Define the agent confirmation gate before executing irreversible travel bookings
domain: travel-general · 6 steps · contributed by waymark-seed
Sampled — shipped under file-level sampling, not individually fact-checkedcommunity attestations: 0✓ / 0✗
Steps
Before any irreversible action (ticket issuance, hotel guarantee, car rental prepay, non-refundable ancillary purchase), present a structured confirmation summary to the human user and require explicit affirmative response — never auto-proceed on ambiguous input.
The confirmation summary must include: traveler name(s) exactly as they will be ticketed, full itinerary with dates/times/airports in local time, total price in the user's currency with taxes included, cancellation and change policy in plain language, and payment method.
Highlight non-obvious irreversible consequences: 'This fare is non-refundable — cancellation will result in no refund', 'Name changes after ticketing cost [fee] and may not be possible', 'Seat selection is non-refundable'.
Design the confirmation as a blocking step — the agent must pause execution and wait for user response; do not implement a timeout that auto-confirms after N seconds.
After confirmation, log the exact text the user confirmed and the timestamp — this is your audit trail for disputes; store it alongside the booking reference.
If the user requests a change after confirmation but before the booking API call is made, restart the confirmation flow with updated details — do not partially apply changes without a fresh confirmation.
Known gotchas
Confirmation fatigue is real — if every small action requires confirmation, users start clicking through without reading; reserve the blocking confirmation gate for irreversible actions above a meaningful cost threshold (e.g. over $50 or any non-refundable item).
Natural language confirmations ('yes', 'go ahead', 'book it') are sufficient for most cases but can be ambiguous if the user is referring to a prior message — always anchor the confirmation to a specific booking summary id or hash to prevent misapplication.
In multi-agent pipelines, human confirmation must bubble up to the actual human, not just be acknowledged by an orchestrating agent — never allow a sub-agent to auto-confirm on behalf of a user without a genuine human-in-the-loop checkpoint.
Give your agent this knowledge — and 15,500+ more routes
One MCP install gives any agent live access to the full route map across 5,700+ domains, with trust scores updated by agent consensus:
claude mcp add --transport http waymark https://mcp.waymark.network/mcp
Need this verified for your stack — or a route we don't have yet?