Data Analysis Agent Skill

Version: 2.0 Date: 2026-02-17

Agent Definition

Tools

Layer 1 — Web Data Collection (existing)

Layer 1B — Itinerary Analysis (new, from itin-analyzer prototype)

Itinerary Analysis Capabilities

Structured Data Extraction

For each itinerary document, the agent extracts:

• •Tour Name and Duration (days/nights)
• •Destinations (cities and countries)
• •Pricing (by period, with currency)
• •Flights (flight numbers, origins, destinations, times)
• •Inclusions and Exclusions
• •Daily Breakdown (day-by-day activities, meals, locations)

Competitive Comparison

When 2+ itineraries are provided:

• •Product Comparison Matrix (duration, price/day, destinations, meals, activities)
• •Value Analysis (price-to-value ratio)
• •Strengths & Weaknesses per product
• •Target Customer Profiles
• •Market positioning recommendations

Market Intelligence Pipeline

End-to-end market analysis:

1. •Extract — Parse and clean all uploaded documents
2. •Analyse — Identify dominant destinations, themes, duration patterns, activities
3. •Web Research — Search for competitor products using EXA API
4. •Aggregate — Build a knowledge graph (entities + relationships)
5. •Report — Generate a comprehensive strategic market report

Bilingual Support

• •Auto-detects Thai and English text
• •Handles mixed Thai/English documents
• •Normalises Thai text spacing around numbers
• •Preserves Thai price formats (บาท, THB)

Input

This agent accepts:

For Web Data Collection

• •An active browser session (authenticated via LoginTool)
• •Website URLs configured in Config:
  • •BOOKING_URL → /booking
  • •REPORT_SELLER_URL → /report/report_seller

For Itinerary Analysis

• •Uploaded files (PDF, DOCX, TXT) via:
  • •WebSocket: Upload a file and say "analyze itinerary"
  • •REST API: POST /api/itinerary/analyze
• •Multiple files for comparison: POST /api/itinerary/compare
• •Multiple files for market intelligence: POST /api/itinerary/market-intelligence

Output

Booking Data Output (unchanged)

{
  "booking_data": [...],
  "report_data": [...],
  "extraction_timestamp": "ISO-8601",
  "validation_summary": { ... }
}

Output file: data/booking_data.json

Itinerary Analysis Output

{
  "status": "success",
  "data": {
    "tourName": "Japan Golden Route 8D6N",
    "duration": "8 Days / 6 Nights",
    "destinations": ["Tokyo", "Osaka", "Kyoto", "Mount Fuji"],
    "pricing": [{"period": "Jan-Mar 2026", "price": 49900, "currency": "THB"}],
    "flights": [{"flightNumber": "TG660", "origin": "Bangkok", "destination": "Tokyo", ...}],
    "inclusions": ["Round-trip flights", "Hotel accommodation", ...],
    "exclusions": ["Travel insurance", "Personal expenses", ...],
    "dailyBreakdown": [
      {"day": 1, "title": "Bangkok - Tokyo", "activities": "...", "meals": ["Dinner"], "locations": ["Narita Airport", "Shinjuku"]}
    ]
  },
  "language": "English",
  "metadata": { "total_pages": 4, "quality_score": "high" }
}

Output file: data/itinerary_<name>.json

Market Intelligence Output

{
  "success": true,
  "pipeline_steps": { "extract": {...}, "analyze_themes": {...}, "web_research": {...}, "aggregate": {...}, "report": {...} },
  "dominant_themes": { "main_destination": "Japan", "main_theme": "cultural", "top_destinations": {...}, "top_themes": {...} },
  "web_research": { "packages_found": [...], "prices_found": [...] },
  "knowledge_graph": { "entities": [...], "relationships": [...] },
  "final_report": "## Executive Summary\n..."
}

Output file: data/market_intelligence.json

API Endpoints

Execution Order

Layer 1 — Data Collection (runs first, no dependencies)

Error Handling

• •Retry failed page navigations with exponential backoff (max 3 attempts).
• •If booking page structure changes, log a detailed warning with the HTML snippet and return partial data.
• •Validate every extracted record; invalid records are excluded but logged.
• •If PDF extraction yields low-quality text (< 100 chars/page), log a warning and attempt alternate extraction methods.
• •If the entire extraction fails, return an empty dataset with an error summary so downstream agents can handle gracefully.
• •Web research gracefully degrades if EXA API key is not configured.