Interface RealtimeChannelVerb

What the verb does and when to use it, for the model.

Who may invoke it.

Verb name as the agent addresses it (e.g. 'AddNote'). Matched case-insensitively.

The native provider tool this verb is ALSO exposed as, when the channel is mounted at session start (e.g. 'Whiteboard_AddNote'). Present so prompts and transcripts that already know the tool name stay valid, and so the dispatcher can route a proxy call to the same executor a native call uses.

JSON Schema of the verb's parameters; the dispatcher validates calls against it.

Conditions that must hold for the verb to succeed, in plain language for the model.

The least exposure the channel must have for this verb's RESULT to be given to the agent, for a verb whose result carries what the channel holds (a component's data, a page's text, a board's contents). Omit it for a verb that only acts and reports success.

Exposure ('none' < 'state' < 'pixels') limits what the model may perceive of a channel; a verb the agent calls could otherwise hand it exactly what the user (or a zero-data-retention policy) held back. When the channel's effective exposure is below this level the dispatcher refuses the call with an 'exposure_restricted' error the agent can read, and the catalog tells it the verb is unavailable and why. This is policy, not redaction: the verb is refused whole, its result is never filtered. 'state' is for results that describe the channel's contents as data; 'pixels' for results derived from what it looks like.