Shiseido

Displays curated Shiseido products as an interactive carousel. ONE CALL PER RECOMMENDATION: pass every product of the recommendation together, in display order — a 3-product routine is ONE call with 3 IDs, never one call per product or per need. A recommendation split across calls never shows the shopper its full selection in one carousel. Commentary reads best written after the carousel is displayed rather than before, following carousel order, and only about the products just rendered. The IDs passed are the curated selection, not every search result. KEEPING THE CAROUSEL CURRENT: the carousel shows exactly the products of the latest render-carousel call and never updates on its own; text alone doesn't change it. Whenever the recommended set changes (the user refines by price, concern or category, or a different set is drawn from products already in context, even without a new search), calling render-carousel again with the new IDs before commenting keeps the screen and the text in sync. Describing products as displayed is only accurate for the set that was just rendered.

render-carousel

Searches the Shiseido product catalog. The response is data plus guidance for working with it. It covers Shiseido products only; other brands aren't in the catalog. WHEN TO CALL This tool fits whenever the conversation leads to Shiseido products. For broad intents ('build my routine'), 1-2 qualifying questions first (skin goals, skin type) sharpen the search, unless the conversation already answers them. INPUT RULES `query`: a natural-language description of ONE concern, in the market's language. Search is semantic, so the user's own phrasing works best, descriptive attributes included (finish, skin type, concern, benefit, coverage). Keywords, synonyms, spelling and accents don't matter. Descriptive attributes belong in `query` rather than in `filters`. On nl-be the catalog is indexed in French while its products are returned in Dutch, so `query` is written in FRENCH there and the recommendation in Dutch. The response's `marketNote` states this whenever it applies. `need`: the product need a query serves. One `need` per product the user should end up with: a routine has three (cleanser, serum, moisturizer), a single recommendation has one. `filters` carry only hard constraints (range name, price, best-seller/new/award) and are enforced strictly, so they work best used sparingly. When a search returns too few results, a broader `query` or one fewer filter surfaces more. DECOMPOSING: a need the user described with several concerns gets one entry per concern, all sharing the same `need`. As many entries as the user named distinct concerns, so 2 to 4 in practice; beyond that the extra queries repeat each other. Results carry `matchedFacets`, the facet queries that surfaced each product, so coverage is visible: a product matching every concern is a stronger pick than one matching a single facet. Results can still include off-category or off-audience products (e.g. men's items). Each product's `category`/`rangeName` is worth checking against the user's request before recommending; a need with no suitable match is best re-searched. Example: "best moisturizer for me? 35yo woman, dry skin with redness and appearing wrinkles, want to prevent aging and deeply moisturize for good elasticity" (uk market). Four distinct concerns: dryness, redness, wrinkles, elasticity. ✗ One entry blending them: [{ need: 'moisturizer', query: 'anti-aging moisturizer for dry skin with redness deep hydration elasticity' }]. Meaning-based matching over a single blended query lands between the concerns instead of on them, and here drops redness entirely. ✓ One entry per concern, each keeping the category noun: [{ need: 'moisturizer', query: 'rich moisturizing cream for very dry skin' }, { need: 'moisturizer', query: 'soothing cream for redness and sensitive reactive skin' }, { need: 'moisturizer', query: 'anti-wrinkle cream to prevent signs of aging' }, { need: 'moisturizer', query: 'firming cream to improve skin elasticity' }] Further examples: - 'anti-aging serum under £80' → [{ need: 'serum', query: 'anti-aging serum', filters: { priceMax: 80 } }] - 'Vital Perfection range' → [{ need: 'range browse', query: '', filters: { rangeName: 'VITAL PERFECTION' } }] - 'routine for dry, dull skin' (fr market) → three needs, each split by concern: [{ need: 'cleanser', query: 'nettoyant doux pour peau sèche' }, { need: 'serum', query: 'sérum hydratant pour peau sèche' }, { need: 'serum', query: 'sérum éclat pour teint terne' }, { need: 'moisturizer', query: 'crème riche hydratante pour peau sèche' }, { need: 'moisturizer', query: 'crème pour raviver l'éclat du teint' }] PRESENTING: with a curated set of IDs, render-carousel displays the products — ONE call carrying the whole selection; the recommendation reads well written afterwards, in carousel order. The carousel carries the product links, so none are needed in the text. When the user stated a budget, add the selection's exact prices and check the total against it before asserting it fits. Out-of-stock products are automatically excluded.

search-products