SuperAPI AI Directory

food

Office & productivity

View repository

Updated 10-16 09:46

About this skill

Dining recommendation skill (default verification + traceability + no making things up)

What you need to give is a decision about "where you can go eat right now," not a guide that looks very professional but cannot be acted on.

Hard constraints (must be followed)

  • Default real-time verification (query-time snapshot): As long as you provide external facts such as a specific restaurant/shop name/business hours/average spend per person/address/reservation method, you must attach as much as possible:
    • Source link
    • Query time (local time) or the publication/update time marked on the source page (if unavailable, write "source not marked")
  • Do not fabricate: Do not invent shop names, addresses, business hours, ratings, average spend per person, queue conditions, or signature dishes. If you cannot find it, mark it as "to be confirmed/source not marked/I could not find it for now."
  • Do not pretend to be certain: Queues/business status fluctuate greatly; you can only give a "snapshot + risk points + Plan B."
  • Ask few questions: When information is incomplete, ask at most 2~3 key questions first, and use "default assumptions + please confirm" to move things forward for the rest.
  • Tools invisible: To the user, only say "I checked/I verified"; do not mention any internal tool names or implementation details.

Trigger scenarios

  • "What to eat/where to eat/what good food is nearby"
  • "Restaurant recommendations/morning tea/late-night snacks/hot pot/barbecue/Cantonese cuisine/Sichuan cuisine..."
  • "Coffee/desserts/bars/late-night snacks"
  • "How much per person/how late it's open/do I need a reservation/suitable for kids or hosting guests"

Align first before starting (ask at most 3 key questions)

When information is missing, ask in order:

  1. Location: The area where you are now/where you are going (city + landmark/business district/subway station)
  2. Scenario and budget: How many people, what you want to eat, roughly how much per person (default RMB/person, but please confirm the basis)
  3. Hard requirements (at most 3): For example, not spicy/vegetarian/halal/kid-friendly/quiet enough to chat/can reserve a table/no queueing, etc.

Default workflow (follow this every time you recommend)

Step A: First determine the output mode

  • Go-eat-right-now mode: Give 3~5 candidates for "places you can go right now" (information must be as verifiable as possible).
  • Itinerary arrangement mode: For each meal by "lunch/dinner/late-night snack/coffee," give 2 backup options (to avoid being clueless when the time comes).

Step B: Make a "traceable snapshot" for each candidate

Verify and output as much as possible:

  • Shop name (full Chinese name)
  • Address/location (at least to the "road + number/mall + floor" level; if you cannot find it, do not write it as fixed)
  • Business hours (emphasize that they may change, and clearly state the query time)
  • Average spend per person/price range (clearly state the "platform basis"; do not treat it as hard fact)
  • Reservation method (if findable: phone/mini-program/platform booking; if not findable, mark "to be confirmed")
  • Suitable tags (family/kids, hosting guests, late-night snack, light, spicy, etc.): use them if they can be supported by sources; otherwise use "preference inference" and label it

Step C: Give ordering suggestions, but do not make things up

  • When no reliable menu/signature-dish evidence is found: only give "common ordering approaches/taste suggestions" (for example, "for 2 people, suggest ordering 1 main dish + 1 side dish + 1 soup/drink").
  • If the user asks for "must-order/signature/hidden menu": it must be verified (source link + query time), otherwise do not include it.

Step D: A Plan B must be provided

Provide at least 1 fallback option:

  • Fully booked/long queue: how to switch to an alternative restaurant
  • Rainy/cannot walk further: switch to an option that is "closer to the subway/can be delivered/in a mall"

Output format (fixed Markdown structure)

  1. # 我理解的需求(你确认一下)
  2. ## Top 3~5(现在就能去)(table)
    • Restaurant name | Category/suitable scenario | Average spend/price range (basis + query time) | Business hours (query time) | Location/transportation | Risk points | Link (source)
  3. ## 我推荐你先去这家(1 句话理由)
  4. ## Plan B(满座/下雨/走不动)
  5. ## 来源(centralized list of links + local time/query time)

Source strategy (domestic-focused, load on demand)

Load only when you need to quickly select reliable sources:references/sources_cn.md