Claude Code v1.0.64

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

UI/UX Enhancements

• Model display in agent info: Agent model selection is now displayed in the agent information
• Better model descriptions: Clear descriptions for each model option:
• Sonnet: "Balanced performance - best for most agents"
• Opus: "Most capable for complex reasoning tasks"
• Haiku: "Fast and efficient for simple tasks"
• Inherit: "Use the same model as the main conversation"

Bug Fixes

• Removed duplicate MCP management code: Cleaned up redundant MCP server management implementation
• Fixed agent file path handling: Improved path resolution for agent files across different locations
• Better error messages: More descriptive error messages for invalid agent configurations

️ Architecture

• Modular agent system: Agents now have a more structured configuration with model preferences
• Improved type safety: Better validation of agent properties including model selection
• Resource cleanup: Proper cleanup of MCP resources when connections are terminated

Compatibility

• Backward compatible: Existing agents without model specifications default to Sonnet
• Survey system: Added force display option via CLAUDE_FORCE_DISPLAY_SURVEY environment variable for testing

Creating an agent with a specific model:

# In your .claude/agents/fast-analyzer.md
name: fast-analyzer
description: Quick code analysis using Haiku for speed
model: haiku
tools: [Read, Grep, Glob]

Your agent prompt here...

Using the inherit model option:

# If you're using Opus in your main conversation
claude --model opus

# This agent will also use Opus when called
name: inherit-example
description: Uses the same model as the parent conversation
model: inherit
tools: [*]

Global model override:

# Force all agents to use Haiku for faster responses
export CLAUDE_CODE_SUBAGENT_MODEL=haiku
claude

This update focuses on giving users more control over agent behavior through model selection, improving the MCP integration, and enhancing the overall agent management experience.