anthropic Claude Code Releases · · 2.1.193

Claude Code v2.1.193: New settings, improved background agents

sdkaiengineer
patch feature

This release introduces a new `autoMode.classifyAllShell` setting to enhance security by classifying all shell commands. It also adds auto-mode denial reasons to transcripts and toasts, and an OpenTelemetry log event for model responses. The update includes several improvements to background agents, fixes for UI and background task issues, and enhanced plugin and MCP authentication.

  • Auto-mode classification for all shell commands
  • Auto-mode denial reasons and response logging
  • Live file path autocomplete and startup notices
  • Automatic memory-pressure reaping for idle background shells
  • Fixed stale UI state after login and background task cancellations
Features (3)
  • Auto-mode classification for all shell commands

    A new `autoMode.classifyAllShell` setting is introduced to route all Bash and PowerShell commands through the auto-mode classifier, enhancing security beyond arbitrary-code-execution patterns.

  • Auto-mode denial reasons and response logging

    Auto-mode denial reasons are now available in the transcript, denial toast, and `/permissions` recent denials. A new `claude_code.assistant_response` OpenTelemetry log event provides the model's response text, configurable via `OTEL_LOG_ASSISTANT_RESPONSES`.

  • Live file path autocomplete and startup notices

    Live file path autocomplete is now available for bash mode (`!`). A startup notice is displayed when MCP servers require authentication, directing users to `/mcp`.

Enhancements (1)
  • Automatic memory-pressure reaping for idle background shells

    Idle background shell commands will now be automatically reaped due to memory pressure. This behavior can be disabled by setting `CLAUDE_CODE_DISABLE_BG_SHELL_PRESSURE_REAP=1`.

Fixes (2)
  • Fixed stale UI state after login and background task cancellations

    Issues with `/model` and other client-data-gated UI components showing stale or empty states immediately after `/login` have been resolved. Spurrious cancellations of background tasks with "N background tasks would be abandoned" messages have also been fixed.

  • Resolved phantom sub-agent creation and agent panel issues

    A phantom "general-purpose (resumed)" sub-agent that incorrectly re-ran the main conversation when backgrounding the main turn has been fixed. Issues with the agent panel hiding sibling agents when viewing a subagent are also resolved.

Read the original announcement →

https://github.com/anthropics/claude-code/releases/tag/v2.1.193