Claude Code v2.1.198: Chrome extension reaches general availability
The Claude in Chrome extension is now public. Background agents launched from claude agents can now commit, push, and open draft PRs in a worktree upon completion instead of pausing to prompt the user.
New features include background agent notifications via the Notification hook (agent_needs_input / agent_completed) and a /dataviz skill with a color palette validator. The built-in Explore agent now inherits the session model, capping at Opus instead of Haiku.
The Gateway now includes Claude Platform on AWS (anthropicAws) as an upstream provider, with failover logic for model-not-found errors. Network interruptions during responses, such as ECONNRESET, now trigger retries with backoff. The master /agents command is removed; sub-agents are managed via Claude or directly in .claude/agents/.
Up next
Claude Cowork sandbox bypass allows root code execution
Researchers bypassed the Claude Cowork sandbox on Windows to gain root access inside the VM. This highlights that local AI tool isolation often creates a false sense of security. Since the RPC flaw neutralizes the container barrier, users should restrict Claude Desktop on critical hosts.
Related stories
- No multiplayer in Claude Code Projects yet, Anthropic says
- Claude Code on Pro lists Haiku 5.5 as the recommended default
- Claude Code on Opus 5.5 refuses to clone existing games
- Your claude.ai skills follow you into Claude Code 2.1.275
- Claude Code projects can burn your limits faster
- Claude Code 2.1.274 warns before memory turns critical
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.