travel-intelligence

name: travel-intelligence description: Travel intelligence pipeline — proactive trip messages at key moments. Covers flight status, weather, airport info, gate updates, check-in reminders, landing intel, daily morning briefings. Triggers on travel, trip, flight, weather, airport, hotel, packing.

Travel Intelligence Pipeline

Fact-driven proactive messages delivered at key trip moments. Each message is assembled from live data (flights, weather, places, facts) and injected into the contact's foreground session — never sent as standalone SMS.

Message Templates

1. Check-in Reminder (T-24h before departure)

✈️ Check-in open for {airline} {flight_number} {origin_code} → {dest_code} · {depart_date} · {depart_time} Seat {seat} · Conf {booking_ref} 👉 {check_in_url}

Data sources:

• ~/dispatch/scripts/fact list --contact "{contact}" --json — pull legs[] for flight, seat, booking_ref
• Check-in URL from static lookup (see Airline Check-in URLs)

Fallback: If seat or booking_ref missing, omit those lines. Always send the check-in link.

2. Packing Weather (T-1 day, morning)

🧳 Packing weather for {dest_city} {day1}: {icon} {high}°/{low}°F — {summary} {day2}: {icon} {high}°/{low}°F — {summary} {day3}: {icon} {high}°/{low}°F — {summary} 💡 {packing_tip}

Data sources:

• ~/.claude/skills/weather/scripts/weather — 3-day forecast for destination coords
• Generate packing tip from forecast (e.g., "Pack layers — 30°F swing" or "Rain gear for Thursday")

Fallback: If weather unavailable, send: "Couldn't pull forecast for {dest_city} — check weather.gov before packing."

3. Pre-departure (T-Xh, X = drive_time + 2h buffer for domestic, +3h international)

🚗 Leave by {leave_by_time} for {origin_airport} Drive: ~{drive_minutes} min via {route_summary} Flight: {airline} {flight_number} departs {depart_time} from {terminal}/{gate} Status: {flight_status} · {delay_info_or_"On time"}

Data sources:

• uv run ~/.claude/skills/flight-tracker/scripts/track.py {flight} --json — status, gate, terminal, delay
• Drive time: Google Maps (coming soon) — use fact-stored estimate as fallback
• ~/dispatch/scripts/fact list --contact "{contact}" --json — origin address, airport

Fallback: If drive time unavailable, omit drive line and say "Head to {origin_airport} — {airline} {flight_number} departs {depart_time}." If flight status unavailable, omit status line.

Leave-by calculation: leave_by = depart_time - drive_minutes - buffer where buffer = 120 min domestic, 180 min international.

4. Gate Update (T-1h, delta-only)

🔄 Gate change: {airline} {flight_number} now Gate {new_gate} ({terminal})

Or if delay:

⏱️ {airline} {flight_number} delayed {delay_minutes} min — new depart {new_depart_time} · Gate {gate}

Data sources:

• uv run ~/.claude/skills/flight-tracker/scripts/track.py {flight} --json

Delta-only rule: Compare current gate/time against last-sent values. If nothing changed, suppress entirely — do not send a "no changes" message. If only minor delay (<10 min), send a single line.

5. Flight Cancellation

🚨 {airline} {flight_number} CANCELLED Was: {origin_code} → {dest_code} at {original_depart_time} 📞 {airline} rebooking: {airline_phone} 👉 {airline_manage_booking_url}

Data sources:

• uv run ~/.claude/skills/flight-tracker/scripts/track.py {flight} --json
• Airline phone/URL from static lookup

Behavior: Send immediately on detection, regardless of quiet hours. This is the one exception.

6. On Landing

🛬 Welcome to {dest_city}! Local time: {local_time} ({tz_name}, {tz_delta} from home) Weather: {icon} {temp}°F — {summary} 🏨 {hotel_name} — {hotel_address} 🍽️ Nearby: {restaurant_1}, {restaurant_2}, {restaurant_3}

Data sources:

• ~/.claude/skills/weather/scripts/weather — current conditions at destination
• ~/dispatch/scripts/fact list --contact "{contact}" --json — hotel name/address
• goplaces nearby --location "{lat},{lng}" --type restaurant --json — top 3 nearby restaurants
• Time zone delta: compute from origin and destination tz

Fallback: Omit any section whose data source fails. At minimum send the welcome line + local time.

7. Daily Morning Intel (each morning at destination)

☀️ {dest_city} · {day_of_week} {date} {icon} {high}°/{low}°F — {summary} 📅 {agenda_or_"No plans on file"} 💡 {tip_of_the_day}

Data sources:

• ~/.claude/skills/weather/scripts/weather — today's forecast
• ~/dispatch/scripts/fact list --contact "{contact}" --json — any stored agenda items
• Tip: contextual (e.g., "Hotel checkout is at {checkout_time}" on last full day)

Frequency cap: Daily for first 7 days, then every-other-day for longer trips.

Fallback: If weather fails, send "Good morning from {dest_city}!" with whatever data is available.

8. Check-out / Last Morning (merged)

🏨 Last day in {dest_city} — checkout by {checkout_time} {icon} {temp}°F — {summary} ✈️ {airline} {flight_number} departs {depart_time} · leave by {leave_by_time} 🧳 Don't forget: chargers, toiletries, safe contents

Data sources: Combination of weather, flight tracker, and facts (checkout time, return flight).

Fallback: If return flight unknown, omit flight line. Always include checkout reminder.

Tool Reference

Airline Check-in URLs

For international carriers, search {airline_name} online check-in at send time.

Fact Schema Reference

Trip data lives in the contact's fact store. Expected structure:

{
  "trip": {
    "destination": "Austin, TX",
    "dest_coords": "30.2672,-97.7431",
    "dest_tz": "America/Chicago",
    "dates": "2026-03-20 to 2026-03-24",
    "purpose": "SXSW"
  },
  "legs": [
    {
      "direction": "outbound",
      "airline": "B6",
      "flight_number": "B6 1234",
      "origin": "BOS",
      "destination": "AUS",
      "depart_time": "2026-03-20T08:30-05:00",
      "arrive_time": "2026-03-20T12:15-06:00",
      "seat": "12A",
      "booking_ref": "XYZABC",
      "class": "economy"
    },
    {
      "direction": "return",
      "airline": "B6",
      "flight_number": "B6 5678",
      "origin": "AUS",
      "destination": "BOS",
      "depart_time": "2026-03-24T17:00-06:00",
      "arrive_time": "2026-03-24T23:30-05:00",
      "seat": "14F",
      "booking_ref": "XYZABC",
      "class": "economy"
    }
  ],
  "hotel": {
    "name": "Hotel San José",
    "address": "1316 S Congress Ave, Austin, TX 78704",
    "coords": "30.2490,-97.7488",
    "check_in": "15:00",
    "check_out": "11:00",
    "confirmation": "HSJ-98765"
  },
  "home": {
    "address": "123 Main St, Somerville, MA",
    "airport": "BOS",
    "drive_minutes_to_airport": 25,
    "tz": "America/New_York"
  }
}

Fields are optional. The pipeline sends whatever it can assemble — partial data beats no data.

Required Fields Checklist

When a trip is first stored, immediately check for these critical fields and proactively ask the user for any that are missing:

Behavior: When storing a new travel fact, scan for missing booking_ref and seat. If either is absent, immediately ask the user — e.g., "got the flight details stored! what's your confirmation code? (i'll need it for the check-in reminder)". Do NOT silently omit critical fields at send time — that's too late.

Key Principles

1. Session-injected, not standalone. Messages are injected into the contact's foreground session via inject-prompt. The user can reply conversationally to update facts (e.g., "actually I'm in seat 14C now").

2. Delta-only at T-1h. Gate update messages suppress entirely if nothing changed. No "everything looks good!" noise.

3. Always send partial data. If weather API is down but flight data works, send what you have. Never skip a moment because one source failed.

4. Leave-by time in pre-departure. Always compute: departure_time - drive_time - buffer. Buffer = 2h domestic, 3h international.

5. Time zone delta on landing. Show the offset from home so the user can mentally adjust ("3h behind home").

6. Frequency cap on daily intel. Days 1-7: daily. Day 8+: every other day. Prevents fatigue on long trips.

7. Quiet hours: 11 PM - 6 AM local time. Queue messages and deliver at 6 AM. Exception: flight cancellations send immediately.

8. Facts are conversational. The user can text trip details naturally ("flying B6 1234 on Thursday, seat 12A") and the session stores them as structured facts for the pipeline to consume.

9. Self-review before sending. Before sending any travel intel message, re-read it and check:

   • Is it under 5 lines? If not, cut the least important line.
   • Does every line contain actionable info? Remove fluff ("have a great flight!" = fluff).
   • Are times in the contact's local timezone?
   • No duplicate info from a previous alert? If the message fails self-review, revise and re-check once.

10. No dedicated CLIs. Use existing skills (flight-tracker, chrome-control, webfetch, places) to fetch weather, drive times, and other data. No weather CLI or drive-time CLI needed — just search the web.