Claude Code plugins become Claude Mods, shipping in weeks

Anthropic has committed to shipping function hooks, its TypeScript plugin layer for Claude Code, within weeks rather than days or months, and is giving the feature a product name: Claude Mods. The announcement came in a September 9 update to its GitHub issue. You can already try it by starting Claude Code with CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1 claude. The interface underneath is still being iterated on, and the team expects fewer breaking changes than in the first week.

The rename doesn’t touch the code. A mod is “just a plugin that uses function hooks, nothing is changing there,” and “function hook” stays the documented implementation primitive that mods are built on. The update says much of the semantics is now fixed. It links the source of the first three built-in mods and a cheat sheet of the affordances in v267 and v268, which the post labels “today / tomorrow”.

What a function hook can do

Today’s hooks are shell commands registered in .claude/settings.json. Each one gets a JSON payload on stdin and answers with an exit code, where 2 blocks the action and sends the error back to Claude. In the September 3 proposal, function hooks are TypeScript functions with full type definitions and LSP support. They compose like Express or Koa middleware: each hook gets a next continuation, and the first one registered wraps all the others.

Side effects go through a parameterized $ object. Anthropic says this is what keeps the system safe while still letting plugins modify Claude Code “very deeply”. Hooks can reach the interface too. Claude Code is built on React throughout, so a hook can change a component’s props or wrap what it renders. A single hook on ui.press catches the same button press in both the terminal and the desktop app. A hook on * sees every event, including every plugin’s own calls on $, and the proposal says that turns an audit log into a single function.

Admins sit on the outside

Registration order is the control model. Admins add hooks at the front of the chain to enforce control and at the end to set defaults. They can also remove affordances from $, and once one is gone, no plugin further down the chain can trigger that side effect. In the workplace, the original post says, admins will be able to restrict even what users can change, “to a very fine level of programmatic detail.”

Anthropic’s case studies are already about keeping secrets out of sight. In one, Claude takes a one-sentence request, then writes, validates and loads a plugin that replaces secrets in tool output before the model reads them. In another, a plugin in Claude Code Desktop hides sensitive values until you hover over them, so you can share your screen without exposing them. The post says its thumbnails are AI-generated but the demos and code are real and run on actual binaries.

From proposal to commitment

The September 3 post said the community’s response “likely dictates whether this ships or not.” Six days later, the team said “insightful, high-signal feedback” had materially shaped the design, and committed to shipping. It also plans to move existing Claude Code features into mod form. That would put built-in behavior on the same primitive that third-party plugins use.

The flag is public now so that more people can try it before release: “We are still iterating and it is only more important for the community to engage and give critical feedback to ensure this ends up in the right shape.”