Anthropic launches Claude Code mods in JS and TS

Plugins can intercept the coding client and redraw parts of the UI. They run with the same machine access as the app, according to the company.

On 1 October 2026 Anthropic started offering Claude Code mods. Claude Code is the coding client in the terminal and the desktop app. A mod is a plugin whose behavior lives in a JavaScript or TypeScript module. The client itself calls that module when an event fires: a tool call (the moment the agent invokes a tool), a prompt going out, a piece of UI being drawn.

That is a different kind of extension. The plugin is no longer only sitting outside the client. It can change what happens next. The company described the feature in its product blog and docs. Addy Osmani’s tutorial, “Getting started with Claude Code mods,” is dated 1 October 2026. GitHub issue #91870 has an update the same day: “We’re live!” This newsroom has not tested the CLI, the command-line program. Anthropic sells Claude Code.

First: what a Claude Code mod is

Claude Code already had outside extensions: settings hooks (a shell command, HTTP call, or prompt in a settings file), skills, and MCP, the Model Context Protocol, a standard way for a model to talk to other programs. A mod is code that runs inside the process. The event arrives, the code can change what continues, and the next mod sees the result. Several mods on the same event run in load order: the first to load sees the event first and the result last.

That order does not lock down the operating system. The official blog says: “Mods run with the same access to your machine as Claude Code itself. They aren’t sandboxed, and you should only install mods from sources you trust.” Osmani’s tutorial describes a different layer: the JS/TS module runs with no DOM and no Node.js; anything outside it goes through the $ API. The process on the machine and the JavaScript environment are not the same thing. You do not need Node.js, a bundler, or a build step. Claude Code loads .js and .ts directly.

What Anthropic announced

The blog says that, on an event, a mod can rewrite the prompt, intercept a tool call, decide a permission, redact secrets in tool output, and change the UI, including buttons. The docs add what settings hooks, skills, and MCP do not do, according to the company’s table: a panel beside the transcript (the session chat log), a strip above the prompt, redrawing a spinner or dialog, stepping into a tool call, a /command with no model turn, and state across hooks.

Some native features are now mods. The example is /diff, which you can turn off in /plugin or replace. The company says it plans to move more native features over. Public code in anthropics/claude-code/mods includes diff, agents-md (loads AGENTS.md), sec-default, and telemetry. The builtin plugin cc-plugin-you-should-know is a side agent, off by default; you turn it on with /plugin enable cc-plugin-you-should-know@builtin. In the official playground, unsupported: token-weather (context-window forecast above the prompt), blast-radius (holds a risky command such as rm -rf or a force push), and replay-theater (/replay of the last turn’s diffs).

How to install it, and where it runs

The module exports register(on). Each on(event, matcher?, hook) registers a hook. The tutorial uses session.start, tool.call, command.run, and ui.render. Minimum version, according to docs checked on 9 October 2026: CLI 2.1.287 or later (claude --version); on Desktop, from 2.1.286 (Claude Code line in /status). Mods are on by default. Install with /plugin install name@marketplace or claude plugin install. Before that, claude plugin validate ./folder lists events and calls without running the code. The flags --bare, --safe-mode, and disableAllHooks stop installed mods, not native ones.

From the “Where mods run” table, the terminal and the Claude Desktop Code tab run hooks and draw. On Desktop, plugins are unavailable under WSL (Windows Subsystem for Linux). In VS Code extension chat, claude -p, the Agent SDK (the kit for building agents with Claude), and a cloud session, hooks can run but there is no drawing. RUMOR: full parity of the VS Code extension with CLI and Desktop.

The risk, according to the company

A mod, the docs say, runs with the user’s permissions: files, processes, network. It sees prompts and tool calls. It can rewrite a prompt or tool call, approve a tool without asking, and spend plan or API usage. Claude Code’s Bash sandbox does not isolate a process the mod starts. A mod cannot change the permission prompt. On Team and Enterprise plans, and on machines with managed settings, the native mod sec-default (cc-plugin-sec-default in /plugin) loads first and, according to Anthropic, stops user mods from overriding permission deny rules. The code is on GitHub. A solo user has to trust the author.

What we still don’t know

RUMOR: an official count of hook events (“about twenty”). It does not appear in the blog or the docs we read. The sources we checked also have no official count of mods in the Claude Directory. The exact publication date of the HTML at claude.com/blog/claude-code-mods did not come with a dateline; the 1 October 2026 launch is pinned by the tutorial, the issue, and the playground samples. The Decoder, by Matthias Bastian, on 3 October 2026, described the system as in-process JS/TS middleware.

Mods are in the CLI and on Desktop as of 1 October 2026. What you install still runs with the same machine access as the client.

Sources

Official

• Anthropic, Customize Claude Code with mods in TypeScript
• Claude Code Docs, mods overview
• Claude Code Docs, create a mod
• Addy Osmani, Getting started with Claude Code mods, 1 Oct 2026
• GitHub, issue #91870
• GitHub, native mods
• GitHub, playground samples
• @ClaudeDevs on X

Press

• The Decoder, Matthias Bastian, 3 Oct 2026