Implement bidirectional conversation between Claude and Subconscious

[cpfiffer]

Summary

Currently, the Subconscious (Letta agent) can send messages to Claude through the hook system, and the Subconscious can see Claude's responses. However, there's no explicit mechanism for structured bidirectional conversation.

Current State

• Subconscious sends messages via letta_message tags in hook output
• Claude can read these messages and respond in regular output
• Subconscious can see Claude's responses (presumably through the hook system)

Proposed Enhancement

Create a more structured bidirectional conversation mechanism that:

• Makes it clear when Claude is addressing the Subconscious vs the user
• Potentially adds a dedicated tool or command for Claude to send messages to the Letta agent
• Improves conversation threading and context management
• Allows for richer interaction patterns between the two agents

Use Cases

• Collaborative problem-solving between Claude and Subconscious
• Subconscious providing real-time feedback or suggestions
• Testing how memory blocks affect multi-agent conversations
• Exploring emergent behavior in agent-to-agent communication

Questions

• Should there be a dedicated tool/command for Claude to message the Subconscious?
• How should we distinguish user-directed vs Subconscious-directed messages?
• What additional context should be passed between agents?