// the find
tourmind-com/Tourmind-Booking-Skills
AI agent skills for end-to-end hotel and flight booking—compare live hotel rates across leading OTAs and suppliers, search and verify real-time airfares worldwide, book stays and flights, and manage reservations, hotel cancellations, and payments via the TourMind API.
A set of 'Skill' definitions (mostly SKILL.md instruction files plus a couple of YAML configs) that teach an AI agent how to call TourMind's proprietary hotel and flight booking API — search, price-check, book, pay, cancel. It's for developers or end users who already have (or are willing to sign up for) a TourMind account and want their agent to book travel through that one vendor's inventory.
The confirmation flow is genuinely well thought out: search never creates an order, price rechecks are required before booking (cached search price isn't the bookable price), and payment/cancellation each need a separate explicit user confirmation — this is the kind of guardrail you actually want before letting an LLM touch a credit card. Token handling has a sane failure mode: a 401 causes the agent to delete the local token rather than retry blindly, and file permissions are locked to 600 on macOS/Linux. There are dedicated test files for auth recovery and payment safety, which is more than most 'agent skill' repos bother with.
There's almost no actual software here — the repo is markdown instructions for an LLM to follow (SKILL.md, reference docs) plus one validation script; 'testing' means checking whether the agent obeys the prompt, not testing a library's behavior. Every booking action is locked to TourMind's own API and requires their token, so this is a vendor SDK dressed as an open-source skill, not something you can point at a different OTA or your own supplier contracts. Pricing and inventory are a black box — no visibility into which underlying suppliers TourMind is aggregating or how its rates compare to booking direct, so 'compare live rates across leading OTAs' can't be independently verified. The README is landing-page marketing (hero image, GIF demos, install-in-1-minute, ClawHub install counter) rather than documentation, which makes it hard to judge engineering quality separately from the sales pitch.