Claude Code users can now customize the agent’s runtime interface with “mods,” plug-ins written in JavaScript or TypeScript that react to events such as prompts, tool calls and screen renders. A mod can change what the developer sees, intercept default behavior or add actions.
Anthropic ships six built-in examples. One optional mod, You Should Know, runs a side agent after substantial work and flags details the main agent may have buried. In one documented session, it warned that dozens of files already contained uncommitted work before a new edit was discarded.
Users can also ask Claude to build a mod in plain language. That makes the human-facing part of an agent workflow programmable, including when a second opinion appears or which next step becomes a one-key action.
The flexibility carries costs. Each supervisor check can reread a large conversation and consume additional tokens. Frequent warnings can become noise, and third-party mods are not sandboxed: they can read and write files, see prompts and spend usage under the user’s permissions. Developers should review code before installation.