Ask an agent to draft the suggestion context
Queues one costgraph-ai run that reads the tenant’s billing shape and writes a first draft of the suggestion context, the brief every later proposal is generated from. A tenant drafts one at a time, so a request made while a draft is in flight answers that run rather than starting a second one. The run is asynchronous and saves nothing: poll it with the returned id, and on done read the drafted text from the run and save it with PUT /virtual-tag-settings/suggestion-context if the person keeps it. Repeat presses inside the same minute collapse into the run already in flight, which is what queued=false reports; the id still points at that run.