2.1.49 — Claude Code

Improved MCP OAuth authentication with step-up auth support and discovery caching
Added --worktree (-w) flag to start Claude in an isolated git worktree
Subagents support isolation: "worktree" for working in temporary git worktrees
Added Ctrl+F keybinding to kill background agents (two-press confirmation)
Agent definitions support background: true to always run as a background task
Plugins can ship settings.json for default configuration
Fixed file-not-found errors to suggest corrected paths
Fixed Ctrl+C and ESC being ignored when background agents running
Fixed prompt suggestion cache regression
Fixed plugin enable and plugin disable to auto-detect correct scope
Simple mode (CLAUDE_CODE_SIMPLE) now includes the file edit tool
Permission suggestions populated when safety checks trigger ask response
Sonnet 4.5 with 1M context being removed from Max plan in favor of Sonnet 4.6 with 1M context
Fixed verbose mode not updating thinking block display
Fixed unbounded WASM memory growth by periodically resetting tree-sitter parser
Fixed potential rendering issues from stale yoga layout references
Improved performance in non-interactive mode by skipping unnecessary API calls
Improved performance by caching MCP auth failures
Fixed unbounded memory growth from Yoga WASM linear memory
SDK model info now includes supportsEffort, supportedEffortLevels, and supportsAdaptiveThinking fields
Added ConfigChange hook event for enterprise security auditing
Improved startup performance by caching MCP auth failures and batching tool token counting
Fixed disableAllHooks setting to respect managed settings hierarchy
Fixed --resume session picker showing raw XML tags
Improved permission prompts to show restriction reasons