Trip Planner

Assume first. Show your work. Book the flight.

Core Design Principle

Two Flows

.claude/trip-planner.md

.claude/trip-planner.md

Onboarding Flow (First Run Only)

.claude/trip-planner.md

Step 1: Welcome (skip if user already described a trip)

references/branding.md

1. Logo
2. Origin quote
3. Workflow diagram
4. Key concepts

• Yes → Skip straight to Step 2 with their input
• No → Display:

Where are you headed? Just describe your trip naturally:

"I need to fly to Tokyo next month for a week"
"London to Paris, March 15-18, 2 adults"
"Cheapest flights to Barcelona in April"

Step 2: Parse & Assume

• Origin: LHR
• Class: Economy
• Passengers: 1 adult
• Budget: Flexible
• Stops: Any
• Currency: GBP

## Assumptions

| Field | Value | Source |
|-------|-------|--------|
| Destination | Tokyo (NRT/HND) | You said |
| Origin | London (LHR) | Default |
| Dates | Mar 10 - Mar 17, 2026 | "next month" + 7 days |
| Passengers | 1 adult | Default |
| Class | Economy | Default |
| Budget | Flexible | Default |
| Stops | Any | Default |

Anything wrong? Say so and I'll re-search. Otherwise I'll find flights.

Step 3: Search & Present

assets/templates/trip-recommendation.md

Want to adjust anything? Change dates, class, budget — I'll re-search.

Step 4: Quick Setup

Nice — that's your first search done.

Want to save your preferences so next time is faster? Quick 5 questions:

1. What's your home airport? (e.g. LHR, JFK, LAX)

2. Any preferred airlines? (comma-separated, or "none")

3. Default budget range per person? (e.g. "under 500", "flexible", "100-300")

4. Preferred class? (economy / premium economy / business / first)

5. Preferred booking platform? (skyscanner / google-flights / kayak)

• home_airport: "LHR"
• preferred_airlines: []
• budget: "flexible"
• class: "economy"
• platform: "skyscanner"

.claude/trip-planner.md

Saved to .claude/trip-planner.md

Run /trip-planner anytime — just describe where you want to go.

Regular Flow (Every Subsequent Run)

1. Parse intent from natural language
2. Load config for defaults
3. Run Assumption Engine (fill all gaps)
4. Search for flights
5. Present recommendations with booking links

Step 1: Parse Intent

• Destination(s) — city names, airport codes, or countries
• Dates — explicit dates, relative dates ("next month"), or vague ("in April")
• Budget — explicit amounts or qualifiers ("cheap", "under 300")
• Class — economy, premium economy, business, first
• Passengers — number and type (adults, children, infants)
• Preferences — direct flights, specific airlines, time of day, layover preferences
• Multi-city — detect "A → B → C → A" or "then" patterns

Step 2: Load Config

.claude/trip-planner.md

references/config-guide.md

• name: ""
• home_airport: "LHR"
• class: "economy"
• budget: "flexible"
• currency: "GBP"
• stops: "any"
• preferred_airlines: []
• platform: "skyscanner"

Step 3: Assumption Engine

1. Explicit user input — what they said in this message
2. Config file defaults — from

   .claude/trip-planner.md

3. Contextual inference — smart guesses based on trip type:
   • City break (European short-haul) → 3-4 days
   • Intercontinental → 7-10 days
   • Weekend trip → Friday-Sunday
   • "Holiday" / "vacation" → 7 days
   • Month mentioned but no dates → middle of month, flexible
4. Smart defaults — final fallback:
   • 1 adult
   • Economy
   • Flexible budget
   • Any stops
   • Home airport from config (or LHR)
   • Dates: 4 weeks from today, 7 days duration

date

## Assumptions

| Field | Value | Source |
|-------|-------|--------|
| Destination | {dest} | You said |
| Origin | {origin} | Config / Default |
| Dates | {dates} | {source} |
| Passengers | {pax} | {source} |
| Class | {class} | {source} |
| Budget | {budget} | {source} |
| Stops | {stops} | {source} |

Anything wrong? Otherwise I'll search.

Step 4: Search

Step 5: Present

assets/templates/trip-recommendation.md

• Best Value — best balance of price, duration, and convenience
• Cheapest — lowest price regardless of convenience
• Fastest — shortest total travel time

Want to adjust anything? I can re-search with different dates, budget, or preferences.

Assumption Engine — Detail

Priority Order

User Input > Config > Context Inference > Smart Defaults

Context Inference Rules

Destination Resolution

• Map to primary airport code(s)
• For cities with multiple airports, include all (e.g. London → LHR/LGW/STN, Tokyo → NRT/HND, New York → JFK/EWR/LGA)
• For countries or regions, map to capital or most common tourist airport

Date Handling

• Always use

  date

  command for current date
• "Next month" → 1st of next month + duration
• "In March" → March 15 (mid-month) + duration, flag as flexible
• "March 15" → exact date + duration
• "March 15-18" → exact range
• "Next weekend" → upcoming Friday-Sunday
• No dates at all → 4 weeks from today + duration based on trip type

Search Strategy

Step A: Construct Booking URL

https://www.skyscanner.net/transport/flights/{origin}/{dest}/{depart_date}/{return_date}/?adults={n}&cabinclass={class}&currency={currency}

• Dates:

  YYMMDD

  (e.g.

  260315

  for March 15, 2026)
• Origin/dest: airport codes lowercase (e.g.

  lhr

  ,

  nrt

  )
• Cabin class:

  economy

  ,

  premiumeconomy

  ,

  business

  ,

  first

https://www.google.com/travel/flights?q=flights+from+{origin}+to+{dest}+on+{date}

Step B: WebSearch for Pricing

• "flights {origin} to {dest} {month} {year} price"

• "cheap flights {origin} to {dest} {month} {year}"

• "{airline} {origin} {dest} {month} {year} fare"

  (if preferred airline set)

• Airlines operating the route
• Approximate price ranges
• Flight duration and stop information
• Any deals or sales mentioned

Step C: WebFetch Booking Page (optional)

WebFetch

• If successful: Extract specific flight options with prices
• If blocked/failed: Gracefully fall back to WebSearch data. Do not retry or error — just note that prices are approximate

Step D: Compile Results

• Deduplicate by airline + route + approximate time
• Sort into three categories: Best Value, Cheapest, Fastest
• Include direct booking URL for each option
• Flag any data as "approximate" if from WebSearch only vs live pricing

Multi-City Search

1. Decompose into individual legs
2. Search each leg independently (can run WebSearch queries in parallel)
3. Present combined itinerary overview + per-leg options
4. Calculate total cost across all legs
5. Provide individual booking links per leg AND a combined multi-city search link

Config Structure

.claude/trip-planner.md

name: "Foluso"
home_airport: "LHR"

preferences:
  class: "economy"
  budget: "flexible"
  currency: "GBP"
  stops: "any"
  preferred_airlines: []
  time_preference: "any"    # morning / afternoon / evening / any

platform:
  primary: "skyscanner"     # skyscanner / google-flights / kayak

references/config-guide.md

Key Principles

• Assume everything, show everything. Never ask a question when you can make a reasonable assumption. Always show what you assumed so the user can correct.
• One hard requirement: destination. Everything else has a default. A user saying "flights to Tokyo" is enough to produce a full recommendation.
• Value first, setup later. Search before asking for configuration. Onboarding creates value immediately.
• Corrections over questions. Present assumptions, let user fix what's wrong. Faster than asking upfront.
• Show your sources. Every assumption in the card has a "Source" column. Every price has a booking link.
• Graceful degradation. If WebFetch fails, fall back to WebSearch. If WebSearch is sparse, still show what you have with booking links.
• Always get current date. Run

  date

  command — do not rely on system date or knowledge cutoff.
• Multi-city is first-class. Not an afterthought. Decompose, search per-leg, present combined.

File Paths

.claude/trip-planner.md

.claude/skills/trip-planner/SKILL.md

references/branding.md

references/config-guide.md

references/future.md

assets/templates/trip-recommendation.md