Version 1.0.86

What: New /context command that displays current token usage as a colored grid How to use:

# In your Claude Code session, type:
/context

Details:

• Shows a visual grid where each square represents a portion of your context window
• Different colors indicate different categories (system prompt, tools, messages, etc.)
• Displays detailed breakdowns including memory files, MCP tools, and custom agents
• Grid size adapts based on model capacity (20x10 for ≥1M token models, 10x10 otherwise)
• Helps identify what's consuming context and why you might hit limits

What: Added security protection for .git/config and .gitconfig files How to use:

# Claude will now ask for explicit permission before editing:
# - .git/config files in repositories
# - .gitconfig files (global git configuration)

Details:

• Prevents potential code execution via git configuration files
• Requires explicit user permission to edit these sensitive files
• Protection covers both repository-level and global git configurations
• Provides clear explanation about why permission is needed

What: Context-aware suggestions when Claude requests file permissions How to use:

# When Claude needs file access, you'll see smart suggestions like:
# For reading: "Allow reading from /directory/** during this session"
# For writing: "Allow all edits in /directory/ during this session"

Details:

• Different suggestions for read vs write operations
• Suggests appropriate permission scope based on the operation
• Can suggest switching to "acceptEdits" mode for write operations
• Provides rules that can be applied to efficiently grant permissions

What changed: Simplified and consolidated the project trust dialog Previous behavior: Multiple separate warnings with complex dynamic options New behavior: Single unified warning with clear execution source listing Impact:

• Clearer messaging about security risks
• Better organization showing what allows execution (MCP servers, settings files, commands)
• Simplified to just "Yes, proceed" or "No, exit" options
• Shows configuration sources in a more digestible format

What changed: Todos are now saved to disk when modified Previous behavior: Todos only existed in memory during the session New behavior: Todos persist to JSON files on disk (though not automatically loaded on restart) Impact:

• Todo changes are saved for potential future recovery
• TodoWrite tool now enforces keeping at least one task in_progress
• Visual feedback simplified - completed tasks always show in green

What changed: TodoWrite tool now controlled by runtime feature flag instead of hardcoded visibility Previous behavior: TodoWrite had hardcoded visibility conditions New behavior: Todo list visibility controlled by todoFeatureEnabled feature flag (defaults to disabled) Implementation: Tool inclusion controlled by dynamic feature flag system rather than hardcoded conditions Impact:

• Todos become opt-in feature rather than always-on
• Tool functionality remains fully implemented but filtered out at runtime by default
• Users must explicitly enable todo feature to see todo list management
• Creates configurable task management system
• Todos still persist to disk and function when enabled

What changed: Improved todo list visualization with intelligent truncation (for internal use) Previous behavior: Simple list display without truncation New behavior: Shows up to 4 most relevant items with counts for hidden items Impact: (Note: These improvements exist in code but users never see them due to display removal)

• Prioritizes showing in-progress tasks
• Displays "... +N done" for multiple completed tasks
• Shows "... +N more" for remaining tasks
• Would keep focus on active work while maintaining context (if visible)

What changed: Operation-specific permission messages Previous behavior: Generic permission messages for all operations New behavior: Different messages for read vs write operations Impact:

• "Allow reading from /directory/" for read operations
• "Allow all edits in /directory/" for write operations
• Clearer about what permissions are being granted
• Better session-based permission options

What changed: Improved handling of file attachments in conversations Previous behavior: Attachments could be separated from their context New behavior: Attachments are properly grouped with associated messages Impact:

• File attachments stay with their relevant assistant or user messages
• Prevents context loss when processing conversations with attachments
• Better message organization for complex conversations

• Issue: Spinner words were stored as a plain array limiting extensibility
• Cause: Original implementation didn't account for future configuration needs
• Resolution: Restructured as an object with words property allowing future expansion
• Affected versions: All versions prior to 1.0.86