Interface representing the AI agent logging channel.
Kept separate from ILogger on purpose: ILogger is the framework's internal diagnostics sink wired through setLogger, whose four severity levels describe the health of the framework itself.
Agent output is a different concern — it records what a model did (reasoning steps, tool calls, completions), not how the framework is behaving, and it is addressed to the user reading the log history rather than to whoever is debugging the engine.
Splitting the interface also keeps ILogger implementations supplied by users source-compatible: adding agent logging to the log history never widens the contract they implement.
agent: (topic: string, ...args: any[]) => void
Logs an AI agent message. Used to record model-driven activity — reasoning, tool invocations, and completions — so it stays distinguishable from framework diagnostics when reviewing the log history.