New Features

Agent Model Configuration

• Configurable AI models for agents: Agents can now specify which Claude model they should use (Sonnet, Opus, or Haiku)
• "Inherit" model option: Agents can inherit the model from the parent conversation, allowing for flexible model usage
• Environment variable override: New CLAUDE_CODE_SUBAGENT_MODEL environment variable to globally override agent model selection

  export CLAUDE_CODE_SUBAGENT_MODEL=haiku
  claude # All agents will use Haiku regardless of their configuration

️ Enhanced Agent Management

• Edit Model option: New menu option "Edit model" when managing agents through the interactive UI
• Model validation: Agent files now validate model specifications against allowed values (sonnet, opus, haiku, inherit)
• Agent file model syntax: Agents can specify their model in the frontmatter:

  name: my-agent
  description: Description of what this agent does
  model: opus  # New field
  tools: [Read, Write, Bash]

MCP (Model Context Protocol) Improvements

• Resource management: Better handling of MCP resources with proper cleanup and state management
• Tool namespacing: MCP tools are now properly namespaced with mcp__<server-name>__ prefix for better organization
• Connection stability: Improved reconnection logic for SSE (Server-Sent Events) based MCP connections

Status Hook Integration

• External status monitoring: New internal status hook system for external tools to monitor Claude Code's status
• JSON-based status communication: Status information can be passed to external commands via JSON
• Configurable timeout: Status hook operations have a 5-second default timeout