Claude Code v0.2.47

# Claude Code v0.2.47 Changelog

• Press Tab to auto-complete file and folder names while typing paths
• Works in interactive mode when entering file paths
• Speeds up navigation and reduces typing errors

• Press Shift+Tab to toggle auto-accept mode for file edits
• When enabled, file edits are automatically accepted without confirmation prompts
• Visual indicator shows when auto-accept is active: ⏵⏵ auto-accept edits on
• Particularly useful for workflows with many trusted file modifications

• Conversations can now continue indefinitely without hitting token limits
• Claude automatically compacts older messages when approaching token thresholds
• Toggle this feature on/off using the /config command
• Preserves conversation context while managing token usage efficiently

• New environment variable MAX_THINKING_TOKENS allows precise control over thinking token allocation
• Set a specific number of thinking tokens instead of relying on keyword detection
• Example: export MAX_THINKING_TOKENS=8000

• Enhanced redaction of sensitive information in error messages and stack traces
• Automatically redacts:
• Anthropic API keys (sk-ant-*)
• AWS keys (AKIA* patterns)
• GCP keys (AIza* patterns)
• GCP service accounts (@.iam.gserviceaccount.com)
• Generic API keys in various formats

• Bug report title generation now creates more technical, descriptive titles
• Better extraction of key error messages for clearer issue tracking
• Fallback to simple title extraction if AI generation fails

• Permission prompts now show numbered options (1, 2, 3) for easier selection
• Type the number to quickly select an option
• Escape key consistently selects "No" across all prompts
• More consistent formatting and better visual hierarchy

• Stream imports updated from default imports to named imports for better compatibility
• Process imports now use node:process prefix for clarity
• Removed unused imports and variables for cleaner codebase
• Added fallback behavior for bug report title generation

# Start typing a path and press Tab
$ claude read /home/user/proj[TAB]
# Completes to: /home/user/projects/

$ claude edit ./src/comp[TAB]
# Shows available completions or completes if unique

# During a conversation, press Shift+Tab
# You'll see: ⏵⏵ auto-accept edits on (shift+tab to toggle)

# All subsequent file edits will be automatically accepted
# Press Shift+Tab again to disable

# Set a specific thinking token limit
$ export MAX_THINKING_TOKENS=15000
$ claude

# Now all thinking operations will use exactly 15000 tokens