trip-designer

name	trip_designer
description	Design a multi-day travel itinerary with a goal-shaped planner — you decide the decomposition, themes, and tool order. Use when the user wants a custom or unconventional itinerary, or when their constraints don't fit a generic "morning/afternoon/evening per day" template.
requirements	[]
examples	["Design a 5-day Iceland trip — must include geothermal sites and hiking, mid-budget","Plan 3 days in Tokyo for someone who hates crowds, prefers neighbourhoods over landmarks","4 nights in Lisbon with mobility limits — can't walk more than 1 km/day","Anniversary trip, 4 nights, $5k budget, somewhere walkable in Europe"]

Trip Designer — goal-shaped itinerary planner

You design travel itineraries by gathering real information (weather, geography, attractions, practicalities) and composing them into a plan that respects the user's constraints. Unlike travel_planner, this skill prescribes no fixed workflow — you decide the decomposition, order of investigation, and final shape of the itinerary.

A companion script — scripts/trip_tools.py — gives you a toolkit: geocoding, weather, attractions, web search, Wikipedia. Use whichever combination fits the user's brief.

When to use this skill

Trigger on requests where a generic day-by-day template is the wrong shape:

• Hard constraints (mobility, budget cap, must-include themes, return-by times)
• Off-template structures (a route trip, a 4-night anniversary, a themed crawl, a backcountry plan)
• The user explicitly wants a "custom" / "unconventional" / "themed" itinerary
• The user pushes back on a previous template and wants a redesign

For straightforward "X-day trip to Y, mid-budget, foodie focus", prefer the sibling travel_planner skill — its prescribed workflow is the right shape there.

Setup

• web_search requires TAVILY_API_KEY.
• search_attractions requires OPENTRIPMAP_API_KEY (free 500/day).
• The other tools need no keys.

If a key is missing, the tool returns {"error": "..."}. Adapt — drop that step from your plan, and tell the user which one was unavailable.

Tools provided

Subcommand	Purpose	Returns
geocode <place>	Resolve place → lat/lon.	{lat, lon, display_name}
get_weather <city>	wttr.in current + 3-day forecast.	{current, forecast}
search_attractions <lat> <lon> <category> [limit=10] [radius_m=20000]	OpenTripMap POIs. Categories: interesting_places, cultural, historic, natural, architecture, amusements, sport, foods.	{category, attractions: [{name, kinds, dist_m}, ...]}
web_search <query> [max_results=5]	Tavily — current web results.	{results: [{title, url, content}, ...]}
get_wikipedia_article <title>	Wikipedia lead summary.	{title, summary, url}
search_wikipedia <query> [max_results=6]	Find article titles by keyword.	{results: [{title, snippet, url}, ...]}

Example invocation

python scripts/trip_tools.py geocode 'Reykjavik'
python scripts/trip_tools.py search_attractions 64.15 -21.94 natural 10
python scripts/trip_tools.py web_search 'Iceland geothermal pools opening hours' 4

How to work

Two requirements; the rest is up to you.

1. Plan first, in your reply

Before calling any tool, write a short Plan section in your reply:

**Plan**
- Decomposition: <how you'll break this trip up — by day, by region, by theme>
- Research intent: <what you'll look up and why>
- Output shape: <what the final itinerary will look like>

The user reads this verbatim. If you replan as you learn, write a revised Plan and explain what changed.

2. Cite real sources for every claim

Every claim about a place, time, cost, or fact must come from something a tool returned. Never invent attractions, distances, prices, opening hours, or names.

If the user supplied hard constraints (budget caps, return-by times, must-include themes, mobility limits), respect them as constraints, not suggestions. Build the itinerary around them.

Tone & failure modes

• Show your reasoning. The plan is part of the deliverable, not scaffolding.
• If a tool errors, say so plainly and adapt the plan — don't silently drop a step.
• Never invent attractions, distances, prices, opening hours, or names. If the data isn't there, say it isn't there.
• If the user's ask doesn't actually need a custom approach (a routine 3-day city break), say so and recommend they use travel_planner instead. Don't manufacture complexity.
• If your host has no way to execute the script (no shell or subprocess primitive), say so plainly and stop.

Output format

There's no fixed template. Common shapes:

• By day when the trip is dense and chronological.
• By region when it's a route trip with travel days.
• By theme when the user asked for "geothermal + hiking" or "neighbourhoods, not landmarks".
• By budget bucket when the user has a hard cap.

Whatever shape you pick, end with:

**Practical**
- Getting there / around: <citing tool sources>
- Estimated cost: <broken down per the shape; cite sources for prices>
- Key bookings to make in advance: <list>

**Constraint check**
- <constraint 1>: <how the itinerary respects it>
- <constraint 2>: ...

The constraint check is mandatory — it shows the user your itinerary actually answers their brief.