跳到正文
cccskills
Claude 官方免费在 GitHub 公开

purchase-research

Teaching a customer how to choose before any candidates appear, when they ask what matters in a category, how two approaches to it differ, or where to begin. Not needed once the customer has named the thing they want, however considered the purchase (that is search work), nor for outfitting a space or an event (planning-goals).

中文简介正在准备中。当前显示原文说明。

查看安装方法

包含的文件(1)

  • SKILL.md3.8 KB

SKILL.md(原文)

安装前可以查看交给 AI 代理的指令内容。

Purchase research

Below, "category" means a kind of whatever this store sells: products, stays, plans, or seating.

The customer has a category in mind and no criteria yet. Teach the criteria from the store's own guide, then set what the store carries against them. The choice, and anything that touches the cart, stays with the customer.

Ask once or not at all

  • When the deciding facts are open (who it is for, how many, what for, a firm budget), spend one turn on two or three short questions in one message, each with its likely answers as chips; ask only what the profile you were given leaves open. This intake happens once.
  • Whatever comes back closes intake: answers, a request to get on with it, or a question of their own. Research on what you have and carry each still-missing fact into the answer as a stated assumption.
  • When the request already names the people, the purpose, and the limits, skip the questions and research.
  • The customer sees either the questions or the answer; do not describe the intake decision or the research work.

Where the criteria and the products come from

  • Open every research turn with one round of calls: search_policies for the store's buying guide on the category, search_products for what it carries, and one or two web_search calls on the category where that tool is registered. Follow with get_product_details on the few candidates the criteria single out.
  • Read the guide even for a familiar category; it is the advice the store puts its name to. Your own knowledge of the category stands in only when no guide comes back.
  • Use web material for the category side only (criteria, terminology, how the trade-off works), restated in your own words. It says nothing about any catalog product and is no grounds for naming a store, brand, or product outside this catalog.
  • List in sources what you retrieved, the store's guides and any web pages used, and nothing when neither came back. Leave out a criterion none of the retrieved guides or records support.

Shape of the answer

  • After an intake turn, answer whole: the criteria and the store's options against them in one turn.
  • For a broad ask with no intake, answer in two steps: present_guide with three to five sections of one criterion each and chips for saying which criteria weigh most, then the shortlist next turn, filtered by their choice. Apply a later steer to the same candidates without sending the criteria again.
  • For a narrow ask whose wording already fixes the top criterion or two, answer in one turn: a compact criteria section, then the options against those criteria, in present_comparison for two to four real candidates and present_products otherwise.
  • Show the store's real position. Present a category with one option as the one item it carries, with chips that name it; say when the guide's advice points at something the catalog lacks; report an out-of-stock candidate as a gap and recommend it to nobody.
  • Name the pick with the criterion that decided it, in one clause. Prose covers the recommendation, the assumption you made, and the one caveat that matters; the components carry the rest.

Scope

  • Research turns read and recommend. Write to the cart or to memory only when the customer asks, under the standing rules.
  • Where the category borders on health, safety, or a licensed profession, keep the criteria about which product to buy.
  • Add a purchase chip only once the customer has seen the criteria.

评价

还没有评价。欢迎分享你的使用感受。

评论

无需登录即可发表。欢迎分享使用感受、提问,或说明在哪些环境下无法运行。

还没有评论。来写第一条吧。

提交后会立即公开。请勿填写邮箱、电话等个人信息。URL 最多 2 个。

0 / 1000 字

此表单受 reCAPTCHA 保护,适用 Google 的隐私权政策和服务条款。

同一仓库的技能

catalog-listings

Claude 官方免费

Creating and improving listing content, covering titles, descriptions, attribute completeness, categorization fixes, image callouts written as text, edits written from material the operator supplies, and content-quality audits and bulk fixes across many listings. Not needed for questions about how a listing is performing.

中文简介正在准备中。当前显示原文说明。

anthropics/commerce-agents33002026年10月20日 更新

commerce-architecture

Claude 官方免费

How the reference commerce agents of either role are structured, covering the loop, where each rule lives, skills, the backend interface, delegates, and model fields. Load when designing or reviewing the structure of a shopping or merchant agent.

中文简介正在准备中。当前显示原文说明。

anthropics/commerce-agents33002026年10月20日 更新

commerce-evals

Claude 官方免费

Authoring and running behavioral evals for a shopping or merchant agent, covering the case shape, authoring rules, code graders and judges, the run pattern, and poisoned fixtures. Load when writing eval cases or rubrics or deciding how a suite runs.

中文简介正在准备中。当前显示原文说明。

anthropics/commerce-agents33002026年10月20日 更新

commerce-merchant-operations

Claude 官方免费

The reference merchant agent, covering its flows, staged changes and host approval, metrics grounding, the analysis delegate, store memory, components, and marketplace rules. Load when building or reviewing an agent for the operators of a store, property, subscriber base, or venue.

中文简介正在准备中。当前显示原文说明。

anthropics/commerce-agents33002026年10月20日 更新

commerce-prompt-caching

Claude 官方免费

The reference agents' cache-stable request assembly, covering the static system and per-request context split, the fixed tool list, the rolling conversation breakpoint, which config fields are prompt bytes, and verification. Load when writing or reviewing a commerce agent's prompt assembly, when cache reads are zero, or when a turn's latency or cost is the question.

中文简介正在准备中。当前显示原文说明。

anthropics/commerce-agents33002026年10月20日 更新

commerce-trust-safety

Claude 官方免费

The rules the reference agents enforce in code for third-party content, writes, grounding, identity, and memory, each with its module, plus two adversarial-eval rules. Load when handling untrusted tool results, guarding a write tool, or scoping what an agent remembers.

中文简介正在准备中。当前显示原文说明。

anthropics/commerce-agents33002026年10月20日 更新

查看 anthropics 的全部技能

报告此技能的问题