linkfox-amazon-alexa-for-shopping

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

Parameter	Type	Required	Description	Default
prompts	string[]	Yes	Conversation prompts. Only 1 element is allowed per call. To ask follow-up questions, make a new call with context summary + new question as prompts[0].	-
format	string	No	Response format: markdown returns a readable report; json returns a structured array.	markdown
url	string	No	Specific Amazon page URL (category, search results, or product detail) to anchor the conversation. Skip when there is no specific page; do not pass a plain homepage URL such as https://www.amazon.com/.	-

Response Fields

Field	Type	Description
stdout	string	Markdown report when format=markdown: per-turn question, Alexa answer, recommended product groups, follow-up questions
data	array	Structured turns when format=json. Each item has prompt, content, products[], followUpQuestions[], screenshot
resultsNum	integer	Number of answered turns (0 = Alexa did not respond)
code / errcode	string / integer	200 on success; non-200 indicates a business error
msg / errmsg	string	ok on success; otherwise an error description
costTime	integer	API latency in milliseconds
costToken	integer	Tokens consumed (only billed on success)
taskId	string	Upstream task identifier for tracing
type	string	Render hint: stdoutWorkbenches for markdown, json for json

Structured data[*] shape (format=json)

Field	Type	Description
prompt	string	The question or follow-up sent for this turn
content	string	Alexa's natural-language answer
products[].title	string	Group title (e.g. "Top picks", "Best for running")
products[].items[].asin	string	Product ASIN
products[].items[].title	string	Product title
products[].items[].url	string	Product detail page URL
products[].items[].cover	string	Product cover image URL
products[].items[].price	string	Current price string (with currency)
products[].items[].originalPrice	string	List price / strikethrough price
products[].items[].score	string	Star rating
products[].items[].ratingsCount	string	Review count
products[].items[].describe	string	Short product blurb
followUpQuestions	string[]	Questions Alexa offers to continue with
screenshot	string	Screenshot URL for this turn

API Usage

This skill calls the LinkFox tool gateway. See references/api.md for the calling convention, request/response shape, error codes, and a curl example. You can also run scripts/amazon_alexa_search.py directly to test it from the command line.

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:

User Says	Scenario
"用 Alexa 帮我推荐...", "亚马逊 Alexa 问下..."	Direct Alexa Q&A
"在亚马逊上聊聊给我推荐 ...", "对话式选品"	Conversational discovery
"顺便再追问一下 / 接着问 ..."	Follow-up (agent summarizes prior result and re-asks in new call)
"在这个页面 / 这个分类下推荐...", "基于这个页面再问一下"	Page-anchored conversation (use url)
"best XX for YY under $Z on Amazon"	Goal + constraint + budget Q&A
"对比 Alexa 给的前两个推荐"	Compare within Alexa's reply
"Alexa 还能继续问什么 / 给我一些追问思路"	Surface follow-up questions

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.

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.

<!-- LF_LARGE_RESPONSE_BLOCK -->

Handling Large Responses

To avoid overflowing the agent context, persist the response to disk and extract only the fields you need:

python scripts/response_io.py run --script scripts/amazon_alexa_search.py --out-dir <DIR> '<params>'
python scripts/response_io.py read <file> --fields "<paths>"   # or --path "<JMESPath>"

Pick --out-dir outside any git working tree (e.g. /tmp/... on Unix, %TEMP%/... on Windows). Persisted responses may contain PII, pricing, or auth-sensitive data — do not commit them. Files are not auto-deleted; clean up when the task is done.

run writes the full response to a file and emits only a schema preview + file path. read projects specific fields, with --limit/--offset for slicing and --format json|jsonl|csv|table for output.

When to prefer this pattern — apply your judgment based on the response characteristics, e.g.:

• High field count per record, or fields you don't need
• Batch/paginated results (multiple items per call)
• Long-text fields (descriptions, reviews, HTML, time series)
• Output reused across later steps rather than consumed immediately

For small, single-use responses, calling the main script directly is fine.

⚠️ The preview is a truncated schema + sample, not the full data. Any field-level decision must read from the persisted file via read.

<!-- /LF_LARGE_RESPONSE_BLOCK -->

---

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