Claude Code adds mods: TypeScript plugins that can redraw and rewire the agent

Anthropic has opened up Claude Code's insides. Version 2.1.287, released on October 1, 2026, adds mods: plugins whose JavaScript or TypeScript code runs inside Claude Code itself and can watch, change, or take over what the agent does. The changelog entry is short ("plugins may now modify deeper behavior"), but the new documentation shows how big the change is. The New Stack also covered the launch.

What a mod is

A mod is a regular Claude Code plugin with one extra piece: a "hooks module", a .js or .ts file that exports a register function. Claude Code calls that function when the mod loads, and the mod uses it to subscribe to events such as a tool call, a submitted prompt, a slash command, or a part of the interface being drawn. Each handler can do one of three things:

• Observe the event and let it continue unchanged
• Rewrite it, for example changing a command before it runs
• Answer it, so Claude Code's normal behavior never runs

The docs list what that makes possible: drawing your own panes and bands with buttons and text fields, restyling parts of Claude Code's own interface such as the spinner, holding a tool call while you ask the user a question, adding /commands that run your code instantly with no Claude turn, and sending a request to a different model. No Node.js, bundler or build step is needed, because Claude Code loads .js and .ts files directly.

Settings hooks, the shell-script kind configured in settings.json, still exist. The difference is that a mod runs inside Claude Code's process and can draw in the interface; a settings hook runs a script from outside.

You can ask Claude to write one

According to the Create a mod guide, you can simply describe the mod you want ("make a mod that shows the current git branch above the prompt"). Claude uses a built-in plugin-authoring skill, writes the files into a per-session folder, and asks whether to enable hot reloading. After you approve, the mod reloads at the end of every turn that changes it. For hand-written mods, claude plugin validate lists every event a mod hooks and every API it calls, and claude plugin test runs automated tests without a session.

Some of Claude Code's own features are now mods too, including the /diff pane and AGENTS.md loading. A new built-in mod, You should know, runs a side agent that flags things you or Claude might miss during longer tasks. It is off by default; the changelog says to enable it with /plugin enable cc-plugin-you-should-know@builtin. Anthropic also published sample mods, such as blast-radius, which holds risky shell commands like rm -rf and shows what they would change before you approve.

The catch: mods are not sandboxed

The docs are blunt about the security trade-off. A mod "runs with your permissions": it can read and write files, read secrets in environment variables and settings files, see every prompt and tool call, rewrite them, approve tool calls before you are asked, and spend your model usage. Turning on Claude Code's sandbox does not contain a process a mod starts. One limit stays in place: a mod cannot restyle or change the permission prompt.

Practical advice for learners:

• Install mods only from authors and marketplaces you trust, the same rule as browser extensions.
• Before installing, run claude plugin validate ./some-mod and read its hooks: and calls: lines.
• To switch every installed mod off for one session, start with --safe-mode; to stop them everywhere, set "disableAllHooks": true in ~/.claude/settings.json.

Mods require Claude Code v2.1.287 or later and are on by default. They work in the terminal and the Desktop app's Code tab; in the VS Code chat panel and claude -p, their hooks run but nothing they draw appears.