All videos
0:00 / 0:00
ai

You've Been Using Claude Code at 10%. Here's the Rest.

Mark Kashef16 June 2026Watch on YouTube

Part of series

Ep. 7 · Claude Code Slash Commands

Geavanceerde Claude Code-functies zoals /goal voor autonome en efficiënte AI-workflows.

View the series

What you'll learn

  • The .claude folder is a hidden directory in every Claude Code project that controls what Claude remembers, is allowed to do, and how it behaves.
  • Commands and skills differ: commands are explicit slash-commands with $ARGUMENTS wildcards, skills auto-invoke based on their descriptions.
  • Rules can be path-scoped so Claude loads only relevant instructions when needed for specific directories.
  • Agents, settings.json and hooks (PreToolUse, PostToolUse) let you set up specialist agents with their own models, permissions and automatic workflows.
  • CLAUDE.md works as a router, not an encyclopedia, by directing to relevant components instead of explaining everything itself.

Frequently asked questions

What is the difference between commands and skills in the .claude folder?
Commands are explicit slash-commands you invoke that use the $ARGUMENTS wildcard. Skills, by contrast, auto-activate based on their description. Commands give you control over when something happens, skills work reactively when their trigger description becomes relevant.
How do path-scoped rules work and why are they useful?
Path-scoped rules load only the instructions needed when Claude is working in a specific directory. This prevents irrelevant rules from staying constantly in context and helps Claude stay focused without processing unnecessary information.
What do PreToolUse and PostToolUse hooks do in settings.json?
PreToolUse and PostToolUse hooks in settings.json automate workflows by executing actions before or after tools are used. This lets you trigger follow-up actions or apply validations automatically.
How do you configure specialist agents with their own models and tools?
You define specialist agents in the .claude folder by assigning them their own settings, available tools, and optionally a different model. This lets you create, for example, a security-auditor agent with restricted tool access, separate from other agents.

Topics

claude-apideveloper-toolsprompt-engineeringai-automationllm-configuration

Read next

Description from the channel

Master Claude Code: https://www.skool.com/earlyaidopters/about Assets in Video: https://markkashef.gumroad.com/l/claude-folder-anatomy-guide --- There is a hidden folder inside every Claude Code project that most people never open. It is called the .claude folder, and it is responsible for everything: what Claude remembers, what it is allowed to do, which skills it can invoke, and which rules it follows. Once you understand it, Claude Code starts to feel like your setup rather than a generic assistant. In this video I walk through every component of the .claude folder using a real hypothetical project called MeetingFlow. You will see how CLAUDE.md works as a router instead of an encyclopedia, how path-scoped rules let Claude load only what it needs, how commands and skills differ and why that difference matters, how to define specialist agents with their own models and restricted tools, and how settings.json handles permissions and hooks all in one place. By the end you will have a complete mental model of how all seven components connect, plus a working demo folder you can adapt for your own project. --- Timestamps 0:00 - The hidden folder inside every Claude Code project 0:16 - Two types of .claude directories and why they are different 1:41 - Introducing MeetingFlow: the hypothetical demo project 1:49 - Walking through the project-level .claude folder live 2:17 - Commands: how process-meeting works with $ARGUMENTS 4:05 - CLAUDE.md: why it is a router, not an encyclopedia 5:57 - Bash commands to navigate your .claude folder fast 7:03 - Rules: path-scoped intelligence that loads only when needed 8:50 - Rules in practice: the MeetingFlow rules folder 10:45 - Commands explained: filename equals slash command 12:10 - The $ARGUMENTS wildcard and how to use it 13:21 - Skills vs commands: the key difference 13:46 - How skills auto-invoke from their description triggers 14:19 - Skills in MeetingFlow: action tracker and auto follow-up 15:20 - Expert tip: disable-model-invocation and context fork 16:37 - Agents: meeting analyst and security auditor 18:43 - Permissions: allow, ask, and deny in settings.json 20:31 - Hooks: PreToolUse and PostToolUse inside settings.json 21:49 - The full MeetingFlow scenario: all seven components together 23:44 - Where to get the guide and bash cheat sheet --- Book a Consultation: https://calendly.com/d/crfp-qz3-m4z #ClaudeCode #Claude #Anthropic #AIProductivity #Automation #ClaudeFolder #CLAUDE #DotClaude #AITools #LLM #DeveloperTools #PromptEngineering #ClaudeCodeTips #AIWorkflow #VibeCoding