linkfox-amazon-reviews-list

Amazon Product Reviews

Fetch and analyze Amazon product reviews to help sellers extract actionable insights from customer feedback.

Core Concepts

This tool retrieves real customer reviews for a given Amazon ASIN across 15 marketplaces. You can control how many reviews to fetch per star rating (1-5 stars, up to 100 each), sort by recency or helpfulness, and apply various filters. Only one ASIN per request; for multiple ASINs, make separate calls.

How to call

• Asynchronous submission endpoint：POST /amazon/reviews/async/submit
• Polling query endpoint：POST /amazon/reviews/async/result
• Full parameters, responses, and error codes: see references/api.md
• Python script：python scripts/amazon_reviews.py '<JSON 参数>' [--inline]; submit or query only once each time and return immediately
• Cost constraints: This tool consumes compute; by default, the same parameter combination is called only once per session, and the script has a 24h local cache. On failure or empty results, do not automatically try different keywords, paginate, or change the ZIP code in continuous probing; if further retrieval is needed, first explain to the user that additional consumption will be incurred.

Asynchronous call flow

1. The first script execution only calls /amazon/reviews/async/submit, saves and immediately returns taskId, and must not wait for results within the same script call. Do not resubmit because the task is still running or a single query failed.
2. The submission parameters are the same as the original review parameters, and provideris not passed; the backend automatically selects Pango or Apify based on domainCode , and the actual provider can be seen from the response's provider .
3. The Agent may continue performing other independent work in the user's request; when results are needed, execute the script again with the original parameters, or pass in {"taskId":"..."}. Each execution calls /amazon/reviews/async/result only once and returns immediately, without looping or sleeping inside the script.
4. Based on current empirical results, Pango usually returns in about 10~30 seconds, and Apify usually returns in about 15~60 seconds; this is not an SLA. The script will return estimatedReadyInSeconds and suggestedNextCheckAfterSeconds; prioritize continuing other work, query again at the suggested time, and do not busy-poll at the API's minimum polling interval.
5. Observe for at most 200 seconds from task creation time. If within the window it is PENDING or RUNNING, keep the original taskId, continue other work, and query again after about 10~20 seconds;SUCCEEDED: read result；FAILED: explain to the user error and stop;CANCELLED: politely explain that the current review service may be receiving many requests, and the task was automatically canceled while waiting for execution resources and no fees were incurred; it is recommended to resubmit later.
6. If after more than 200 seconds it is still PENDING or RUNNING , return POLL_TIMEOUT and stop automatic querying, letting the user choose: continue querying the same taskId, or stop querying and abandon this result. Do not automatically submit a new task. There is currently no manual cancellation endpoint; the backend will only automatically cancel PENDING tasks that time out while waiting for execution resources and have not yet called the provider. Tasks that have already entered RUNNING may still complete and incur fees.
7. Unread tasks are retained for at most 4 hours from submission; after the first read reaches the terminal state SUCCEEDED、FAILED or CANCELLED , the backend immediately deletes the task. The script saves successful results to a local file, and the task cache only retains "claimed" and the file path. The original parameters and taskId share the task record within the same working directory; executing again within the 24h local cache validity period will return ALREADY_RECEIVED、resultFile and this time costToken: 0, the file should be read, and the backend should no longer be queried or resubmitted. If the file does not exist, it must not automatically re-fetch.
8. When the query returns "task does not exist or has expired", stop and do not automatically create a new paid task. Only when the user explicitly requests a retry may it be resubmitted.
9. Submission,PENDING、RUNNING、FAILED、CANCELLED and POLL_TIMEOUT must not be recorded as actual consumption according to the estimated formula; the actual charge is only recognized as the X-Cost-Token/costTokenreturned in the response when the result is successfully claimed for the first time, and the Agent must not calculate or supplement it on its own.

The script only executes the above asynchronous process, and locally caches the unfinished taskId for subsequent queries.

Output strategy (script default behavior)：

• Alwayswrite the complete response to <cwd>/linkfox/<YYYY-MM-DD>/<session>/data/linkfox-amazon-reviews-list-<timestamp>.json（<cwd> is the working directory when the script is executed, which in Claude Code is the current project directory;<session> is taken from the environment variable SESSION_ID, automatically aggregated according to the user task;Writing to /tmp is prohibited, and an error is reported if the current directory is not writable)
• Response body ≤ 8 KB: after saving to disk, print the complete JSON to stdout
• Response body > 8 KB: after saving to disk, stdout outputs only the summary (top-level fields, common counts such as total/costToken, the length of the largest list field + the first 3 samples)
• Add --inline to force full printing to stdout (also saved to disk)

Data reading recommendations: first look at the summary to determine whether it is sufficient; when specific fields are needed, prefer to use jqorConvertFrom-Json to extract on demand from the saved json file, avoiding putting the entire JSON into the context.

Resolving authentication and computing power issues

When the following abnormal situations occur, use references/onboarding.md to guide problem resolution:

Abnormal situations

• API Key not configured: the environment variable is not configured LINKFOX_AGENT_API_KEY, and LINKFOXAGENT_API_KEY。
• Response 401 or 402 status code
• Response indicates insufficient computing power or balance: the message contains "insufficient computing power balance/insufficient billing/insufficient balance/quota exceeded/insufficient balance/package expired/recharge required/please recharge", or content with a similar meaning.

Parameter Guide

Parameter	Type	Required	Scope	Description	Default
taskId	string	Yes	Async result	Task ID returned by async submit	-
asin	string	Yes	Async submit	Amazon product ASIN	-
star1Num	integer	No	Async submit	1-star reviews to fetch (0-100)	10
star2Num	integer	No	Async submit	2-star reviews to fetch (0-100)	10
star3Num	integer	No	Async submit	3-star reviews to fetch (0-100)	10
star4Num	integer	No	Async submit	4-star reviews to fetch (0-100)	10
star5Num	integer	No	Async submit	5-star reviews to fetch (0-100)	10
sortBy	string	No	All	recent (newest) or helpful (most helpful)	recent
formatType	string	No	All	all_formats or current_format	all_formats
domainCode	string	No	Async submit	Marketplace code (see Supported Marketplaces); use com for US	com
filterByKeyword	string	No	Async submit	Filter reviews by keyword (max 1000 chars)	-
reviewerType	string	No	Async submit	all_reviews or avp_only_reviews (verified only)	all_reviews
mediaType	string	No	Async submit	all_contents or media_reviews_only	all_contents

Star Count Defaults

• If no star count fields are provided, star1Num to star5Num all default to 10.
• If any star count field is provided, unspecified star counts default to 0.

Supported Marketplaces

Marketplace	Code
United States	com
Canada	ca
United Kingdom	co.uk
Germany	de
France	fr
Italy	it
Spain	es
Japan	co.jp
India	in
Australia	com.au
Brazil	com.br
Mexico	com.mx
Netherlands	nl
Sweden	se
United Arab Emirates	ae

Use domainCode for every supported marketplace. Always confirm the user's intended marketplace.

Usage Examples

1. Fetch US reviews (Amazon.com)

{"asin": "B08N5WRWNW", "domainCode": "com", "star1Num": 10, "star2Num": 10, "star3Num": 10, "star4Num": 10, "star5Num": 10, "sortBy": "recent"}

2. Fetch negative reviews with keyword filter (Germany)

{"asin": "B08N5WRWNW", "domainCode": "de", "star1Num": 30, "star2Num": 30, "filterByKeyword": "quality", "reviewerType": "avp_only_reviews"}

3. Fetch 5-star reviews with media (Japan)

{"asin": "B08N5WRWNW", "domainCode": "co.jp", "star5Num": 50, "star1Num": 0, "star2Num": 0, "star3Num": 0, "star4Num": 0, "sortBy": "helpful", "mediaType": "media_reviews_only"}

4. Fetch only 3-star reviews (explicit star mode)

{"asin": "B0FP5C63HZ", "domainCode": "com", "star3Num": 100}

Display Rules

1. Present data clearly: Show reviews grouped by star rating with key fields: rating, title, text, date, verified status, helpful count.
2. Summarize when appropriate: For many reviews, provide a theme/pain-point summary before listing individuals.
3. Highlight actionable insights: Call out recurring complaints in negative reviews; note praised features in positive reviews.
4. Vine and verified labels: Clearly indicate Vine Voice and verified purchase status.
5. Media indicators: Note when reviews include images or videos.
6. Response normalization: Normalize rating and helpful-count fields for consistent display when the raw response uses marketplace-specific text formats.
7. Error handling: When a query fails, explain the reason based on the response message and suggest adjusting parameters.
8. Single ASIN limitation: If the user asks about multiple ASINs, make separate requests for each.

Important Limitations

• One ASIN per request: Only a single ASIN can be queried at a time.
• Per-star cap: Each star rating returns max 100 reviews per request.
• Parameter scope: filterByKeyword, reviewerType, mediaType are available on async submit, including domainCode: "com".
• No historical snapshots: Reviews are fetched in real-time.
• Review text language: Reviews are returned in their original language as posted.

User Expression & Scenario Quick Reference

Applicable — Tasks involving Amazon product reviews:

User Says	Scenario
"Show me the reviews for this ASIN"	Direct review lookup
"Get US reviews for B08N5WRWNW"	Marketplace-specific lookup
"What are customers complaining about"	Negative review analysis
"Get me all the 1-star reviews"	Star-filtered retrieval
"Any common issues in the bad reviews"	Pain point mining
"What do people like about this product"	Positive review analysis
"Find reviews mentioning 'battery'"	Keyword-filtered reviews
"Show me reviews with photos"	Media-filtered reviews
"Verified purchase reviews only"	Reviewer-type filtering
"Help me analyze competitor reviews"	Competitor review research
"Product improvement suggestions from reviews"	Actionable insight extraction

Not applicable — Needs beyond product review data:

• ABA search term data / keyword research (use ABA Data Explorer instead)
• Sales estimation or revenue analysis
• Listing copywriting or A+ content creation
• Advertising / PPC strategy
• Pricing strategy or profit margin calculations

Boundary judgment: If "product research" or "competitor analysis" boils down to reading customer reviews for specific ASINs, this skill applies. If it involves search volume, keyword rankings, sales estimates, or market sizing, it does not.

Computing power consumption rules

Billing is dynamically based on the planned number of pages to crawl. Let P = Σ min(ceil(各星级请求评论数 / 10), 10), then:

• Billing Token:21000 × P。
• Actual computing power consumption = ceil((21000 × P) / 2000) = ceil(10.5 × P), rounded up separately for each request.
• When no star rating quantities are passed, 1~5 stars each crawl 10 items by default,P = 5, consuming 53 computing power.
• After any star rating quantity is passed, unpassed star ratings use 0; when all five star ratings explicitly pass 0 ,P = 0, the supplier is not called and 0 computing power is consumed.
• If the request succeeds but the result is empty, or is empty after keyword, verified purchase, or media filtering, billing is still based on the planned number of pages; if some subtasks fail but the overall task succeeds, billing is still based on all planned pages; if the call fails completely, there is no charge.

Common consumption: 1 page 11 computing power, 2 pages 21 computing power, 5 pages 53 computing power, 10 pages 105 computing power.

Important: the fee is calculated based on the requested number of pages to crawl, not on the number of comments ultimately returned. Before requesting multiple star ratings or a large number of comments, you must explain the estimated number of pages and computing power consumption to the user.

The above formula is only used for estimation before calling. Whether a fee is actually charged and the amount charged shall be based solely on the X-Cost-Token/costToken returned upon the first successful claim result; submission, processing, failure, or timeout must not be counted as consumed based on the estimated value.

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 the references. Do not interrupt the user's flow.

---

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