Replace ad-hoc AGENTIC-N comments with standardized TODO-NNN format across
the entire codebase. Each TODO has:
- Running number (TODO-001 through TODO-011)
- Priority level (high/medium/low)
- Phase reference (0, A.1, A.4, B, cleanup)
- Clear step-by-step instructions an AI agent can follow
TODO index:
TODO-001: Kill switch maintenance banner (providers.tsx)
TODO-002: Feedback button in settings page
TODO-003: Accessibility focus trap for modals
TODO-004: Clone routine template instead of mutating in-place
TODO-005: Wire real LLM enrichment for context messages
TODO-006: Centralize backend URL configuration
TODO-007: MCP tool integration tests (common-plat)
TODO-008: Wire trackEvent() calls into routes + components
TODO-009: Unit tests for AI context generation
TODO-010: Import PRODUCT_ID from product-config (5 route files)
TODO-011: Wire error boundary to telemetry
Added consolidated TODO Index table at top of AGENTIC_AI_ROADMAP.md
for agent scanning. 219 backend tests pass, no code changes.
$contains generates CONTAINS() SQL in Cosmos (string match), not
ARRAY_CONTAINS() needed for string[] fields. Use findMany + in-memory
post-filter for correct behavior across both providers.
