[honzajavorek]

The expert course now builds on the fixed academy-amazon-scraper starter repo instead of the archived course. Solution code follows the starter's layout and code style, including the Actor template alignment: Dockerfile in the root, @crawlee/cheerio imports, and context pushData/addRequests. demo-actor is renamed to match the starter, and the filter Actor parses prices like $1,080.59 correctly. Closes #3122.

Still to do before this is ready:

• Run each solution on top of the starter repo and confirm it works (code was only checked against the repo's source, not executed)
• Check that screenshots still match Amazon and Apify Console: images/product-offers.jpg, solutions/images/adding-webhook.jpg, webhook-succeeded.png, dataset-preview.png, result-items.jpg, create-new-task.jpg, creating-task.png
• Decide whether the tasks lesson should explicitly ask readers to create the academy-amazon-scraper-task task, which the API & client solution relies on
• Revisit the solution code once the starter moves to Crawlee 4 and Apify SDK 4 (Port to Crawlee 4 and Apify SDK 4 apify-projects/academy-amazon-scraper#3)

🤖 Generated with Claude Code

https://claude.ai/code/session_01M4cTCT4ARqx59nHGuyeXAw

[apify-service-account]

✅ Preview for this PR (commit d07bf39e) is ready at https://pr-3137.preview.docs.apify.com (see action run).

[honzajavorek]

I ran the solutions locally on top of the starter (apify run, keyword headphones):

• V Handling migrations: works. Killed a run midway and resumed it; the tracker counts match the dataset (71/71).
• VI Rotating proxies: works. The saved session pool shows maxUsageCount: 5 and maxErrorScore: 1.
• VII Saving run stats: works. totalSaved matches the item count, errors are recorded, and dateHandled, numberOfRetries, and currentPendingRequests are filled in.
• I Integrating webhooks: the filter Actor works on a real dataset (cheapest offer per ASIN, $1,080.59 parses correctly). I didn't test the webhook itself on the platform.
• IV API & client: not run, because it needs the task on the platform.

Issues:

1. VII: errorHandler: async ({ error, request }) only works because Crawlee 3 still adds a deprecated error to the context. The documented signature is async ({ request }, error) => ….
2. VII adds persistStateKey: 'AMAZON-SESSIONS' without explaining it, and solution VI doesn't have it.
3. V imports ASIN_TRACKER from constants.js but never shows its definition (export const ASIN_TRACKER = 'ASIN-TRACKER';).
4. The new asinTracker.js and Stats.js fail the starter's npm run lint: console.log, default exports, and import order.
5. Starter, not this PR: the default keyword iphone returns only renewed iPhones with "No featured offers available", so a run with the default input produces an empty dataset.

Amazon blocks the search page after many runs in a row. A few runs failed on all retries, and a short pause fixed it.

[honzajavorek]

Follow-up: I tested I and IV on the platform. The starter was deployed as is, filter-actor and actor-caller were built from the code and schemas in the docs, and the keyword was headphones.

• I Integrating webhooks: works after one fix. The webhook fired after the scraper succeeded and started filter-actor with the scraper's dataset ID. The filter run returned 16 items for 16 ASINs, each with that ASIN's lowest price (120 scraped offers, 0 mismatches).
• IV API & client: works after one fix. Both modes write OUTPUT as text/csv with only the requested fields and exactly maxItems rows (10 with the client, 5 with the API).

Issues:

1. I and IV: new Actors run with limited permissions by default, and the docs don't mention it.
   • filter-actor fails with 403 insufficient-permissions when it reads the scraper's dataset. Declaring the input as a dataset resource fixes it without full permissions. Both properties are needed; resourceType alone still gets a 403:

     "datasetId": { "title": "Dataset ID", "type": "string", "description": "…", "editor": "textfield", "resourceType": "dataset", "resourcePermissions": ["READ"] }

   • actor-caller fails with 403 insufficient-permissions when it starts the task. It worked after switching it to Settings > Permissions > Full permissions. The lesson should tell readers to do that.
2. IV: task.call({ memory }) doesn't set memory. The first argument of call() is an input override, so the task got {"keyword":"headphones","memory":4096} as input and ran with the default 1024 MB. Use task.call(undefined, { memory }). The API version passes memory correctly, and that task run got 4096 MB.
3. IV: nothing in the course tells readers to create a task named academy-amazon-scraper-task, and the solution only works if one exists. I created it by hand with input {"keyword": "headphones"}.
4. IV: the prose says the result is saved under OUTPUT.csv, but the code saves it under OUTPUT.
5. I: the payload template appears only in the screenshot. It needs to be {"datasetId": {{resource.defaultDatasetId}}} with no quotes around the variable. With quotes, the filter Actor receives the literal string {{resource.defaultDatasetId}} and fails. Writing the template as text would make it copyable and harder to get wrong.
6. Minor: the client's CSV starts with a byte order mark (BOM), and the API's CSV doesn't.

Amazon didn't block any of the five platform runs.

---

Generated by Claude Code