> ## Content Index
> Fetch the complete content index at: https://claudefox.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Claude Code 2.1.323 adds an experimental Code Mode and version ranges for plugins
- URL: https://claudefox.com/claude-code-2-1-323-code-mode-plugin-engines/
- Published: 2026-11-13T10:10:35.000Z
- Updated: 2026-11-20T03:29:29.000Z
- Description: Claude Code 2.1.323 adds Code Mode, an opt-in experiment where Claude writes JavaScript that makes several tool calls in one step, and an engines field that lets plugins declare which Claude Code versions they support. The Code Mode docs are now live: here is how to turn it on and its limits.
- Author: hupicake
- Tags: Tools

*Updated November 20, 2026: Updated: the Code Mode documentation, missing when we first wrote this, is now live. Added setup steps, requirements, permission behaviour and limits.*

Anthropic released Claude Code 2.1.323 on November 12\. The release brings two changes worth knowing about: an experimental Code Mode, and a way for plugin authors to say which Claude Code versions their plugin works with. Code Mode now has its own documentation page, which this article draws on; we checked it on November 20, and some behaviour it describes depends on later versions.

## What Code Mode does

Normally Claude asks for one tool call at a time, such as Bash, Read or Edit, and reads each result before deciding the next step. In Code Mode, Claude writes a short piece of JavaScript that makes the tool calls, and Claude Code runs that code in your session. One piece of code can run commands in parallel, feed one result into the next call, and return only the part of the output Claude needs. The docs give this example of a retry-limit change:

```
const files = await $`rg -l "function retryPolicy" src`.text();
const target = files.trim().split("\n")[0];
await edit(target, "maxAttempts: 3", "maxAttempts: 5");
({ edited: target, tests: $`npm test -- retryPolicy`.nothrow() });
```

You still see the individual tool calls, such as an edit's diff, not the code; Ctrl+O opens the transcript, where each piece of code is a collapsed Code row. The docs suggest it for one change repeated across many files, a search whose results need filtering, or several independent commands.

## How to turn it on

Check that you are on 2.1.323 or later with claude --version, and run claude update if not. On the stable channel the update may not reach 2.1.323 yet; wait, or switch to the latest channel. Then start a session with the environment variable set (in Windows PowerShell, set $env:CLAUDE\_CODE\_EXPERIMENTAL\_CODE\_MODE = "1" and then run claude):

```
CLAUDE_CODE_EXPERIMENTAL_CODE_MODE=1 claude
```

To keep it on, put the same variable in the env block of \~/.claude/settings.json. Claude Code ignores it in a project's settings files, so a cloned repository can't switch Code Mode on for you. The first time, Claude Code asks you to confirm, restarts and says the session can only be resumed with Code Mode on, so keep the variable set for claude --continue or --resume. To turn it off, remove the variable or set it to 0\. Administrators can block it organisation-wide with 0 in managed settings, but cannot switch it on for users that way. If it fails to start, a line saying Code Mode isn't available gives the reason, which /status also shows.

## Requirements and limits

- Models: Claude Opus 5.5, Sonnet 5.5, Haiku 5.5 or Fable 5.2, on a Claude subscription or Console account.
- Not available through a gateway, Amazon Bedrock, Claude Platform on AWS, Google Cloud's Agent Platform, Claude Models on Google Cloud (Anthropic managed), Microsoft Foundry, or for organisations using the HIPAA configuration or zero data retention. On Windows, Git for Windows must be installed.
- Interactive terminal sessions on the native build only. It does not work with -p, the Agent SDK, the VS Code panel, the Desktop app or cloud sessions.
- Some settings stop it from starting, including an API key helper, OpenTelemetry export and DISABLE\_TELEMETRY.
- Code that Claude waits on stops after ten minutes. In auto or bypassPermissions mode, code can run in the background for up to two hours. Press Esc to stop running code, or use /tasks and x for background code.
- Remote Control, Claude in Chrome and computer use are unavailable in a Code Mode session.

## Permissions work differently

Each tool call inside the code still goes through your deny and ask rules and your PreToolUse hooks under its usual tool name. In auto mode, though, the classifier reviews the piece of code before it runs, not each call, and can't see values computed at run time, such as a path taken from search output. To approve code before it starts, add an ask rule for REPL, the tool that runs the code. PreToolUse hooks that parse shell commands should check each part of a command, because a call can arrive as something like cd 'app' && git push. Shell commands follow your sandbox settings, but the JavaScript itself and its file, web and MCP calls run outside the sandbox.

## Plugins can now declare supported versions

A plugin's plugin.json can now include an engines field with a semver range under the claude-code key, as in the docs' example "engines": { "claude-code": ">=2.1.400" }. Plugins outside their stated range are not installed, updated or loaded, and install and update pick the newest plugin version that fits. There is a catch for older clients. Versions 2.0.15 through 2.1.75 reject a manifest containing the field with an Unrecognized key error, so leave it out if you still support them. claude plugin validate also now type-checks a plugin's TypeScript hooks module and reports findings as warnings, which fail the run under --strict.

## Smaller changes worth acting on

- Claude Code now warns at startup about Bash allow rules with a \* inside a variable's value, such as Bash(PORT=\* npm run dev).
- CLAUDE\_CODE\_DISABLE\_AGENT\_PROGRESS\_SUMMARIES=1 turns off subagent progress summaries and the requests that write them.
- Running /compact a second time no longer summarises away a message Claude had not answered yet.
- Sign-in fixes include sessions on a named profile moving to another sign-in in the background, and an expired claude.ai login showing "Credit balance too low".

## Related on claudefox

- [What a Claude Code mod can do on your machine, and the guard many solo users don't get](https://claudefox.com/what-a-claude-code-mod-can-do-and-the-guard-solo-users-dont-get/)
- [Floorplan: a Claude Code and Codex plugin that maps your codebase in plain English](https://claudefox.com/floorplan-plain-english-codebase-map-claude-code-codex/)
- [BLA (bla-agents): a CLI and MCP server that trips AI reverse-engineering limits on release files](https://claudefox.com/bla-agents-mcp-cli-ai-reverse-engineering-defense/)

## Our take

In our view, Code Mode targets the round trips where Claude reads every intermediate result just to pick the next obvious step, so bulk edits and filtered searches could need less waiting. The trade-off is review: you audit a program rather than single calls, and auto mode's classifier sees less. Try it on a repository you know well, with ask rules for anything that pushes, deletes or deploys. Plugin authors should add an engines range now, unless they still serve the old versions that reject it.

Source: [Claude Code v2.1.323 release notes (GitHub)](https://github.com/anthropics/claude-code/releases/tag/v2.1.323?ref=claudefox.com), [Run tool calls from code with Code Mode (Claude Code docs)](https://code.claude.com/docs/en/code-mode?ref=claudefox.com), [Plugins reference: plugin.json fields (Claude Code docs)](https://code.claude.com/docs/en/plugins-reference?ref=claudefox.com), [Claude Code changelog](https://code.claude.com/docs/en/changelog?ref=claudefox.com)

This story first appeared in the [claudefox daily briefing for November 13, 2026](https://claudefox.com/daily-2026-11-13/).