anthropic Claude Code Releases · · 2.1.196

Claude Code v2.1.196: Organization defaults, better session management, and security

sdkaiengineermedia
patch security

This release introduces organization-level default model settings for admins and improves session management with readable names and clickable attachments. Security is enhanced by preventing untrusted workspaces from auto-approving configurations. These updates benefit all users, especially administrators managing shared environments, by improving usability and security posture.

  • Support for organization default models
  • Enhanced security for untrusted workspaces
  • Improved session identification and file attachment handling
  • Fixed background job conversation deletion and prompt re-running
  • Corrected rate-limit warning and telemetry counting
Security (1)
  • Enhanced security for untrusted workspaces

    The `claude mcp list`/`get` commands no longer spawn `.mcp.json` servers that are self-approved via committed `.claude/settings.json`. Untrusted workspaces will now show a "⏸ Pending approval" status.

Features (1)
  • Support for organization default models

    Admins can now set organization-wide default models via the organization console. This default is displayed in the `/model` command as "Org default" or "Role default" when no specific model has been selected by the user.

Enhancements (2)
  • Improved session identification and file attachment handling

    Sessions now start with readable default names, making them easier to identify. Additionally, chat attachments are now clickable, allowing users to reveal files in their respective file explorers.

  • Improved reliability for background sessions and agents

    Long-running commands and workflows are now more reliable, surviving process stops, restarts, or updates, including on Windows. Workers killed by daemon restarts will automatically resume from their last checkpoint.

Fixes (2)
  • Fixed background job conversation deletion and prompt re-running

    Resolved an issue where background jobs could permanently delete their conversations and re-run the original prompt due to a transcript probe misreading a real transcript. The file is now set aside instead of being deleted.

  • Corrected rate-limit warning and telemetry counting

    Addressed a bug where the rate-limit warning would flicker off prematurely and rate-limit telemetry was over-counted when multiple parallel requests hit a usage limit simultaneously.

Read the original announcement →

https://github.com/anthropics/claude-code/releases/tag/v2.1.196