Get Agent Discovery

The whole discovery record, plus an honest coverage report.

EVERY field comes back, answered or not - an unanswered one carries status: "open" rather than being absent, so a caller renders a stable list instead of one that grows as the interview proceeds.

coverage is the part worth reading: open_musts names what still blocks a launch, amber_fields names the working assumptions, and can_launch is the single boolean. It is not a percentage, on purpose - a framework with silent gaps is worse than no framework, because the gaps become confident copy three weeks later.