Tool names selected by the classifier.
The wall-clock time in ms of the classification step, excluding the agent response.
OptionaltoolThe optional per-tool scope, keyed by tool name. Each value is the raw
classifier output for that tool, which prepareStep validates against
the tool's scopeSchema before applying any narrowing. Tools without
scopeSchema, or with no scope from the classifier, keep their full
description and inputSchema for the turn.
Token usage of the classification, summed over its one LLM call, or two when the first answer fails validation.
Result returned by classifyUserIntent.