Anthropic has released Claude Code v2.1.271, which introduces several new capabilities for remote sessions and improves the security of sandboxed environments.
Claude Code v2.1.271 Released — Fast mode for remote sessions and enhanced sandboxing
New Features and Improvements
- Fast mode for remote sessions: Added fast mode support for Claude Code Remote sessions (both cloud and self-hosted runners).
- Enhanced UI controls: Added mouse support to the
/configpanel in fullscreen mode, including scrolling and clicking to change settings. - Granular sandboxing: Added per-command
allowed_domainsfor Bash, PowerShell, and Monitor in auto mode with sandboxing to restrict command access to specific hosts. - Expanded subagent control: Added
omitClaudeMdto agent frontmatter and--agentsJSON, allowing custom and plugin subagents to run without requiring localCLAUDE.mdfiles. - Plugin installation safety: Added
--accept-command <sha256>toclaude plugin installandupdateto ensure exact command execution. - Improved pricing support: Added support for a
multiplierup to 10 inmodelPricingsettings for internal chargeback rates.
Bug Fixes
- Policy and credential management: Fixed issues where cached organization policies were reused after account switches or where policies failed to refresh during active sessions.
- Sandboxed command fixes: Resolved issues where Bash permission checks skipped files under certain wildcard conditions or failed to recognize specific command flags.
- Session stability: Fixed various issues related to resumed sessions (via
/resumeor/continue) where tools, context windows, or artifacts were not properly restored. - MCP handling: Fixed multiple issues related to Model Context Protocol (MCP), including OAuth registration errors, tool search mismatches, and improper disconnection handling.
- Terminal and UI glitches: Fixed rendering issues in terminal displays, including incorrect status updates for long-running tasks and UI flickering on Windows.