claude-howto: a visual guide to Claude Code patterns

TL;DR: claude-howto is a guide to Claude Code, not a tool that plugs into it. The repository description promises a visual, example-driven walk from basic concepts to advanced agents, with copy-paste templates you can drop into a session. The directory layout backs that up: ten numbered chapters, an index, a quick reference and a learning roadmap, with 43,274 stars and an MIT license.

What it is

The owner is a single GitHub account, luongnv89, and the repository was created on 2025-11-07, which makes it a young project with an unusually fast climb. It now has 43,274 stars and 5,395 forks, and it was pushed to three days ago.

The three declared topics are claude-code, guide and tutorial, and the implementation language is Python. That last detail is mildly surprising for something that reads as documentation, and it is explained by the repository containing a scripts/ directory and a local-progress/ folder, which are tooling for the guide rather than for your agent.

The project publishes its own site at luongnv.com/claude-howto, so the repository and a hosted version of the same material are available side by side.

How it is organised

The chapter directories are numbered, which is the clearest signal about how the material is meant to be consumed:

Chapter	Subject
01-slash-commands/	The slash command surface
02-memory/	Persistent memory for a session or project
03-skills/	Reusable skills
04-subagents/	Delegating to subagents
05-mcp/	Model Context Protocol servers
06-hooks/	Hooks that fire around tool use
07-plugins/	Packaging things up as plugins
08-checkpoints/	Checkpoints and session state
09-advanced-features/	The advanced corners
10-cli/	Working from the command line

The numbering is a curriculum, not a directory listing. Reading 01 through 10 in order gives you the building blocks before the combinations, which is a different experience from searching a wiki for the one page you need.

Sitting alongside the chapters are the files that make a large guide usable. INDEX.md is the table of contents, QUICK_REFERENCE.md is the cheat sheet, LEARNING_ROADMAP.md tells you what to learn in what order, and CATALOG.md inventories everything in the repository. CLAUDE.md and STYLE_GUIDE.md cover how the project wants its own agent sessions to behave, and clean-code-rules.md extends that to code you write with it.

Getting to the useful part quickly

Clone it and look at the top level first:

git clone https://github.com/luongnv89/claude-howto.git
cd claude-howto
ls

The two fastest reads are the index and the quick reference:

cat INDEX.md
cat QUICK_REFERENCE.md

Then the prompts/ directory, which is where the copy-paste templates the description mentions actually live. resources/ and resources.md hold supporting material, and slides/ is there if you would rather present the material than read it.

Because the chapters are separate directories, you can also work through one topic at a time. Nothing forces you to read the whole thing before you start using it, which matters when the guide covers features you have no immediate use for.

The translations and the tooling

The repository carries translated copies in ja/, uk/, vi/ and zh/. A guide that translates itself is a decent signal that its author expects a wide reader base.

The project also brings its own linting setup, visible in .markdownlint.json, .cspell.json and .pre-commit-config.yaml, plus a CHANGELOG.md that records changes over time. There is a coverage.xml at the root and a .gitissue.yml for issue templates. None of this is visible in the API metadata, but it is the difference between a wiki someone exported and a project someone maintains.

When to use it

It is worth your time when you are new to Claude Code and want a map, when you are experienced with one feature and want to learn the adjacent ones, or when you need a working template rather than a description of one. The copy-paste angle is the real draw. A lot of agent configuration is fiddly enough that seeing a correct example saves more time than any explanation does.

When to skip it

If you already run Claude Code daily and have your own setup, the early chapters will not teach you anything and the later ones overlap with what you already know. If you are on a different agent tool, a good part of the material does not transfer. And if what you want is a single specific answer, QUICK_REFERENCE.md answers that in a minute while the whole guide takes an afternoon.

Source and Accuracy Notes

• Project page: https://github.com/luongnv89/claude-howto
• Source repository: luongnv89/claude-howto (verified via GitHub API html_url)
• License: MIT (verified via GitHub API license.spdx_id)
• Stars at time of writing: 43274