home / skills / jinfanzheng / kode-sdk-csharp / food

food skill

safe

/examples/Kode.Agent.WebApiAssistant/skills/food

This skill provides real-time, source-backed restaurant recommendations with operating hours, price ranges, and booking options to guide immediate dining

npx playbooks add skill jinfanzheng/kode-sdk-csharp --skill food

Review the files below or copy the command above to add this skill to your agents.

Files (3)

SKILL.md

3.9 KB

---
name: food
description: >-
  餐饮推荐/吃什么/哪里吃/咖啡酒吧甜品夜宵。用于用户问“附近有什么好吃的/餐厅推荐/人均多少/营业到几点/要不要预约/适合亲子或请客”等。
  默认做“查询时间快照”（当地时间）并给来源链接；不编造店名/地址/营业时间/人均/评分；信息不足先问2-3个关键问题。
---

# 餐饮推荐 skill（默认查证 + 可追溯 + 不瞎编）

你要给的是“现在就能去吃”的决策，不是写一篇看起来很专业但没法行动的攻略。

## 强约束（必须遵守）

- **默认实时查证（查询时间快照）**：只要你给出具体餐厅/店名/营业时间/人均/地址/预约方式等外部事实，就必须尽量附：
  - 来源链接
  - 查询时间（当地时间）或来源页面标注的发布时间/更新时间（拿不到就写“来源未标注”）
- **不编造**：不虚构店名、地址、营业时间、评分、人均、排队情况、招牌菜。没查到就标“待确认/来源未标注/我暂时没查到”。
- **不要装确定**：排队/营业状态波动大；你只能给“快照 + 风险点 + Plan B”。
- **少问问题**：信息不全时最多先问 2~3 个关键问题，其余用“默认假设+请确认”推进。
- **工具不可见**：对用户只说“我查了一下/我核对了”，不提任何内部工具名或实现细节。

## 触发场景

- “吃什么/哪里吃/附近有什么好吃的”
- “餐厅推荐/早茶/宵夜/火锅/烧烤/粤菜/川菜…”
- “咖啡/甜品/酒吧/夜宵”
- “人均多少/营业到几点/要不要预约/适合亲子或请客”

## 开始前先对齐（最多问 3 个关键问题）

缺信息时按顺序问：
1. **位置**：你现在/你要去的区域（城市 + 地标/商圈/地铁站）
2. **场景与预算**：几个人、想吃什么、人均大概多少（默认 RMB/人，但请你确认口径）
3. **硬性要求（最多 3 条）**：例如不辣/素食/清真/带娃友好/安静能聊天/能订位/不排队等

## 默认工作流（每次推荐都按这个走）

### Step A：先定输出模式

- **立刻去吃模式**：给 3~5 家“现在就能去”的候选（信息必须尽量可核验）。
- **行程里安排模式**：按“午/晚/夜宵/咖啡”给每餐 2 家备选（避免到点抓瞎）。

### Step B：对每个候选做“可追溯快照”

尽量查证并输出：
- 店名（中文全称）
- 地址/定位（至少到“路+号/商场+楼层”层级；查不到就别写死）
- 营业时间（强调可能变动，写清查询时间）
- 人均/价位（写清“平台口径”，不当成硬事实）
- 预约方式（如可查到：电话/小程序/平台订位；查不到就标“待确认”）
- 适配标签（亲子/请客/宵夜/清淡/辣等）：能用来源支撑就用；否则用“偏好推断”并标注

### Step C：给点单建议，但不乱说

- 没查到可靠菜单/招牌证明时：只给“常见点法/口味建议”（例如“2人建议点1主菜+1配菜+1汤/饮品”）。
- 用户要求“必点/招牌/隐藏菜单”：必须查证（来源链接+查询时间），否则不写。

### Step D：必须给 Plan B

至少给 1 套兜底：
- 满座/排队久：备选店怎么切换
- 下雨/走不动：换“离地铁更近/可外卖/商场内”的方案

## 输出格式（固定 Markdown 结构）

1. `# 我理解的需求（你确认一下）`
2. `## Top 3～5（现在就能去）`（表格）
   - 店名｜品类/适合场景｜人均/价位（口径+查询时间）｜营业时间（查询时间）｜位置/交通｜风险点｜链接（来源）
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`（集中列链接 + 当地时间/查询时间）

## 来源策略（国内为主，按需加载）

需要快速选取可靠来源时再加载：`references/sources_cn.md`

Overview

This skill helps users decide where to eat right now by returning short, verifiable restaurant recommendations with source links and a current-time snapshot. It prioritizes actionable options you can visit immediately, avoids fabricating facts, and asks up to three quick questions if key details are missing. The goal is a fast, practical decision: which place to go, what to expect, and a Plan B.

How this skill works

When asked for dining suggestions, the skill first checks for missing context (location, party/budget, hard constraints) and may ask up to two or three clarifying questions. It then produces 3–5 nearby candidates (or meal-based pairs for itinerary mode), each accompanied by verifiable fields: name, address level, business hours, price guidance, reservation method, risk notes, and a source link with the local query time. If information is unavailable it marks fields as 'to be confirmed' rather than inventing details.

When to use it

You want a quick list of restaurants you can visit right now in a specific area.
You need answers about opening hours, average cost, or whether to reserve.
You’re choosing by scene (family-friendly, date night, late-night, coffee, desserts).
You need a Plan B for full/closed venues or bad weather.
You want traceable sources for any factual claims about a restaurant.

Best practices

Provide location (city + landmark/metro or neighborhood) up front for the fastest results.
State party size and budget or accept the default assumption (RMB/person unless confirmed).
List up to three hard constraints (e.g., vegetarian, no spicy, kid-friendly).
Treat all hours and crowd/queue comments as snapshots; check links before leaving.
Accept ‘to be confirmed’ markers when sources don’t provide a specific fact.

Example use cases

User asks: 'Any good dinner spots near X metro station for 2 people, budget ~RMB 100/person?'
User asks: 'Where for late-night snacks open after 11pm in downtown?'
User asks: 'Recommend family-friendly restaurants near the mall, need high chairs.'
User wants coffee/dessert suggestions around a business district for a 3pm meeting.
User asks: 'Should I reserve for a Saturday night dinner at popular local restaurants?'

FAQ

What if you can't find definite opening hours or menu/price?

I will mark those fields as 'to be confirmed' and provide trusted links and the query time so you can verify before going.

Will you invent restaurant names or addresses if sources are missing?

No. I never fabricate facts. If a detail isn’t verifiable I flag it and may ask a quick follow-up question.

How many clarifying questions will you ask before recommending?

At most two or three concise questions (location, party/budget, and hard constraints). Other assumptions will be noted and need your confirmation.