Version 1.0.89

What: Filter debug messages by category using --debug=pattern syntax How to use:

# Show only MCP-related debug messages
claude --debug=mcp <command>

# Show all debug messages except verbose ones
claude --debug=!verbose <command>

# Show multiple specific categories
claude --debug=mcp,server1,api <command>

Details:

• Supports both include patterns (show only these) and exclude patterns with ! prefix (hide these)
• Automatically categorizes messages based on prefixes, MCP server names, and other patterns
• Cannot mix include and exclude patterns in the same filter
• Helps reduce noise when debugging specific components

What: A new React-based wizard system that provides consistent navigation for multi-step processes How to use: The wizard framework is currently used internally for the agent creation flow, providing a more intuitive step-by-step experience with:

• Forward/backward navigation with clear visual indicators
• Jump to specific steps when needed
• Step history tracking for complex navigation patterns
• Persistent data across all wizard steps

Details:

• Replaces the previous state machine approach for agent creation
• Provides consistent UI with step counters and navigation instructions
• Supports cancellation at any point with Escape key
• Designed to be reusable for future multi-step features

What: Permanent permissions granted during tool use are now automatically saved to settings files How to use: When you grant permanent permissions to a tool during execution, they are now automatically saved to the appropriate settings file (.claude/settings.json for project, ~/.claude/settings.json for user). No manual configuration needed. Details:

• Supports all permission types: file paths, tool rules, and directories
• Respects settings hierarchy (project vs user settings)
• Eliminates the need to re-grant permissions in future sessions
• Includes comprehensive debug logging for troubleshooting

What: Internal diagnostics system that tracks the last API request and maintains an error log How to use: This feature works automatically in the background, maintaining a rolling log of the last 100 errors and tracking the most recent API request for debugging purposes. Details:

• Helps diagnose issues when errors occur
• Maintains structured clones of API requests for analysis
• Error log has a 100-item limit with automatic rotation
• Primarily for internal debugging and support

What: 18 additional bash commands now require appropriate file permissions How to use: Commands like cat, head, tail, diff, awk, and others now trigger permission checks:

# These commands now check read permissions for the specified paths
cat /etc/passwd
head -n 10 /var/log/system.log
diff file1.txt file2.txt

Details:

• New commands include: cat, head, tail, sort, uniq, wc, cut, paste, column, tr, file, stat, diff, awk, strings, hexdump, base64, nl
• All new commands require read permissions only
• Provides clear descriptions when permission prompts appear
• Enhances security by ensuring all file access is properly authorized

What changed: Ripgrep now recovers partial results when interrupted or hitting limits Previous behavior: Lost all results if ripgrep was terminated or hit buffer limits New behavior: Preserves and returns valid partial results, trimming potentially incomplete last lines Impact: More reliable search results in large codebases where ripgrep might hit resource limits

What changed: Default API timeout increased from 60 seconds to 600 seconds (10 minutes) Previous behavior: API calls would timeout after 1 minute New behavior: API calls have 10 minutes before timing out Impact: Prevents timeouts for complex Claude interactions, especially with slower models or lengthy operations

What changed: Timeout errors now provide actionable guidance Previous behavior: Generic timeout error messages New behavior: Displays "Request timed out (check API_TIMEOUT_MS)" with clear next steps Impact: Users can easily identify and resolve timeout issues by adjusting the API_TIMEOUT_MS environment variable

What changed: TodoWrite tool now requires dual-form descriptions for all todo items Previous behavior: Todo items only had content and status fields:

{"content": "Fix authentication bug", "status": "pending"}

New behavior: Todo items require three fields including a new activeForm:

{
  "content": "Fix authentication bug",
  "activeForm": "Fixing authentication bug",
  "status": "pending"
}

Impact:

• Forces clearer articulation of both task goals and current actions
• Improves communication between AI and user about task progress
• Enables status spinner to show contextual progress messages (though todo lists remain hidden since v1.0.86)
• All three fields (content, activeForm, status) are now required for TodoWrite tool usage
• Note: While activeForm enhances the schema, users still cannot see todo lists due to display removal in v1.0.86

What changed: Added detection for musl-based Linux distributions like Alpine Linux Previous behavior: Assumed all Linux systems used glibc New behavior: Detects musl libc by checking for library files and ldd output Impact: Improves compatibility with Alpine Linux and other musl-based distributions

What changed: Ripgrep timeout is now configurable via environment variable Previous behavior: Fixed 10-second timeout for ripgrep operations New behavior: Uses BASH_DEFAULT_TIMEOUT_MS environment variable, defaulting to 120 seconds Impact: Allows adjustment for different system capabilities and codebase sizes

• Issue: When users granted "permanent" permissions through the UI, they weren't saved to disk
• Cause: Missing persistence layer between permission granting and settings files
• Resolution: Added comprehensive persistence system that writes permissions to appropriate settings files
• Affected versions: All versions prior to 1.0.89