Prompt History Repository

Medium complexity with dual execution context (backend and mobile), this repository must be delivered before end-to-end scenario evaluation testing can occur, as the Scenario Evaluation Edge Function depends on hasPromptBeenSent for its core deduplication guarantee. Coordinate delivery with the edge function team to ensure the backend query interface is finalized early in the sprint cycle. Testing requires simulating multi-user, multi-scenario states to validate deduplication logic under concurrent scheduler runs. The watchActivePrompts streaming interface needs UI integration testing with the notification inbox or badge system.

Plan for QA data-seeding scripts to populate realistic prompt history states across test accounts, as empty-database test runs will not surface timing or deduplication issues.

Backed by a Supabase table (e.g., scenario_prompt_history) with columns for user_id, scenario_id, activity_id, status (sent/acted/dismissed), and sent_at timestamp. The hasPromptBeenSent query uses a composite lookup on (userId, scenarioId, activityId) — index this combination for scheduler performance during bulk evaluation runs across large chapters. recordSentPrompt should use upsert logic to handle race conditions from concurrent edge function invocations. watchActivePrompts subscribes to Supabase Realtime filtered by userId and status not in (dismissed, acted).

markActedUpon is called by the deep-link handler post-navigation; markDismissed is called by UI gesture handlers. Both mutations should be fire-and-forget on the mobile side with structured error logging. On the backend, both must be awaited within the scheduler's try/catch block to confirm state before the next evaluation cycle.