Yihan
ea022aa512
The 2026-08-03 re-audit cleared four of five skills; hotel-search kept its MEDIUM with identical wording and confidence, despite SECURITY.md having shipped before the audit ran. Socket attaches the alert to SKILL.md, and hotel-search was the skill whose in-SKILL.md disclosure was thinnest — a six-line pointer at SECURITY.md — while booking-assistant, which carries the same closed-source backend plus guest PII, states everything inline and passed. So state it inline here too: operator, endpoint, the closed-source caveat said plainly, the exact fields each tool sends, what the skill cannot do, and the standalone deal-finder as the no-backend alternative. The other four are deliberately left alone — they pass, and every edit re-triggers an audit.
Pricewin Skills Hub
Find the cheapest hotel deal in one command. AI-agent skills that compare live prices across Booking.com, Agoda, Google Hotels & OpenTravel and hand back ranked best-value / cheapest / quality picks with direct booking links.
⭐ Most popular skill:
pricewin-hotel-deal-finder— standalone, no MCP server, no API keys.npx skills add https://github.com/Price-Win/pricewin-skills-hub --skill pricewin-hotel-deal-finder
🏨 Tokyo • Aug 12–15 • 3 nights • 2 guests
━━━━━━━━━━━━━━━━━━━━
🥇 BEST VALUE Shinjuku Granbell Hotel ✅ agoda $118 · booking $131 → save $13
🥈 CHEAPEST APA Hotel Shinjuku ✅ google $94
📊 18 hotels | agoda, booking, google, opentravel • prices in USD
Why people install it:
- 🥇🥈🥉 best-value / cheapest / quality picks, side by side
- 4 sources, all normalized to USD, with clickable links to the cheapest OTA
- Works for any city worldwide — even bot-hardened ones (Shanghai, Hangzhou, Bangkok) via a stealth Patchright daemon
- No MCP server, no API keys — just
node+npx
Skills for AI agents to search hotels, compare OTA prices, and surface the best booking deals. Two families:
- MCP skills — drive the Pricewin MCP server (
pricewin). - Standalone skill —
pricewin-hotel-deal-finderruns its own browser + API stack, no MCP server required.
Repo: git@github.com:Price-Win/pricewin-skills-hub.git
Skills
| Skill | Type | Trigger | Sources / Tools |
|---|---|---|---|
pricewin-hotel-search |
MCP | search/find hotels live for travel dates | search_hotels_live → poll_search_results (Agoda + Booking.com + Traveloka, plus opentravelResults OpenTravel listings) |
pricewin-price-comparison |
MCP | compare prices, check room rates for a named hotel | search_hotels_live → poll_search_results, get_ota_hotel_detail, get_hotel_detail, get_cancellation_policy |
pricewin-booking-assistant |
MCP | recommend room, book + pay, manage a reservation | get_hotel_detail → create_booking, check_booking_status, recreate_payment_link, request_cancel_token → cancel_booking |
pricewin-hotel-deal-finder |
Standalone | hotel deals / price comparison for travel dates | Browser automation (Patchright) over Booking.com + Agoda + Google Hotels, plus the OpenTravel public API — no MCP server |
pricewin-hotel-deal-finder (standalone)
Self-contained skill: a long-running Patchright (stealth Chromium) daemon scrapes Booking.com, Agoda, and Google Hotels, while a single HTTPS call hits the OpenTravel public API. One command does everything:
node bin/search.js "<city>" <checkIn YYYY-MM-DD> <checkOut YYYY-MM-DD> <adults> en-us
- OpenTravel API base:
https://api.opentravel.one(override withOPENTRAVEL_API_BASE_URL) - Setup:
skills/pricewin-hotel-deal-finder/install.sh(Node deps + Chromium, ~200MB one-time) - Cache:
~/.cache/pricewin-hotel-deal-finder/selectors.json
See skills/pricewin-hotel-deal-finder/SKILL.md for full usage and rules.
Roadmap: hotels and flights ship today (flights via skills/pricewin-flight-search/, MCP-driven); rental-car price comparison is on the way — the same one-command, multi-source, USD-normalized deal-finding, extended across your whole trip. (Powered by PriceWin — tìm giá tốt nhất across 50+ travel platforms.)
Install (agents)
The skill follows the agentskills.io SKILL.md standard and
publishes to both ecosystems. node_modules and Chromium are not bundled — the
install hook fetches them on first install (needs network + node/npx on PATH).
# OpenClaw / ClawHub
clawhub skill publish ./skills/pricewin-hotel-deal-finder \
--slug pricewin-hotel-deal-finder --name "PriceWin Hotel Deal Finder" --version 0.7.0 --tags travel,hotel
openclaw skills install pricewin-hotel-deal-finder # users / agents
# Hermes — install straight from this repo (github: source pulls the whole dir)
hermes skills install github:Price-Win/pricewin-skills-hub/skills/pricewin-hotel-deal-finder
# Manual / dev
git clone git@github.com:Price-Win/pricewin-skills-hub.git
cd pricewin-skills-hub/skills/pricewin-hotel-deal-finder && ./install.sh
Security & data handling
Every skill ships a SECURITY.md next to its SKILL.md — what it executes, what
leaves the machine, and who operates the backend:
| Skill | Runs code? | Data it sends | Disclosure |
|---|---|---|---|
pricewin-hotel-deal-finder |
Yes — local Patchright daemon | search query only (city, dates, guests) | SECURITY.md |
pricewin-hotel-search |
No | search query | SECURITY.md |
pricewin-price-comparison |
No | search query + queryText |
SECURITY.md |
pricewin-flight-search |
install.sh only (MCP registration) |
route + dates | SECURITY.md |
pricewin-booking-assistant |
No | guest PII (name, phone, email) — real bookings | SECURITY.md |
The deal-finder's local daemon is loopback-only, token-authenticated (0600
state file), rejects non-loopback Host headers, and keeps the Chromium sandbox
on. No skill ever handles card numbers: payment happens on the provider's page.
Privacy policy: https://price.win/en/privacy-policy.
FAQ
Is Booking.com or Agoda cheaper? Neither wins universally — it depends on the hotel, dates, and region. Independent tests (CNBC's review of ~200,000 hotel searches) found Agoda had the lowest rate about 34% of the time globally, and it tends to win in Asia-Pacific while Booking.com pulls ahead more often in Europe. That is exactly why pricewin-hotel-deal-finder queries both (plus Google Hotels and OpenTravel) in a single run, shows the per-night gap, labels the cheapest source per property, and links you straight to it — so you never have to guess.
How many sources does it compare, and how fast? Up to 4 sources (Booking.com, Agoda, Google Hotels, OpenTravel), all normalized to USD. A cached city returns in ~30–60 seconds; a brand-new city pays a one-time Agoda discovery cost of 2–4 minutes, then joins the cache.
Do I need API keys or an MCP server? No. pricewin-hotel-deal-finder is standalone — node + npx is all it needs. The only external dependency is Chromium (~200 MB), fetched once by the install hook.
Does it work for bot-hardened cities? Yes. A stealth Patchright (stealth Chromium) daemon handles anti-bot detection, so cities like Shanghai, Hangzhou, and Bangkok resolve where vanilla scrapers get blocked.
How are prices normalized? Booking.com returns USD natively; Agoda, Google Hotels, and OpenTravel geo-lock to VND by IP and are converted via a live FX rate. Every price in the output is USD.
What does one command return? Ranked best-value 🥇 · cheapest 🥈 · quality 🥉 picks, each hotel name a clickable link to its cheapest OTA, plus a "More good deals" list grouped by source and a USD footer of exactly which sources returned data.
Last verified: 2026-07-08 · reviewed quarterly.
MCP skills — prerequisites
- Pricewin MCP server running (
pricewin-mcp, stdio transport, auto-launched by Claude) - Backend API accessible
{
"mcpServers": {
"pricewin": {
"command": "node",
"args": ["/path/to/pricewin-mcp/dist/index.js"]
}
}
}