daolv-hotel-booking

# Daolv Hotel Booking

Provide reliable hotel planning and booking support with structured MCP calls and decision-ready outputs.

## Workflow

1. Capture booking intent before calling tools
- Extract: destination, check-in date, nights, adults/children, room count, budget, purpose (business/family/leisure), required amenities, preferred/avoided brands.
- If key constraints are missing, ask only the minimum follow-up questions.

2. Prime tags once per task
- Call `ai-go-hotel.getHotelSearchTags` once.
- Cache returned tags for the rest of the conversation.
- Use those tags to build `hotelTags.requiredTags`, `preferredTags`, `excludedTags`, and optional budget constraints.

3. Search hotels with normalized parameters
- Call `ai-go-hotel.searchHotels` with:
  - `place`
  - `placeType`
  - `originQuery`
  - optional `checkInDate`, `stayNights`, `adultCount`, `size`, `starRatings`, `hotelTags`, `countryCode`, `distanceInMeter`, `withHotelAmenities`, `language`
- Prefer `size=8-12` for first pass; narrow to top 3-5 in final output.
- Respect live schema behavior:
  - `checkInDate` invalid/past/empty may fallback to tomorrow
  - `price` is an object (use `price.lowestPrice` + `price.currency`)
  - some fields can be null or missing
- `placeType` can be normalized from user language:
  - 城市/city → 城市
  - 机场/airport → 机场
  - 景点/attraction → 景点
  - 火车站/railway station → 火车站
  - 地铁站/metro → 地铁站
  - 酒店/hotel → 酒店

4. Enrich finalists with room-level details
- For each shortlisted option, call `ai-go-hotel.getHotelDetail` (prefer `hotelId` when available).
- Pass dates with `checkInDate` / `checkOutDate` format `YYYY-MM-DD`.
- Handle fallback and edge behavior:
  - invalid/empty dates may auto-correct
  - failures may return plain text (not structured JSON)
  - `roomRatePlans` can be very large; render only top rows by relevance/price
- Extract actionable room/price data, cancellation policy, breakfast inclusion, and important constraints.

5. Return...