🧳 Avatar Intelligent Travel Assistant

Positioning

A one-stop intelligent travel service platform capable of automatically routing user queries to the appropriate sub-skills to provide complete travel solutions, including flights, hotels, and itinerary planning.

Core Philosophy: The user only needs to state their needs. Avatar intelligently determines which sub-skill (or combination of sub-skills) to invoke, without the user needing to worry about the specific underlying details.

Intelligent Routing Framework

Intent Recognition

When a user's query involves travel needs, Avatar analyzes the query content to identify the user's true intent:

Common Intent Types

Routing Decision

Based on the intent analysis, Avatar decides how to invoke the sub-skills:

• Single-Skill Invocation: When the intent is clear and requires only one skill.
• Combined Invocation: When the user expresses a combined need (e.g., "recommend flights and hotels together").
• Sequential Invocation: When the result of a previous query is needed for the next one (e.g., search for a flight first, then recommend hotels at the destination).

Sub-skill Descriptions

Avatar integrates the following sub-skills, each remaining independent and orchestrated by Avatar.

🏨 hotel_recommendation

Skill Path: avatar_travel_assistant/references/hotel_recommendation/SKILL.md

When to Invoke:

• The user explicitly mentions "hotel," "accommodation," or "stay."
• The query involves a combined search that includes accommodation.
• The user specifies a destination city and check-in details.

Core Capabilities:

• Complex queries based on city, points of interest (POI), hotel name, or brand.
• Intelligent expression in the role of a senior hotel consultant (v4.0 expression layer rewrite).
• Dual-path service: intelligent recommendations vs. precise Q&A.

Important Notes:

• Adheres to a strict evidence-based principle (all recommendations must be 100% sourced from hotel data tags).
• No duplicate hotel recommendations.
• User-centric logic (categorization is tailored to the query's intent, not a fixed template).

✈️ flight_recommendation

Skill Path: avatar_travel_assistant/references/flight_recommendation/SKILL.md

When to Invoke:

• The user explicitly mentions "flight," "airline," or "fly."
• The query involves a combined search that includes travel.

Core Capabilities:

• Invokes the Voyager flight search API.
• Supports single-segment and multi-segment journey searches.
• Provides data support for estimation and recommendation.

Usage Conventions:

• It is the default skill for all flight-related queries.
• Data support for domestic routes is strong; international routes may be limited.

🗺️ trip_itinerary_planner

Skill Path: avatar_travel_assistant/references/trip_itinerary_planner/SKILL.md

When to Invoke:

• The user explicitly uses keywords like "plan," "itinerary," "guide," or "X-day X-night."
• The request involves a multi-day or multi-city travel arrangement.
• The request requires route optimization, budget estimation, or a detailed schedule.
• When a simple combination of intents is insufficient, this skill should be prioritized to handle the entire planning need.

Core Capabilities:

• Intelligent multi-day, multi-destination itinerary planning.
• Route optimization and budget estimation.
• Detailed daily schedules (attractions, dining, transportation).
• Scene adaptation: family trips, honeymoons, backpacking, team-building, etc.
• Real-time information: weather, traffic conditions, policy alerts.

Decision Priority: When the user expresses a strong intent like "plan a 3-day trip," route to this skill with high priority instead of combining other skills like flight + hotel. This skill can independently handle the entire planning process.

Decision Principles

When faced with a user query, Avatar should use the following principles for routing:

1. Clear Intent, Single Skill

Example: "A flight from Beijing to Shanghai tomorrow." Analysis:

• Clear intent: Flight search.
• Conclusion: Invoke flight_recommendation.

2. Clear Intent, Combined Skills

Example: "For the May Day holiday, I'm going to Chengdu. Recommend flights and hotels together." Analysis:

• Intent 1: Flight search (flights + Chengdu).
• Intent 2: Hotel recommendation (hotels + Chengdu).
• The word "together" suggests a combined need.
• Conclusion: Invoke flight_recommendation + hotel_recommendation in parallel.

3. Sequential Invocation (Flight then Hotel)

Example: "Find a flight to Chengdu, and then recommend a hotel after I arrive." Analysis:

• First, execute the flight search to confirm destination availability.
• Then, recommend hotels based on the flight results.
• Conclusion: Invoke flight_recommendation → hotel_recommendation sequentially.

4. Prioritize Itinerary Planning

Example: "Help me plan a 3-day trip to Chengdu." Analysis:

• Keywords "plan," "itinerary" + duration "3-day trip."
• A clear multi-day travel planning intent.
• Conclusion: Prioritize invoking trip_itinerary_planner (to handle all planning aspects centrally).

5. Default to Recommendation

Example: "Recommend some good value flights." Analysis:

• A general flight request.
• Conclusion: Default to flight_recommendation for a baseline recommendation.

Design Principles

1. Intelligent, Subjective Routing: Decisions are based on the semantic meaning of the query, not on hard-coded rules.
2. Preserve Sub-skill Integrity: The SKILL.md for each sub-skill is kept intact. Avatar only acts as a router and result integrator.
3. Agile-Friendly: The routing logic can be adjusted flexibly without being tied to specific code implementations.
4. Seamless User Experience: The user is unaware of the underlying sub-skills and interacts only with Avatar.

Typical Use Case Analysis

Cautions & Notes

1. Handle Skill Dependencies: The SKILL.md files of sub-skills may reference each other. As the central router, Avatar must understand these dependencies.
2. Coordinate Combinations: When invoking multiple sub-skills in parallel, manage the output order and logical dependencies. Use sequential invocation when necessary.