Version 2.1.40

New CLI subcommands for managing authentication outside interactive mode.

claude auth login              # Sign in to your Anthropic account
claude auth login --email me@example.com  # Pre-populate email on login page
claude auth login --sso        # Force SSO login flow
claude auth status             # Show authentication status
claude auth status --text      # Human-readable output
claude auth logout             # Log out from your Anthropic account

• Replaces the need to use /login and /logout slash commands
• --json flag outputs machine-readable status (default)
• --text flag outputs human-readable status

Auth command group (search for "Manage authentication", "Sign in to your Anthropic account")

New interface for selecting and copying code snippets from Claude's responses.

• Press Ctrl+Q to open the snippet picker
• Select from multiple snippets when available
• Status bar shows snippet count: "ctrl+q to copy · N snippets"

• Keybinding changed from Ctrl+Y (copy single) to Ctrl+Q (picker)
• Works with SSH sessions
• Displays "Select a snippet to copy:" when multiple snippets exist

Snippet picker keybinding (search for "chat:snippetPicker", "ctrl+q to copy")

Administrators can restrict which models users can select.

Configure in managed settings:

{
  "availableModels": ["opus", "sonnet-4"]
}

• Accepts family aliases ("opus" allows any opus version)
• Accepts version prefixes ("opus-4-5" allows only that version)
• Accepts full model IDs
• Empty array restricts to default model only
• Clear error messages when restricted: "Model 'X' is not available. Your organization restricts model selection."

Model allowlist setting (search for "Allowlist of models that users can select", "restricts model selection")

Prevents launching Claude Code inside another Claude Code session.

• Detects when running inside an existing Claude Code session
• Displays clear error explaining the issue
• Provides bypass instructions via CLAUDECODE environment variable

Nested session check (search for "cannot be launched inside another Claude Code session")

/rename can now auto-generate kebab-case session names from conversation context.

/rename          # Auto-generates name from conversation
/rename my-name  # Still accepts manual names

• Generates 2-4 word kebab-case names (e.g., "fix-login-bug", "add-auth-feature")
• Shows "Could not generate a name: no conversation context yet" if no messages exist
• Falls back to manual name requirement if generation fails

LLM name generation (search for "Generate a short kebab-case name", "no conversation context yet")

Streams now abort automatically if no data is received for extended periods.

• Warning logged when stream is idle
• Automatic abort after configurable timeout
• Throws "Stream idle timeout - no chunks received" error

Idle timeout handling (search for "Streaming idle timeout", "no chunks received")

New setting to allow Go programs to verify TLS certificates via trustd agent.

Configure in sandbox settings:

{
  "enableWeakerNetworkIsolation": true
}

• Required for Go programs (gh, gcloud, terraform, kubectl) using MITM proxies with custom CAs
• Enables access to com.apple.trustd.agent mach service
• Security trade-off: Opens potential data exfiltration vector through trustd

Trustd agent setting (search for "Enable weaker network isolation", "com.apple.trustd.agent")

Network sandbox rules now allow binding to all interfaces (:) instead of just localhost.

Network rules (search for '(allow network-bind (local ip ":"))')

• Glob patterns are now expanded for deny paths on Linux
• Better handling of symlink-based deny path attacks (mounts /dev/null)
• Cleanup of bwrap mount points after sandbox teardown
• More nuanced logging for skipped deny paths

Linux sandbox improvements (search for "Cleaned up bwrap mount point", "Expanded glob pattern", "Mounted /dev/null")

Error messages now suggest alternative models when the selected model fails:

• 429 errors suggest switching to a fallback model
• 404 errors indicate model availability issues and suggest alternatives

Model error handling (search for "to switch to", "to pick a different model")

The Read tool now properly rejects special files (devices, pipes, sockets) with clear error messages.

File type check (search for "not a regular file (it may be a device, pipe, or socket)")

When running as a team member, pending plan approval requests from teammates can be auto-approved based on team context.

Auto-approval logging (search for "Auto-approved plan from", "auto-approving")

New notification when only one Claude Code instance can connect to VS Code at a time.

Connection note (search for "Only one Claude Code instance can be connected to VS Code")

Automatic migration from bypassPermissionsModeAccepted to skipDangerousModePermissionPrompt setting.

Migration code (search for "tengu_migrate_bypass_permissions_accepted")

• Fixed read tool to reject special file types (devices, pipes, sockets) instead of hanging (search for "not a regular file")
• Improved memory learning prompt to be more specific about saving corrections and preferences (search for "consider saving that to memory")

The following changes are internal and do not affect user behavior:

• Removed fine-grained-tool-streaming-2025-05-14 beta header
• Removed cache marker fallback logging ("No stable tool found for cache marker")
• Removed "IDE disconnected" status indicator
• Removed proxy server reconnection logging
• Removed "Team/Task files are allowed for writing" logging
• Added maxVersion capping for auto-updater
• Improved interrupt handling with stream mode logging