Claude Code Update Enhances Security and Stability

What happened

Anthropic has released an update to its Claude Code, introducing a range of security and usability improvements. A new warning is now included in the /schedule command when cloud tasks involve unrequested connectors, along with instructions for their removal. Additionally, a prompt is added to confirm before sending a user's email address through a web or shell tool call.

The update also introduces the onFailure: "block" option for Agent SDK hook callbacks, which stops an action from proceeding if the callback fails. A notice is added in the Code tab of the Claude Desktop when an organization's remote managed settings have not yet loaded, and another is shown once they are available.

Why it matters

These changes reflect a broader effort to enhance security, usability, and stability in the Claude Code environment. By ensuring that user data is handled more carefully and that system responses are more predictable, the update helps maintain trust and reliability in the platform.

Key details

The release includes fixes for several bugs, such as the disconnection of project and local MCP servers after the first turn in some desktop sessions and the premature exit of the claude auth status command during slow token refreshes. Additionally, the update addresses issues with the handling of scheduled tasks, including ensuring that the next run time and link are included in results after creation or modification.

Other key features include the addition of new metrics and events for OpenTelemetry, which track specific data about user interactions and system performance. The cost report for the Claude apps gateway admin API now provides detailed spending data, and the system now retains daily records of developer IdP groups for use in cost reporting.