Anthropic and Trump Officials Near Model Deal

Highlight

Claude Code Adds Artifacts for Live Previews

You can now preview in-progress work from a Claude Code session as a live, interactive artifact built from your full session context. The artifact stays linked to the session, updates as work progresses, and can be shared with teammates without exporting code or running a separate dev server. See the artifacts announcement for details on what's supported.

Worth Knowing

Agent Teams Drop Setup Step

With CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 set, the TeamCreate and TeamDelete tools are gone -- every session has one implicit team and you spawn teammates directly through the Agent tool's name parameter. The team_name parameter is still accepted but ignored, so existing prompts keep working.

Auto Mode Blocks Destructive Git and IaC Commands

Auto mode now blocks git reset --hard, git checkout -- ., git clean -fd, and git stash drop when you didn't ask to discard local work, blocks git commit --amend on commits the agent didn't make this session, and blocks terraform destroy, pulumi destroy, and cdk destroy unless you asked for the specific stack. Subagent spawns are also evaluated by the classifier before launch.

Set Any Setting From the Prompt With /config

You can now run /config key=value from any prompt to change a setting inline -- for example /config thinking=false -- and it works in interactive mode, -p, and Remote Control. Run /config --help to see the shorthand keys, and the toggle UI now uses Enter or Space to change and Esc to save and close.

Permission Rules Can Now Match Tool Parameters

Permission rules support a new Tool(param:value) syntax with * wildcards, so you can write rules like Agent(model:opus) to block Opus subagents specifically. This gives you finer control over what subagents and tools can do without blocking the tool wholesale.

Nested .claude Directories Now Resolve by Proximity

When agents, workflows, output-styles, or skills exist in multiple .claude/ directories, the one closest to your working directory wins on name collisions, and nested skills load when you're working on files there. On a skill name clash, the nested skill appears as <dir>:<name> so both stay available.

Suppress Mobile Push Notifications When You're at the Machine

Set CLAUDE_CLIENT_PRESENCE_FILE to a marker file and Claude Code will suppress mobile push notifications while that file exists. Useful if you run remote sessions and don't want your phone buzzing when you're already at the keyboard.

Under the Hood

Subagent Transcripts, Mid-Stream Drops, and Background Tasks

Viewing a subagent's transcript now shows tool results and live progress, messages sent while a subagent finishes its turn are no longer dropped, and backgrounding a running subagent with Ctrl+B no longer restarts it. Mid-stream connection drops now preserve partial responses instead of showing a raw error, and background tasks started by a teammate are no longer killed when the teammate finishes a turn.

Faster Startup and Reliable Writes on Network Drives

A 2.1.169 regression that added ~120ms to every launch in fresh environments is fixed, and startup no longer blocks with a blank terminal for up to 15 seconds when the account settings fetch is slow. Write and Edit no longer produce 0-byte or truncated files on network drives and cloud-synced folders.

From Anthropic

In the News