Njengah/claude-code-cheat-sheet: a command reference for Claude Code

What the cheat sheet actually is, and who it is for

The repository is a Markdown reference for Claude Code, Anthropic's terminal coding agent. The README describes it as a guide that takes a reader "from basic to advanced user without wasting time", and the file layout matches that claim: a README, a subagents.md page, a subagents/ directory, an images/ directory and a medium/ directory. There is no source tree, no build configuration and no package manifest at the top level. The primary language field is unknown, which is consistent with a repository that is mostly prose.

The intended reader is someone who has already installed Claude Code and wants to stop looking things up one at a time. The material is organised into six numbered levels, from Basic Commands through Master Commands, with each level covering a different surface: installation and navigation early, tools and permissions in the middle, and session and configuration controls later. That progression is the useful part. A new user can read Level 1 and start working; an experienced user can jump to the permission flags.

It is not a tutorial for people who have never opened a terminal. The README assumes Claude Code access on an account, and every example is a shell command. It is also not a substitute for the official documentation, because a third-party cheat sheet drifts.

How the reference is organised and where the risk sits

The structure is a flat sequence of fenced code blocks grouped under headings, with inline comments explaining each command. That format is easy to scan and easy to copy, and it is the reason this kind of repository gets bookmarked. The levels are a reading aid rather than a technical boundary: the same command family can appear at more than one level, and the README repeats installation snippets in both the Quick Start and Level 1.

The risk is version drift. The repository was last pushed on 2026-05-27, and its only tagged release, v1.0.0, dates from 2025-07-07. Claude Code is a fast-moving CLI, so any page that mirrors its flags will lag. The README itself flags this indirectly by labelling the project Beta in its own title, and by marking the npm install path as a "Deprecated npm fallback" while the native installers are described as recommended. A reader who copies a flag from a cheat sheet into a CI script is trusting that the flag survived the intervening releases.

There is one more structural point worth stating plainly. Because the repository contains no executable code, there is nothing to test and nothing that can fail at runtime. The failure mode is informational: a command that no longer exists, or a permission flag whose behaviour changed, silently breaks a workflow that the reader believed was documented.

Installing Claude Code and taking the first commands from the sheet

The cheat sheet does not install itself. It documents how to install Claude Code, and the README gives platform-specific one-liners. On macOS, Linux or WSL the recommended path is the native installer script, followed by sign-in and an interactive session.

curl -fsSL https://claude.ai/install.sh | bash
claude auth login
claude

After the script completes, the README says to sign in with claude auth login or to start claude and follow the browser prompts. Running claude with no arguments opens the interactive REPL at the root of the project you want the agent to work on.

Windows users get separate commands for PowerShell and CMD, and the README also lists package managers: brew install --cask claude-code on macOS and winget install Anthropic.ClaudeCode on Windows. The npm route, npm install -g @anthropic-ai/claude-code, is present but explicitly labelled deprecated in the README, so treat it as a fallback rather than the default.

Once installed, the sheet's first useful entries are the verification commands and the print mode, which runs a single prompt and exits instead of opening a session.

claude --version
claude auth status
claude -p "explain this function"

The first two confirm that the binary is on your PATH and that your account is signed in. The third is the entry point for scripting: the README also shows piping content in, as in cat logs.txt | claude -p "explain", and continuing the most recent conversation with claude -c. If you intend to use Claude Code inside a pipeline, print mode is the mode to learn first, because it terminates instead of holding a session open.

Session handling and permission flags are the parts worth reading twice

The middle levels of the cheat sheet cover ground that is easy to get wrong in practice. Session management gets its own block: claude --continue picks up the most recent session in the current directory, claude --resume accepts either a session ID or a name, and claude -n auth-refactor names a session at startup. Inside a running session the README lists /rename, /branch and /compact, where /compact summarises the current history to free context and accepts optional instructions.

Permission control is the other cluster. The README shows claude --allowedTools with patterns such as "Bash(git log:*)" and "Bash(git diff:*)", and claude --disallowedTools with patterns such as "Bash(rm:*)" and "Bash(sudo:*)". Those two flags let you pre-approve read-only git commands while keeping destructive ones blocked, which is the configuration most teams converge on. Then there is claude --dangerously-skip-permissions, which the README itself annotates as dangerous. That annotation is correct and worth repeating: the flag removes the confirmation step, and the cheat sheet offers no guidance on sandboxing the environment in which you use it.

Keyboard shortcuts are listed compactly, including Shift+Tab to cycle permission modes, Esc Esc to clear a draft or open the rewind menu when the input is empty, and Ctrl+T to toggle the task list. These are the entries most likely to go stale, because terminal keybindings change without any announcement in a changelog.

Where the cheat sheet stops being the right tool

Two limitations are visible from the repository itself. First, there is no verification layer. No test suite, no CI configuration and no script checks that the documented commands still exist in a current Claude Code build. The v1.0.0 tag from 2025-07-07 is a snapshot, not a maintained compatibility promise, and the last push on 2026-05-27 tells you when the text was last edited, not which CLI version it was checked against.

Second, the coverage is uneven. Subagents get a dedicated page and directory, which suggests that area received attention. Configuration files, MCP server setup and CI integration are not visible in the README excerpt, and the table of contents points to a Contributing section and a License section rather than to deeper operational guides. If your question is about wiring Claude Code into a build pipeline or restricting it inside a container, this repository is unlikely to answer it.

The wrong-tool case is straightforward. If you need authoritative behaviour for a flag before putting it in production automation, the cheat sheet is a starting point for the search, not the source of truth. Read it to learn what exists, then confirm against the CLI on your machine with claude --help or claude --version.

Compared with the official documentation and with running claude --help

The obvious alternative is Anthropic's own Claude Code documentation, which the cheat sheet mirrors in condensed form. The difference in approach is maintenance model, not content. Official documentation is versioned with the product and updated by the people who ship it; this repository is a personal compilation that advances when its author or contributors push changes. The README's own framing, describing the author's testing experience, makes that relationship explicit.

A second alternative costs nothing and is always current: the built-in help. Running claude --help, or the /help slash command inside a session, prints the commands your installed build actually supports. That output cannot be out of date relative to your binary, which is precisely the failure mode a static cheat sheet cannot avoid.

The sensible division of labour is to use the cheat sheet for discovery and the CLI for confirmation. The sheet is better than raw help output at one thing: it groups commands by experience level and explains what each one is for in a sentence. Help output lists flags; it does not tell you that /compact is the command you want when context fills up.

Licence, maintenance and what an upgrade costs you

The repository is MIT licensed. In practical terms that permits reuse, modification and redistribution with attribution and the licence text, and it disclaims warranty. Because the repository contains no code, the licence mostly governs the prose and the images. If you fork the cheat sheet into an internal wiki, keep the LICENSE file and the copyright notice with it. Nothing here is legal advice; read the LICENSE file at the repository root for the actual terms.

Maintenance cost is low in the sense that there is nothing to upgrade. There is no dependency to bump, no runtime to patch and no breaking change to absorb. The cost sits on the other side: keeping the content accurate is manual work, and the gap between the last push on 2026-05-27 and any Claude Code release after that date is a gap in the reference. The single v1.0.0 release from 2025-07-07 suggests the project does not follow a regular release cadence, so you should not expect tagged updates to signal when the content has been revised.

For a team, the realistic workflow is to pin a copy of the sheet to a known CLI version, note that version in the copy, and re-check it when you upgrade Claude Code. The commands most worth re-checking are the ones you script: print mode flags, output formats and the tool permission patterns.

Editorial conclusion

Adopt this cheat sheet if you already have Claude Code installed and want a single page covering commands, keyboard shortcuts and session handling, and check the repository's last push date against your installed CLI version before trusting any specific flag. Do not adopt it as tooling: there is nothing to install, no test suite and no release process beyond one v1.0.0 tag from 2025-07-07, so the README is the product. Before relying on it, run claude --version and claude auth status locally and confirm that any command you intend to script, particularly claude --dangerously-skip-permissions, still exists in your build.

Frequently asked questions

Official sources

1. License: MIT
2. Njengah/claude-code-cheat-sheet on GitHub
3. Project website
4. README
5. Releases