Traffic Query

Overview

Use this skill to gather and summarize domestic China flight and train information. Prioritize getting a complete, verifiable list over giving a fast but fuzzy answer. Prefer Browser Relay for already-open result pages; use public pages only when necessary.

Core workflow

1. Extract the search constraints.
2. Choose the best source.
3. Collect raw results.
4. Normalize into a clean list.
5. Summarize the best options and any caveats.

1) Extract constraints

Identify and restate internally:

• mode: flight / train / either
• origin
• destination
• departure date
• one-way / round-trip
• filters: direct only, earliest, cheapest, specific airline/train type, arrival deadline, station/airport preference

If the user omits a critical field, ask only for the missing field.

2) Source priority

Choose sources in this order:

Flights

1. User-opened results page via Browser Relay if available
2. Airline / airport official pages
3. Ctrip / Trip.com / similar OTA results pages
4. Flight status/reference sites for schedule confirmation only

Trains

1. User-opened 12306 or OTA results page via Browser Relay if available
2. 12306 web pages
3. Ctrip / Trip.com rail pages
4. Other public timetable pages for backup only

3) Retrieval strategy

When Browser Relay is available

Use the connected browser tab first. It is best for:

• logged-in pages
• dynamic results
• anti-bot heavy pages
• extracting all rows from an already-filtered result page

When only public web is available

Use browser automation or web fetch to reach a public results page. If the site is dynamic and does not expose clean results, say so clearly instead of inventing missing rows.

When using Agent Browser / browser automation

• Prefer result pages with explicit route/date parameters.
• Re-snapshot after each page/state change.
• Be careful with city-level routes versus exact airport/station routes.
• Confirm whether the page is showing BJS vs PKX, SHA vs PVG, etc.

4) Normalize output

Always normalize each result into the same fields where available.

Flights

• airline
• flight number
• origin airport
• destination airport
• departure time
• arrival time
• duration
• direct / stopover
• price
• cabin if visible
• source

Trains

• train number
• train type (G/D/C/Z/T/K etc.)
• origin station
• destination station
• departure time
• arrival time
• duration
• seat classes / remaining tickets if visible
• price
• source

If a field is unavailable, leave it blank in notes rather than guessing.

5) Response format

When the user asks for all matching options, provide:

1. brief scope line
2. full list
3. short highlights: earliest / cheapest / best balance
4. caveats about source quality or incomplete fields

Recommended compact format for chat:

• 航司/车次｜出发-到达｜时长｜直飞/经停或车型｜价格

Example:

• 南航 CZxxxx｜07:30-10:45｜3h15m｜直飞｜¥980
• 高铁 Gxxxx｜08:00-16:32｜8h32m｜二等座 ¥553

Reliability rules

• Do not claim results are complete unless the page clearly exposes the full list.
• Distinguish schedule info from bookable price info.
• Say when prices may vary by seat/cabin, region, or login state.
• If only partial data is available, explicitly label it as partial.
• Prefer honesty over false precision.

Fallback guidance

If the user wants a complete real-time list and public scraping is unreliable:

• ask them to open a filtered result page in browser
• connect via Browser Relay
• extract directly from that page

References

For source-specific notes and caveats, read references/sources.md.