Rp Source Sfcc

rp-source-sfcc

Technique guide for reading a public SFCC/SFRA storefront. It ships no capture code: rp-discovery follows its "no adapter exists → capture manually, same output contract" path, and you write the capture code for each migration as generated project code under migrations/<project>/src/, with its outputs in migrations/<project>/data/sfcc-discovery/. ⚠️ Nothing site-specific (code, ids, values, fixtures) is ever added to this skill.

Recognize the platform

• Pipeline URLs: /on/demandware.store/Sites-<site>-Site/<locale>/<Controller>. The homepage og:url often uses one, which gives you the site id and the default locale.

• Static assets: /on/demandware.static/.... Image renditions: /dw/image/v2/<REALM>_<ENV>/....

• OCAPI (/s/<site>/dw/shop/..., /s/-/dw/data/...) answers 401 without credentials. Don't ask for Business Manager or OCAPI credentials in this flow; record them as a later phase. Use sourceMode=public_content.

Files to read first

1. robots.txt. Obey it for every request, and take the sitemap index from it.

2. Sitemaps (index → sitemap_N.xml). These list products, categories and <image:loc> originals. ⚠️ Sitemaps are incomplete. Pre-order products and variant SKUs are commonly missing, some listed products aren't searchable, and some *.html entries are content assets or malformed. Never treat the sitemap as the catalog.

3. The homepage, plus a few category and product pages. Read the header, navigation, footer and hreflang links. The homepage may have no hreflang, so learn the locale routes from inner pages before you crawl.

Public storefront endpoints (SFRA)

The product set is the union of the grid tiles from every category (including categories linked only from the navigation) and the sitemap's product URLs. Key products by SFCC product id, not by URL: locale renditions have their own canonicals.

Signals that are easy to misread

• Product types. Variation attributes create SKUs (Wix variant options). SFCC product options (e.g. engraving) change the price, not the SKU, so they are Wix modifiers. Keep them in separate fields.

• Option visibility. An option can be attached to a product but hidden on the storefront: its markup is present but hidden. SFRA has no standard visibility flag. Find the store's own custom field by comparing products where the option is shown and where it isn't. If you can't find one, record visibility as unknown and never drop the option.

• Stock. The values are inStock, preorder (with inStockDate, which only means a date when the status is preorder), and out of stock (often no status at all, only a message). lowStockCount and maxOrderQuantity are hints, never authoritative quantities.

• Prices. price.list is the compare-at price. promotions[] names promotion ids but never their rules. callForPrice hides the price and add-to-cart: that needs a human decision.

• Images. The original is the static path: drop /dw/image/v2/<REALM>_<ENV> and the query string. Larger sw/sh renditions are upscales.

• Ratings can be an aggregate number with no reviews behind it, which can't be imported as reviews.

• Videos may be opaque ids whose host is not exposed.

Regional settings (currency, language, country)

Record the store's regional facts from the public pages, read-only. Time zone is not exposed publicly: leave it unknown (not-exposed).

• Currency: the price currency code on product pages: the Product-Variation JSON price (price.sales.currency) or the PDP JSON-LD offers.priceCurrency. Sample products from every locale route you saw.

• Language and locale region: the default locale in the pipeline URL (Sites-<site>-Site/<locale>/, e.g. en_US) and the hreflang links. The language is the two-letter part (en) and locale is the en-US tag. Other hreflang languages remain adapter discovery evidence rather than fields in the neutral locale record.

• Several price currencies (locale routes priced in different currencies): preserve the list in SFCC discovery evidence and leave paymentCurrency unknown with reason ambiguous, unless the person names the base one, which is then a supplied value. Never pick one from the default locale alone.

• A value you cannot read is unknown with a reason (not-exposed, unreadable, ambiguous), never a guess or a default.

Build the file with buildSourceRegional (adapter sfcc; evidence names the page or field the value came from) and write it with writeSourceRegional from lib/source-regional.js to discovery/source-locale.json. There is no promote step for this public path.

Scan the other pages

Product and category pages are covered by the capture above. Everything else (about, shipping, returns, FAQ, legal, store locator, forms, gift certificates, loyalty) is usually linked only from the header and footer.

Seeds:

• the homepage;

• every link harvested from product and category pages;

• every default-locale sitemap URL the catalog capture did not claim (content assets and Page-Show pages often appear only there).

Rules:

• Stay on the same host, follow robots.txt, and skip the alternate-locale prefixes.

• Admit content routes: Page-Show?cid=<id> (SFRA content assets), Page Designer pages, Stores-Find (store locator), and ordinary *.html or extensionless content paths.

• Skip only these: assets; query variants of an already-seen page; catalog routes the capture owns (Product-*, Search-*); and action or AJAX routes (Cart-*, Checkout*, Account-*, Login-*, Order-*, Wishlist-*, *-Ajax*, form POST targets). ⚠️ Never skip a whole controller family that isn't on this list. Record it for review instead.

• Give every URL a disposition (captured, covered-by-catalog, skipped:<reason> or failed:<status>) in data/sfcc-discovery/page-inventory.json. Re-runs re-attempt failed.

For each captured page, read its content and propose the Wix destination: a static page, CMS items, a form, pickup locations, a gift-card product, or a manual or dropped item. Policy prose such as free-shipping thresholds, delivery times, VAT wording or the return window is evidence for Wix settings, not settings.

• Quote it verbatim into data/sfcc-discovery/policy-rules.json, with status: candidate.

• Never fill in a value the text doesn't state.

• Have the user confirm each rule, including threshold boundaries, shipping option titles and delivery estimates. Store their verbatim answer and set status: confirmed before anything is written to Wix.

Be polite and resumable

• Keep one throttle for every request (start around 60 per minute), use a bounded timeout, and retry 429/503 honoring Retry-After.

• Checkpoint as you go, so a re-run resumes and re-attempts only failures. Refresh prices and stock just before import.

• Decode URLs tolerantly: a malformed % must not abort the run.

Hand-off (existing artifacts only)

Each stage writes only its own part, through the contracts that already exist:

URL state, each file in its own role:

• base-paths.json: entity-level route intent from mapping urlPolicy.

• url-ledger.ndjson: one append-only row per public record (source URL, planned and actual target).

• redirects.ndjson: only when both relative URLs are known and differ.

• unresolved.ndjson: only URLs still waiting on a route decision, a target lookup or manual review.

• A preserved URL (same relative path) closes with no redirect.

Cutover items:

⚠️ No cutover UI reads these yet. They are preserved for it, not wired to it.