Implement stale-cache indicator and manual refresh
epic-organization-feature-flags-ui-task-006 — Add a stale-cache indicator banner to the admin screen that appears when the last sync timestamp from FeatureFlagCache exceeds the configured TTL. Display the time-since-last-sync in human-readable form. Provide a manual refresh button that triggers a forced fetch from FeatureFlagRepository and clears the stale indicator on success.
Acceptance Criteria
Technical Requirements
Execution Context
Tier 2 - 518 tasks
Can start after Tier 1 completes
Implementation Notes
Derive isStale from a computed Riverpod provider that watches FeatureFlagCacheProvider and returns a bool based on timestamp delta. This ensures the banner reactively appears/disappears without widget-level polling. For the 'X minutes ago' display, use a StatefulWidget with a single Timer.periodic(60s) that calls setState to recompute the elapsed label—do not create multiple timers. The banner should use MaterialBanner for native accessibility support (it handles live region semantics on Android).
For iOS VoiceOver, add a Semantics wrapper with liveRegion: true. Confirm the amber warning color token exists in the design system before hardcoding; if absent, add a --color-warning-bg and --color-warning-text token to styles.css.
Testing Requirements
Widget tests: (1) banner is absent when cache is within TTL; (2) banner appears when cache exceeds TTL; (3) 'X minutes ago' text reflects correct elapsed time; (4) tapping refresh calls forceRefresh exactly once; (5) on success, banner is removed and flag list provider is invalidated; (6) on failure, banner remains and error snackbar is shown; (7) button is disabled during in-flight refresh to prevent double calls. Use fake async (fakeAsync from flutter_test) to simulate TTL expiry without real time delays.
The feature flag admin screen allows persisting changes to organization_configs. If the role guard is implemented only client-side (checking role state in Riverpod), a user who manipulates their local role state could toggle flags for their organization without proper server-side authorization, potentially exposing features prematurely.
Mitigation & Contingency
Mitigation: Implement server-side authorization for flag update operations using Supabase RLS UPDATE policies that check the user's role in the memberships table. The client-side guard is UX only; the database enforces the actual restriction.
Contingency: If an unauthorized update is detected, audit the RLS policies and add a Supabase Edge Function as an authorization middleware for flag toggle operations, rejecting requests from non-admin role JWTs.
Developers on other feature teams may use FeatureGate incorrectly — for example, wrapping business logic rather than UI, or using it before flag initialization completes — leading to features that are visible but non-functional or cause runtime errors when flags are queried in a loading state.
Mitigation & Contingency
Mitigation: Add assert statements in FeatureGate's build method that throw in debug mode if the provider is still in a loading state. Write developer documentation with a clear usage contract: FeatureGate is UI-only; logic gating must use the provider's isEnabled method directly. Include lint examples in the codebase.
Contingency: If misuse is found in code reviews, add a custom Dart lint rule via custom_lint that flags FeatureGate usage outside of the widget tree, and conduct a codebase audit to find existing violations.
If the audit log is stored in the same organization_configs table without pagination or archival strategy, high-frequency flag changes during pilot testing could produce an unbounded number of rows, degrading query performance on the admin screen.
Mitigation & Contingency
Mitigation: Store audit log entries in a separate feature_flag_audit_log table with an index on (organization_id, changed_at DESC). Implement cursor-based pagination in the repository and limit the initial load to 50 entries.
Contingency: If table size becomes a performance concern, add a Supabase scheduled function to archive entries older than 90 days to cold storage, and add a database index on changed_at for range queries.