feat/chatbot context handoff

[XuanHanTan-School]

[Copilot]

Pull request overview

This pull request implements AI context handoff functionality for chatbot escalations to human support agents. When the AI chatbot needs to escalate a customer issue to a human agent, it now captures and passes along three key pieces of context that are invisible to the customer but critical for the support agent.

Changes:

• Added three new nullable database fields to call_bookings table: aiSolutionsSummary, escalationReason, and conversationContext
• Updated backend types, validation schemas, and execution logic to handle the new AI context fields throughout the data flow
• Modified AI system prompts to instruct the model to provide context when scheduling calls
• Enhanced the admin UI to display AI context information in the call details panel

Reviewed changes

Copilot reviewed 17 out of 19 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
8cbc_backend/drizzle/0017_next_whirlwind.sql	Database migration adding three new text columns for AI context
8cbc_backend/drizzle/meta/0017_snapshot.json	Drizzle snapshot reflecting the new database schema
8cbc_backend/database/schema.ts	Schema definition for the three AI context fields in call_bookings table
shared/types/confirmation.types.ts	TypeScript interfaces updated to include AI context fields for call scheduling
shared/types/api.types.ts	CallBooking interface extended with AI context fields
8cbc_backend/controllers/calls/calls.controller.ts	Validation schema updated to accept AI context fields in booking requests
8cbc_backend/services/langchain/tools/call.tools.ts	LangChain tool schema extended with AI context parameters
8cbc_backend/services/langchain/system-prompt.ts	System prompt updated with instructions for providing AI context during escalations
8cbc_backend/services/realtime/realtime.config.ts	Voice assistant instructions updated for AI context handoff
8cbc_backend/services/langchain/modal.types.ts	Zod validation schema for schedule call operation includes AI context
8cbc_backend/services/langchain/execution/operations.ts	Booking creation logic updated to persist AI context fields
8cbc_frontend/contexts/ChatbotContext.tsx	Chat context passes AI context fields from AI responses to backend
8cbc_frontend/components/voice/voice-overlay.tsx	Voice overlay passes AI context fields to backend
8cbc_frontend/components/ui/modals/schedule-call/schedule-call-modal.tsx	Schedule call modal receives and passes through AI context fields
8cbc_frontend/components/bookings/CallDetailsPanel.tsx	UI component displays AI context information with Bot icon for support agents
8cbc_frontend/app/(admin)/admin/page.tsx	Minor unrelated whitespace fix and title clarification
8cbc_frontend/bun.lock	Bun lockfile update with configVersion field
8cbc_backend/bun.lock	Bun lockfile update with configVersion field

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The example date "2024-01-15" should be updated to use a date from 2026, as we are currently in 2026. Using an outdated year in the example could confuse the AI model or make the examples less relatable.

Suggested change

	- Example: "User concerned about 3 unauthorized transactions totaling $450 on 2024-01-15. Transaction IDs: TXN-123, TXN-124, TXN-125. Card ending 4532. User noticed charges this morning, card is still in possession."
	- Example: "User concerned about 3 unauthorized transactions totaling $450 on 2026-01-15. Transaction IDs: TXN-123, TXN-124, TXN-125. Card ending 4532. User noticed charges this morning, card is still in possession."

[Copilot]

There's a discrepancy between the system prompt documentation and the schema implementation. The system prompt states these fields are "REQUIRED" (lines 329, 335, 340) with instructions like "you MUST provide context" (line 327), but the Zod schema defines all three fields (aiSolutionsSummary, escalationReason, conversationContext) as optional() at lines 137-154 in call.tools.ts. This could lead to inconsistent behavior where the AI might not always provide these fields despite the prompt strongly indicating they are mandatory. Consider either making these fields required in the schema, or adjusting the prompt language to say "STRONGLY RECOMMENDED" instead of "REQUIRED" to match the optional schema.

[fisherman-23]

works well, lgtm