"Secret codes for Claude"

"Claude Code Custom Commands"

"Every Claude Code Command in 13 minutes"

Do these stop your scroll and have you save the video or screenshot the commands? It's such an easy rabbit hole to fall into. The comments fill up with "I didn't know this existed!" and "game changer!" But how many actually go test and use these commands?

I fell into this pattern too. I'd see a command shared, screenshoot it and before I know it, I have a stash of commands in my gallary, not knowing how they work.

Until one slow Sunday morning, I decided to ask my Claude Code, what these commands mean and more importantly, if they apply to me and my business. I'm not a developer, I'm a mompreneur who wants to get the best work done in the easiest way. This is what I got back: most commands don't matter for you. A small handful do. Here's how to tell them apart.

First, how Claude Code actually works

Before you can make sense of commands, it helps to understand there are three separate levers controlling how Claude responds to you:

• Which model you're using (think of this as which "brain" is working - more capable, or faster and lighter)

• How hard it thinks on any given question (you can ask it to reason more carefully when it matters)

• The Effort setting (how much thinking it applies across a whole session)

These three work together, and understanding this saves you from chasing every "productivity hack" command that shows up in your feed.

The models, simply:

• Haiku - fast, light, good for simple tasks

• Sonnet - the default, balanced, handles most things well

• Opus - the most capable, slower, worth it for complex decisions

For most of what you're doing - writing content, working on your strategy, day-to-day tasks - Sonnet is fine. You don't need to change models unless you're doing something that genuinely requires a higher ceiling.

The commands worth actually learning

think and ultrathink

Type these words in your message when you want Claude to reason more carefully before answering. Not every question needs this. Most don't.

Use think or ultrathink when:

• You're making an offer or pricing decision

• You're working out a launch sequence with multiple moving parts

• You're confused about strategic direction and can't see the way forward

• The wrong answer would cost you weeks of work

Don't use them for writing social posts, saving files, or anything with a clear answer. You'd be spending extra processing on something that doesn't need it.

The rule I follow: if I'd normally sleep on a decision, it's worth ultrathink. If I'm executing a plan I've already made, standard Sonnet is enough.

/clear

This resets your conversation without losing your setup. Your voice skill, your references, your frameworks - all of that stays intact. What gets cleared is the conversation thread itself.

Use it when a session has gone in too many directions and you want a focused fresh start. It's different from starting a new session - that's like closing the whole office and reopening it. /clear is wiping the whiteboard while staying in the room.

/advisor

This brings in a secondary, more powerful model that your primary model can quietly consult when it needs a second opinion. You don't manage it - your AI decides when to ask for help.

Run /advisor before a strategy session where you want deeper reasoning without switching to Opus for every single exchange. It's like having a senior consultant available in the background, on standby. So cool!

/insights

Run this at the end of each month. It generates a report on how you've been using your AI setup - what's working, where things are slowing you down, what to change. The tool helps you improve the tool. 🤓

/context

When a session has been running a long time and responses start feeling shallow or slow, run this. It shows a breakdown of what's filling up your context window - memory files, skills, conversation history. Helps you diagnose what's taking up space so you can decide whether to clear it or continue.

What to skip (and why)

Caveman mode - there's a third-party plugin that compresses Claude's responses to save tokens. It strips out explanations, nuance, and depth. The savings are real but smaller than claimed, and what you lose is exactly what makes Claude useful for content and strategy work. Skip it.

Switching models constantly - the temptation is to use Opus for everything because it's the most capable. But Sonnet with the right context (your voice skill, your references, good foundations) handles most work well. Save the model switch for when you genuinely need the ceiling.

Every new command you see posted - someone finds something, it gets amplified, and it feels like you're missing out. The commands above are the ones I actually use in a real business, not a demo environment. If a command doesn't solve a problem you actually have, you don't need it.

The bigger point

The best AI setup isn't the one with the most features turned on. It's the one that fits how you actually work - reliably, without you having to manage it constantly.

I spent months chasing better prompts, testing different tools, trying every new thing that showed up in my feed. The shift came when I stopped optimising the tool and started building the foundation. Once your AI knows who you are - your voice, your business, your clients, your strategies - most of the tinkering becomes unnecessary.

Commands are useful when they solve a real problem you have. If you're learning them because someone said they were game-changing, pause. Ask first: do I actually have this problem?

That clarity will save you more time than any command.