You may want a frequently checked metric beside the Claude Code conversation, or a custom check before a tool runs. Mods bring those changes inside Claude Code. Announced by Anthropic on October 1, 2026, they reach beyond a setting that changes the tone of replies.
A mod is extension code running inside Claude Code
Mods consist of JavaScript or TypeScript event handlers: small functions called when something happens, such as a tool call, a submitted prompt or an interface being drawn. A handler can observe the event, change it or replace its normal behavior.
This allows a pane beside the conversation, an information band above the prompt, buttons or custom commands. A mod can rewrite a prompt before it reaches the model, or hold and modify a tool call. One Anthropic example displays how full the context window is—the information carried into the model for the session. The useful idea is to bring working information into view at the point where you need it.
Not every preference needs a mod
For shorter answers, an output style is the more direct setting. For repeated instructions governing one kind of task, use a skill. A permission mode separately controls which tool actions need approval.
Consider a mod when you need an interface element or intervention in the event flow. Installing executable extension code just to shorten replies adds maintenance and access for a small communication preference. Starting with the smallest feature that meets the need keeps the setup easier to understand.
Check the version and start with one change
Current documentation requires Claude Code 2.1.287 or later in the terminal, or the embedded Claude Code 2.1.286 in Desktop’s Code tab. Check with claude --version in the terminal or /status in a local Desktop session. Visual support in the VS Code chat panel starts with extension 2.1.299; sessions without feature-flag fetching require 2.1.304.
- Describe one change. For example: “Build a small mod showing the number of tool calls; leave other behavior unchanged.” Claude can prepare the code.
- For a ready-made mod, review the publisher and plugin files. Mods install as plugins, managed through
/pluginin the terminal. - Before installation, run
claude plugin validate ./some-mod, replacing the example path with the actual plugin folder. Itshooks:andcalls:lines list the declared events and operations without running the mod. - After loading, check that one intended behavior. If you install or update the plugin externally while a session is open, use
/reload-plugins; it also loads when Claude Code next starts.
Look beyond the visible interface
Mods run with your user account’s access and are not sandboxed. They can read and write files, start processes, make network requests, see the conversation and consume model usage. Review those capabilities along with the appearance. Your organization can also restrict which mods load.
Mods do not run in Desktop WSL sessions, where plugins are unavailable. With Remote Control from a phone, the code runs in the session on your computer and its custom interface stays in the local terminal. Disable the relevant plugin when you no longer need it, keeping the workspace limited to customizations that serve the work.
