Inbox
The inbox is the short list of things that need a human decision, delivered as action cards and ranked by priority.
How it works
Each card names one thing that changed and what to do about it. The type says why it's there: a fix_contradicted alarm when interviews keep corroborating a signal you already shipped a fix for, a new_recommendation to decide, a live_call when a specific participant is worth hearing live — with the verbatim quote and whatever contact info you already had, a cross_study_signal worth reading, a research_gap to act on, a follow_up_suggestion when a shipped fix's evidence went quiet and a deliberate follow-up study is worth considering, a theme_contradiction where the evidence disagrees with itself, study_saturation when a study stops learning, an interview_failure to investigate, or a low_balance warning. Priority is a tier from 1 to 10 — lower is more urgent; contradicted fixes and recommendations rank first.
A card is a pointer, not the object itself: it targets the signal, recommendation, gap, or study it's about.
What you can do
Resolve a card as followed (you handled it) or dismissed (not worth acting on). Or snooze it for up to 90 days — it returns to the inbox when the time elapses. Resolving a card never changes the object it points at; deciding a recommendation is its own action.
When it's quiet
An empty inbox means nothing needs you — not that nothing is happening. Studies keep collecting and analysis keeps evaluating; the inbox only speaks when a decision is worth your time.
From your agent
list_action_cards— the pending inbox, ranked.resolve_action_card— mark a card followed or dismissed.snooze_action_card— defer a card for a duration.