what you want, who it's for, what to avoid, and what good looks like — and show an example rather than describing one. A vague answer is almost always the result of a vague ask. 4 entries How do I understand a codebase nobody at my company understands anymore? Understanding code you didn't write Point Claude at one file or one name and ask in plain language. It reads the actual code and explains it, so an unfamiliar system becomes a series of small answered questions instead of one impossible one. Reading changes nothing, so it's the safest way to start. 2 entries Why does Claude forget things, and what should I put in front of it? Context and the context window Claude only knows what's in front of it — your messages, the files it has read, and its own replies. That space is finite, so a long session starts dropping earlier details. The fix is deliberate: give it what matters, and start fresh between unrelated tasks. 9 entries When do I clear the chat vs keep going? Running a session well A session is one continuous conversation with Claude — everything it can still see. Start a new session for a new task. Use /compact when a long thread is getting forgetful but isn't finished, and /clear when you're moving on to something unrelated. 8 entries How do I hand Claude something big without it building the wrong thing? Tasks too big for one prompt Spot the size first — a task is big when you can't say in one sentence what done looks like. Then plan before you act, argue with the plan while it's still free, slice the work into pieces you can check, and interrupt the moment it drifts. 2 entries Which Claude model should I use? Choosing a model Claude comes in four tiers — Haiku, Sonnet, Opus, and Fable. Sonnet is the default and right for most work. Drop to Haiku for quick, simple jobs; reach up to Opus or Fable when a problem is genuinely hard. You switch mid-conversation. 4 entries How do I stop Claude doing something I didn't approve? Permissions, plan mode, and staying in control Claude asks before it changes a file or runs a command. Start in Ask permissions mode, where it checks with you every time. Plan mode goes further — Claude proposes an approach and touches nothing until you approve. You loosen the rules as trust builds. 10 entries Claude is asking permission. How do I know whether to say yes? Deciding what to approve Read what it will actually do, then ask two questions. How far do the consequences reach if this is wrong — just this chat, or other people? And can I take it back? Reversible and contained means approve and move on; irreversible or wide means slow down and read. 2 entries How do I see what Claude changed before it's saved — and undo it? Seeing and undoing what Claude changed Claude Desktop shows every edit as a visual diff before it's saved — the old text and the new side by side, with removals and additions marked in color — and each change has its own accept and reject button. Nothing lands until you accept it. 4 entries How do I know if the AI is making this up? When AI is wrong AI models compose plausible text rather than look facts up, so they can state something false with complete confidence. Confidence tells you nothing. The fix is verification — give the model the source material, and check any specific fact, figure, or citation against it. 5 entries How do I stop re-explaining my project every time? CLAUDE.md and memory Put a file called CLAUDE.md in your project and Claude reads it at the start of every conversation — your stack, your conventions, what never to touch. Ask Claude to draft one, or run /init. Memory is the separate, private notes Claude keeps itself. 4 entries
Working knowledge
What you'll hit in your first month.
What are slash commands and how do I make my own? Slash commands A slash command is a saved prompt you run by typing `/name` in the chat. Claude Code ships with built-in ones like `/review` and `/clear`, and you can add your own by saving an instruction you keep re-typing into a file with a name. 4 entries What is a Claude Skill and when do I write one? Skills and plugins A skill is a folder of written instructions Claude reads on its own when the situation matches — your team's standards captured once instead of re-explained every session. A plugin bundles several skills, commands, and connectors together so a whole setup installs in one step. 4 entries What's a subagent and when should I delegate to one? Subagents A subagent is a second Claude that your main Claude spawns to handle one focused job — reading forty files, running a search — in its own separate context window. It reports back only the conclusion, so your main conversation stays short and on topic. 3 entries How do I connect Claude to my other tools? Connecting Claude to your other tools MCP (Model Context Protocol) is the standard way to give Claude access to tools it doesn't have built in — your Google Drive, your database, your ticketing system. You add a connector once, and from then on Claude can read from and write to that tool during a normal conversation. 8 entries How do I make Claude think harder (or go faster)? How hard Claude thinks Claude can reason step by step before answering, which meaningfully improves accuracy on hard, multi-step problems. In the Claude app that's the Extended Thinking toggle; in Claude Code it's the `/effort` setting. A separate `/fast` dial trades a little cost for quicker replies. 6 entries What is git and why does Claude keep talking about commits? Version control (git) Git is a tool that records every change made to a project so any earlier version can be brought back exactly. It is what makes letting an agent edit your files reasonable — if a change turns out wrong, you restore the last good state in seconds. 11 entries How do I review what an agent produced without reading every line? Reviewing AI-written work Review by checking behaviour and intent rather than reading every line — confirm the change does what you asked, exercise it, and spend your attention on the parts touching money, data, or permissions. Automated reviewers can sweep the routine remainder for bugs. 5 entries Why is my AI bill this high and how do I cut it? What AI costs Most of an AI bill is the context you send, not the answers you get back — the input side is roughly 90% of the cost. The fix is sending less: a narrower working folder, fresher sessions, and caching the parts you reuse. 5 entries How do I get my team actually using this? Rolling Claude out to a team Start small and specific — a handful of people, on Desktop rather than the terminal, each with one task worth changing their habits for. Settle the data rules before day one, then capture what works so one person's good prompt becomes everyone's. 7 entries Can I put company data into AI? Company data, privacy, and rules Usually yes, on an approved company account — but not for everything. Sort your material into public, internal, confidential, and regulated, and set a different rule for each. A blanket ban doesn't stop AI use; it just moves it somewhere you can't see. 5 entries What are Artifacts, Projects, and Deep Research? The Claude app surfaces They're three features of the Claude app, not three products. Artifacts is a side panel where Claude builds something you can see and use; Projects is a workspace that remembers your context; Deep Research runs many searches and returns a cited report. 7 entries How does my team actually run this day to day? Your team's operating guide An operating guide is the layer above individual tasks: who owns a decision, what material your team may share, and which check a piece of work must pass before it leaves the building. There is one below for each of nine teams. 9 entries What do frontend, backend, API, and server actually mean? Software words you'll hear Most software is one program asking another for something. The frontend is what you see on your own device; the backend is the logic and data running on a server; an API is the agreed way to ask. Everything else is vocabulary around that shape. 12 entries
Advanced
When you outgrow the basics.
Can I have several Claudes working at once? Agent teams and parallel work Yes. An agent team splits one request across several Claudes working in parallel — one per file, page, or question — and then gathers their findings into a single result. You start one by describing the goal in plain language, not by writing code. 7 entries How do I make Claude run without me sitting there? Hooks, schedules, and loops Three mechanisms, each with a different trigger. A hook fires on an event — after every edit, before every command. A schedule fires on the clock — every Monday at 9am. A loop repeats on an interval until the condition you set is met. 10 entries How do I make agent-written software actually reliable? Building reliably with agents Reliability comes from the system around the agent, not from a better prompt. Specify the outcome precisely, split the work into pieces small enough to check, and make something verify each piece — a test, a second agent, a fixed rule — before it counts as done. 6 entries How does AI search my own documents? Search over your own documents (RAG) RAG (retrieval-augmented generation) means the system searches your documents first, pulls out the few most relevant passages, and hands those to the model to answer from. The answer is grounded in your actual source material instead of the model's training memory. 3 entries Is AI going to take my job? What AI is doing to work Not your whole job — but AI is absorbing specific tasks, and every job is a bundle of tasks. The useful question is which of yours are exposed, which get amplified, and which hold their value because a person has to be accountable for them. 5 entries Can I put Claude inside something I'm building, not just talk to it? Building Claude into your own software Yes, at three levels. The raw API gives you one question and one answer. Headless mode runs Claude as a command inside a script. The Agent SDK hands your program the whole Claude Code engine — the loop, the tools, the permissions — as a library. 3 entries