● What changed: Claude Code can now write JavaScript that calls its own tools instead of invoking them one at a time.
● Status: experimental and opt-in. It is off until you turn it on.
● Version: shipped in v2.1.323 on 12 November.
● Also new: plugins can declare an engine version range, so a plugin can refuse to load on a version it was not built for.
Anthropic shipped an experimental Code Mode in Claude Code v2.1.323 on 12 November. Instead of calling one tool, reading the result, and deciding what to call next, Claude writes a short JavaScript program that calls the tools itself and returns one result.
The difference matters on multi-step work. A task that needs ten tool calls normally costs ten round trips, with the full output of each one landing back in the context window. Written as code, the loop runs once and only the answer comes back. That is fewer tokens spent and less context consumed on intermediate output nobody needs to read.
The trade is control. A tool call is inspectable before it runs; a program that calls tools is harder to review line by line, which is presumably why Anthropic has shipped it off by default and labelled it experimental. The plugin engine version range landing in the same release points the same way — it is a guard against a plugin running against an API that moved underneath it.
Common questions
Is Code Mode on by default?
No. It is experimental and opt-in, so Claude Code behaves exactly as before until you enable it.
What problem does it actually solve?
Token and context cost on multi-step tasks. Ten separate tool calls put ten sets of intermediate output into the context window; one program that calls those tools returns only the final answer.
What is the engine version range for?
It lets a plugin declare which Claude Code versions it supports, so it fails cleanly rather than behaving unpredictably on a version it was not built against.
Sources