You installed Claude Code. You ran some prompts. And then you hit the wall that almost everyone hits: the official docs describe what each feature does, but don’t show you how to combine them into something that actually saves you hours.
claude-howto is a GitHub repository that exists specifically to solve that problem. It’s a modular, example-based guide for Claude Code — slash commands, project memory, subagents, hooks, MCP and plugins — structured as a progressive learning path with copy-paste templates you can drop directly into your projects. It’s trending on GitHub with over 11K stars.
What’s in the Repo
The structure is organized into 10 modules that build on each other:
- Slash commands — production-ready
.mdtemplates for/review,/commit,/pr,/optimizeand others, with examples of how to chain shell commands using the!commandsyntax so Claude gets live data, not just instructions - Project memory —
CLAUDE.mdtemplates for different project types, covering code standards, deploy context and team conventions that persist across sessions - Skills — reusable workflows that you install once and invoke from anywhere with
/skill-name - Subagents — specialized agents (code-reviewer, test-engineer, security-auditor, api-documenter) that the main agent can delegate tasks to
- Context7 integration — MCP configuration to bring updated, version-specific documentation directly into context
- Hooks — event-driven shell scripts that run on Claude Code events like
PreToolUseandPostToolUse, to automate quality checks, formatting and security validations - MCP servers — ready-to-use configurations for GitHub, Postgres, Kubernetes, Figma and others
- Plugins — packaged collections (pr-review, devops-automation, documentation) that combine commands, agents, hooks and MCP into one installable package
- CLI reference — complete reference of flags and configuration for the Claude Code CLI
- Learning path — a self-assessment you can run with
/self-assessmentdirectly from Claude Code to get a personalized roadmap
Why the Learning Path Structure Matters
Most Claude Code users are leaving a lot of its power on the table — not for lack of motivation, but because the feature surface is large and the connections between them aren’t obvious just from reading the docs.
Do you learn MCP before hooks? Skills before subagents? claude-howto answers that question with a sequenced 11-13 hour path that takes you from zero to orchestrating multi-agent pipelines. The self-assessment hook is a particularly smart touch: you run it inside Claude Code itself, get a diagnosis of where your gaps are, and receive a personalized roadmap.
How to Get Started in 15 Minutes
The repo is designed so you can get value immediately without reading everything first:
# Slash commands (15 min)
cp 01-slash-commands/*.md .claude/commands/
# Project memory (15 min)
cp 02-memory/project-CLAUDE.md ./CLAUDE.md
# Install a skill (15 min)
cp -r 03-skills/code-review ~/.claude/skills/
Just those three steps already change how your sessions behave — Claude will have context about your project standards, know how to invoke a code review in production, and have a set of slash commands for the most common workflows.
The Combination That Changes Everything
The repo includes concrete examples of what happens when you connect features together. A /review-pr command that:
- Loads project memory with your code standards
- Fetches the PR via GitHub MCP
- Delegates to a
code-reviewersubagent - Delegates to a
test-engineersubagent - Synthesizes both analyses into a complete review
That’s not a toy example. It’s a reproducible workflow that your whole team can standardize — and claude-howto provides the templates to build it.
MIT licensed, actively maintained and synced with every Claude Code release. Current version is v2.2.0 (March 2026), compatible with Claude Code 2.1+.
