Search Main Tech Stores

Preflight — Hebrew term resolution

Israeli retailer search returns nothing useful with raw English category nouns. Before touching any retailer:

1. Check data/hebrew-category-map.json for the product class.
2. If the class isn't mapped, invoke discover-hebrew-term first to derive and cache it.
3. Build the search query by combining: <hebrew category> <brand in Latin> <model in Latin>. Example: עמדת טעינה Anker Prime A2343, not Anker Prime A2343 charging tower.
4. If the map lists multiple Hebrew variants (he_terms array), be ready to try each — different retailers tag the same class differently.

Backend choice

Ask the user once per session, default Playwright.

How to find each retailer's search

Do not trust hardcoded URL patterns — they drift. Each retailer has a different search system that may change without notice.

The reliable approach: navigate to the retailer's homepage, find the search input, type the query, submit, then read the results. With Playwright:

1. browser_navigate to the retailer homepage.
2. browser_evaluate or browser_snapshot to find the search form (usually input[name="q"], input[type="search"], or #qSearch).
3. browser_type the Hebrew query into that input, with submit: true.
4. Wait for results, then browser_snapshot and extract product cards.

data/retailers.md has the last-known-good URL patterns as a starting hint, but treat them as candidates to verify, not contracts. If a documented pattern returns 404 or redirects to homepage, fall back to the homepage-form approach above and update data/retailers.md with what you discovered.

Workflow

1. Preflight — Hebrew term resolution (above).
2. For each of the four retailers, in parallel where possible:
   • Submit the Hebrew query (homepage-form approach if URL pattern is unreliable).
   • Capture the top 5 product cards: title, price, link, in-stock indicator.
3. Try Hebrew variants. If a retailer returns nothing for the primary he_terms[0], retry with he_terms[1] and he_terms[2] before declaring it absent.
4. Apply VAT rules from data/shopping-rules.md. On KSP always use the higher (regular, non-Eilat) price.
5. Rank low-to-high by VAT-inclusive ILS price.

Fallback chain (if a retailer returns nothing)

For each retailer that comes back empty after trying all he_terms:

1. Try the same retailer with the brand + model only (no Hebrew category noun) — sometimes retailer search rejects the noun but accepts the SKU.
2. Try search-google-il with <brand model> site:<retailer-domain> — Google often has a better index of the retailer's catalog than the retailer itself.
3. Skip the retailer and note "no results across N variants" in the output.

Output format

## {product} — main Israeli tech stores
Hebrew query terms tried: {he_terms used}
Backend: {Playwright | Tavily}

| # | Price (₪, inc. VAT) | Product | Retailer | Link |
|---|---------------------|---------|----------|------|
| 1 | ...                 | ...     | ...      | ...  |

### Notes
- Any VAT adjustments applied
- Any retailer that errored / timed out / had no results across all he_terms
- Any obvious outliers
- Any URL pattern in data/retailers.md found stale (with the new pattern, if discovered)

Rules

• Cap at 5 results per retailer.
• Skip retailers that error; note them rather than failing the whole run.
• Never visit non-whitelisted domains.
• If both backends fail, report which one failed and offer to retry with the other.
• If the user-facing comparison would be more efficient via Zap, suggest search-zap instead and only continue if the user confirms.