There's something built into Claude Code that barely anyone knows about — and the ones who do barely use it to its full potential. It's like having a personal concierge that knows every slash command, every hook, every configuration option, and every nuance of how Claude Code works.
You don't have to google it. You don't have to watch another tutorial. You just tag it, ask whatever you want, and it does the heavy lifting for you.
This is the Claude Code Guide sub-agent — a built-in sub-agent whose sole role is to be an expert on the latest version of Claude Code. Here's how to unlock it across five levels of power.
What Is the Claude Code Guide Sub-Agent?
Claude Code ships with built-in sub-agents out of the box. The Claude Code Guide is one of them — and it's the most useful for anyone learning or troubleshooting Claude Code itself.
Its job is simple: know everything about Claude Code. It reads the official documentation, understands every hook event, every CLAUDE.md pattern, every slash command, and every configuration option in the latest release.
🤖 Why It Doesn't Touch Your Context Window
When you invoke the Claude Code Guide, it spins up as a separate sub-agent. It fetches documentation and processes up to 28,000 tokens of source material entirely on its own — without consuming your main conversation's context window. You get the answer. Your session stays clean.
You invoke it by tagging it in your prompt. Let's walk through what you can actually do with it.
What You'll Need
- Claude Code — any recent version (the Guide sub-agent is built in)
- 5–10 minutes — that's genuinely all this takes to change how you learn and troubleshoot
No setup required. No installs. Just type and tag.
Level 1: Ask Any Claude Code Question
The most basic use: ask it anything about how Claude Code works. Instead of hunting through docs or watching tutorial after tutorial, go directly to the source.
A simple example — how do you switch models mid-session?
@claude-code-guide How do I change my model mid-session? For example, switching from Sonnet to Opus mid-conversation?The sub-agent fetches the relevant documentation, processes it, and comes back with the exact answer: use the /model slash command, switch to Opus, and continue. Clean and precise.
You can go broader too. Ask it conceptual questions that are hard to search for:
@claude-code-guide How do CLAUDE.md files work? Can I have a global CLAUDE.md, and then a CLAUDE.md within folders, and then a local version? How does that work? Can you explain it to me like I'm 10 years old?The response breaks down the hierarchy, explains what each level controls, gives example directory structures, and even throws in advanced tips — like using modular rules across multiple files. This alone is enough to stop most people from getting stuck.
Level 2: Diagnose Your Own CLAUDE.md
Level two is where things get more powerful. Instead of just asking abstract questions, you attach your actual files and ask the Guide to diagnose them.
Common problem: you wrote a CLAUDE.md, but Claude keeps ignoring your instructions or you keep having to repeat yourself. Hand it over and ask what's wrong.
@claude-code-guide [attach your CLAUDE.md]
I keep having to repeat myself to Claude about how I want my code formatted and where to save things. Why isn't my CLAUDE.md helping me? What's missing?The Guide reads your file as an expert who knows exactly what a good CLAUDE.md looks like. Its typical diagnosis: most people write a README, not a CLAUDE.md. The file describes the project but doesn't actually tell Claude how to behave.
A good CLAUDE.md audit from the Guide will flag:
- Missing code formatting rules — how you want code structured, indented, named
- No file organization rules — where things should and shouldn't be saved
- No patterns or anti-patterns — what to do and what to actively avoid
- Absent negative prompting — telling Claude what not to do, not just what to do
It then gives you specific examples of what to add — not generic advice, but actual snippets you can copy directly into your file.
Level 3: Understand Claude Code Hooks in Depth
Hooks are one of the most powerful and least understood features of Claude Code. They let you trigger custom scripts and automations at specific points in a session — but the JSON configuration is easy to get wrong, and the docs are dense.
Ask the Guide to explain the full picture:
@claude-code-guide What types of hooks are available in Claude Code? I want to understand what actions I can attach to automations. Can I trigger something every time a file is created, when a session starts, or when Claude edits something? Walk me through what's possible.The response maps out all 17 hook events in Claude Code — including:
- Session start / end — fire automations when a conversation begins or closes
- Pre-tool-use / post-tool-use — intercept before or after any tool runs
- Pre-compact / post-compact — trigger actions around context compaction
- Prompt submission — react every time you send a message
Once you understand the full menu, you start seeing creative applications everywhere. Hooks turn Claude Code from a conversational tool into a true automation engine.
💡 The JSON Pitfall
The most common mistake with hooks isn't picking the wrong event — it's getting the JSON structure wrong. The Guide will walk you through the exact format you need for your settings.json, with working examples for each hook type.
Level 4: Build a Custom Slash Command (and a Hook)
This is where the Guide stops explaining and starts building. You describe what you want — it creates the files, sets up the hooks, and leaves you with something that actually works.
Example: a custom /stand-up command that summarises your last session into three bullet points for your team chat.
@claude-code-guide I want to create a custom slash command called /stand-up that looks at what I worked on in my last session and writes a 3-bullet summary I can copy and paste into my team stand-up chat. Show me how to create the file for this. Actually — do it yourself and set up a relevant hook.A few minutes later, you have:
- A
/stand-upslash command file in your.claude/commands/folder - A session-start hook configured in your
settings.json - A macOS notification that fires every time a new session begins
The Guide checks the latest documentation to make sure the syntax matches the current version of Claude Code — not a tutorial written six months ago that might be out of date.
⚠️ Version Drift Is Real
Claude Code updates frequently. Blog posts, YouTube tutorials, and community guides lag behind. The Claude Code Guide always references the current documentation — which means it won't give you deprecated syntax or outdated patterns.
Level 5: Use It as a Live Debugger
The final level is the most impressive. Instead of asking conceptual questions or building new things, you hand the Guide your broken configuration and ask it to fix it.
Real scenario: a hook that's supposed to archive session plans with meaningful names is saving them as random animal names (dizzy-purple-flamingo.md, jumpy-velvet-penguin.md). The hook is broken and the plans folder is a mess.
@claude-code-guide [attach hook file] [attach plans folder]
My archive hook saves plans with random names like dizzy-purple-flamingo, jumpy-velvet-penguin, etc. I want to archive plans with a short description of what they're about (e.g. "add-user-auth") — read the first lines of each plan to figure out the description — and then organise them into a past-plans folder sorted by date. Diagnose what's wrong and fix it.Here's what happens in under a minute:
- The Guide reads the official hook documentation to understand the current spec
- It diagnoses the broken hook — in this case, a missing wrapper command in the JSON config
- It rewrites the
settings.jsonwith the corrected hook configuration - It renames every plan file from a random animal name to a meaningful descriptor
- It organises them into a date-sorted folder
✅ The Result
Open a new terminal, run your hook, and there are no more errors. The plans are named correctly, sorted by date, and the hook fires cleanly. End-to-end fix from a single prompt.
Why This Changes the Way You Learn
Most people learn Claude Code by watching YouTube videos, reading blog posts, or googling specific errors. The problem is that most of that content is:
- Derivative — most tutorials reference the same handful of sources
- Outdated — Claude Code ships updates frequently; old content doesn't keep up
- Generic — written for a general audience, not your specific project
The Claude Code Guide reads the actual documentation. It knows your files. It understands the latest version. And it responds to your specific situation — not a hypothetical one.
Whether you're just starting out or already running advanced agentic workflows, the Guide is the fastest path from stuck to unstuck.
Go Deeper with Claude Code
Join a community of builders mastering Claude Code together — hooks, CLAUDE.md patterns, custom sub-agents, and everything in between. Early access is open now.
Join the Academy →Key Takeaways
- The Claude Code Guide sub-agent is built into every Claude Code install — no setup needed
- It runs as a separate sub-agent, so it doesn't consume your context window
- Tag it in any prompt to ask questions, diagnose files, or build configurations
- It always references current documentation — never outdated blog posts
- At Level 5, it can read your broken files, diagnose the problem, and fix it live
- Use it over YouTube tutorials for anything Claude Code-specific — it's faster and more accurate
