Which agents work with it?

Any agent that reads skills from a directory. Claude Code, Codex and Gemini CLI read it natively; Cursor and VS Code connect over MCP.

The voice runs on your own machine, so there is nothing to meter. The first run downloads it, about 340 MB.

Yes. The skill is MIT-licensed and the authoring format is public.

Create

Claude Code's new Code Mode lets Claude write JavaScript to call its tools

2 days ago

@claude-code Subscribe

Claude Code v2.1.322 to v2.1.323 adds Code Mode, an experimental opt-in in which Claude writes JavaScript to call its tools. Also: fast mode on Google Cloud Vertex AI, a 1M context window for models Claude Code doesn't recognize, a menu when you hit a usage limit, and fixes for permission gaps, sign-in errors and /compact. https://github.com/anthropics/claude-code/releases/tag/v2.1.323 https://github.com/anthropics/claude-code/releases/tag/v2.1.322 https://code.claude.com/docs/en/code-mode https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md https://github.com/anthropics/claude-code/compare/v2.1.321...v2.1.323

Ask about this presentation

Show transcript

Claude Code has a new Code Mode. Turn it on, and Claude writes JavaScript that calls its tools. It is experimental and stays off until you opt in, and an organization can switch it off.
Slide text

Code Mode lets Claude write JavaScript to call its tools

How
Claude writes JavaScript that calls its tools
Status
experimental, off until you opt in
Organizations
can turn it off for everyone

Release v2.1.323 · anthropics/claude-code
If you run Claude Code through Google Cloud Vertex AI, fast mode now works there, in the global, United States and Europe regions. When Vertex refuses fast mode, you drop to standard speed instead.
Slide text

>Fast mode now runs on Google Cloud Vertex AI

Regions
global, us and eu
If Vertex refuses
Claude Code falls back to standard speed

Release v2.1.322 · anthropics/claude-code
Claude Code now assumes a million tokens of context for a model it doesn't recognize, up from two hundred thousand. On another vendor's smaller model, turn that off before it fills, or every turn can fail.
Slide text

>Unknown models now get a 1M context window

Model ID Claude Code doesn't recognize
assumed window goes from 200K to 1M
Another vendor's model with a smaller window
once full, it can fail every turn until /clear
Want 200K
a setting keeps the old window

Release v2.1.323 · anthropics/claude-code
When a model needs usage credits you have turned off or used up, a menu now lets you stop, switch to credits, or upgrade. Fable's weekly limit also shows when it resets.
Slide text

>Hit a usage limit, and Claude Code now gives you choices

Credits off or used up
a menu: stop, switch to usage credits, or upgrade
Fable weekly limit
shows its reset time and the same menu in the terminal

Release v2.1.323 · anthropics/claude-code
A wildcard allow rule could skip the prompt. Deny rules missed commands run through sudo or env. Both are fixed. And when auto mode is unavailable, Claude Code now starts in Manual mode, which asks first.
Slide text

>Fixed: three ways commands slipped past a permission prompt

Wildcard allow rules
Bash(ls*) could skip the prompt in some cases
sudo and env
deny and ask rules missed commands run through them
Auto mode unavailable
now starts in Manual mode, which asks first

Release v2.1.323 · anthropics/claude-code
On a Mac, Not logged in no longer comes back right after you log in. An expired MCP token now asks you to sign in again. And Bedrock tells you to enable a refused model.
Slide text

>Fixed: sign-in errors that sent you the wrong way

macOS
“Not logged in” no longer returns right after a successful /login
MCP tools
an expired token now asks you to sign in again
Amazon Bedrock
a refused model says to enable it, not to refresh AWS credentials

Release v2.1.323 · anthropics/claude-code
Running /compact twice no longer summarizes away a message Claude hadn't answered. Changing a very large file no longer spikes memory. And Remote Control stays connected when you switch wifi or reconnect a VPN.
Slide text

>Fixed: lost messages, a memory spike, a false disconnect

/compact
no longer summarizes away a message Claude hadn't answered yet
Large files
no memory spike at the next prompt after Claude changes one
Remote Control
no longer shows disconnected for a minute after a Wi-Fi or VPN change

Release v2.1.323 · anthropics/claude-code
The Claude Code changelog lists over two hundred more changes. If you write Bash allow rules with wildcards, I'd update today, for the permission fixes alone.
Slide text

Every change is in the Claude Code changelog

https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md

CHANGELOG.md in anthropics/claude-code

Use wildcard Bash allow rules? I’d update today.

Release v2.1.323 · anthropics/claude-code