Anthropic updates Claude Code with expanded agent workflows

Anthropic has released updates for Claude Code, specifically versions v2.1.268 and v2.1.269, introducing enhanced capabilities for AI agent workflows and improved security protocols.

Version 2.1.269 expands the maximum number of concurrent agents in a workflow to 256, allowing for massive parallel processing of independent tasks. To manage this increased scale, a new ‘agent map’ has been added to VS Code, providing a visual interface to monitor agent status, progress, and transcripts. Additionally, a new ‘plugin eval’ feature allows developers to run evaluation suites and generate reproducible results in JSON and HTML formats.

Version 2.1.268 focused heavily on stability and security, including 96 total changes. Key improvements include fixes for security loopholes regarding permission rules for symbolic links and Bash commands, and a fix to prevent secrets from being exposed in error messages. Other updates addressed high CPU usage during idle sessions, resolved issues with third-party Anthropic-compatible endpoints, and implemented a 300-second timeout for WebFetch operations to prevent infinite waiting.