Anthropic has released a new feature called mods that allows developers to customize Claude Code's core features and interface through small JavaScript or TypeScript functions, according to an announcement published Thursday on The New Stack. The capability goes beyond existing customization options like settings and persistent instructions, giving developers the ability to rewrite prompts before they reach the model, block or alter tool calls, manage permission requests, swap out interface components, or introduce completely new functionality. Mods represent a shift from adjusting how Claude Code operates within its existing framework to reshaping the tool's fundamental behavior and appearance.
The feature became available by default in Claude Code version 2.1.287 and later, with mods functioning as event hooks that stay active throughout a session. Because mod code runs continuously, it can retain information between events and interact with Claude Code as actions occur, making it possible to build interface elements that refresh in real time, trigger processes, introduce commands, or expose new tools to the model. Anthropic has already deployed mods internally for features including AGENTS.md support and the /diff pane, and has made their source code available in the Claude Code repository. Developers can distribute mods as plugins through the same plugin system, including from GitHub repositories. Within hours of the announcement, users began building specialized mods, including one that passes credentials to Claude without storing the secret in conversation history.
Claude Code creator Boris Cherny described mods as a way for developers to reshape Claude's appearance and behavior through a simple prompt, with the option to package those customizations as plugins that others can install. "Each person works differently, so there's no reason why everyone should have an identical Claude experience," Cherny stated on X. Addy Osmani, a member of technical staff at Anthropic, acknowledged that developers could already customize Claude Code through various mechanisms, but said that mods grant them control over Claude Code's own behavior and interface, with modules that respond to events during a session and can change or replace what Claude Code does. "That makes mods a way to fit Claude Code to how you work," Osmani wrote in a technical blog post.
The report explains that previous customization controls could influence the instructions Claude follows, its permissions, and the tools it can access, but couldn't modify Claude Code's own features or the rest of its interface. Mods address this limitation by intercepting events inside Claude Code, allowing developers to alter what happens at key moments in a session. In one example, Osmani detailed an approximately 80-line mod called Token Weather that monitors how much of Claude's context window has been used and shows the result in a band above the prompt, with the display shifting from "Clear" at 18% of a 200,000-token window to "Showers" at 67% and "Storm" at 81%, while also displaying recent usage and how many tokens the latest turn added. The ability to track live information such as tool call counts, runtime duration, and token consumption, then present a summary once a task completes, illustrates how mods can surface data that wasn't previously visible during Claude Code's operation.
Mods are now accessible through the Claude Code CLI and desktop app, and can be installed or shared via plugin marketplaces, including GitHub repositories. The report notes a security consideration: because mod code executes locally with access through Claude Code to the developer's machine, Anthropic recommends reviewing the source and installing only code from trusted publishers. For teams, mods inherit Claude Code's existing plugin controls, while Team and Enterprise users and machines with managed settings load Anthropic's sec-default guard ahead of user-installed mods, which by default prevents those mods from overriding permission-deny rules. The feature had been in development publicly for at least a month, with Anthropic first proposing the concept on GitHub on September 3 under the technical name "function hooks" before rebranding it as "Claude Mods" six days later. The extensibility model invites a marketplace dynamic where developers can share customizations, potentially fragmenting the Claude Code experience across different use cases while allowing each developer to optimize the tool for their specific workflow. Whether enterprises will embrace user-installed code with local machine access, or whether standardized mod libraries will emerge as de facto defaults, remains an open question that will shape how deeply this customization layer penetrates professional development environments.