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 v2.1.319 moves slow MCP tool calls to the background

27 minutes ago

@claude-code Subscribe

Claude Code v2.1.319: MCP tool calls that a server runs as tasks now move to the background, and Claude hears the result when they finish. Also: permissions.denyInsteadOfPrompting for unattended runs, /claude-api latency-optimize, auto mode behind LLM gateways now asks or denies, self-paced /loop ends without a fallback wakeup, and three permission-check security fixes. https://github.com/anthropics/claude-code/releases/tag/v2.1.319 https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md https://github.com/anthropics/claude-code/compare/v2.1.318...v2.1.319

Ask about this presentation

Show transcript

Claude Code now supports MCP tasks in the terminal. When an MCP server runs a tool call as a task, Claude Code moves it to the background at once, and Claude hears the result when it finishes.
Slide text

Claude Code moves slow MCP tool calls to the background

MCP tasks
a tool call the server runs as a task moves to the background right away
When it ends
Claude is told the result
Where
the terminal, including Bedrock, Vertex, Foundry and LLM gateways

CHANGELOG.md · anthropics/claude-code
A new setting, deny instead of prompting, is for runs nobody is watching. A tool call that would sit waiting on a permission prompt is denied instead, and the turn keeps going. That covers subagents and teammates too.
Slide text

>Unattended runs can skip the permission prompt

Setting
permissions.denyInsteadOfPrompting
Effect
a tool call that would wait on a prompt is denied, and the turn continues
Applies to
terminal sessions, subagents and teammates

CHANGELOG.md · anthropics/claude-code
If you build on the Claude API, the /claude-api skill gets a latency optimize command. It finds where your code makes people wait, then tries faster settings against your own eval.
Slide text

>/claude-api latency-optimize finds where your app makes people wait

Finds
where your Claude API code makes people wait
Tries
faster settings against your eval

CHANGELOG.md · anthropics/claude-code
Behind an LLM gateway, or with telemetry off, auto mode now asks about or denies calls the server can't check. It no longer falls back to the built-in classifier. Bedrock, Vertex and Foundry are unchanged.
Slide text

>Auto mode behind an LLM gateway now asks or denies

Affects
auto mode with telemetry or feature flags off, including behind LLM gateways
Now
calls the server can't check are asked about or denied, with no built-in classifier fallback
Unchanged
Bedrock, Vertex and Foundry

CHANGELOG.md · anthropics/claude-code
A self-paced /loop now ends when a turn finishes without scheduling its next run. It no longer sets a fallback wakeup for you. An environment variable keeps the old behavior.
Slide text

>Self-paced /loop ends when a turn doesn't reschedule

Before
a turn that finished without rescheduling still got a fallback wakeup
Now
the loop ends there
Old behavior
an environment variable keeps it

CHANGELOG.md · anthropics/claude-code
Three security fixes. Creating a worktree could escape the sandbox. A crafted environment prefix could hide part of a Bash command from permission checks. And a sed command could skip the check in auto mode.
Slide text

>Fixed: three ways past the permission checks

Worktrees
a sandbox escape when creating a worktree
Bash
a crafted environment prefix could hide part of a command from permission checks
sed
a sed command could skip the permission check in accept-edits and auto mode

CHANGELOG.md · anthropics/claude-code
Slash commands typed after a stray space now run. /diff works in a repo with a file named HEAD. And claude -p exits with an error when auto mode refused its actions, so scripts stop reporting a pass.
Slide text

>Fixed: ignored slash commands, blank diffs and false passes

Slash commands
typed after a leading space, tab or newline, they now run instead of going to Claude as text
/diff
no longer shows "Diff unavailable" when the repo has a file or folder named HEAD at its top
claude -p
exits 1 in auto mode when its actions were refused for lack of a safety check

CHANGELOG.md · anthropics/claude-code
The Claude Code changelog lists over a hundred more changes. If you leave Claude Code running on its own, I'd update today for the security fixes and the new deny setting.
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

Leave Claude Code running alone? I'd update today.

CHANGELOG.md · anthropics/claude-code