ai-travel-agent（艾旅行社）

Travel Agent

Setup — Credentials

Requires: A SerpAPI key. SerpAPI (serpapi.com) provides access to Google Flights, Google Hotels, and Google Maps Directions. A free tier is available with 100 searches/month. Sign up at: https://serpapi.com

Load the user's SerpAPI key before running any search. Check these locations in order:

• ~/.serpapi_credentials → SERPAPI_KEY=...
• ~/.travel_agent_config → SERPAPI_KEY=...

These are the only local files this skill reads. If neither file is found, ask the user where their SerpAPI key is stored, or whether they would like to create one at https://serpapi.com.

All flight, hotel, and transport searches are sent to serpapi.com using the user's own API key. This skill does not store, transmit, or proxy the key anywhere else.

Step 1 — Gather Trip Details (always first)

Before searching anything, ask a small set of quick questions. Keep it conversational — one message, not a form. Tailor which questions to ask based on what the user hasn't already told you.

Questions to ask (as needed):

• 👥 How many people? (adults + children)
• ✈ Flying from which airport? (if no default configured)
• 💺 Cabin class? (Economy / Premium Economy / Business / First)
• 💰 Rough budget? (per person, or total — flights + hotel)
• 🏨 Hotel preferences? (e.g. beach, city centre, boutique, pool, specific star rating)
• 🎒 Trip vibe? (relaxing beach, city/culture, adventure, food-focused, nightlife, mix)
• 📍 One place or travel around? If moving around — roughly how many stops, or leave it to you?
• 🗓 Any flexibility on dates or duration?

Skip questions already answered. Keep to 4–6 questions max in a single message.

Step 2 — Determine Mode

Mode A: Dates ✅ + Destination ✅

• If single stop → go to Step 3
• If multi-stop → go to Step 2M

Mode B: Dates ✅ + Destination ❓

1. Read references/seasonal-destinations.md
2. Suggest 3–4 destination options that match the dates (season fit) and the user's vibe
3. For each option, include: why it's good then, rough flight time, vibe summary, and whether it suits single or multi-stop
4. Wait for user to choose, then return to Mode A or 2M

Mode C: Destination ✅ + Dates ❓

1. Check for an installed calendar skill — try google-calendar, outlook-calendar, or any other calendar skill available. This is an optional convenience: it checks the user's calendar for free windows to suggest good travel dates. If no calendar skill is found (or the user prefers not to use it), fall back to asking the user directly for their available date windows. Never access calendar data without the user's awareness.
2. Read references/seasonal-destinations.md for best time to visit that destination
3. Recommend 2–3 date windows ranked by: season fit + calendar availability
4. Present clearly: "Window 1: [dates] — great timing (dry season), you're free. Window 2: ..."
5. Wait for user to confirm dates, then go to Mode A or 2M

Step 2M — Multi-Stop Itinerary Planning

Use when the user wants to travel around a country or region rather than stay in one place.

2M-1: Propose a Route

1. Read references/multi-stop-routes.md for regional route suggestions
2. Based on total duration, vibe, and number of stops, propose a day-by-day route:

- Suggest how many nights per stop (balance: enough time to explore, not rushed) - Recommend a logical geographic order (minimise backtracking) - Note highlight of each stop (what it's known for)

1. Present as a clear itinerary skeleton, e.g.:

   Days 1–3: Tokyo (arrive, recover, explore)
   Days 4–6: Kyoto (temples, geisha district, day trip to Nara)
   Days 7–9: Osaka (food scene, day trip to Hiroshima)
   Day 10: Fly home from Osaka (KIX)

1. Invite the user to adjust: swap stops, add/remove places, change durations

2M-2: Confirm the Route

Once the user is happy with the skeleton, confirm:

• All stop names and exact dates
• Which city to fly into (may differ from first stop)
• Which city to fly home from (may differ from last stop — open-jaw flights)
• Preferred transport between stops (flight / train / bus / hire car)

2M-3: Search All Legs

Run searches for each component in this order:

Flights:

• Inbound: home airport → first city
• Between stops: if flying (run search_flights.py per leg)
• Outbound: last city → home airport
• Note: for open-jaw (fly into City A, home from City B), search each leg as one-way

Hotels:

• Run search_hotels.py for each stop with correct check-in/check-out dates

Internal transport between stops — run BOTH scripts in parallel:

# Ground transport (trains, buses, ferries)
python3 scripts/search_transport.py \
  --from "City A, Country" --to "City B, Country" \
  --key $SERPAPI_KEY

# Domestic flights
python3 scripts/search_flights.py \
  --from [IATA_A] --to [IATA_B] --date YYYY-MM-DD \
  --adults N --class economy --currency GBP \
  --key $SERPAPI_KEY

Present all options in a single combined comparison, grouped by type:

• ✈ Flights — fastest door-to-door if airports convenient; include airport transfer time in assessment
• 🚄 Train — usually best for city centre → city centre under ~4h; no airport faff
• 🚌 Bus — cheapest, good for budget travellers
• ⛴ Ferry — if applicable (islands, coastal routes)

For each option show: operator, duration, price, booking link. Recommend the best option based on journey time including transfers, price, and convenience — don't just list them. Note when advance booking is essential (e.g. Trenitalia, Eurostar, Renfe, Shinkansen).

Present the full itinerary in one structured summary — see Step 5M.

Step 3 — Search Flights (single destination)

Run scripts/search_flights.py:

python3 scripts/search_flights.py \
  --from [IATA] --to [IATA] \
  --date YYYY-MM-DD [--return-date YYYY-MM-DD] \
  --adults N [--children N] \
  --class economy|premium_economy|business|first \
  --currency GBP \
  --key $SERPAPI_KEY

For round trips, also run the return leg separately for more options.

Present results as 3–4 options:

• 💰 Cheapest — lowest price, even if slower
• ⚡ Fastest — fewest stops / shortest duration
• ⭐ Best value — balance of price, airline quality, legroom
• 👑 Premium — best class/airline if budget allows

For each option show: airline + flight number, depart/arrive times, duration, stops, price, aircraft, legroom, any delay warnings.

Step 4 — Search Hotels (single destination)

Run scripts/search_hotels.py:

python3 scripts/search_hotels.py \
  --location "[CITY] [preference]" \
  --check-in YYYY-MM-DD --check-out YYYY-MM-DD \
  --adults N [--children N] \
  [--stars N] [--amenities "pool,beach"] \
  --sort highest_rating \
  --currency GBP \
  --key $SERPAPI_KEY

Present results as 3 tiers:

• 🎒 Budget — good rating, lower price
• 🏨 Mid-range — solid choice, best balance
• 🌟 Splurge — top rated, premium amenities

For each: name, stars, rating, price/night, total cost, key amenities, location notes, booking link.

Step 5 — Iterate (single destination)

After presenting options, invite refinement:

• "Swap to the BA flight instead?"
• "Find something closer to the beach?"
• "What's available in business class?"
• "Show me budget hotels only?"

Re-run the relevant script with adjusted parameters.

Step 5M — Present Full Multi-Stop Itinerary

Once all flights and hotels are searched, present a complete trip summary:

🌍 [TRIP NAME] · [DATE RANGE] · [N] people

DAY 1–3 | CITY A
✈ [Inbound flight] | [time] → [time] | £XX
🏨 [Hotel name] ★★★★ | £XX/night | £XX total

[Train/bus to City B — Xh, book via Trainline: URL]

DAY 4–6 | CITY B
🏨 [Hotel name] ★★★★ | £XX/night | £XX total

✈ [Internal flight if applicable]

DAY 7–9 | CITY C
🏨 [Hotel name] ★★★ | £XX/night | £XX total

✈ [Outbound flight] | [time] → [time] | £XX

💰 TOTAL ESTIMATED COST: £XX
   Flights: £XX | Hotels: £XX | (Transport: £XX est.)

Invite iteration on any leg:

• "Upgrade the hotel in City B"
• "Find a cheaper internal flight"
• "Swap City C for City D"
• "Add an extra night in City A"

Step 6 — Lock In & Calendar

Once the user confirms everything:

1. Summarise full trip with all costs
2. Ask if they'd like the dates blocked in their calendar
3. If yes: check which calendar skill the user has installed (e.g. google-calendar, outlook-calendar, or any other calendar skill) and use that. Do not assume Google Calendar. This access is explicitly user-requested at this step — the user has just asked to add events to their calendar.

- If no calendar skill is found, offer to generate an .ics file the user can import into any calendar app (Google, Apple, Outlook, etc.)

1. Create events for each component:

- Each flight leg: departure datetime → arrival datetime, airline + flight number in title - Each hotel stay: check-in → check-out date, hotel name + address - Spanning "Trip to [destination]" event across full duration

1. Include booking links in all event descriptions

Notes

• Always show prices in the user's preferred currency (default GBP if not set)
• SerpAPI free tier = 100 searches/month — multi-stop trips use more quota; heads up if >10 searches in one session
• Open-jaw flights (fly into A, out of B) often work out cheaper than returning to origin — suggest when relevant
• See references/serpapi.md for full API reference (flights, hotels, transport/directions)
• See references/seasonal-destinations.md for destination/timing guidance
• See references/multi-stop-routes.md for regional itinerary templates