Claude Code Introduces "Mod" Feature | Now Fully Customizable
機械翻訳 / Machine-translated
A
AI Friends
4時間前・11分で読める
機械翻訳 / Machine-translated
@aifriends
AI Friends（https://aifriends.jp）のクロスポスト公式アカウント。AIツールの紹介・使い方・できることを、中学生でもわかるやさしい日本語で届けます。
What You'll Learn in This Article
On October 1, 2026, "Mod" — a new extension feature — was added to Claude Code, the AI coding tool developed by Anthropic. With this feature, users can change how the tool behaves and looks to suit their own preferences.
A Mod is a small program written in TypeScript. Developers can now freely adjust how Claude Code works by writing just a few lines of code.
Even more surprisingly, you can ask Claude Code itself to create a Mod for you. Simply tell it what kind of feature you want, and the AI will automatically build the Mod.
Mods enable the following capabilities.
First, prompt rewriting. You can modify messages sent to the AI mid-stream or automatically append specific instructions. For example, you could have it automatically add "Always respond in English" to every request.
Next, tool execution control. You can display warnings before dangerous commands are run, or automatically approve certain operations — helping prevent the AI from going off the rails.
Custom UI additions are also possible. You can create your own control panels with buttons, input forms, and dedicated display panels. For instance, you could add a dedicated panel for visually reviewing code changes.
Finally, information protection is another important feature. Sensitive data such as API keys and passwords can be automatically hidden before the AI reads them.
Anthropic has published several official Mods. Here are some notable examples.
Blast Radius acts as a safety mechanism when executing dangerous commands. For example, if a file deletion command like rm -rf is about to run, a warning appears on the right side of the screen showing exactly which files will be affected — helping prevent accidental operations.
Replay Theater is a Mod that records file change history. By typing the /replay command, you can watch changes play back step by step. Being able to review what was changed and how makes it useful for learning and verification.
Token Weather is a Mod that visualizes AI processing volume. After a task runs, it displays how many tokens were consumed using weather icons — sunny means plenty of headroom, rainy means you're approaching the limit.
Using Mods is straightforward. Run the /plugin command in either the CLI version or desktop app of Claude Code. A list of available Mods will appear, and you can choose and install whichever ones you like.
In Claude Code version 2.1.287 and later, the Mod feature is enabled by default. No special configuration is required.
You can also create your own Mods. Write your code in TypeScript and register it in Claude Code's plugin directory, and you're done. Even if you're not a strong programmer, you can ask Claude Code itself to "build a feature like this," and the AI will generate the code for you.
Note, however, that Mods created on the spot disappear when the session ends. If you want to use a Mod continuously, copy the created Mod's folder to another location and reinstall it as a regular plugin.
The arrival of the Mod feature has the potential to bring significant change to the world of AI coding tools.
Until now, competing tools such as Cursor and GitHub Copilot have offered a degree of customization. But Claude Code's Mods go a step further by allowing users to reach into the tool's internal behavior — enabling deep-level customization such as prompt rewriting and tool execution control.
Mods can also be distributed as plugins, making them shareable across teams and organizations. According to an article by Gijutsu-Hyohron, it is also possible to apply to have your own Mods listed in Claude's official plugin directory.
In other words, an ecosystem where the developer community collaboratively builds and shares Mods could emerge. As more useful Mods appear, Claude Code's capabilities are likely to expand rapidly.
While the Mod feature is convenient, there are important points to be aware of — namely, security risks.
Mods operate with the same permissions as Claude Code itself. This means a Mod can freely access files and data on your computer. Moreover, since Mods do not run in a sandbox (an isolated safe environment), a malicious Mod could cause serious damage if it makes its way onto your system.
Anthropic officially advises users to only install Mods obtained from trusted sources. Mods of unknown origin or with poor ratings should be avoided.
Caution is also needed when creating your own Mods. Writing code that causes unintended behavior could damage your own system. Even when asking Claude Code to create a Mod, always review the content of the generated code.
/plugin command (version 2.1.287 and later)This article is a cross-post from AI Friends.