In Part 1, we covered the fix that works for everyone: a values page you give the AI every time. In Part 2, we looked at 7 mods that enforce it. Now let's set one up.
This guide uses the Claude desktop app, so you won't need a terminal. We'll build the Values Prompt from Part 2, the mod that adds your values page to every content request. You won't write code: you'll write down what you believe and ask Claude to do the rest.
A rough version takes about 15 minutes once your values page exists. The page itself can take as long as you want to give it, and a short first draft is fine. You can improve it every week.
Step 1: Write your values page
Do this first, because it's useful even if you never build a mod. Make a folder on your computer called Marketing, and inside it a plain text file called values.md. Here's a starting point to change until it sounds like you:
Who we are: A family-owned roofing company in Bismarck, ND. We follow Christ, and it shows in how we treat people, not in how often we say it.
How we talk: Like a neighbor over the fence. Plain words and short sentences.
What we never do: Invent urgency, like "only 3 spots left" when it isn't true. Shame or scare anyone into buying. Quote a customer without permission. Promise results we can't guarantee. Use our faith as a sales pitch.
What we always do: Tell the truth about price and timing. Say when we're not the right fit.
That line about faith deserves a moment. Faith that shapes how you work is a witness, a living example people can see. Faith used to close a sale is something else, and customers can tell the difference. Write the line that keeps you on the right side of it.
Step 2: Open the Code tab and check your version
Open the Claude desktop app and switch to the Code tab, then open your Marketing folder there. Mods need a recent version of Claude Code: Anthropic's documentation says 2.1.286 or later in the desktop app. To check, type /status in the box where you'd normally type a message, and look at the row labeled Claude Code. If it's older, update the app. Mods are on by default after that.
Step 3: Ask Claude for the mod
Type this into the Code tab:
Build me a mod that adds values.md to the start of every prompt asking for a post, email, ad or blog. Show a small "Values loaded" note above the prompt box.
Claude Code has a built-in skill for writing mods, so it knows how. It writes the mod, installs it and reloads it while you watch. If it asks a question, like which kinds of requests should count as marketing, answer in plain words.
Step 4: Make sure it loaded
You should see the "Values loaded" note above the box where you type. That's why we asked for something visible: you can see the mod working instead of hoping it is. If the note isn't there, tell Claude what you see and ask it to fix it.
Step 5: Try to make it fail
Ask for exactly what your values rule out:
Write a Facebook post for our fall roof inspection special. Make it urgent so people book this week.
If the mod works, you'll get a post that's warm and clear and leaves out the fake deadline, because your values now ride along with the request. If it comes back full of "act now," tell Claude what happened and ask it to fix the mod. Try a couple more: one email, one blog intro. Each test shows you something to tighten in values.md.
Step 6: Ask Claude to explain what it built
You can't read the code, and you don't need to. You do need to know what it does. Ask:
Explain in plain English what this mod does, what it can read or change on my computer, and how to turn it off.
You're not taking Claude's word on faith here. You're getting a plain description you can hold the mod to, and if the answer mentions anything you didn't ask for, like reaching the internet, ask Claude to remove it.
Before you install anyone else's mod
Building your own is the easier place to start, because you choose what it's for. Installing a mod someone else wrote takes more care, and this is the part not to skip.
A mod isn't walled off from the rest of your computer. According to Anthropic, once a mod loads it can read and write your files, read your secrets (such as API keys, the passwords one piece of software uses to talk to another), see every prompt you send and every action Claude takes, change them, approve an action before you're asked, and use up your Claude usage.
That's a lot of trust to give a stranger, so here's how to be careful:
Get it from a source you know. Mods install as plugins (packages of add-ons for Claude Code) from a marketplace, which is a catalog of plugins. Stick to ones you trust, like the Claude directory or one your own team runs.
Ask Claude to read it first. A mod you find online usually comes as a folder of files you can download. Before you install it, open that folder in the Code tab and ask Claude to explain what it does and what it can reach, just like Step 6. If anything surprises you, don't install it, and for anything important, ask someone who reads code for a second opinion.
Know how to turn it off. Type /plugin in the message box, open the Installed tab, and disable or uninstall any mod there. Mods you build yourself show up in the same list.
If your business is on a Claude Team or Enterprise plan, the owner controls which marketplaces your team can use from the admin console, and Anthropic loads a built-in safety mod first that stops installed mods from doing certain risky things. Decide which mods your team may use before someone installs one on their own.
Where this leaves you
Start with the page, not the tool. Put your values in writing, give them to Claude every time, in a Project if you use the chat or a mod if you use Claude Code, and test what comes out. Build your own mods before you install anyone else's.
And remember what the mod is for. It's a reminder you set up while you were thinking clearly, so it can catch you when you're tired. The faithfulness behind it is still yours to bring.
If you'd rather have someone set this up with you, it's part of the AI training we do for business owners. Book a call and we'll look at how you use AI today and what's worth changing.
Read the series: Part 1: Put your values into the AI · Part 2: 7 mods that keep your marketing honest · Part 3 (you are here)