从 AdMapix API 拉取广告创意、应用、榜单和收入预估等数据,原样返回结构化 JSON。
数据分析
Meituan
试用根据美团可见的配送费、包装费、满减门槛、ETA、评价等公开信息,给出一个明确的下单、换店、凑单、再等等或不点的判断。
它能做什么
只看美团页面或截图里能看见的公开信息——商家主页、购物车明细、可见的优惠文案——然后给出一个明确动作:现在下单这家、换一家、加一样真正用得上的东西凑满减、再等等,或者直接划掉。判断时会拆开算配送费、包装费、门槛差额、ETA、距离、评价信号、退款摩擦,看表面折扣扣完费用和凑单品之后还省多少。最终实付价、按地址算的 ETA、账户券是否可用,这些都明确标出来让用户自己核对。不登录、不读账户、不动购物车、不结账、不付款。
什么时候用它
- 两家附近午餐店之间二选一
- 要不要顺手加一份小菜凑满减
- 判断一张团购券有没有隐藏门槛
- 读截图里的优惠信息,看值不值得下单
技能文档
Meituan (美团)
Use this skill when the user needs a decision about a Meituan food-delivery or local-life deal, especially when the question is shaped by delivery time, distance, minimum order, packaging fee, visible discounts, merchant reviews, refund practicality, or whether a promotion is actually worth chasing.
This is a public decision-support skill. It should produce an action, not a menu recap.
Safety Contract
These rules override every workflow below.
- Use only public visible information or details supplied by the user, such as screenshots, copied cart lines, public merchant pages, or visible offer text.
- Do not log in, read order history, read account coupons, claim red packets, save cookies, change addresses, change cart contents, submit orders, or pay.
- Stop at login, CAPTCHA, identity checks, address selection, coupon wallets, order pages, checkout, or payment screens.
- If final payable price depends on address, account coupons, membership, payment method, inventory, or checkout state, mark it as user-only verification.
- Redact personal data if it appears unexpectedly, including phone numbers, addresses, names, order IDs, and payment details.
- Do not instruct the user to bypass platform rules, obtain unauthorized discounts, or evade anti-abuse systems.
When To Use
Use this skill for:
- choosing among Meituan merchants for a meal or local-life purchase
- deciding whether a full-reduction threshold is worth crossing
- checking whether a visible group-buy or coupon offer has hidden friction
- weighing cheaper price against slower delivery or weaker merchant trust
- translating screenshots or copied cart details into a clear ordering decision
Prefer another skill when:
- the user wants broad cross-platform shopping comparison rather than a Meituan-specific decision
- the user wants account/order troubleshooting, refund escalation, or private coupon lookup
- the task requires authenticated account access or irreversible platform actions
Input Discipline
Capture the few inputs that change the call:
- category or exact item
- candidate merchants or offer screenshots
- urgency, such as hungry now, lunch break, dinner planning, appointment, or no rush
- budget ceiling and minimum acceptable quality
- visible subtotal, delivery fee, packaging fee, threshold discount, ETA, distance, rating, and review cues
Ask at most one short follow-up when missing context would flip the recommendation. If the user wants speed, assume ETA matters more. If the user wants lowest cost, still protect them from fake-cheap fee traps.
Workflow
1. Classify The Decision
Choose the active mode:
- merchant compare: pick the best store among visible alternatives
- threshold check: decide whether to add an item for a discount
- deal sanity check: judge a group-buy, coupon, or local-life offer
- risk triage: decide whether a weak merchant is still acceptable
- screenshot readout: extract only visible facts, then recommend what to verify
2. Gather Visible Evidence
For each candidate, capture what is visible:
- merchant name and category fit
- rating, sales cue, recent review signals, and photo/title consistency
- distance, ETA, delivery mode, and deadline fit
- item subtotal, minimum order gap, delivery fee, packaging fee, service fee, and visible threshold discount
- add-on item usefulness if a threshold is involved
- repeated review risks such as delay, wrong item, small portion, hygiene, stale food, spills, or refund friction
Re-snapshot or ask the user for updated visible details after changing location, filters, merchant, selected items, coupon view, or delivery method.
3. Compute Checkout Reality
Do not judge by headline discount alone.
Compare:
- natural basket subtotal before artificial add-ons
- fee stack: delivery, packaging, service, tableware, or pickup friction when visible
- threshold gap and whether the add-on is useful
- net saving after the useful add-on and fee stack
- ETA difference and the user's urgency
- merchant trust difference and downside if the order disappoints
Use this language when appropriate:
这是门槛价,不是自然到手价。便宜是便宜,但得靠凑单。看着省,配送费和包装费把优惠吃掉了。今天这单时间比省几块钱更值钱。
4. Apply Decision Rules
- If two options differ by only a small amount, prefer faster delivery and stronger merchant trust.
- If one option is more than 15-20 minutes slower and the user is time-constrained, treat the discount as weak unless the saving is material.
- Add an item for a threshold only when the item is useful and the net saving remains positive after fees.
- For low-value solo meals, moderate merchant risk can be acceptable if the user is not deadline-sensitive.
- For shared meals, expensive orders, hygiene-sensitive food, gifts, appointments, or work breaks, weak trust is a strong reason to switch.
- A suspiciously deep discount with weak reviews, mismatch photos, or many complaint patterns should lose unless the user explicitly accepts the risk.
- If live price or ETA is missing, give a directional call and list the exact visible fields needed to confirm it.
5. Recommend One Move
End with one clear action:
- order this merchant now
- switch to another merchant
- add one specific useful item to cross the threshold
- do not add anything just for the promotion
- wait or search again
- skip this store
Do not end with 都可以 unless the tradeoff is genuinely flat; even then, choose a default based on the user's stated priority.
Output
Use this structure unless the user asks for something shorter:
Recommended Move
Say the action in one sentence, including the winning merchant or offer when known.
Checkout Reality
Show the real tradeoff: visible subtotal, fees, threshold gap, discount, useful add-on logic, ETA, and whether the headline saving survives.
Risk Check
Call out merchant trust, review patterns, refund friction, hygiene, delay, mismatch, or deadline risk.
Confidence And Gaps
State confidence as high, medium, or low based on visible evidence. Name the missing fields that could change the call.
Before You Order
List user-only checks: final payable amount, address-based ETA, account coupon eligibility, item options, stock, refund or after-sales rule, delivery note, and payment.
Example Prompts
午饭 30 分钟内要吃到,黄焖鸡和麻辣烫两家怎么选?这家满 35 减 12,但配送费 7 元,要不要为了满减再加一个小菜?这家便宜 6 元但晚 25 分钟,今天赶会,值得吗?帮我比较这三个美团商家,只给我该点哪家和下单前核对项。这家评价有点一般,但离我近,适合现在点吗?截图里这个团购券看起来便宜,帮我判断有没有隐藏门槛。
Package Surface
This published package is intentionally Markdown-only. It should not ship browser automation, login helpers, cookie storage, order-history code, account-coupon code, or payment tooling. If future versions add tools, they must preserve the Safety Contract and make any account-state capability explicit, consent-based, and separate from this public default.
Quality Bar
Do:
- optimize for the user's immediate decision
- separate headline discount from real cost
- treat ETA, merchant trust, and refund practicality as first-class decision inputs
- make uncertainty visible instead of inventing live price, coupon, stock, or ETA data
- tell the user exactly what to verify manually before ordering
Do not:
- log in or read account pages
- claim account coupons or final payable price are known unless the user supplies visible evidence
- save cookies, orders, addresses, phone numbers, or account data
- mutate cart state, submit an order, confirm an order, or pay
- recommend threshold add-ons that the user does not actually need
常见问题
- 会登录我的美团账号吗?
- 不会。只使用公开可见的信息,或用户主动提供的截图、复制下来的购物车明细等。遇到登录页、验证码、领券页、结账页、付款页会直接停住。
- 能告诉我最终实付多少钱吗?
- 不能。最终实付价取决于收货地址、账户券、会员、支付方式、库存和结算状态,所以会明确标记为需要用户自己核对的项。
- 会帮我下单或付款吗?
- 不会。只会给到一个推荐动作和下单前需要核对的字段,不会改动购物车、提交订单或完成支付。
相关技能
以 AI 机器人身份加入视频会议,提供语音、虚拟形象与屏幕共享四种模式。
把自然语言描述转为结构化 JSON,并由 mcp-diagram-generator MCP 服务生成 Draw.io、Mermaid 或 Excalidraw 图表文件。
在本地磁盘以分类纯 Markdown 文件保存需要长期留存的事实,与智能体内置记忆并存。
诊断生产力系统反复失效的根因,给出最小干预——容量测算、瓶颈定位、可靠的本地记录。
编写、调试与调优 Playwright 测试,涵盖定位器策略、追踪诊断与 CI 友好的超时配置。
harrylabsj 的更多技能
浏览全部技能淘宝与京东、拼多多、唯品会同款比价,输出可执行的购买建议。
在下单前拆解拼多多低价:百亿补贴、拼团门槛、店铺风险、退货条款一次过。
定位技能瓶颈,把练习日志转化为可执行的下一次训练方案。
用通俗中文解读常规体检和化验报告,标注异常、分清轻重、给出下一步动作。
为盒马鲜生购物规划清单、评估鲜度、推荐配送时段,并整理手动下单步骤。
为会话中的观察、规律和发现建立带标签、日期与生命周期的可检索记录。