Provides the `function_tool` decorator that converts Python functions into agent-callable tools with automatic JSON schema generation, strict mode enforcement (additionalProperties: false), custom error handling, and Pydantic model support. This is a reusable pattern for defining tool interfaces in any agent framework, not just CAI.

Implements the `Handoff` abstraction for routing between agents, with input validation via Pydantic models, `on_handoff` hooks, strict schema enforcement, and support for tool-based handoffs. This pattern is reusable for multi-agent orchestration in any conversational AI system.

Defines the `SpanProcessor` interface and concrete implementations (e.g., `BatchSpanProcessor`, `SimpleSpanProcessor`) for exporting trace data. This follows the OpenTelemetry processor pattern and can be extracted for use in any tracing system that needs batched or immediate export of spans.

Implements a global `ToolRegistry` that auto-registers tools from module imports, categorizes them by agent type, and provides API-key gating. The registry pattern with category-based filtering is reusable for any plugin or tool system that needs dynamic discovery and conditional availability.

Provides session context compaction and handoff generation for multi-agent conversations. It extracts security-relevant findings (IPs, CVEs, artifacts) from message history using regex patterns, builds an XML-style handoff block, and persists state via environment variables. This pattern is reusable for any multi-turn agent system that needs to pass summarized context between agents.

Implements a stateless input router for the TUI that parses user commands and returns `RouteDecision` dataclasses indicating target terminals, agent names, and command types. It supports CLI, chat, broadcast, and parallel patterns. The pure-logic router pattern is reusable for any multi-window or multi-agent UI system.