hotels

Search travel sites for hotels and vacation rentals, return a cost comparison table, then save results.

When this skill is invoked:

Step 1 — Collect Inputs

First, check whether travel_plans/ exists and contains any destination folders.

• If existing destination folders are found, use AskUserQuestion to ask the user:
  • "Which destination would you like to search?" with one option per existing folder (using the folder name as the label) plus a "New destination" option
  • If they pick an existing destination, pre-fill the destination field from the folder name and skip asking for it again
  • If they pick "New destination", prompt for the destination as normal

Then collect any remaining missing details via AskUserQuestion:

• Destination / neighborhood / property name — skip if pre-filled from an existing folder
• Check-in date (YYYY-MM-DD)
• Check-out date (YYYY-MM-DD)
• Number of guests
• Number of rooms (default 1 if not specified)

Experiences — ask the user which experiences they plan to attend, using multiSelect: true. Offer 3–4 destination-appropriate options based on the destination (e.g., for Orlando: Theme Parks, Disney World, Universal Studios, International Drive, Beaches). Always include an "Other / No preference" option. The user may select multiple.

Use the selected experiences to:

• Prioritize hotels that are close to or affiliated with the chosen experiences (e.g., Disney-area resorts for Disney World, beachfront for beaches)
• Note distance or relevance to each experience in the results
• Filter out properties that are far from all selected experiences when better options exist

Step 2 — Search for Hotels and Rentals

Step 2a — Run Playwright headless search

Use the Bash tool to run:

node playwright/search.js hotels --destination "<DESTINATION>" --depart <YYYY-MM-DD> --return <YYYY-MM-DD> --travelers <N> --rooms <N>

This writes results to travel_plans/[destination-slug]/hotels/processed=[YYYY-MM-DD]/results.md and updates summary.md automatically. Wait for the command to complete.

Step 2b — Read results and identify gaps

Read the written results.md file. Identify any rows marked N/A or sites that returned an error.

Step 2c — Fill gaps with Chrome MCP

Silent mode: Execute all Chrome MCP browser steps without narrating individual actions (clicks, scrolls, typing). Use the tools quietly and only report the final prices found or a single error message if a site is blocked.

For each site with an N/A result, use mcp__claude-in-chrome__* tools to search that site in Chrome and retrieve results. Refer to the per-site navigation guides in sites/ for UI steps:

1. Google Hotels — https://www.google.com/travel/hotels → sites/google-hotels.md
2. Expedia — https://www.expedia.com/Hotels → sites/expedia.md
3. Kayak — https://www.kayak.com/hotels → sites/kayak.md
4. Costco Travel — https://www.costcotravel.com/Hotels → sites/costco-travel.md
5. VRBO — https://www.vrbo.com → sites/vrbo.md
6. Airbnb — https://www.airbnb.com → sites/airbnb.md

For each selected option, note:

• Property name
• Type (Hotel / Condo / Vacation Rental)
• Star rating or guest rating
• Price per night
• Total cost for the full stay
• Proximity or relevance to selected experiences (e.g., "On-site Disney access", "0.5 mi from Universal", "Beachfront")
• Direct link to the listing (if visible)

Step 2d — Update results file with MCP data

If any MCP searches succeeded, edit results.md to replace the N/A rows with the new data.

For each site, find the top 1–4 highest-quality or most preferred options. Prioritize by:

1. Experience alignment — properties close to or purpose-built for the selected experiences rank first
2. Guest rating (highest rated first — e.g. 8.5+ on Expedia/Kayak, 4.5+/5 on Google)
3. Star class (prefer 3-star and above when available at reasonable price)
4. Value (best rating-to-price ratio)

Avoid surfacing low-rated properties (below 7.0/10 or 3.5/5) unless nothing better is available.

If a site still cannot be searched after MCP attempt, leave it as "N/A" and continue.

Step 3 — Output Results Table

Present the results as a markdown table (up to 8 best options across all sites, prioritized by experience alignment):

Site	Property	Type	Rating	Per Night	Total Stay	Experience Fit	Link
Google Hotels	...	Hotel	⭐4.5	$X	$X	Disney area, 0.2 mi to park	Book
Costco Travel	...	Hotel	⭐4.2	$X	$X	I-Drive, near Universal	Book
VRBO	...	Vacation Rental	⭐4.8	$X	$X	Disney Springs area	Book
Airbnb	...	Condo	⭐4.7	$X	$X	Near Universal	Book

Step 4 — Save Results

Derive a folder-safe destination name from the destination input: lowercase, spaces replaced with hyphens (e.g. "San Diego, CA" → san-diego-ca).

4a — Write search results

Write results to travel_plans/[destination-slug]/hotels/processed=[YYYY-MM-DD]/results.md (create directories if needed):

## Hotels — Searched: [current date and time]

### Inputs
| Field | Value |
|---|---|
| Destination | [destination] |
| Check-in | [check-in date] |
| Check-out | [check-out date] |
| Guests | [number] |
| Rooms | [number] |
| Experiences | [comma-separated list of selected experiences] |

### Results
[paste the full results table here]

4b — Update summary.md

Update travel_plans/[destination-slug]/summary.md (create if it doesn't exist):

• Set the Latest Prices row for Hotels to the best rate found today
• Append a new row to the Price History > Hotels table with today's date, best rate, and the change vs. the previous search
• Add an entry to the Search History list pointing to the new processed directory