Model behavior: direct questions must be answered before/alongside actions, never replaced by them

Feedback on model behavior (Claude Fable 5 in Claude Code, orchestration session).

Observed pattern: when the user asks a direct question mid-task (e.g. 'why did X happen?', 'why did you do Y?'), the model sometimes leads its turn with tool calls/dispatches and either buries the answer after the actions or answers only partially, treating the in-flight work as higher priority than the user's question.

Expected: a direct user question is ALWAYS answered first, explicitly and completely, in the response text — before or clearly ahead of any new actions. Acting without answering (or answering as an afterthought below a wall of tool activity) reads as the model ignoring the user. This should hold even when the question arrives mid-turn while background work is running.

Repro shape: long multi-agent orchestration session; user interjects 'why did you kill that agent / why so long?'; model's next turn opens with new agent dispatches and monitors, with the explanation interleaved or after the fact. User had to demand the answer explicitly.

Request: strengthen the answer-the-question-first behavior in the harness prompting/model training so direct questions preempt action-first turns.