POST /v1/sessions/{session_id}/next, check action before deciding what to display. A warning does not automatically end the conversation.
Possible warnings
Fallbacks and stops
degraded_reasons may include semantic_service_unavailable (semantic analysis unavailable), unresolved_previous_turn (previous question not resolved), or missing_user_text (a reply was absent when expected). summary_only_context remains in the schema, but a summary-only add_context is now a normal operation and is not flagged as degraded. The only current stop_reason is no_question_available: no question remains after hard exclusions. The turn limit and objective achievement produce warnings, not hard stops.
Example:
action: "ask" with warnings: ["max_turns_reached"] means a question is available despite the limit. Your application decides whether to ask it or stop.