Anthropic Lets Developers Reshape Claude Code With Mods

Anthropic is rolling out a new customization layer for Claude Code that goes well beyond existing configuration options. Through mods—lightweight JavaScript or TypeScript functions—developers can now intercept and modify how Claude Code operates at a fundamental level, including rewriting prompts before they reach the model, blocking or altering tool calls, managing permissions, and redesigning interface components.

Previously, Claude Code users could adjust behavior through settings, persistent instructions stored in CLAUDE.md, hooks, and MCP servers, along with cosmetic tweaks to status lines and output formatting. These mechanisms shaped Claude's instructions, permissions, and available tools, but they could not alter Claude Code's core features or interface structure itself.

Reshaping Claude's appearance and behavior

Claude Code creator Boris Cherny announced the feature on X Thursday, framing mods as a way for developers to tailor Claude's look and function through straightforward prompts, then package those modifications as installable plugins for others. "Each person works differently, so there's no reason why everyone should have an identical Claude experience," Cherny wrote.

One practical application involves displaying live information while Claude Code executes tasks. A mod could surface the count of tool calls Claude has made, elapsed time, and token consumption, then display a summary when the work finishes. Within hours of the announcement, developers had already begun experimenting with specialized applications—one engineer created a mod for passing credentials to Claude without storing the underlying secret in the conversation record.

https://x.com/bcherny/status/2105756563302723721?ref_src=twsrc%5Etfw

How mods function and integrate

The concept has been developing publicly for over a month. Anthropic first proposed the idea on GitHub on September 3 under the technical designation "function hooks," requesting developer input on an API enabling JavaScript and TypeScript functions to intercept Claude Code events. Six days later, the company announced plans to release the feature under the rebranded name "Claude Mods," while preserving function hooks as the underlying technology.

Mods are now active by default in Claude Code 2.1.287 and later versions. Because a mod's code persists throughout a session, it can retain information between events and respond to Claude Code activity in real time. This capability enables developers to build interface elements that refresh dynamically, initiate background processes, introduce new commands, or expose additional tools to the model. Anthropic has already integrated mods into features such as AGENTS.md support and the /diff pane, publishing their source code in the Claude Code repository.

Addy Osmani, a member of technical staff at Anthropic, acknowledged in a technical blog post that developers already had multiple ways to customize Claude Code. However, he emphasized that mods deliver control over Claude Code's own behavior and interface through modules that respond to session events and can modify or replace Claude Code's actions. "That makes mods a way to fit Claude Code to how you work," Osmani wrote.

Osmani illustrated the concept with Token Weather, an approximately 80-line mod that monitors context window consumption and displays the result in a band above the prompt. As Claude processes more files, the display transitions from "Clear" at 18% of a 200,000-token window, to "Showers" at 67%, and "Storm" at 81%, while also reporting recent usage and tokens added in the latest turn.

Distribution and security considerations

Since mods distribute as plugins, developers can publish them through the existing plugin system, including from GitHub repositories. This approach introduces security implications: mod code executes locally with access to the developer's machine through Claude Code, so Anthropic advises reviewing source code and installing only from trusted sources.

For organizational use, mods inherit Claude Code's current plugin controls. Team and Enterprise users and machines with managed settings receive Anthropic's security-default guard before user-installed mods load. By default, this guard prevents mods from overriding permission-deny rules; in unguarded environments, mods can override certain permission decisions.

Mods are available now in the Claude Code CLI and desktop application, and can be installed or shared through plugin marketplaces, including GitHub repositories.