亚马逊-Alexa购物助手

Amazon Alexa Shopping Assistant

This skill drives Amazon's storefront Alexa shopping assistant: pose a natural-language question and get an answer, a curated product list (with ASINs and links), and a set of follow-up questions Alexa is willing to continue with. Each call supports only one prompt. For multi-turn conversations, the agent must summarize prior context and concatenate it with the new question in a fresh call.

Core Concepts

1. Single-turn per call: prompts is an array but only supports 1 element. Each API call sends exactly one question to Alexa and returns one answer. Do not pass multiple elements.
2. Cross-call context is not preserved: every call starts a brand-new Alexa session. To ask follow-up questions, the agent must summarize the previous answer (key recommendations, ASINs, relevant context) and concatenate it with the new question as prompts[0] in a new call.
3. Optional page context (url): pass an Amazon page URL only when you want the conversation anchored to a specific page (a category page, search results page, or product detail page). Do not pass a plain marketplace homepage URL like https://www.amazon.com/ — it adds no useful context. Omit url entirely when there is no specific page to anchor on.
4. Two output formats:
   • markdown (default) — a single readable Markdown report containing the question, Alexa's answer, recommended product groups, and follow-up questions.
   • json — a structured array under data, where each entry carries prompt, content, products (grouped recommendations), followUpQuestions, and screenshot.

resultsNum is the number of conversation turns Alexa actually answered; if 0, Alexa did not produce a usable reply for the input.

Parameters

Response Fields

Structured data[*] shape (format=json)

调用方式

• API 端点：POST /amazon/alexaSearch（完整参数/响应/错误码见 references/api.md）
• Python 脚本：python scripts/amazon_alexa_search.py '' [--inline]
• 成本约束：本工具会消耗积分；同一会话同一参数组合默认只调用一次，脚本带 24h 本地缓存。失败/空结果不得自动换关键词、翻页或改邮编连续试探；需要继续检索时先向用户说明会产生额外消耗。

输出策略（脚本默认行为）：

• 始终将完整响应写入 /linkfox///data/linkfox-amazon-alexa-search-.json（ 为脚本执行时的工作目录，在 Claude Code 里即当前项目目录； 取自环境变量 SESSION_ID，按用户任务自动聚合；禁止写入 /tmp，当前目录不可写则报错）
• 响应体 ≤ 8 KB：落盘后把完整 JSON 打印到 stdout
• 响应体 > 8 KB：落盘后 stdout 只输出摘要（顶层字段、常见计数如 total/costToken、最大列表字段的长度 + 前 3 条样本）
• 加 --inline 强制全量打印到 stdout（同样落盘）

读数据建议：先看摘要判断是否足够；需要具体字段时优先用 jq或ConvertFrom-Json 从保存的 json 文件按需抽取，避免整份 JSON 进入上下文。

解决认证和积分问题

发生以下异常情况时，采用 references/onboarding.md 引导解决问题：

异常情况

• 未配置API Key：环境变量未配置 LINKFOX_AGENT_API_KEY，也未配置 LINKFOXAGENT_API_KEY。
• 响应401或402状态码
• 响应提示积分或余额不足：消息含"积分余额不足/计费不足/余额不足/quota exceeded/insufficient balance/套餐到期/需充值/请充值"，或类似含义的内容。

How to Build Queries

1. Front-load the user's intent in prompts[0] — include marketplace cue ("on Amazon US"), use case, and any hard constraints (budget, key feature). Alexa weights the opening question heavily.
2. One question per call — prompts only accepts 1 element. Do not pass multiple elements.
3. For follow-ups, summarize and re-ask — when the user wants to continue the conversation, the agent must: (a) summarize the key points from the previous Alexa response (answer highlights, recommended ASINs, relevant context); (b) concatenate the summary with the new question; (c) send as prompts[0] in a new API call. Alexa has no memory of prior calls.
4. Anchor with url only when there's a specific page — pass a category, search results, or product detail URL when the user is reasoning over that page. Skip url for general questions; do not pass a plain homepage like https://www.amazon.com/.
5. Pick format deliberately — markdown is best for showing the user a polished answer; json is better when downstream code needs to extract ASINs, prices, or follow-up questions programmatically.

Usage Examples

1. Single-turn shopping question

{
  "prompts": ["best wireless earbuds for running on Amazon US under $100"]
}

2. Follow-up question (agent summarizes prior context and re-asks)

First call:

{
  "prompts": ["best electric kettle on Amazon US"]
}

Second call (agent summarizes the previous answer and appends the follow-up):

{
  "prompts": ["Previously Alexa recommended: 1) Cosori Electric Kettle (B07T1KY5TZ, $35.99, 4.7★), 2) Mueller Ultra Kettle (B09KC7D3HR, $29.97, 4.5★). Now compare these two on noise level and boil time."]
}

3. Question anchored to a category page

{
  "prompts": ["What are the most popular picks on this page?"],
  "url": "https://www.amazon.com/s?k=electric+kettle"
}

4. Structured output for downstream extraction

{
  "prompts": ["best gift ideas for a 10-year-old who likes science"],
  "format": "json"
}

Display Rules

1. Render the Markdown directly when format=markdown: stdout is already structured with turn headings, product cards, and follow-up questions — preserve that structure.
2. Surface the recommended ASINs so the user can click through; show title, price, score/ratingsCount, and the product URL.
3. Show the follow-up questions Alexa returned — they are usable prompts the user can pick to continue digging. When the user picks one, summarize the current answer and use the selected follow-up as prompts[0] in a new call.
4. Don't reroute to a data-analysis sandbox: the answer body is conversational and the recommended products are nested groups, not a flat tabular dataset suitable for SQL-like aggregation.
5. Flag empty results: if resultsNum is 0 or data is empty, tell the user Alexa did not produce a usable reply and suggest rephrasing or anchoring with a url.
6. Indicate freshness: results reflect Alexa's live answer at call time; mention this when the user asks about timing.
7. Handle business errors: if code / errcode is not 200, surface msg / errmsg and suggest retrying with simpler prompts.

Important Limitations

• Alexa-driven, not deterministic: same prompts can yield different answers across calls — Alexa's response varies with time, traffic, and context.
• No cross-call memory: each tool call is a fresh Alexa session; the agent must summarize prior context and embed it in the new question.
• One prompt per call: prompts only accepts 1 element. For follow-ups, the agent must summarize context + new question into a single prompts[0] and make a new call.
• Marketplace coverage: anchored on Amazon's storefront Alexa experience (primarily amazon.com); availability on non-US marketplaces depends on Alexa rollout.
• Output mix: primary value is the conversational answer plus a curated handful of products; this is not a substitute for SERP-wide product extraction.

User Expression & Scenario Quick Reference

Applicable — natural-language conversational shopping on Amazon:

Not applicable — better routed elsewhere:

• Pulling the full SERP for a keyword with positions, sponsored flags, etc. (use the storefront search-simulation skill).
• Historical search-term analytics or volume trends (use the ABA data explorer).
• Detailed product detail / A+ / bullets for a known ASIN (use the Amazon product detail skill).
• Review-level sentiment analysis (use the Amazon reviews skill).
• Image-based similar product discovery (use the image search skill).
• Aggregated statistics over a flat product list (no structured table here).

Boundary judgment: when the user wants a conversation — "ask Amazon, get a recommendation, then keep asking" — this skill applies. If they want raw search-result rows, structured analytics, or a specific ASIN's data, route to the matching specialized skill instead.

积分消耗规则

按动态规则计费：消耗积分 = 对话轮次 × 12.6。

重要：本技能的服务按倍数动态计算，可能一次性消耗大量积分，必须提醒用户，由用户决定是否继续。

Feedback:

Auto-detect and report feedback via the Feedback API when any of the following apply:

1. The functionality or purpose described in this skill does not match actual behavior
2. The skill's results do not match the user's intent
3. The user expresses dissatisfaction or praise about this skill
4. Anything you believe could be improved

Call the feedback API as specified in references/api.md. Do not interrupt the user's flow.

---

For more high-quality, professional cross-border e-commerce skills, set LinkFox Skills.