Run any Skill in Manus with one click

grocery-comparison

Smart grocery price comparison agent for Asian international students. Compares prices across Weee, Asian markets (H Mart, 168, etc.), and mainstream stores (Kroger, Costco, Sam's Club, Aldi). Considers user taste preferences, distance, and cultural factors. Use when user asks to compare grocery prices, find best deals, or get shopping recommendations.

Run Skill in Manus

Overview

Install command

npx skills add https://github.com/KnoWhiz/DoWhiz --skill grocery-comparison

Copy and paste this command into Claude Code to install the skill

Source

KnoWhiz/DoWhiz

Stars18

Forks4

UpdatedApril 8, 2026 at 18:50

File Explorer

2 files

SKILL.md

readonly

Smart Grocery Comparison Agent

This skill helps users (primarily Asian international students) find the best value groceries by comparing prices across multiple channels while considering personal preferences.

Core Principles

Not just cheapest, but best value for YOU - Consider taste, distance, delivery options
Cultural awareness - "American cakes are too sweet" is a valid preference
Multi-channel - Online (Weee, Yami) + Physical stores (H Mart, 168, Costco, Kroger)
Explain trade-offs - "$2 cheaper at 168 but 25 min drive vs Weee delivers free"

Culinary Knowledge (核心差异化)

This is the key differentiator: understand what dish the user wants to make, and recommend the RIGHT ingredient, not just the cheapest.

菜肴 → 食材要求映射

当用户提到想做某道菜时，要理解这道菜对食材的具体要求：

菜肴	推荐部位/食材	关键要求	不推荐
糖醋小排	肋排/小排 (spare ribs)	小块、带少量肉、骨肉比例均衡	大骨头、纯瘦肉
排骨汤	筒骨/大骨 (pork neck bones, soup bones)	大块、骨髓丰富、适合长时间炖	小排（肉多骨少，不出味）
红烧排骨	肋排	中等大小、肥瘦相间	太大或太小的块
红烧肉	五花肉 (pork belly)	层次分明、皮肉脂比例好	太瘦的部位
回锅肉	后臀尖/二刀肉	肥三瘦七、连皮带肉	纯瘦肉、五花（太肥）
水煮肉片	里脊/梅花肉	瘦肉为主、易切薄片	带筋的部位
火锅肥牛	肥牛卷 (beef for hotpot)	薄切、脂肪花纹	牛排级别（太厚）
牛肉面	牛腩/牛腱 (beef shank, brisket)	带筋、炖后软糯	里脊（太柴）
清蒸鱼	鲈鱼/鳜鱼/石斑	鲜活、1-1.5磅	冷冻鱼（口感差）
酸菜鱼	草鱼/黑鱼片	肉厚、易片	带刺多的鱼

文化口味偏好

根据用户的文化背景，理解他们可能的口味偏好：

背景	典型口味偏好	注意事项
中国大陆	美式甜品太甜、偏好鲜味和咸鲜、重视食材新鲜度	Hershey's 巧克力偏甜、American cakes 糖霜过重
北方人	偏咸、喜面食、口味重	南方菜可能觉得淡
南方人 (粤/闽)	偏清淡、重视食材原味、甜品偏清甜	川菜太辣、北方菜太咸
川渝	能吃辣、麻辣口味、重油	不辣的菜可能觉得没味道
台湾/港澳	口味介于大陆南北之间、接受度较广	对食材品质要求高
韩国	喜辣、发酵食品、烤肉文化	泡菜是刚需
日本	清淡、鲜味为主、注重摆盘	对新鲜度要求极高
东南亚	酸辣甜复合、椰奶、鱼露	香料是刚需
ABC/华裔	可能偏美式口味、也可能保留家庭传统	需要具体了解

美式产品避雷

中国用户常见的"踩坑"产品，可以主动提醒：

产品类型	问题	推荐替代
美式蛋糕 (Walmart/Kroger bakery)	糖霜太甜、奶油假	85°C、Paris Baguette、或自己做
Hershey's 巧克力	有酸味（butyric acid）、太甜	Lindt、Ghirardelli、或亚洲品牌
American cheese (Kraft singles)	口感像塑料、不是真芝士	真正的 cheddar、mozzarella
Wonderbread 白面包	太软太甜	亚洲面包店、或自己买吐司
美式香肠 (hot dog)	调味不同、太咸	亚超的台湾香肠、或自己买肉馅
Spam	太咸	梅林午餐肉 (亚超有)

Data Source Availability Status (Tested 2026-04-08)

数据源	状态	抓取方法	备注
Weee	✅ 可用	browser-use	成功抓取产品名称、价格、单位
Yami	✅ 可用	browser-use	已验证，价格/产品信息可抓取
Google Maps	✅ 可用	browser-use	成功获取店铺评分、评论、营业时间
Kroger	✅ 可用	官方API	已配置 OAuth2，支持产品搜索+店铺查询+价格
H Mart	⚠️ 部分可用	browser-use	主页可访问，搜索功能需进一步测试
Sam's Club	⚠️ CAPTCHA	cookies + browser-use	有 `.samsclub_cookies.json`，但 headless 被检测，需 human_approval_gate
Costco	❌ 被限制	需登录+cookie	headless browser 被 block
小红书	❌ IP blocked	需 Azure Asia	Error 300012 "IP at risk"，需中国 IP

Cookies 文件位置 (Staging/Production)

~/server/DoWhiz/DoWhiz_service/.samsclub_cookies.json  # Sam's Club 会员登录
~/server/DoWhiz/DoWhiz_service/.notion_cookies_liuxt.json  # Notion (测试用)

小红书访问方案 (TODO)

小红书对非中国 IP 有严格限制。解决方案：

方案 1: Azure Asia VM (推荐)

# 在 Azure East Asia / China North 创建小型 VM
# 专门用于抓取小红书
# 通过 API 调用或 SSH tunnel 访问

方案 2: 代理服务

使用中国代理 IP 池
或用户自己的 VPN 账号

方案 3: 用户自助

让用户自己搜索小红书，提供关键词
用户粘贴相关笔记链接，agent 解析

推荐抓取策略：

首选: Weee + Yami + Google Maps (直接 browser-use)
需 API: Kroger (grocery_cli kroger search)
需 cookies: Sam's Club (有 cookies，但需处理 CAPTCHA)
需特殊处理: 小红书 (Azure Asia VM 或用户自助)

UI Questionnaire Component

问卷 UI 组件位于 website/src/components/intake/GroceryPreferencesQuestionnaire.jsx

功能特性：

6步向导式流程
单选/多选按钮组
滑块控件（甜度、辣度）
拖拽排序（优先级）
响应式设计

使用方式：

import GroceryPreferencesQuestionnaire from './components/intake/GroceryPreferencesQuestionnaire';

<GroceryPreferencesQuestionnaire
  onComplete={(formData) => {
    // 保存到用户memory
    console.log('User preferences:', formData);
  }}
  onCancel={() => {
    // 取消问卷
  }}
  initialData={{}} // 可传入已有偏好数据
/>

Quick Start: Check User Preferences First

ALWAYS check user preferences before giving grocery recommendations:

# Get user's saved preferences (zip code, stores, dietary restrictions, etc.)
grocery_cli preferences get <user_id>

If preferences not found or incomplete, you have two options:

Ask in conversation - If you need critical info (like location), ask directly
Point to questionnaire - For comprehensive setup: https://dowhiz.com/grocery/onboarding

Use your judgment: if user's question can be answered with partial info, answer first and suggest completing preferences later.

Response Requirements

0. Price Accuracy (CRITICAL)

价格准确性是用户信任的基础。 错误的价格会让用户失去对我们的信任。

实时验证流程 (REQUIRED)

在给出价格推荐之前，必须使用 browser-use 实时验证价格：

# 1. 搜索产品
IN_DOCKER=true browser-use open "https://www.sayweee.com/en/search?keyword=orion+choco+pie"
browser-use state  # 获取搜索结果

# 2. 点击具体产品页面获取准确价格
browser-use click <product_index>
browser-use state  # 获取产品详情页的准确价格

# 3. 截图作为证据（可选但推荐）
browser-use screenshot /tmp/price_verification.png

验证要点：

不要只依赖 web search 返回的价格（可能是缓存/过期数据）
必须进入产品详情页确认当前价格
注意区分原价和促销价
检查是否有会员专属价格

查询时间戳 (REQUIRED)

所有价格推荐必须包含查询时间：

## 价格比较 (查询时间: 2026-04-08 13:30 EST)

| 渠道 | 价格 | 链接 |
|------|------|------|
| Weee | $4.49 | [购买](url) |

免责声明 (REQUIRED)

每次价格推荐的结尾必须添加免责声明：

---
⚠️ **价格说明**: 以上价格为查询时的实时价格，可能随时变动。最终价格以实际页面为准。
促销价可能需要满足特定条件（如会员、满减等）。

处理价格不确定性

如果无法实时验证价格（如实体店），明确标注数据来源和时效：

| 店铺 | 价格 | 数据来源 | 可信度 |
|------|------|---------|--------|
| 168 Asian Mart | ~$3.99/lb | 小红书@用户 (2026-03-15) | ⚠️ 可能有变化 |
| Weee | $4.49 | 实时查询 (2026-04-08) | ✅ 已验证 |

1. Purchase Links (REQUIRED)

Every product recommendation MUST include clickable purchase links:

渠道	链接获取方法	格式
Weee	browser-use 抓取搜索结果	`[产品名](https://www.sayweee.com/en/product/...)`
Yami	browser-use 抓取	`[产品名](https://www.yamibuy.com/en/p/...)`
Kroger	`grocery_cli kroger search` 返回的 URL	`[产品名](https://www.kroger.com/p/...)`
H Mart	搜索页链接	`[Search H Mart](https://www.hmart.com/search?q=...)`
实体店	Google Maps	`[店名](https://maps.google.com/?q=...)`

示例输出:

## 五花肉比价 (查询时间: 2026-04-08 13:30 EST)

| 渠道 | 价格 | 数据来源 | 购买链接 |
|------|------|---------|---------|
| [168 Asian Mart](https://maps.google.com/?q=168+Asian+Mart+Madison+Heights+MI) | ~$3.99/lb | 小红书 (3/15) | 实体店 |
| [Weee](https://www.sayweee.com/en/product/12345) | $5.49/lb | ✅ 实时验证 | [立即购买](https://www.sayweee.com/en/product/12345) |
| [Costco](https://maps.google.com/?q=Costco+Ann+Arbor) | $3.29/lb | 用户报价 (3/20) | 需会员 |

---
⚠️ **价格说明**: 以上价格为查询时的实时价格，可能随时变动。最终价格以实际页面为准。

2. Questionnaire Link (回复结尾)

在回复结尾附上问卷链接，让用户选择是否完善偏好:

---
💡 完成偏好问卷可获得更精准推荐: https://dowhiz.com/grocery/onboarding

3. Preference Updates (Agent Judgment)

当用户在对话中透露偏好信息时，自主判断是否值得保存:

明确的偏好 (如 "我不吃猪肉") → 保存到 memory
一次性需求 (如 "今天想吃辣的") → 不需要保存
位置信息 (如 "我住安娜堡") → 保存到 memory

保存格式: 更新用户 memory 文件的 ## Grocery Preferences 部分

Conversational Preference Collection

如果用户没有完整偏好且你需要关键信息才能回答，可以在对话中自然地询问。

原则:

不要一次问太多问题
只问回答当前问题必需的信息
让用户选择是回答还是跳过

示例:

用户: 排骨哪里买便宜？
Agent: 为了给你推荐最近的店，请问你住在哪个城市/zip code？
       (或者直接告诉我你方便去的超市，我帮你比价)

Legacy: Detailed Questionnaire Reference

以下问题仅供参考，用于理解完整的偏好维度。实际收集通过 UI 问卷完成。

可收集的偏好维度:

位置: city, zip_code
交通: has_car, max_drive_minutes
购物方式: online/offline/both
会员: costco, sams_club, kroger_plus
常买品类: meat, seafood, vegetables, snacks, hotpot...
口味: sweetness_tolerance, spice_tolerance, american_sweets_opinion
饮食限制: vegetarian, no_pork, lactose_intolerant, gluten_free
预算: price_first, value_focused, quality_first

详见 website/src/components/intake/GroceryPreferencesQuestionnaire.jsx

Shopping Habits Reference

以下内容描述用户可能的购物习惯，用于理解上下文:

交通选项:

没有车，靠公共交通/走路
有车，15分钟内
有车，30分钟内
有车，1小时内也可以

User Preferences Schema

偏好数据由 UI 问卷收集并存储在后端。Agent 通过 grocery_cli preferences get <user_id> 读取。

Memory 文件格式示例:

## Grocery Preferences

### Profile
- background: chinese_mainland
- city: Ann Arbor, MI
- zip_code: 48109
- household_size: 2

### Transportation
- has_car: true
- max_drive_minutes: 30
- prefers: both

### Memberships
- costco: true
- sams_club: false
- kroger_plus: true

### Main Categories
- primary: [meat, vegetables, snacks, hotpot]
- secondary: [condiments, instant_noodles]

### Taste Profile
- sweetness_tolerance: 2  # 1-5，美式甜品太甜
- spice_tolerance: 4  # 能吃辣
- american_sweets: too_sweet

### Exclusions
- dietary: []  # 无饮食限制

### Budget & Priorities
- budget_mindset: value_focused  # price_first/value_focused/quality_first
- priority_order: [price, quality, convenience, variety]

### Known Baselines (user-reported prices)
| Item | Store | Price | Unit |
|------|-------|-------|------|
| 五花肉 | 168 Asian Mart | $3.99 | /lb |
| 老干妈 | 168 Asian Mart | $3.29 | 280g |

Data Sources (Priority Order)

Tier 1: API/Automated (Most Reliable)

Kroger API (via grocery_cli)

IMPORTANT: Use grocery_cli for Kroger queries - it handles OAuth automatically.

# Search for products (with prices at nearest store)
grocery_cli kroger search "pork belly" --location 48109 --limit 5

# Find nearby Kroger stores
grocery_cli kroger stores 48109 --radius 15 --limit 5

# Get user's saved preferences
grocery_cli preferences get <user_id>

Example output:

Kroger Products for "pork belly":

1. Kroger® Fresh Natural Pork Belly Boneless
   Brand: Kroger
   Size: 1.5 lb   Price: $12.99

2. Smithfield Fresh Pork Belly
   Brand: Smithfield
   Size: 2 lb   Price: $15.49 (Sale: $12.99)

Available data: Product name, brand, price, sale price, size, store location

Weee (via browser-use)

# Weee doesn't have public API, use browser automation
IN_DOCKER=true browser-use open "https://www.sayweee.com/en/search?keyword=pork%20belly"
browser-use state  # Get product listings
browser-use screenshot /tmp/weee_search.png

Parsing Weee results: Look for price patterns like $X.XX and unit patterns like /lb, /ea, /pack

Tier 2: Browser Scraping (Medium Reliability)

H Mart Online

IN_DOCKER=true browser-use open "https://www.hmart.com/search?q=pork+belly"
browser-use state

Yami (Yamibuy)

IN_DOCKER=true browser-use open "https://www.yamibuy.com/en/search?keywords=pork+belly"
browser-use state

Costco (Requires login)

# Check if user has saved Costco cookies
if [ -f ~/.costco_cookies.json ]; then
  IN_DOCKER=true browser-use cookies import ~/.costco_cookies.json
fi
IN_DOCKER=true browser-use open "https://www.costco.com/CatalogSearch?keyword=pork+belly"
browser-use state

Tier 3: Social Media & Reviews Mining (For Local Asian Markets)

For stores without online presence (168 Asian Mart, Way1/Huaxing, local markets), use social media and review platforms to gather price intelligence:

Google Maps Reviews

# Search for store and read reviews mentioning prices
IN_DOCKER=true browser-use open "https://www.google.com/maps/search/168+Asian+Mart+Madison+Heights+MI"
browser-use state  # Find the store listing
browser-use click <store_listing_index>  # Click to open details
browser-use scroll down  # Scroll to reviews
browser-use state  # Look for reviews mentioning prices

# Search pattern in reviews:
# - "prices are good/cheap/expensive"
# - "$X.XX" patterns
# - "cheaper than H Mart"
# - Comparisons with other stores

What to extract from Google Maps:

Overall price sentiment (cheap/moderate/expensive)
Specific price mentions in reviews
Comparisons with other stores
Recent review dates (prioritize recent ones)

小红书 (Xiaohongshu) Price Discovery

⚠️ CURRENT STATUS: Direct scraping blocked

Xiaohongshu blocks datacenter IPs (Azure, AWS, etc.) with error 300012 "IP存在风险"
Fallback: Use web search (Google/Bing) to find Xiaohongshu content indexed by search engines
Alternative: Use Zhihu (知乎) or other review platforms for similar user-generated content

# BLOCKED - Direct access no longer works from cloud IPs
# IN_DOCKER=true browser-use open "https://www.xiaohongshu.com/search_result?keyword=168亚洲超市+密歇根+价格"

# ALTERNATIVE: Use web search to find indexed Xiaohongshu content
# Search engines have limited Xiaohongshu indexing, but some content is available

# Alternative searches:
# "密歇根 华人超市 价格"
# "Ann Arbor 买菜 攻略"
# "168超市 什么便宜"
# "美国亚超 比价"

小红书搜索关键词模板：

场景	关键词
特定商品	`{城市} {商品名} 价格`
超市攻略	`{城市} 华人超市攻略`
比价笔记	`{超市名} vs {超市名}`
省钱技巧	`美国买菜省钱`
产品评价	`{超市名} {商品名} 好吃吗`
踩坑避雷	`{超市名} 避雷不要买`
推荐必买	`{超市名} 必买清单`

小红书内容解析要点 - 不只是价格！

维度	要提取的信息	示例
价格	具体价格、单位、是否促销	"$3.99/lb, 特价"
分量/规格	包装大小、几人份	"一盒够4个人吃火锅"
品质评价	新鲜度、口感、质量	"很新鲜"、"肉质紧实"
加工难度	是否需要处理、骨头多少	"骨头很大不好切"、"已经切好了"
口味描述	甜/咸/辣、是否正宗	"偏甜"、"和国内味道一样"
性价比	相比其他店如何	"比H Mart便宜但分量小"
适合人群	适合谁、不适合谁	"适合北方人口味"、"ABC可能觉得太咸"
踩坑提醒	需要注意的问题	"保质期短"、"需要提前解冻"

小红书"问一问"AI功能：

# 小红书有AI问答功能，可以直接提问
IN_DOCKER=true browser-use open "https://www.xiaohongshu.com/search_result?keyword=山姆+五花肉"
browser-use state
# 找到"问一问"入口或搜索结果中的AI总结
# 可以直接问："山姆的五花肉怎么样？骨头多吗？"

产品评价综合报告示例：

## 山姆 五花肉 详细评价

### 基本信息
- 价格: $3.29/lb (会员价)
- 规格: 整块约5-6lb，需要自己切
- 产地: 美国

### 用户评价摘要 (来自小红书 12篇笔记)
- **优点**: 
  - 价格便宜，性价比高
  - 肉质新鲜，脂肪层分明
- **缺点**:
  - 骨头比较大，需要自己处理
  - 整块太大，1-2人家庭吃不完
  - 需要自己切片，不如168切好的方便
  
### 适合人群
- ✅ 3人以上家庭
- ✅ 会做饭、有切肉工具的人
- ✅ 喜欢批量买然后冷冻的人
- ❌ 1-2人小家庭（分量太大）
- ❌ 不想自己处理的人

### 加工建议
- 买回来先切成小块分装冷冻
- 或者让山姆肉柜帮忙切（部分店提供）

### 综合推荐
如果你是3人以上家庭，有时间处理，山姆的五花肉性价比最高。
如果你是1-2人，建议去168买已切好的，虽然贵$0.70/lb但省事很多。

WeChat Articles & Groups (微信公众号/群聊截图)

# Search for WeChat articles about local grocery shopping
IN_DOCKER=true browser-use open "https://www.google.com/search?q=site:mp.weixin.qq.com+密歇根+华人超市+价格"
browser-use state

微信渠道信息来源：

当地华人公众号的超市攻略
留学生群里分享的价格截图
团购群的拼单信息

Reddit & Local Forums

# Search Reddit for local grocery discussions
IN_DOCKER=true browser-use open "https://www.reddit.com/r/AnnArbor/search/?q=asian+grocery+prices"
browser-use state

# Alternative: Google search with site filter
IN_DOCKER=true browser-use open "https://www.google.com/search?q=site:reddit.com+Ann+Arbor+168+Asian+Mart+prices"

Yelp Reviews

IN_DOCKER=true browser-use open "https://www.yelp.com/biz/168-asian-mart-madison-heights"
browser-use scroll down  # Scroll to reviews
browser-use state

Tier 4: Web Search (Supplementary Channel)

When direct platform access is blocked or data sources are insufficient, use web search as a supplementary channel:

# General product reviews and comparisons
# Use search queries like:
# - "山姆 排骨 评价" (Sam's Club ribs reviews)
# - "Costco meat quality reviews 2026"
# - "[product] 测评 知乎" (product review on Zhihu)
# - "site:zhihu.com [product] 推荐" (search Zhihu specifically)

Best search queries by intent:

Intent	Query Pattern	Example
Product quality	`{store} {product} 评价/review`	"山姆五花肉评价"
Price comparison	`{product} 价格对比`	"Costco vs Sam's 排骨价格"
User recommendations	`{store} 必买/值得买`	"山姆必买清单肉类"
Avoid list	`{store} 不要买/避雷`	"Costco 肉类避雷"
Cooking tips	`{product} 怎么做/食谱`	"美国超市排骨红烧"

Useful sources accessible via web search:

知乎 (Zhihu): site:zhihu.com - Chinese Q&A platform with detailed product discussions
Reddit: site:reddit.com - English-language discussions on r/Costco, r/samsclub, local city subreddits
Food blogs: General queries often surface recipe sites with ingredient recommendations
腾讯新闻/网易: Aggregated content sometimes includes 小红书 summaries

Note: Web search provides indexed snapshots - content may be outdated. Prioritize recent results (check dates).

Tier 5: User-Reported Data (Baseline Prices)

For stores where social media mining doesn't yield results:

Check user's baseline prices in their memory file
Ask user to report if data is stale (>7 days)
Offer to help track via receipt scanning (future feature)

Social Media Price Discovery Workflow

When a user asks about a store without online pricing (like 168 Asian Mart), follow this discovery workflow:

Step 1: Quick Google Maps Check

# Fast sentiment check from reviews
IN_DOCKER=true browser-use open "https://www.google.com/maps/search/{store_name}+{city}+{state}"
browser-use state
# Click on store → scroll to reviews → extract price mentions

Step 2: Web Search for Reviews (Fallback for 小红书)

# Direct 小红书 access is BLOCKED from cloud IPs
# Use web search to find indexed content instead:

# Search Zhihu (知乎) - often has similar user-generated content
# Search pattern: site:zhihu.com {store_name} {product} 评价

# Or general web search for Chinese grocery reviews:
# Query: "{store_name} {city} 评价 买菜"

Parsing 小红书 Results:

Look for posts with 价格/价签/账单 in title
Check images for price tag photos
Note the post date (ignore posts >6 months old)
Read comments for updated prices

Step 3: Synthesize Price Intelligence

After gathering data, create a summary:

## Price Intelligence: 168 Asian Mart (Madison Heights, MI)

### Data Sources
- Google Maps: 47 reviews mentioning prices (avg sentiment: "affordable")
- 小红书: 12 relevant posts (most recent: 2026-03-15)
- Yelp: 23 price mentions

### Price Findings
| Item | Reported Price | Source | Date |
|------|---------------|--------|------|
| 五花肉 | $3.99/lb | 小红书@密歇根吃货 | 2026-03-15 |
| 老干妈 | $3.29 | Google review | 2026-02-20 |
| 整体印象 | "比H Mart便宜20-30%" | Multiple sources | - |

### Confidence Level
- High: 多个来源一致的价格
- Medium: 单一来源但近期
- Low: 老旧数据或模糊描述

### Recommendation
基于社媒数据，168的价格整体比H Mart便宜，尤其是中国特色商品。
建议将此信息保存到你的memory作为baseline。

Step 4: Update User Memory

If useful price data is found, offer to save to user's memory:

我从小红书和Google Maps找到了168的一些价格信息：
- 五花肉 ~$3.99/lb
- 老干妈 ~$3.29
- 整体比H Mart便宜20-30%

要我帮你保存到memory作为baseline吗？下次比价可以直接用。

Comparison Workflow

Step 1: Understand the Request

Parse user query to extract:

Product: What they want to buy (e.g., "五花肉", "pork belly", "老干妈")
Quantity: How much (optional, affects bulk store recommendations)
Urgency: Need it today vs can wait for delivery

Example queries:

"帮我比一下五花肉哪里便宜"
"Where should I buy pork belly?"
"Weee上的旺旺和168比哪个划算"

Step 2: Gather Price Data

# Pseudocode for data gathering
async def gather_prices(product: str, user_prefs: dict):
    results = []
    
    # 1. Check user's baseline data first
    baseline = check_user_baselines(product, user_prefs)
    if baseline:
        results.extend(baseline)
    
    # 2. Query online sources
    if "Weee" in user_prefs.preferred_stores:
        weee_price = await scrape_weee(product)
        results.append(weee_price)
    
    if "Kroger" in user_prefs.preferred_stores:
        kroger_price = await query_kroger_api(product, user_prefs.zip_code)
        results.append(kroger_price)
    
    # 3. Browser scrape other online stores
    for store in ["H Mart", "Yami"]:
        if store in user_prefs.preferred_stores:
            price = await browser_scrape(store, product)
            results.append(price)
    
    return results

Step 3: Normalize and Compare

Unit normalization is critical:

Raw	Normalized
$3.99/lb	$8.80/kg
$4.99/500g	$9.98/kg
$12.99/pack (2lb)	$6.50/lb = $14.33/kg

def normalize_price(price: float, unit: str) -> float:
    """Convert to $/kg for comparison"""
    conversions = {
        "lb": 2.20462,  # 1kg = 2.20462 lb
        "oz": 35.274,   # 1kg = 35.274 oz
        "g": 1000,      # 1kg = 1000g
        "kg": 1,
    }
    # Parse unit and convert
    ...

Step 4: Apply User Preferences

Scoring formula:

score = base_price_score 
      + taste_penalty      # -100 if American-style and user avoids
      + distance_penalty   # -10 per 10 min drive over threshold
      + delivery_bonus     # +20 if delivers to user's zip
      + freshness_bonus    # +15 for fresh vs frozen

Step 5: Generate Recommendation

Output format:

## 五花肉 (Pork Belly) 比价结果

### 推荐: 168 Asian Mart
- 价格: $3.99/lb ($8.80/kg)
- 距离: 15分钟车程
- 优点: 最便宜，品质好，新鲜
- 缺点: 需要开车

### 其他选项:

| 渠道 | 价格 | 单价 | 配送 | 备注 |
|------|------|------|------|------|
| Weee | $5.49/lb | $12.10/kg | 免费送货 | 方便但贵$1.50/lb |
| H Mart | $4.49/lb | $9.90/kg | 需自取 | 20分钟车程 |
| Costco | $3.29/lb | $7.25/kg | 需自取 | 需买整块5lb+，会员价 |

### 建议
考虑到你住在Ann Arbor，168开车15分钟可以接受。如果你这周不想出门，Weee虽然贵$1.50/lb但免费送货，5lb以内差价不到$8，可能值得。

如果你要买大量(>5lb)，Costco最划算，但需要会员卡。

Common Products (Chinese/English Mapping)

中文	English	Category	Typical Stores
五花肉	Pork Belly	Meat	168, H Mart, Costco, Weee
猪蹄	Pork Feet/Trotters	Meat	168, H Mart (rare at Costco)
老干妈	Lao Gan Ma Chili	Condiment	168, Weee, Yami, H Mart
旺旺仙贝	Want Want Rice Crackers	Snack	Weee, Yami, 168
康师傅方便面	Master Kong Instant Noodles	Noodles	Weee, 168, Yami
王老吉	Wong Lo Kat Herbal Tea	Beverage	168, Weee, H Mart
豆腐	Tofu	Produce	All stores
韭菜	Chinese Chives	Produce	168, H Mart, 99 Ranch
火锅底料	Hot Pot Base	Condiment	Weee, 168, H Mart

Store Profiles

Online Delivery

Store	Delivery	Min Order	Coverage	Specialty
Weee	Free >$35	$35	Major metros	Chinese + Pan-Asian groceries
Yami	$5.99 or free >$49	None	Nationwide	Asian snacks, beauty
Instacart (Kroger)	$3.99+	$10	Nationwide	Mainstream groceries
Costco Delivery	Varies	$35	Members only	Bulk items

Physical Stores (Michigan Ann Arbor Area)

Store	Address	Drive Time	Best For
168 Asian Mart	Madison Heights	~25 min	Chinese groceries, best prices
H Mart	Troy	~30 min	Korean + general Asian
Kroger	Multiple	5-15 min	American groceries, good sales
Costco	Ann Arbor	15 min	Bulk buying
Sam's Club	Ypsilanti	20 min	Bulk buying
Aldi	Multiple	5-10 min	Budget European goods

Handling Edge Cases

Product not found online

我在Weee上没找到这个产品。根据你之前报告的价格，168有卖$X.XX。
你要我帮你记录一下其他店的价格吗？下次可以直接比较。

Price data is stale

你上次记录168的价格是2周前了。这些价格可能有变化：
- 五花肉 $3.99/lb (记录于 2026-03-18)

你最近去过168吗？可以帮我更新一下价格。

User asks about non-Asian product

鸡蛋在这些店的价格大概是：
- Costco: $X.XX/18个 (需会员)
- Aldi: $X.XX/12个 (最便宜非会员选项)
- Kroger: $X.XX/12个 (查看本周flyer可能有折扣)

注意：亚洲超市的鸡蛋通常比Costco/Aldi贵，除非你需要特定品种（如皮蛋、咸蛋）。

Weekly Deals Integration

Parsing Promotional Emails

If user forwards promotional emails to DoWhiz:

# Extract deals from email body
def parse_promo_email(email_body: str, store: str):
    # Look for patterns like:
    # "Pork Belly $2.99/lb (reg $3.99)"
    # "五花肉 特价 $2.99/磅"
    deals = extract_deals(email_body)
    
    # Store in user's deal tracker
    for deal in deals:
        save_deal(user_id, store, deal.product, deal.sale_price, deal.valid_until)

Proactive Recommendations

If user has subscribed to weekly recommendations:

## 本周值得买 (Week of 2026-04-01)

基于你的偏好和各店促销:

1. **五花肉** - 168特价 $2.99/lb (原价$3.99，省25%)
   - 比Weee便宜$2.50/lb
   - 建议：多买点冻起来

2. **旺旺仙贝** - Weee买二送一
   - 相当于$3.33/包，比168便宜
   
3. **老干妈** - Yami有8折码 SPRING20
   - 但算上运费不如168划算，除非你要买其他东西凑$49免邮

Receipt OCR Integration (Future)

When receipt scanning is available:

# User uploads receipt image
receipt_ocr --image /tmp/receipt.jpg --output json > /tmp/receipt_data.json

# Parse and update price database
update_prices_from_receipt /tmp/receipt_data.json --store "168 Asian Mart"

Search Query Templates by Scenario

Scenario 1: Unknown Store Price Level

Goal: Understand if a store is generally cheap or expensive

Google Maps: "{store_name} {city}" → read reviews for price sentiment
小红书: "{store_name} 价格 贵不贵"
Reddit: "site:reddit.com {city} {store_name} prices cheap expensive"

Scenario 2: Specific Product Price

Goal: Find what a specific item costs at a store

小红书: "{store_name} {product_name_zh} 价格" (e.g., "168超市 五花肉 价格")
Google: "{store_name} {product_name} price reddit OR yelp"

Scenario 3: Store Comparison

Goal: Compare two stores

小红书: "{store_A} vs {store_B}" or "{store_A} {store_B} 比较"
Reddit: "{city} {store_A} vs {store_B}"
Google: "site:reddit.com OR site:yelp.com {store_A} cheaper than {store_B}"

Scenario 4: Best Store for a Category

Goal: Find which store is best for meat/produce/snacks

小红书: "{city} 买肉 哪家便宜" or "{city} 亚洲零食 推荐"
Reddit: "{city} best asian grocery for {category}"

Scenario 5: Current Deals/Sales

Goal: Find current promotions

小红书: "{store_name} 打折 {current_month}" (e.g., "Weee 打折 四月")
Store's Instagram/Facebook (if they have one)
Local WeChat groups (ask user if they're in any)

Data Freshness Guidelines

Source	Typical Freshness	Trust Level
小红书 posts <1 month	High	⭐⭐⭐⭐
小红书 posts 1-6 months	Medium	⭐⭐⭐
Google reviews <3 months	High	⭐⭐⭐⭐
Google reviews 3-12 months	Medium	⭐⭐⭐
Reddit posts <6 months	Medium-High	⭐⭐⭐
Yelp reviews	Variable	⭐⭐
User's own baseline	Depends on date	⭐⭐⭐⭐⭐

Rule of thumb:

Prices can change 5-15% over 6 months
Sale items can change weekly
Always note the date when reporting social media prices

Handling No Results

If social media search yields no useful price data:

我在小红书和Google Maps上没找到{store_name}的具体价格信息。

可能的原因：
1. 这家店比较小众，网上讨论少
2. 搜索关键词可能需要调整

建议：
1. 你下次去{store_name}的时候可以拍几张价签，我帮你记录
2. 或者你知道大概价格的话可以告诉我，我保存到你的memory
3. 我可以先用H Mart/Weee的价格作为参考（通常本地小店会便宜10-20%）

需要我用哪种方式帮你？

Tips for Effective Use

First time setup: Ask user for zip code and store preferences
Build baseline: Encourage user to share prices they see in stores
Track over time: Prices fluctuate; historical data helps identify deals
Consider total trip: Factor in gas cost for distant stores
Batch recommendations: "If you're going to 168 anyway, also grab X, Y, Z"
Social media first: For local Asian markets, always check 小红书 first - it's the best source for Chinese grocery price info
Save discoveries: When social media yields good price data, offer to save to user's memory

Example Conversations

Basic Price Check

User: 老干妈在哪买便宜
Agent: [Checks Weee, user's 168 baseline, Yami]

老干妈 (280g装) 比价：
- 168 Asian Mart: $3.29 (你上次报告的价格)
- Weee: $3.99 
- Yami: $4.49

建议去168买，便宜$0.70。如果你这周有其他东西要在Weee买，可以顺便加购，差价不大。

Complex Comparison

User: 我想买火锅食材，比较一下168和Weee

Agent: [Comprehensive comparison]

火锅食材比较 (168 vs Weee):

| 食材 | 168 | Weee | 推荐 |
|------|-----|------|------|
| 肥牛卷 | $8.99/lb | $9.49/lb | 168 |
| 羊肉卷 | $12.99/lb | $11.99/lb | Weee |
| 豆腐 | $1.99 | $2.49 | 168 |
| 金针菇 | $1.79 | $1.99 | 168 |
| 海底捞底料 | $5.99 | $5.49 | Weee |

总计 (估算一顿4人份):
- 168: ~$35 + 开车25分钟
- Weee: ~$38 + 免费配送

建议：如果你有时间去168，能省$3+。如果懒得出门，Weee贵不了多少，而且不用自己搬。

Cleanup

Always close browser sessions after scraping:

browser-use close

name	grocery-comparison
description	Smart grocery price comparison agent for Asian international students. Compares prices across Weee, Asian markets (H Mart, 168, etc.), and mainstream stores (Kroger, Costco, Sam's Club, Aldi). Considers user taste preferences, distance, and cultural factors. Use when user asks to compare grocery prices, find best deals, or get shopping recommendations.