Anthropic updates Claude Code with skill synchronization and security fixes

Anthropic has released updates for Claude Code, specifically versions v2.1.275 and v2.1.276, introducing several functional improvements and security enhancements.

A major feature in v2.1.275 is the synchronization of skills and plugins enabled on a claude.ai account directly to terminal sessions using the same account. This allows users to maintain a consistent environment between the browser and the command line, though users can opt out via specific configuration settings.

To improve security, the method for installing plugins via npm has changed. The system now uses ‘npm pack --ignore-scripts’ to fetch plugins and verify integrity, preventing the execution of arbitrary install scripts. This serves as a supply chain security measure for AI agent extensions. Additionally, a bug that exposed passwords or tokens in URLs within logs and marketplace lists has been addressed.

Version v2.1.276 was released shortly after to fix a regression in v2.1.275 that caused all requests to fail with 400 errors when using the ANTHROPIC_BASE_URL in proxy or gateway environments. Other updates include a new ‘Send Now’ key (ctrl+enter) to batch queued messages and the beta launch of Projects for desktop and web versions.