Version 0.2.44

# Claude Code v0.2.44 Changelog

• New thinking mode activation: Simply say "think", "think harder", or even "ultrathink" to have Claude make a plan before executing
• Claude will enter a planning mode where it thinks through the implementation steps before coding
• After presenting the plan, you'll be prompted to exit plan mode and proceed with implementation

• Improved file path autocomplete: The autocomplete system now provides intelligent file path suggestions as you type
• Fuzzy file search: File paths are matched using fuzzy search with relevance scoring
• Directory navigation: Autocomplete shows parent directories and supports navigating through the file tree
• Smart filtering: Test files are deprioritized in autocomplete results
• Visual improvements: Better formatting and scrolling for autocomplete suggestions with a maximum of 10 visible items

• Added gd2() function to delete API keys from macOS Keychain when needed
• Improved security for API key management on macOS

• New suggestion types: Supports both command suggestions (starting with /) and file path suggestions
• Keyboard shortcuts:
• Tab: Accept the current suggestion
• Up/Down arrows: Navigate through suggestions
• Enter: Accept selected suggestion
• Escape: Clear suggestions
• Smart completion: Automatically completes common prefixes when multiple matches exist
• Visual indicators: Selected suggestions are highlighted with appropriate colors

• Triple-click detection: New F91() function detects rapid clicks (3 clicks within 1 second) to trigger special actions
• Auto-accept mode: New toggle system for automatically accepting certain tool confirmations (can be enabled/disabled per session)

• Optimized file indexing: Background file indexing for faster autocomplete responses
• Caching: File list caching to reduce filesystem operations
• Batch operations: Multiple tool calls can now be batched for better performance

• Non-interactive session support: Better handling of non-interactive sessions (e.g., when running from scripts)
• Improved error handling: More robust error handling for API authentication failures
• Code organization: Refactored autocomplete logic into separate, focused functions for better maintainability

• Fixed issues with permission mode handling
• Improved handling of root/sudo privileges with --dangerously-skip-permissions flag
• Better error messages for API key validation failures

• Updated internal version tracking and telemetry
• Refactored suggestion system to use a more modular architecture
• Improved TypeScript/JavaScript code analysis for better language detection

Note: This version focuses heavily on improving the interactive experience with intelligent autocomplete and the new thinking mode for better planning before implementation.