Claude Code v0.2.120

Enhanced CLI Tool Filtering

The permission system for CLI tools has been improved with better parsing of allowed and disallowed tool arguments:

• New comma-separated tool parsing: The --allowed-tools and --disallowed-tools CLI arguments now support comma-separated values for specifying multiple tools in a single argument

  # Allow multiple tools with comma separation
  claude-code --allowed-tools "Read,Write,Edit" 
  
  # Disallow specific tools
  claude-code --disallowed-tools "Bash,WebSearch"
  
  # Can still use multiple arguments
  claude-code --allowed-tools Read --allowed-tools Write

Async MCP Client Discovery

• Converted the MCP (Model Context Protocol) client discovery function from a wrapped promise to a native async function, improving code clarity and maintainability

Stream Import Optimization

• Consolidated stream imports: Removed redundant stream import and instead imports PassThrough directly from the stream module when needed

Process Import Cleanup

• Simplified process imports by importing cwd directly from node:process instead of importing the entire process module

The main changes revolve around the tR5 function (previously Sy0) which handles tool permissions:

// New implementation with tool name parsing
function tR5({ allowedToolsCli, disallowedToolsCli, permissionMode }) {
  let parsedAllowed = fc1(allowedToolsCli),    // Parse comma-separated values
      parsedDisallowed = fc1(disallowedToolsCli);
  return Fb0(
    {
      mode: permissionMode,
      alwaysAllowRules: { cliArg: parsedAllowed },
      alwaysDenyRules: { cliArg: parsedDisallowed },
    },
    Db0(),
  );
}

Where fc1 is the new parsing function that:

1. Handles null/undefined inputs gracefully
2. Splits comma-separated values
3. Trims whitespace from each tool name
4. Filters out empty strings

This enhancement makes it easier to specify multiple tools in scripts and command-line usage without needing to repeat the flag multiple times.