Implements a lightweight hook system with register_hook() and trigger_hooks() for UserPromptSubmit, PreToolUse, PostToolUse, and Stop events. Demonstrates the observer pattern for adding cross-cutting concerns (permissions, logging) without modifying the core agent loop. Self-contained, framework-agnostic, and reusable in any Python application requiring event-driven extensibility.

Defines a TOOL_HANDLERS dictionary mapping tool names to functions, with tool schemas for input validation. Shows how to safely extract tool_use blocks from LLM responses, execute corresponding handlers, and format tool_result outputs. Isolated pattern that can be extracted with minimal modification; only requires defining handler functions and schemas. Demonstrates secure tool dispatch for LLM-mediated API interactions.

Provides a MessageBus class with file-backed JSONL mailboxes for persistent agent communication. Implements request/response correlation using message IDs, condition variables for efficient signaling, and file locking for cross-process safety. Self-contained component usable in any Python application needing reliable, durable messaging between processes or threads. Demonstrates robust inter-agent communication patterns.