For business owners who use Claude Code every single day.
Claude Mods
Do This First.
Mods are live, and most people get them wrong. They install someone else's. Do this first instead. Get Claude to read your own chats. Mine found four questions I ask every single day.
The prompt
Claude reads your chats and tells you which mods you need
My mods
Seven of my thirteen, running in my Claude Code
Build yours, safely
No code to write, and the traps to check first
Written 12 October 2026. Mods went live on 1 October and still change between releases.
What A Mod Is
What Claude Mods Are
Skills, hooks and MCP servers sit outside Claude Code. A mod sits inside, for the whole session.
Here's the quick version. Skills, hooks and MCP servers all sit outside Claude Code. Claude calls them when it needs them. A mod sits inside. It runs for the whole session. It sees everything as it happens. And it can change it.
Claude Code
Outside tools get called, do one job and leave. A mod is there from the first message to the last.
On your screen
It can put something in front of you and keep it there.
Remembers
It can hold on to things while you work.
Stops and asks
It can stop Claude and check with you first.
Instant commands
It can add a command that runs straight away, with no waiting on the model.
Above the prompt
A strip that sits just above where you type. It stays there and updates as you work.
Mine shows three next steps, and what is left before I close the chat.
Why should you care? Because you run a real business. And every day you lose minutes to the same small things.
A mod answers those before you ask.
This is not a side door. Anthropic now builds parts of Claude Code the same way. The diff view is a mod, and they have said more of their own features will move across.
The Prompt
Get Claude To Read Your Own Chats
One prompt. It finds what you keep asking, and it builds nothing.
I started with my own chat history. No guessing. I asked Claude to read my own messages and tell me what I keep doing by hand. Here's the prompt. Copy it as it is.
The Claude mods audit prompt
Read my last 20 sessions in each project. Only count the messages I typed. Skip tool results, system messages and anything run on a schedule. Find three things: → Questions I keep asking you → Things I keep correcting → How I check your work For each idea, give it a name, quote my exact words, and count how many messages and sessions it came up in. Say where it would show on screen and what it would show. Rank the ideas by how much time they would save me. Say if something I already have covers it. Ideas only. Build nothing.
How to run it
Open Claude Code in the folder you work from most.
Paste the prompt. It reads the messages you typed. Not Claude's answers. Yours.
Read the list it gives you. Pick one idea. Then ask Claude to build that one.
The three things it looks for
For each idea it gives a name, quotes your exact words and counts how often it came up. Then it ranks them by how much time they would save you.
The last line matters most. Ideas only. Build nothing.
Because you pick. Then it builds. A count tells you a topic came up. It does not tell you why. You know why. So you pick.
The Audit
Four Questions I Ask Every Day
67 sessions. 440 messages. 5 projects. What came back surprised me.
I asked Claude to read my own messages. Not its answers. Mine.
67
sessions
440
messages
5
projects
Four rejections in 440.
What came back surprised me. I almost never reject what Claude does. Four times, in 440 messages. So I didn't need more guards.
Four questions kept coming up
What's next?
13 times
"Okay done. What's next?"
Answered now by next-steps
What's left?
19 times
"Aanything left to do?"
Answered now by loose-ends
Where's that handoff?
24 times
"Give me the updated handoff here"
Answered now by handoff-shelf
Did it work?
19 times
"I ran that - dod it work?"
Not built. I skipped this one.
It even quoted me back to myself, typos and all.
9 ideas
ranked, each backed by my own words
7 built
I picked. Then Claude built.
Your list will not look like mine. That's the point. Run the prompt and see what you keep asking.
Enjoying the guide?
Enter your details to unlock the remaining insights. We'll also send you the complete guide as a reference.
Mine, Running
Thirteen Mods. All Mine.
Seven of them, one card each. Every one is a Claude Code plugin I had Claude build.
Here's mine. Seven of the thirteen. Start with the first one. It is the best of the lot.
next-steps
Start hereAbove the promptEvery time Claude finishes, three next steps show up above the prompt. I press one, two or three. I never sit there thinking, what do I do next?
loose-ends
Above the promptA list of what's left before I close the chat. Files I haven't saved. Things I haven't rebuilt. When it's all done, it goes green.
handoff-shelf
Side pane, /handoffsI type /handoffs. And there's every handoff from the last two weeks. One press copies it.
client-doc-lint
Stops and asksThis one saved me the other day. I went to publish a page. A mod stopped it. It had a dash I don't allow, and it wasn't set up for phones. Fix, publish anyway, or cancel. I hit fix.
job-watch
Strip and a pingWhen I've got jobs running in the background, a strip shows them. It pings me when one finishes. So I've stopped typing, how about now?
did-we-already
Note before Claude startsIf I ask about a tool I've already looked at, it tells me before Claude even starts. Seen that one. Already checked.
redaction
/redact onIt hides emails, phone numbers, dollar figures and client names on my screen. Claude still sees them. You don't. So I can share my screen and leak nothing.
Thirteen Claude mods. All mine. Not one from a stranger.
And every one of them is a Claude Code plugin. They install the same way as any other plugin. Don't copy this list. It answers my questions, not yours.
Build One
You Don’t Write The Code
Say what you want on screen, in plain words. Claude writes it. Three files, no build step.
Now, the build. And you don't write the code. You pick one idea from your audit and describe it.
Check your version
Mods need a recent Claude Code. Run claude --version. You want 2.1.287 or later. The desktop app carries its own copy, so update that too.
Say what you want on screen
Claude Code comes with a skill for writing mods. You say what you want, in plain words. Claude writes it.
Say yes to hot reload
It asks once if you want hot reloading. Say yes. Then each change shows up in the chat you already have open, as soon as Claude finishes its turn.
Move it, then install it once
A mod Claude writes sits in a temporary folder for that one chat. When it works, move it into your own plugin folder and install it. One install in the terminal covers the desktop app too.
What asking for one looks like
Build me a mod. After every answer, show three likely next steps above the prompt. Let me pick one with the 1, 2 or 3 key.
Swap in the idea your audit ranked top.
It's a Claude Code plugin with one extra file. A mod that remembers things adds a fourth, a small types file. Claude writes that one too.
Installing the one you keep
$ claude plugin marketplace add ./my-mods $ claude plugin install next-steps@my-mods
Swap in your own folder and mod name. Any chat that was already open needs /reload-plugins before it sees the new mod.
Three Moves
Watch, Change, Answer
Every mod does one of three things when something happens.
Every mod does one of three things when something happens. Once you know the three, you can describe any mod you want in one sentence. Tap each one.
It watches, and lets it through.
Something happens
Claude edits a file
The mod
The edit goes through. The list above the prompt gains one line.
loose-ends. Mine watches which files the chat changes. It blocks nothing. It just keeps the list of what's left up to date.
Hooks Or Mods
A Hook Can Say No. A Mod Can Ask.
Hooks are still fine for a simple block. Here is where they run out.
You might be thinking, we already have hooks. True. And if you just need to block something with a script, a hook is still fine.
Hook
Runs once, then it's gone
- A script that fires on one event.
- It can let a command through, block it or change it.
- It can't put its own question in front of you.
- Still fine for a simple block.
Mod
Stays all session
- Code that lives inside Claude Code from the first message to the last.
- It can watch a command, change it or stop it.
- It can stop and ask you, with buttons.
- It can show things on screen and remember them.
I hit that wall.
I wanted Claude to check with me before it squashed a long chat down. A hook could only say no. A mod can ask. So I built it as a mod.
The test is one question. Do you need it to ask you, show you something or remember something? If not, write the hook and move on.
The Traps
Check It Before It Goes In
A mod has more power than a skill. One command shows you what it can touch.
Now the traps. Because there are a few. A mod runs with the same access Claude Code has. And it is written by whoever published it, not by Anthropic.
That's more power than a skill. So here's my rule.
No stranger's mod goes on my machine until it's been checked.
There's a command for it
$ claude plugin validate ./client-doc-lint hooks: tool.call{tool=Artifact} calls: $.env.get, $.fs.read, $.ui.ask ✔ Validation passed
It lists the events the mod listens to and the calls it makes. This one listens for a page about to be published. It reads one setting, reads a file and asks me a question.
If a mod you don't know is using the internet, find out why before you install it. And remember what the check is. It shows what a mod can do. It does not prove the mod is safe.
A broken guard lets things through
What happens. If a mod's check crashes, Claude Code skips that check and the command runs anyway. So a guard that breaks is no guard.
Do this. For any mod that guards something, ask Claude for a catch that blocks when the check fails. Then test that it does.
A reload wipes memory
What happens. A reload wipes whatever your code was holding.
Do this. Keep anything that matters in Claude Code's own store. That survives.
Nobody home
What happens. Run Claude on a schedule with nobody there, and a mod that asks a question gets no answer. The ask fails straight away.
Do this. Decide what the mod does when no one's home. Mine pass straight through.
It's new. It changes.
What happens. Mods change between releases. On 10 October an update broke one of mine. One command stopped working, and the validate check did not catch it.
Do this. I re-check every mod after an update. That one now has a test.
And if a mod ever misbehaves, there is an off switch. Start Claude Code with claude --safe-mode and that session runs without your mods.
When To Use One
Hook, Skill, MCP Or Mod
Four tools. One question sorts them.
So when do you reach for a Claude mod? Not every idea from your audit should be one. Take each idea on your list and ask what you want it to do.
Use a mod
Something on your screen, something that remembers, or something that stops Claude and asks you first. That's a mod.
A mod has the most power of the four. So it has the most to check. If a hook, a skill or an MCP server can do the job, use that.
MyMyClaudeClaudeCodeCodeisisbuiltbuiltaroundaroundhowhowIIwork.work.YoursYoursshouldshouldbebebuiltbuiltaroundaroundyou.you.Update,Update,runrunthetheaudit,audit,pickpickthethethingthingyouyouaskaskeveryeveryday,day,andandbuildbuildthatthatoneonefirst.first.
Jump back to a section
Read next
Primary sources
Checked against these on 12 October 2026. Mods change between releases. If this page and the docs disagree, trust the docs.