The batching feature was timing out due to run_in_executor not properly
executing the sync function from async context. Simplified by calling
the sync function directly, which is fine since we process one batch
at a time anyway.

Additional improvements:
- Add Ctrl+C signal handling to properly kill all agents in parallel mode
- Replace console.print with proper logger calls throughout
- Remove emojis from logging output for cleaner multi-agent logs
- Add publish_to_thought_stream tool that auto-registers on startup
- Tool uses raw requests instead of atproto client for simplicity
- Tool accepts single string (max 5000 chars) instead of list
- Upsert tool on startup to ensure latest version is always used

Batching now works correctly with any batch_size setting.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

- Add processing_queue flag to prevent concurrent queue processing
- Check processing flag before attempting to process queue
- Cancel and clear flush task when processing begins
- Add proper cleanup in finally block to always reset flag
- Improve shutdown handling to cancel pending tasks
- Process remaining messages on shutdown before exit

Fixes issue where multiple timeout flushes would trigger simultaneously,
causing the same batch of messages to be sent to the agent multiple times.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

- Add automatic retry with exponential backoff (2 retries) for ReadTimeout
- Catch ReadTimeout specifically to distinguish from other errors
- Continue bridge processing on agent errors instead of crashing
- Show cleaner error messages with error type name
- Add retry attempt logging for debugging

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

- Add agent name prefix to all log messages for multi-agent identification
- Suppress noisy HTTP logs from httpx/httpcore/letta client libraries
- Improve agent connection display with cleaner formatting
- Add better debug logging for message flow tracking
- Consolidate verbose logs to debug level for cleaner output
- Make agent status more prominent with bold formatting

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

Enhanced run_agent.sh to accept multiple agent names, enabling parallel
execution of agents in both bridge and listener modes. Agents are
automatically validated and run concurrently.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

- Add queue_flush_timeout configuration for batched message processing
- Implement automatic queue flushing after timeout to prevent message delays
- Increase character limit from 1000 to 5000 for blip content
- Update setup wizard to include listener configuration for periodic messaging
- Enhance run_agent.sh script to support both bridge and listener modes
- Configure herald and pedant agents with batch_size: 30 and queue_flush_timeout: 120s
- Add agent self-filtering to prevent feedback loops
- Improve error handling and logging throughout the system

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

- Added message_source field explaining the thought stream system
- Added context field indicating messages are from other users
- Provides agents better understanding of message origins and purpose