This Claude cheat sheet puts the commands, skills, Projects, prompt patterns and shortcuts you actually use on one page, open and free, no PDF and no sign-up. It covers the Claude apps, Claude Code in the terminal, and skills. The last table maps everything to ChatGPT and Gemini so it works as a general AI cheat sheet too.
1. Claude apps basics
| Feature | What it is | When to use it | Tip |
|---|---|---|---|
| Projects | A workspace that keeps instructions and files attached to every chat inside it. | Recurring work: one client, one product, one course. | One Project per context. Unrelated context is noise. |
| Custom instructions | Project-level text Claude reads before every chat. | Tone, audience, format rules, things to never do. | Write it like a brief for a new hire. |
| Knowledge files | Documents uploaded to a Project for every chat to reference. | Style guides, specs, past examples, skill files. | Upload the source of truth once; update the file when it changes. |
| Artifacts | Standalone outputs (documents, code, pages) in a side panel you can iterate on and export. | Anything you will edit, reuse or share. | Ask for edits, not a regeneration, so the good parts survive. |
| Model picker | Switches between the available Claude models. | Faster models for drafts; the most capable for hard reasoning and code. | Strongest model when a wrong answer is expensive. |
For the full setup, read how to use Claude Projects.
2. Claude Code essentials
Claude Code is the terminal agent: it reads your repository, runs commands and edits files with your approval. Commands change between releases, so check /help for the current list.
| Action | How | Notes |
|---|---|---|
| Start a session | Run claude in the project folder. |
It works from the current directory. |
| Set up a project | /init |
Drafts a CLAUDE.md for the repo. Edit it afterwards. |
| CLAUDE.md | Markdown file in the repo root, read at the start of every session. | Build and test commands, conventions, folder layout, things to never touch. |
| .claude/skills | One subfolder per skill, each with a SKILL.md; in the repo or your home folder. | Where marketplace skills go in Claude Code. |
| Get help | /help |
Current commands and shortcuts for your version. |
| Clear context | /clear |
Fresh conversation, same folder. Use between unrelated tasks. |
| Compress context | /compact |
Summarises the session to free context while keeping decisions. Use when long sessions slow down. |
| Leave |
exit or Ctrl+C twice. |
Edits stay on disk; check git status. |
| Approve actions | Answer the permission prompt before each command or edit. | Refusing is free; undoing a bad shell command is not. |
Two habits beat any command: commit often so every change is reversible, and move anything you keep repeating into CLAUDE.md or a skill. The guide to Claude Code skills has the folder layout and examples.
3. Skills: what, where, how
| Question | Answer |
|---|---|
| What is a skill? | A reusable Markdown instruction file: a name, a short description of when it applies, and the detailed process to follow. |
| Where does it live? | Claude apps: as a Project knowledge file or in custom instructions. Claude Code: .claude/skills/<name>/SKILL.md in the repo or your home folder. |
| How does it trigger? | By match: when your request fits the description, Claude loads the full instructions. In Claude Code you can also call it by name with a slash. |
| Why not a long prompt? | A prompt is pasted every time and drifts. A skill is written once, versioned, shared and applied automatically. |
| What makes a good one? | Narrow purpose, precise trigger description, fixed output format, two or three worked examples. |
| Can I buy them? | Yes. The Claude skills marketplace sells finished .md skills with a README, one-time purchase, 30-day money-back, working in Claude, ChatGPT, Gemini and Copilot. |
4. Twelve prompt patterns
One line each, bolt onto any request. Combine two or three, not all twelve.
| Pattern | One-line version |
|---|---|
| Role | "Act as a [role] with [years] of experience in [field]." |
| Audience | "Write for [who], who knows [X] but not [Y]." |
| Format lock | "Return exactly: [headings or fields], nothing else." |
| Constraint | "Under [N] words, [tone], no [thing to avoid]." |
| Example anchor | "Match the style of this example: [paste]." |
| Step by step | "Work through this in numbered steps before giving the answer." |
| Ask first | "Before answering, ask me up to 3 clarifying questions." |
| Options | "Give me 3 different approaches with one-line trade-offs, then recommend one." |
| Critic pass | "Now critique your answer as a sceptical [expert] and fix what you find." |
| Source discipline | "Only use the attached material. If it is not there, say so." |
| Iteration | "Keep everything except [section]; rewrite only that." |
| Checklist close | "Finish with a checklist of what I must verify myself." |
5. Ten copy-paste starter prompts
Fill in the brackets and go. They work in Claude, Claude Code, ChatGPT and Gemini. More are on the ChatGPT prompts page.
6. Common mistakes and fixes
| Mistake | What happens | Fix |
|---|---|---|
| One Project for everything | Irrelevant files and rules pollute every chat; answers drift. | Split by context. One Project per client, product or course. |
| Pasting the same brief every chat | Inconsistent results, wasted time, small variations creep in. | Move it to custom instructions or a skill file once. |
| No output format | Long prose when you needed a table, or vice versa. | Use the format lock pattern. Name the fields and the order. |
| Regenerating instead of editing | You lose the 80 percent that was already right. | Say what to keep and what to change. Edit the artifact. |
| Trusting facts without a source | Confident but wrong details slip in. | Attach the material and ask Claude to say when it is not there. |
| Letting Claude Code run without a CLAUDE.md | It guesses your conventions and build commands. | Run /init, then edit the file so it is true. |
| Never compacting a long session | Responses slow down and lose earlier decisions. |
/compact at natural checkpoints, /clear between unrelated tasks. |
| Writing a skill that does everything | It fires at the wrong times or not at all. | Narrow purpose, precise trigger description, fixed output. |
7. ChatGPT and Gemini equivalents
The mapping is functional, not exact.
| Claude | ChatGPT | Gemini | What it is for |
|---|---|---|---|
| Project | Project or custom GPT | Gem | A persistent workspace or persona with its own instructions and files. |
| Custom instructions (Project) | GPT instructions or Custom Instructions | Gem instructions | Standing rules that apply to every chat in that space. |
| Knowledge files | Knowledge (files uploaded to a GPT or Project) | Files attached to a Gem or chat | Reference documents the model can draw on. |
| Skill file (.md) | Pasted into GPT instructions or uploaded as knowledge | Pasted into Gem instructions | A reusable process document. The same .md file works in all three. |
| Artifacts | Canvas | Canvas | A side-by-side editor for documents and code you iterate on. |
| Claude Code | Codex CLI and coding agents | Gemini CLI | A terminal agent that reads and edits a codebase. |
| CLAUDE.md | AGENTS.md (convention used by several tools) | GEMINI.md | A repo-level instruction file the agent reads first. |
Every file in the Claude skills collection is plain Markdown, so it pastes into a custom GPT or a Gem in a minute.
Frequently asked questions
Is this Claude cheat sheet free?
Yes. Everything on this page is open, there is no PDF to download and no email to give. Bookmark the page or print it from your browser if you want a paper copy.
Does the cheat sheet work for ChatGPT and Gemini too?
Most of it does. The prompt patterns, starter prompts and common mistakes apply to any model. The equivalents table at the end maps Claude features (Projects, skills, artifacts) to their ChatGPT and Gemini counterparts so you can translate the rest.
What is the difference between a Claude skill and a prompt?
A prompt is a one-off instruction you type. A skill is a reusable instruction file (usually a Markdown file with a short description) that Claude loads when a task matches. You write it once and it applies every time instead of being retyped.
Do I need Claude Code to use skills?
No. Skills run in the Claude apps through Projects and knowledge files, and in Claude Code through the .claude/skills folder. The same .md file works in both places, and the content can be pasted into a custom GPT or a Gemini Gem as instructions.
Which slash commands does Claude Code have?
The ones you can rely on include /help, /clear, /compact and /init. The list changes with releases, so type /help inside a session for the current set rather than trusting any static list, including this one.
The bottom line
You do not need to memorise Claude: five app concepts, four Claude Code commands, one skills folder and a dozen prompt patterns cover most days. Bookmark this page, and turn any instruction you have typed three times into a skill so you never type it a fourth.
Browse all Claude skills at KissMySkills.