trip-schedule

name	trip-schedule
description	Use when planning a sourced, budget-constrained trip in mainland China, including attractions, train or flight comparisons, hotel alternatives, local transportation, route maps, and itinerary HTML.

Trip Schedule

Start every trip

Before collecting data, obtain all six required values:

1. Origin city
2. Destination
3. Total budget in CNY
4. Departure date and time
5. Total trip duration
6. Number of travelers

Immediately before formal generation, ask whether the user wants one-shot or interactive generation. Do not infer this choice.

Create a new per-trip output directory. Do not use an older trip directory as a live data cache.

Check dependencies

Run:

python scripts/trip_schedule.py health --json

Explain every provider that is unavailable. Never install a CLI, Skill, MCP server, browser component, or Python package without explicit user approval. Read provider-policy.md before configuring or changing a provider.

Collect mainland-China sources

1. Query the migrated 12306 provider for train schedules and availability.
2. Discover an installed flight tool first; use the configured fli provider when available. Keep train prices null unless a reliable price source returned them.
3. Query the configured Xiaohongshu wrapper for destination recommendations.
4. Query the configured hotel wrapper for total-stay prices and locations.
5. Preserve source URLs, query times, freshness, confidence, warnings, and failure categories.
6. Stop a provider on CAPTCHA, login challenge, or platform verification. Do not bypass it.

Review the sourced media evidence and write attraction_candidates.json using the contract in data-contracts.md. Do not invent an attraction, ticket price, opening hour, or source. Let AMap resolve each candidate to a real POI.

When the user asks for restaurant recommendations, review sourced dining evidence and write structured restaurant candidates using the contract in data-contracts.md. Keep the evidence extraction separate from scoring: the agent extracts restaurant names, cuisine tags, budget/person, source URLs, and travel-stage estimates; the independent planning.restaurant_scoring module ranks those candidates from the structured fields. Do not couple restaurant ranking to the Xiaohongshu wrapper, 12306 provider, or AMap provider internals.

Generate

Write the six inputs and selected mode to request.json, then run:

python scripts/trip_schedule.py generate \
  --request request.json \
  --attraction-candidates attraction_candidates.json \
  --output-root trip-output

For one-shot mode, return the balanced primary plan plus economy and time-saving alternatives when valid candidates exist.

For interactive mode, present only option IDs from interactive-state.json. Collect the user's choice, write:

{"selection_ids": ["an-option-id-that-was-offered"]}

Then resume:

python scripts/trip_schedule.py resume \
  --workspace trip-output/<trip-directory> \
  --selection selection.json

Repeat for intercity transport, hotel area, and attractions. Never accept a record supplied by the selection file; accept only IDs already present in the workspace.

Report quality and cost

Treat the budget as a hard constraint with a 10% contingency. If no plan fits, show the smallest deficit and the main cost drivers.

State which values are live, estimated, missing, or stale. Show provider failures and fallbacks from provider-report.json. Do not describe a partial plan as complete.

Query and recommend only. Do not book, submit identity information, do not pay, or do not bypass security controls.

Present the itinerary

Start the loopback-only itinerary server:

python scripts/trip_schedule.py serve trip-output/<trip-directory>

Return the local URL and the artifact paths. Do not copy AMap credentials into HTML, JSON, logs, or chat.

Update memory

After a completed or useful partial run, update only the installed Skill's de-identified region strategies and provider reliability counters. Do not store trip dates, origin city, budget, traveler count, selected hotel, ticket results, cookies, or credentials.

Disclose the update in one localized sentence equivalent to:

Updated Trip Schedule's de-identified strategy memory.