Make Claude Code Work Your Way With Mods

Anthropic added mods to Claude Code on October 1. A mod can put a new readout on your screen, add a button, rewrite a request before Claude sees it, or swap out one of Claude Code's own features. You don't have to write one yourself. Describe the mod you want in a Claude Code session, and Claude writes it and switches it on for you.

What a mod is

Claude Code writes and runs code on your computer to build whatever software you describe. You use it in a terminal, the plain text window where you type commands, or in the Code tab of the Claude desktop app, and mods work in both.

Every time Claude Code does something, like running a tool or drawing part of the screen, it sends out a signal Anthropic calls an event. A mod is a few lines of code that listens for one of those events and can watch it, change it or take it over. The code is JavaScript or TypeScript, two common programming languages, but Anthropic's guide says you don't need to learn either to try one.

Mods come bundled inside plugins, the packages of add-ons Claude Code already used, so you share and install them the same way: by typing /plugin in Claude Code, or from Anthropic's Claude directory. Anthropic has already rebuilt some of Claude Code's own features as mods, including the view that shows what Claude changed in your files, and says more will follow.

Three things a beginner could ask for

A gauge for how full the conversation is getting. Claude can hold only so much of a conversation at once, and when it runs out of room, Claude Code compresses the conversation into a summary. Anthropic's docs say detailed instructions from early on can get lost when that happens. Anthropic's sample mod Token Weather draws a one-line forecast above the box where you type: Clear when there's plenty of room, then Cloudy, Showers and Storm as the space fills up, and finally "Compact soon." The guide also suggests a cost or usage-limit meter.

A pause before anything risky. The Blast Radius sample steps in when Claude tries to run a command that deletes files or wipes out recent work. It holds the command, shows what it would touch, and asks you first, with Proceed and Cancel buttons. Anthropic notes that some commands can slip past it, and points to Claude Code's permission rules for a hard block.

A quiz on what Claude just built. A community pack called ModSmith, shared on GitHub after launch, includes a mod called quiz-after. When Claude finishes a task that changed something, it shows two or three questions about what was built, with buttons to reveal the answers. Another mod in the pack shows what Claude assumed along the way.

How to try one

You need Claude Code version 2.1.287 or later, and typing claude --version in your terminal shows which version you have. Mods are on by default from that version, so there's no setting to change.

The simplest route is to ask. Start Claude Code and describe what you want in plain words, like this idea from Anthropic's guide: "Make me a mod that sends me a pop-up when a long task finishes." Claude asks once whether to turn on hot reloading, which lets the mod update in place while you work. Approve it, and Claude writes the mod and switches it on when it finishes replying, ready for more tweaks. A mod made this way lasts for that session, and Anthropic's docs say to copy its folder somewhere of your own to keep it.

The other route is installing a set of mods someone else made. Anthropic shares its three samples on GitHub, as they are and without support. ModSmith installs all six of its mods with two /plugin commands typed into Claude Code, listed on its GitHub page, and its README notes that the quiz uses a little extra of your usage.

The safety catch

Mods run with the same access to your computer that Claude Code has, like any program you install, and Anthropic says they aren't sandboxed (walled off from the rest of your machine). According to the docs, a mod that has loaded can read and write any file your account can reach, read secrets like API keys kept in your settings, see every prompt you send, approve an action before you're asked, and spend your usage by calling a model on your plan. It can't change the permission prompt Claude Code shows you.

Anthropic's advice is to install mods only from sources you trust. Before installing one, you can run claude plugin validate on its folder, which lists the events it listens for and what it asks Claude Code to do, such as reading a file or making a network request, without running it. If a mod misbehaves, you can disable it from the /plugin screen, or start Claude Code in safe mode to switch off your installed mods for that session.

Until now, changing Claude Code mostly meant editing its settings or writing it instructions. With mods, a beginner can ask Claude for the guard or gauge they wish it had, and Claude builds it on the spot.

Anthropic rebuilding its own features as mods reads to me as a sign it expects people to shape Claude Code themselves, one small add-on at a time.

For anyone learning to build software with Claude, the quiz mod is the one I like most. It's easy to end up with a project Claude built that you couldn't explain back, and a few questions after each task keep you closer to what's in your own app.

Sources

• Claude blog: Customize Claude Code with mods
• Anthropic: Getting started with Claude Code mods
• Claude Code Docs: Mods overview
• Claude Code Docs: Create a mod
• Claude Code Docs: Plugin security and trust
• Claude Code Docs: How Claude Code works (context and compaction)
• Anthropic's sample mods on GitHub
• ModSmith on GitHub