Rp Import Codegen

This skill turns the schema, mapping, and setup decisions into implementation files under the active migration project.

Required inputs

• migrations/<project>/source-schema.json
• migrations/<project>/mapping/mapping-plan.json
• migrations/<project>/setup/setup-plan.json
• migrations/<project>/setup/setup-requirements.json when setup affects write paths

Prefer the machine-readable artifacts above. Markdown review files are secondary renderings for humans, not the primary codegen contract.

Source read contract

Generating a correct reader requires platform-specific knowledge — auth model, pagination, rate limits, and REST quirks. That knowledge lives in the matching source adapter skill, not here, so this skill never names a platform. Resolve the adapter from the platform field in source-schema.json via the naming convention rp-source-<platform> (e.g. platform: "wordpress" → rp-source-wordpress) and read its “Read contract” section. The operational facts should already be recorded in source-profile.md; use the adapter to fill any gaps rather than guessing. rp-execute-import runs the reader you generate and stays platform-agnostic — so the platform specifics must be baked into this generated code, not deferred to execution.

Adding a new source platform therefore requires no change to this skill: a new rp-source-<platform> adapter is enough.

CSV sources (platform: "csv")

When source-schema.json.platform === "csv", generate a file reader instead of an HTTP reader and read rp-source-csv → “Read contract”. The differences that matter:

• Vendor lib/csv-parse.js into src/lib/ and import it, exactly as a WordPress reader vendors wp-http.js. Do not re-emit parsing: the sampler and the reader must agree on what the file contains. There is no auth, no pagination, and no rate limiting to generate.
• Grouping is the reader’s core job. Replay sourceMeta.sourceFiles[].layout: with continuation: "blank-key" a blank key extends the current group; with a sectioned pattern the discriminator column routes parent vs child rows and parentRefColumn resolves a child to its parent (it may hold id:123 or a SKU, and the child is not guaranteed to follow its parent).
• Iterate the file set by role, treating sourceFiles[].partOf entries as continuations of the same logical stream and honoring each part’s own column order.
• Materialize column-values entities: collect the distinct values of the source column and their ancestors into their own entity file, emitted depth-ascending so a parent category exists before its child, plus the linking relation. Do not invent Wix ids at extract time — the ImportCrosswalk resolves them at import time.
• Apply sourceMeta.dialect.emptyPolicy through the shared coerceEmpty helper so a required Wix field is never fed an empty string the source did not have.
• CSV values are plain text; write them as-is with no entity-decoding step.

Target write contract

Symmetrically, do not re-derive the Wix write surface here. It is identical for every migration and is pre-verified in the rp-target-wix internal resource (see CONVENTIONS.md), which ships shared Wix runtime code: verified request builders, executors, and reusable execution logic for retries, throttling, checkpoints, and reporting. Vendor copies of the shared runtime modules from rp-target-wix into the project (like the source transport) and generate thin project-specific write specs and transforms that call those shared primitives/runtime functions. Never hand-emit Wix endpoints/bodies inline — that path repeatedly shipped wrong shapes (lowercase Ricos plugin enums, {tag:{…}} tag bodies, media.wixMedia.image.id featured images) that only failed at execution.

Validate by real call, not by doc example. MCP doc checks confirm an endpoint exists; they do not confirm the request shape works — public examples have been wrong (e.g. Ricos plugin enums shown lowercase that 400). Trust a Wix request shape only once a real call (or tests/target-wix/contract-test.js in live mode, run from the repo root) has succeeded. Treat the adapter’s // VERIFIED: shapes as the source of truth over any docs example.

Adding a new source platform requires no change here, and the Wix surface change for all migrations is a one-place edit in rp-target-wix (caught by its contract test), not a per-project regeneration.

Workflow

1. Read the discovery, mapping, and setup artifacts, plus the source adapter’s read contract and the rp-target-wix write contract. If a mapping entity includes targetRef, validate its selected writer and reliability against rp-target-wix/scripts/domain-knowledge.js summarize-entities; do not remap source entities independently.
2. Generate machine-readable execution artifacts:
   • execution/execution-manifest.json
   • execution/llm-handoff.json
   • when SAFE_MODE=true or DRY_RUN=true, execution/review/code-safety-review.md
   • optional supporting execution subplans only when the runtime truly needs them
3. Generate setup code that can verify/provision Wix prerequisites by executing the machine setup artifacts through the shared setup runtime.
4. Generate source reader code that can enumerate and fetch source entities. If the adapter ships a shared transport module (auth, pagination, throttling, retries), vendor a copy of it into the project (e.g. migrations/<project>/src/lib/) and import from it instead of re-emitting that plumbing — the reader should hold only per-project orchestration. The reader must extract source records to durable files on disk; it must not require the whole source dataset to live in memory before import begins.
5. Generate transform code that maps source records into Wix-shaped objects — thin glue over wix-build.js, which already owns slug sanitizing and the money/price/variant rules; see “Never hand-write slug sanitizing” below. Wix Stores products: Catalog V3 only. Catalog V1 is not supported by this workflow — there is nothing to generate for it. The only destination a migration ever writes to is a V3_CATALOG site (guaranteed at provisioning and gated before any write — see replatform → “Headless site creation” and rp-execute-setup), so generated code may assume V3 unconditionally: no version detection, no branch, no V1 shape, no V1 fallback for a failing V3 write. A V1 site is a blocker the run halts on, never a case codegen handles. Concretely: never emit Catalog V1-style top-level price, sku, or variant inventory fields. For simple products, emit one variant under variantsInfo.variants[] with variant-level price, sku, and physical properties. When the source product is subscription-based and the source payload exposes explicit recurring cadence in structured product data, emit native product.subscriptionDetails with at least allowOneTimePurchases and one subscriptions[] entry carrying title, description, frequency, interval, and autoRenewal. Do not emit placeholder-only metadata or skip those products on create when that cadence can be inferred deterministically; live create coverage for this shape was verified on July 26, 2026. Product HTML descriptions land in plainDescription, which Wix converts to rich content server-side — it is HTML, not a plain-text flattening, so there is no fidelity loss to accept. Do NOT call the Ricos conversion endpoint on the product path: it costs one HTTP round-trip per product ahead of a bulk create, and that burst is what the endpoint throttles with a 403. Two traps: plainDescription is silently ignored when description is also set (set exactly one), and it is capped at 16,000 characters — longer bodies need truncation or an info section, recorded in mapping-gaps.json. Ricos conversion remains correct for blog posts, where richContent really is a Ricos document. Media scope should be relationship-driven by default: generate imports only for media referenced by mapped entities, not for the entire unattached source library. When the target adapter says an entity can ingest external URLs directly (for example Wix Stores product media), generate that entity-native path instead of routing those files through the slower generic Media Manager import flow.
6. Generate thin project-specific write specs and import orchestration code that pass those Wix-shaped objects into the shared rp-target-wix runtime. Generated code should describe what to write and in what order, not how to implement retries, throttling, checkpointing, or audit logging. Prefer the selected entity’s preferredWrite; if it is not verified-live, surface an execution warning before consent. Native mappings must have a known writerId, a direct REST plan that calls notifyMissingWriter, or an explicit unsupported/gap fallback.
7. Generate or wire the deterministic execution-state preparation step before any import write. The generated runner should call the shared execution-state preparation contract, or the execution instructions should run skills/replatform/scripts/execution-state-prepare.js before the generated import entrypoint.
8. Generate the runnable setup/extraction/import entrypoints (see below) — the artifacts rp-execute-setup and rp-execute-import actually run. This is required, not optional. Then run the sample-preview gate (see below) before the execution-plan approval.
9. Generate execution/review/import-plan.md plus execution/review/import-plan.freshness.json using skills/replatform/scripts/artifact-freshness.js write .... The freshness metadata must include hashes for source-schema.json, mapping/mapping-plan.json, setup/setup-verification.json, generated import code, and the target contract ledger revision. If codegen follows a newly promoted write contract, regenerated code and this metadata must reflect that promotion.
10. Document any manual code follow-up still required.
11. Generate any post-import remediation helpers that are required to reach the accepted mapping fidelity when the source capture cannot express the relationship inline during the first create pass.

Post-codegen code-safety review checkpoint

When either SAFE_MODE=true or DRY_RUN=true for the active project, codegen must produce a mandatory review artifact before the execution approval gate:

• migrations/<project>/execution/review/code-safety-review.md

This review is performed by the agent, not delegated to the user. It verifies the generated code itself. It must check:

• every generated write path that can carry email/phone data passes safeModeOptions into the shared Wix runtime or direct REST wrapper
• every shared/native writer reached by generated code actually consumes those safeModeOptions and routes the request body through the shared sanitizer rather than silently ignoring the parameter
• no generated writer path silently bypasses safe mode because of a missing function parameter or a direct call that skips the shared sanitizer
• dry-run uses the same generated write code path as live import, with Wix calls skipped only at the shared wix.send boundary
• dry-run reporting does not label would-send placeholders as live-created/imported site objects
• mapping-declared safeModeReplacements[] are reflected in the generated write specs and runner wiring, and any entity that still carries outbound contact data without matching replacement-path coverage is treated as a review failure rather than left for the user to reason about manually

If the review finds a gap, execution approval must remain pending until the generated code is corrected and the review artifact is regenerated with a passing verdict. The user’s role at this checkpoint is final go/no-go approval after the agent has already completed the review and surfaced the findings.

Codegen boundary

This skill owns the project-specific layer only.

It should generate:

• setup plan renderings and setup runner wiring
• source readers
• transforms
• per-entity write specs
• import ordering and dependency wiring
• project-local config loading
• execution review artifacts
• post-codegen code-safety review artifacts when safe mode or dry-run is enabled

It must not regenerate for each migration:

• raw Wix auth/client plumbing
• generic retry loops
• throttling behavior
• audit log shape
• compact execution report shape
• checkpoint store mechanics
• generic bulk-write orchestration

Those behaviors belong in rp-target-wix.

Final handoff expectations

The generated plan and downstream execution path must make the final state legible to the user. Treat these as required handoff details, not optional niceties:

• Surface the dashboard URL for the destination site.
• Surface the editor URL only when editor work was actually performed or the next required step is explicitly in the editor.
• Surface the preview URL only when public route/site verification is relevant to the completed work.
• Distinguish clearly between:
  • catalog/data imported
  • website/homepage built
• Do not imply that a successful Stores import means the site’s homepage or full website experience exists. A migration can finish with valid catalog/product/cart/checkout routes while the homepage is still blank or unbuilt.
• If Shopify quick mode relies on public collection feeds such as /collections/{handle}/products.json to recover category membership, bake that into the generated extractor/importer or emit an explicit remediation helper and call it out in execution/review/import-plan.md.

Runnable setup/extraction/import entrypoints — the artifacts are the execution path (required)

rp-execute-setup and rp-execute-import run the migration by executing these artifacts, never by the agent hand-issuing Wix MCP calls. So codegen must emit real, runnable entrypoints:

• src/setup/run-setup.js or equivalent setup entrypoint — reads the machine setup artifacts and executes them through the shared setup runtime.
• src/extract/run-extract.js or equivalent reader entrypoint — reads the source APIs and writes durable extracted files under the project. Extraction is a separate step from destination writes, so large migrations can be resumed or re-imported without re-reading the whole source.
• src/import/run-import.js or equivalent import entrypoint — reads the extracted files, applies transforms, and writes to Wix in dependency order through the vendored shared write runtime. It must:
  • load project-local config files, then process env, for all expected env-style values (never hardcoded; fail fast if absent — do not fall back to any agent/MCP auth),
  • run the notification-mute preflight before the first entity write whenever mute is in effect (see “Notification-mute preflight” below),
  • consume extracted source files from disk rather than materializing the entire source in memory,
  • apply idempotent dedupe keyed by source ID, using either a client-controlled source-id field on the target or the authoritative local state/crosswalk/crosswalk.ndjson durable sourceId -> targetId crosswalk for native Wix entities with server-assigned IDs, reading existing target state under the rules in “Reading existing target state” below — an empty read is not an empty site,
  • honor DRY_RUN=true and --dry-run for the safe-validation pass; --dry-run takes precedence over config and enables the shared Wix runtime’s dry-run mode.
  • --sample may be supported as a narrower dry-run-style validation mode, but it must not replace --dry-run as the primary safe-validation control.
  • honor deterministic selective resume flags: --entity <entity>, --source-type <subtype>, --missing-only, --failed-only, and --deferred-only. --missing-only, --failed-only, and --deferred-only are mutually exclusive. These flags must drive the same import path as a full run after selecting a stable record set; do not generate one-off recovery drivers.
  • emit a stable runId at process start and include it in every audit event, dry-run request capture, placeholder crosswalk row, and summary artifact.
  • emit execution/live-import-summary.json and execution/completion-report.json through the shared completion-report runtime, with entity/subtype completeness counters for extracted, in-scope, attempted, imported, already present by crosswalk, deferred, failed, and skipped-out-of-scope records.
  • stop before dependent phases when required upstream entities fail. For example, do not create products after product-category failures unless the execution plan explicitly marks category assignments as non-blocking.
  • pass safeModeOptions to every relevant writer path, including direct REST fallback paths and unverified native writers that can carry contact values.
• The dry-run is the same import code path with Wix calls skipped at the shared Wix boundary — not a separate driver. Generated code must construct the same Wix client, call the same writer helpers, run SAFE_MODE sanitization, and pass request metadata (phase, operation, entity, sourceId, verification, and when needed responseShape) into wix.send so the runtime can capture would-send requests and return live-compatible dry-run placeholders.

Extraction format requirements:

• write extracted data to project-local files, not process memory
• chunk by entity and page/batch so a large source does not become one giant file
• write a manifest that lets the import step discover which entity files exist and in what order to consume them
• make the extracted files deterministic enough for resume, replay, and debugging

Notification-mute preflight (spec 0012, required)

When mute is in effect — always for WIX_SITE_STRATEGY=new (unconditional, regardless of WIX_MUTE_NOTIFICATIONS), and for existing sites only when WIX_MUTE_NOTIFICATIONS=on — the generated import script must contain a mandatory preflight step, before the first entity write, that asserts the site is muted:

• Primary: getSiteMuteState(wix) (rp-target-wix, VERIFIED 2026-08-04) returns muted: true.
• Fallback (only if the status read is unavailable): an idempotent muteSiteNotifications(wix, { reason }) call — passing the same project-identifying reason as setup (RePlatform migration — <project>), because a re-mute overwrites the recorded reason (last caller wins).

If the preflight call fails or reads muted: false and the re-mute fails, the script aborts before any write with a clear error routing back to setup — exit non-zero, no degraded mode, no --skip-mute-preflight-style flag, no warning-and-continue. For new-site projects the preflight is emitted unconditionally and never appears as a skippable option; for existing-site projects with WIX_MUTE_NOTIFICATIONS=off (the default), no preflight is emitted. The preflight logs its re-verification and outcome into the run’s execution/progress log (keyed by runId) — terminal reports read this recorded state, never strategy/config inference. In dry-run mode the preflight follows the shared dry-run contract like any other Wix call: the intent is captured and the state read is skipped (stateKnown: false), not asserted as muted. The generated script must never call unmuteSiteNotifications — unmute is an explicit owner request outside the import path. A preflight mute-verification failure is recorded through the standard rp-telemetry recorder as an error event (with error_code) like any other run event — no new telemetry surface.

Use the target’s BULK write path (required)

A generated importer must write through the target’s bulk endpoint whenever one exists. Per-record creates are acceptable only when the target has no bulk equivalent, or for a deliberate single-record contract probe. This is not an optimization to add later: at 1000 products, per-record creates are 1000 round trips where bulk is ~11, and the latency difference is the difference between a minute and half an hour.

Derive the batch shape from the endpoint’s own limits, all of them at once. Bulk endpoints routinely cap several dimensions simultaneously and exceeding any one rejects the entire request — Wix bulk product create caps products (100), variants (1000), options (100), modifiers (100) and infoSections (100) per request, so with 2 options per product the options cap binds at 50 products, not 100. Batch with ndjson.readBatchesByLimits and the limits/cost helpers the target adapter exports (BULK_PRODUCT_LIMITS, storesProductBulkCost); never batch on record count alone.

Three properties of bulk responses that generated code must handle explicitly, because each one silently corrupts a report if ignored:

• Bulk is not atomic. A 200 can contain per-item failures. Walk every results[] entry; never infer success from the HTTP status.
• Correlate by the response’s own index field (itemMetadata.originalIndex for Wix), not by response position, and verify that every input is accounted for. A mis-correlated result crosswalks the wrong target id onto a source record.
• Count the “undetailed failures” bucket. Servers drop failure detail past a threshold; those are still failures and must appear in the completion report.

Dedupe before building the batch — a skipped record must never reach the API — and record the crosswalk per successful item, not per batch, so an interrupted run resumes correctly.

Record streams are NDJSON, single documents are JSON (required)

Every file that holds a stream of records must be newline-delimited JSON (.ndjson), one record per line — never a { "records": [ … ] } array. Vendor rp-target-wix/lib/ndjson.js into the project (like wix-writers.js) and use it; do not hand-roll line splitting, which gets chunk boundaries and CRLF wrong.

This applies to:

• data/source-extract/<entity>.ndjson — the extractor’s output
• the sourceId -> targetId crosswalk
• audit logs (already NDJSON)

It exists because every downstream stage does the same three things with these files, and an array is the wrong shape for all of them:

• scan — countRecords counts lines; nothing is parsed and nothing is held in memory. A JSON array must be fully parsed to be counted.
• batch — a bulk endpoint’s page is readBatches(file, 100). Use readBatchesBy when the target caps more than one dimension: Wix bulk product create allows ≤100 products AND ≤1000 variants per request, which is { maxCount: 100, maxCost: 1000, cost: p => p.variantsInfo.variants.length }.
• cursor / resume — readSlice(file, { offset, limit }) skips what is already done without rebuilding it into objects.

Two more properties that matter in practice: a producer can append records as it finds them instead of buffering the whole entity, and an interrupted write leaves a valid readable prefix — a truncated JSON array is unparseable, so a crash mid-extract loses everything.

Do not line-delimit single documents. A manifest, mapping-plan.json, decisions.json, execution-manifest.json, preview-result.json or completion-report.json is one object; it stays .json. NDJSON buys nothing there and makes it unreadable.

Generated importers must stream these files (for await (const batch of readBatches(...))), not readAllRecords them. readAllRecords is an escape hatch for genuinely small streams (a six-record category list) and is named to make its misuse on a large stream obvious.

Projects generated before this rule can be moved forward with convertLegacyJsonFile(jsonPath, ndjsonPath) rather than re-extracting.

Reading existing target state (required)

An idempotent importer has to know what is already on the site before it writes. Every bug in this section is the same bug: a read that returns nothing looks exactly like a site that contains nothing, and nothing-on-the-site is the branch that writes. None of them throw, so none of them show up in a dry-run.

The adapter’s query* executors already return the array. queryStoresCategories, queryStoresProducts, queryContacts, queryCoupons and queryOrders unwrap the response before returning it, so the value is categories / products / etc. Generated code must use it directly:

The second form is what shipped, and reading .categories off an array yields undefined, which the usual || [] turns into an empty array. It silently disabled a category dedupe index, silently disabled a product name-match safety net, and made a setup verification report 0 categories on a site that had 25 — all without one error line.

Never cursor-page through those executors. Unwrapping discards pagingMetadata, so the cursor a loop needs is already gone; a loop built on them cannot advance past page one, and reading .pagingMetadata off the returned array is the same undefined as above. Prefer the adapter’s sweep primitives, which own the loop and the failure semantics:

• queryAllStoresCategories(wix), queryAllStoresProducts(wix), queryAllDataItems(...)

When a sweep is needed for an entity that has no queryAll* primitive yet, generate the loop against the raw response and add the primitive to rp-target-wix rather than leaving the loop in project code:

An incomplete sweep must throw, not fall through. If any page fails, or the loop hits its page ceiling with a cursor still outstanding, the generated code must abort the import with a message naming the sweep. It must not continue with the partial index, and must not treat “the sweep failed” as “nothing exists” — that is precisely the state in which a re-run re-creates the entire catalog it already imported. The single most expensive failure in this whole pipeline is a duplicate import, and it arrives through an empty net.

A match must be ADOPTED into the crosswalk, not skipped. When a safety net (name match, slug match, source-id field) finds that the target entity already exists, record it in the crosswalk with its target id and revision and count it as reused. A bare continue that skips the write without recording the id looks correct — nothing is duplicated — but every later phase that resolves ids from the crosswalk then silently drops the record. Concretely: the product was already on the site, so it was skipped, so it had no crosswalk row, so the category-link phase could not resolve its id and it ended up in no category at all. Index the net as name -> { id, revision }, not as a Set of names, so the id is available to adopt.

An ambiguous match is the one case that must not be adopted: if the source key is not unique (e.g. four source products share a title), the net cannot tell which existing entity corresponds to which source record. Import rather than guess, and report the ambiguity in the completion report.

Never hand-write slug sanitizing (required)

A source handle is not already a valid Wix slug. Wix rejects anything outside [a-z0-9-], and because slug validation happens before the batch is applied, one bad slug fails the entire bulk request — 100 products lost for one character. Shopify mints underscores from decimal titles (“pH 5.5” → ph-5_5), so this is routine input, not an edge case.

It is already solved: wix-build.js exports toWixSlug and applies it automatically through the coerce: 'slug' rule on product.slug in wix-target-spec.js. Generated transforms that call the build layer (see the src/import/transforms/ note under “File targets”) get it for free and must not re-derive it — a per-project copy is how the underscore bug reached a live site in the first place.

Two properties to preserve when a generated transform sets a slug explicitly:

• Sanitize in the build layer, not the writer, and keep both values. URL preservation needs the original sourceSlug alongside the plannedTargetSlug actually derived from it (see the URL preservation rules under “Codegen rules”), which a silent rewrite inside the writer would falsify. normalizeStoresProductV3 therefore passes a slug through untouched.
• toWixSlug throws when a value sanitizes to empty (an all-non-latin title, for example). That is a signal to supply a deterministic fallback — the record’s source id — not to omit the slug and let Wix derive one, which breaks URL preservation with no trace.

Sample-preview gate

Between the extractor being generated and the execution-plan approval, show the user what their data actually became. The mapping review checkpoint validates intent; this validates structure — how source rows turned into entities — before any full run.

Required whenever the source cannot be read back from a live API — in particular every platform: "csv" run, where a misread layout silently produces the wrong entity split.

1. Generate the extractor first (src/extract/run-extract.js).
2. Run it in sample mode (--sample) to materialize a small data/source-extract/ slice.
3. Write two artifacts under migrations/<project>/preview/:
   • preview-summary.md — a short human-readable structure preview: how rows became grouped entities (e.g. one product with its variants and images), per-entity record counts, and which columns landed in which Wix fields.
   • preview-result.json — { "status": "pending", "decidedAt": null, "decidedBy": null, "entityCounts": {...}, "warnings": [] }.
4. Pause and ask the user to validate the structure. On accept, set status: "accepted" with decidedAt/decidedBy; on reject, set status: "rejected" and set approvals.mapping.status back to pending so the router returns to rp-mapper.
5. Record the artifacts on the codegen checkpoint (checkpoints.codegen.artifactRefs + lastCompletedStep: "codegen.sample-preview").

This is a codegen sub-gate, not an orchestration phase: it adds no state to orchestration-state.js. While preview-result.json says pending, the router keeps routing back to this skill instead of advancing to the execution-plan gate — that routing behavior is the enforcement, so the artifact must be written honestly. The gate rides on the existing extract→import split and does not change how import writes to Wix.

Project-local config files

Generated code should treat migrations/<project>/config/ as the canonical home for all values that are otherwise expected as environment variables. Use simple .env syntax and load these files before reading config:

• config/wix.env always exists. Default DRY_RUN=false, except when the user has explicitly asked to start, create, prepare, or run the migration in dry-run mode; in that case scaffold or preserve DRY_RUN=true:

  bashCopy

  WIX_SITE_STRATEGY=
  WIX_SITE_ID=
  WIX_AUTH_TOKEN=
  WIX_MUTE_NOTIFICATIONS=
  DRY_RUN=false
  SAFE_MODE=true
  SAFE_MODE_PHONE_NUMBER=+972 50 0000000

• config/source.<platform>.env exists after the source platform is known. For WordPress:

  bashCopy

  WP_BASE_URL=
  WP_USERNAME=
  WP_APPLICATION_PASSWORD=
  WP_MEDIA_URL_REWRITE_FROM=
  WP_MEDIA_URL_REWRITE_TO=
  WC_CONSUMER_KEY=
  WC_CONSUMER_SECRET=

  For CSV sources this is config/source.csv.env, which is not secret-bearing — all keys are optional hints (CSV_INPUT_ROOT, CSV_DELIMITER, CSV_ENCODING, CSV_VENDOR, CSV_MEDIA_URL_REWRITE_FROM, CSV_MEDIA_URL_REWRITE_TO). Generated readers must resolve input paths against CSV_INPUT_ROOT (falling back to the project directory) rather than baking absolute paths into generated code.

Codegen rules:

• Generate a small dependency-free config loader in the runnable entrypoint or src/lib/.
• Load config/wix.env and the selected source config before constructing source/Wix clients.
• Default DRY_RUN to disabled. Treat true, 1, yes, and on as enabled and false, 0, no, and off as disabled. --dry-run must override config and enable dry-run. --no-dry-run may be supported to override DRY_RUN=true.
• When scaffolding a project that is in dry-run mode, generated review artifacts must say that leaving dry-run later requires explicit user approval for any step other than new-site creation, and that such overrides should be avoided when a dry-run or report is sufficient.
• Preserve an explicit DRY_RUN=true from project config when regenerating code or config. Do not reset it to false during later codegen passes.
• Default safe mode to enabled when SAFE_MODE is missing or blank. Honor SAFE_MODE=false when the user set it before mapping: do not require safeModeReplacements[], do not replace contact values, and do not write safe-mode email recovery rows.
• When safe mode is enabled, require SAFE_MODE_PHONE_NUMBER, default missing/blank values to +972 50 0000000, and make the generated config explicit.
• Real process environment variables may override file values.
• Blank values in config files must not overwrite non-empty process env values.
• If a required key is still missing after loading file + env, fail fast with the key name, not a downstream 401.
• In dry-run, missing or blank WIX_AUTH_TOKEN and WIX_SITE_ID are would_block_live findings, not blockers, unless a generated local artifact requires the site ID as a stable namespace. Do not mint a Wix CLI token solely for dry-run.
• WIX_MUTE_NOTIFICATIONS (spec 0012) resolves by strategy when blank: new → on, existing → off; record the resolved value explicitly (mirrored from orchestration/decisions.json). WIX_MUTE_NOTIFICATIONS=off together with WIX_SITE_STRATEGY=new fails codegen validation — halt with the config conflict rather than generating artifacts (the same rule fails in rp-setup-discovery; fail fast at whichever runs first). Note the preflight emission rule below ignores the config for new sites anyway — it is unconditional.
• Treat WIX_SITE_STRATEGY as required. WIX_SITE_ID becomes required no later than the point where generated code needs to construct Wix clients or destination-specific artifacts. For WIX_SITE_STRATEGY=new, codegen should fail with a clear message to create/select the new Wix site first rather than assuming an existing site flow.
• For RePlatform new site + headless, that unblock message must point back to the Wix CLI headless scaffold in replatform → “Headless site creation” (npm create @wix/new@latest headless). The account-level Projects API is deprecated for this workflow.
• WIX_AUTH_TOKEN is the canonical Wix auth key in project-local config and holds the site write credential for import. With CLI-scaffolded headless sites this is a short-lived CLI token sent as a Bearer token — the generated Wix client must send Authorization: Bearer <token> (plus wix-site-id). Do not assume a raw, non-expiring API key; mint the token at write time.
• Never log secret values. It is okay to log that a key is present/missing.
• Dry-run must write request captures under state/attempts/wix-request-captures.ndjson and dry-run placeholder target IDs only under state/crosswalk/dry-run-crosswalk.ndjson or in memory. It must never append simulated target IDs to state/crosswalk/crosswalk.ndjson.
• Do not generate debug output that dumps config file contents, environment snapshots, or request headers carrying credentials.

WooCommerce subscription product codegen

For WooCommerce Stores products, subscription products are a native Wix Stores product subtype when the target metadata marks product.subscriptionDetails writable for Catalog V3 create. Codegen must read the stores/product fieldContracts[] metadata and vendor/use the matching rp-target-wix/lib/wix-writers.js helpers:

• STORES_SUBSCRIPTION_CONTRACT
• normalizeStoresProductSubscriptions
• validateStoresProductSubscriptionDetails

Generate deterministic native subscription mapping only from recognized structured WooCommerce subscription fields. Accepted first-pass source keys include:

• _subscription_period_interval
• _subscription_period
• _subscription_length
• _subscription_trial_period
• _subscription_trial_length
• _subscription_sign_up_fee
• _subscription_price

Equivalent public REST fields may be used when discovery normalized them into the source schema, but do not infer cadence from product titles, descriptions, prose, shortcode blobs, or unrecognized plugin metadata.

Mapping rules:

• Treat source type: "subscription" or recognized subscription metadata as the product subtype signal.
• Parse billing interval as an integer >= 1.
• Parse billing period from day, week, month, or year, then emit Wix frequency DAY, WEEK, MONTH, or YEAR.
• Use subscription price when present, otherwise the normal product price, and still emit the Catalog V3 variant-level money object required by the shared writer.
• Parse optional length/trial/signup-fee fields only when they are clean structured values. If a field is present but malformed and needed for the selected mapping, defer the record rather than guessing.
• Build subscription descriptions deterministically from source subscription labels or product names, then pass them through normalizeStoresProductSubscriptions so the Wix description <= 60 contract is satisfied before write.
• Run validateStoresProductSubscriptionDetails in the generated dry-run path and before live create for every product carrying subscriptionDetails.

Do not generate a generic subscription skip gate. If a subscription product lacks structured cadence or required values, emit an explicit deferred record such as unsupported_subscription_shape, missing_subscription_cadence, or invalid_subscription_interval, and include that reason in the execution summary. A deferred subscription is a counted import outcome, not a silent skip.

Token minting — always route through scripts/mint-token.sh. Every migration project must include scripts/mint-token.sh. The script reads WIX_SITE_ID from config/wix.env, runs npx @wix/cli@latest token --site "$WIX_SITE_ID", captures the token (shape: OauthNG.JWS.<base64>.<base64>.<sig> — single line, no JSON wrapper), and writes it directly to config/wix.env as WIX_AUTH_TOKEN without printing the value. Run via Bash:

Do NOT run npx @wix/cli@latest token raw in a Bash tool call — it prints the credential to stdout which lands in the transcript. Always use mint-token.sh.

During scaffolding, copy scripts/mint-token.sh from the canonical skills location — do not generate it from scratch:

The preferred pre-import flow is:

1. verify wix whoami,
2. run bash scripts/mint-token.sh (writes WIX_AUTH_TOKEN silently),
3. run the import — the generated client reads WIX_AUTH_TOKEN and sends it as Authorization: Bearer <token> with wix-site-id.

Localhost media sources

If the source profile shows localhost, 127.0.0.1, or another private-only source URL, generated media import code must not assume Wix can fetch those URLs. Wix Media import is URL-based (rp-target-wix import-from-URL primitive), so live media import needs a public URL reachable by Wix servers. This is optional and, as far as we know today, only affects media import. Entity-native background ingestion from external URLs follows the same reachability requirement: if the target API ingests a source URL itself, that URL must still be publicly reachable by Wix.

Codegen/runtime should support one of these explicit paths:

• Use a public HTTPS tunnel/source URL for live media import. For ngrok on macOS:

  bashCopy

  brew install ngrok
  ngrok config add-authtoken "<YOUR_AUTHTOKEN>"
  ngrok http 8090
  export WP_BASE_URL=https://<id>.ngrok-free.app

• If the source REST responses still contain local media URLs, generate a configurable rewrite from the local base URL to the public tunnel base URL. For WordPress, use WP_MEDIA_URL_REWRITE_FROM and WP_MEDIA_URL_REWRITE_TO; when those are blank, it is acceptable to rewrite localhost/private origins to public WP_BASE_URL.

• Or generate/allow a media-skip/defer mode and document that media-dependent references such as hero images, galleries, and downloadable files will be absent until media is imported.

Surface the selected path in execution/review/import-plan.md and the execution plan before any live write. Do not let a dry-run with localhost media URLs imply live Wix Media import is ready.

File targets

Write code under the project-local source tree:

• migrations/<project>/execution/
• migrations/<project>/src/setup/
• migrations/<project>/src/extract/
• migrations/<project>/src/import/
• migrations/<project>/src/lib/ — vendored shared modules (the source adapter’s transport module, rp-target-wix/lib/ndjson.js, rp-target-wix/lib/wix-writers.js, rp-target-wix/lib/wix-target-spec.js, rp-target-wix/lib/wix-build.js), copied here so the project runs standalone with no external deps
• migrations/<project>/src/extract/readers/
• migrations/<project>/src/import/transforms/ — thin glue only. Vendor and call wix-build.js; do not re-derive the canonical→Wix payload layer here. Money objects, slug sanitization, the empty physicalProperties trap, choice-by-name variant references, the compare-at-must-exceed-price rule and the regular/sale price resolution are all settled in wix-target-spec.js + wix-build.js and regression-locked by tests/mapping/wix-build-oracle-test.js against 220 payloads from two live-verified imports. That layer is vendor-independent, so a per-project reimplementation only re-introduces bugs that were already found and fixed. This directory should hold at most the mapping from THIS project’s reader field names onto the canonical vocabulary — and once the reader emits canonical records directly, nothing at all.
• migrations/<project>/src/import/write-specs/
• migrations/<project>/src/setup/run-setup.js — setup execution entrypoint
• migrations/<project>/src/extract/run-extract.js — extraction entrypoint. It must accept --sample (extract a small slice only); that is what the sample-preview gate runs.
• migrations/<project>/src/import/run-import.js — the runnable import entrypoint (required; see “Runnable setup/extraction/import entrypoints” above). --dry-run drives the safe-validation pass through the same import code path.
• migrations/<project>/data/source-extract/ — extracted <entity>.ndjson record streams plus a manifest.json (the manifest is a single document, so it stays JSON)
• migrations/<project>/state/crosswalk/crosswalk.ndjson — authoritative local sourceId -> targetId crosswalk for native Wix entities
• migrations/<project>/state/attempts/write-attempts.ndjson — append-only write attempt journal for resume/reconciliation
• migrations/<project>/state/url-preservation/base-paths.json — entity-level public route intent from mapping urlPolicy
• migrations/<project>/state/url-preservation/url-ledger.ndjson — append-only URL preservation upsert log, one latest row per sourceStableKey + sourceRelativeUrl
• migrations/<project>/state/url-preservation/redirects.ndjson — planned redirect rows when source and target relative URLs are both known and differ
• migrations/<project>/state/url-preservation/unresolved.ndjson — public source URLs waiting on target route configuration, target lookup, or manual review
• migrations/<project>/logs/ — audit/error logs emitted by the shared runtime
• migrations/<project>/execution/review/import-plan.md
• migrations/<project>/execution/review/code-safety-review.md when SAFE_MODE=true or DRY_RUN=true
• migrations/<project>/execution/recovery-log.json — append-only standard recovery actions emitted by every resumed, partial, missing-only, failed-only, or deferred-only run

Recommended machine-readable artifacts:

• migrations/<project>/execution/setup-plan.json
• migrations/<project>/execution/execution-manifest.json
• migrations/<project>/execution/llm-handoff.json
• migrations/<project>/execution/live-import-summary.json
• migrations/<project>/execution/completion-report.json

Generated write spec contract

Generated writer code should be thin. It should primarily define deterministic per-entity write specs, then pass Wix-shaped objects into the shared runtime.

Each write spec should define, as applicable:

• entity
• mode: create | update | upsert | bulk_create | bulk_upsert
• create
• update
• lookup
• bulk
• batchSize
• concurrency
• retryPolicy
• throttlePolicy
• auditKeys
• crosswalkAuthority: "local"
• cmsMirror: "none" | "download" | "upload" | "download-and-upload"
• crosswalkStrategy
• reconciliationStrategy
• dependencyRefs
• verificationLevel: verified | unverified
• safeMode.replacePaths[]: resolved request-body paths copied from mapper safeModeReplacements[], with { kind: "email" | "phone", path } entries. Paths use the shared safe-mode grammar (field, field[], field.items[]) and are relative to the request body passed to the shared writer builder. The mapper is required to surface these replacements in both machine artifacts and review markdown; codegen should treat missing safeModeReplacements[] for an entity with outbound contact fields as a mapper contract failure, not as a cue to infer or invent replacements silently.

The transform layer should output Wix-shaped objects only. Request-envelope building and retry/reporting logic belong in the shared runtime.

When safe mode is enabled, generated import code must derive one safeModeOptions object per source record and pass it to every Wix writer call. The options must include origin entityType, deterministic origin entityId, safeModePhoneNumber, and the write spec’s resolved replacement paths. entityId must be the source ID or deterministic sourceStableKey, never the target Wix ID. If no deterministic origin identity exists, fail before writing.

Generated code must not query Wix by original email or phone values while safe mode is enabled. Any destination lookup keyed by contact data must use the same mock value that the writer will send, with local crosswalk state remaining the primary idempotency authority.

Execution artifact contract

Codegen should emit machine-readable execution artifacts alongside runnable code.

• execution/setup-plan.json: execution-ready rendering of setup work
• execution/execution-manifest.json: authoritative ordered task graph covering setup, extraction, transform/write execution, artifact refs, checkpoint ids, and plan version
• execution/llm-handoff.json: whether unresolved codegen/execution decisions remain

This manifest is the primary downstream contract for rp-execute-setup and rp-execute-import.

For every import task that handles public routed entities, include:

Do not set applyRedirects to true; the current phase records redirect plans but does not configure Wix redirects or site routes.

Import completeness accounting contract

Generated import runners must treat completeness accounting as a first-class runtime output. For each mapped entity class and meaningful subtype, the runner must produce an entityCompleteness row with these counters:

• extracted
• inScope
• attempted
• imported
• alreadyPresentByCrosswalk
• deferred
• failed
• skippedOutOfScope
• unexpectedSkipped

Subtype is required whenever source records with the same entity map through materially different write paths or deferral rules, for example simple vs subscription Stores products. deferred means intentionally not attempted with a recorded reason. failed means attempted and not successfully written. unexpectedSkipped is for in-scope mapped records that were skipped outside an accepted out-of-scope rule.

The generated runner must call or vendor skills/replatform/lib/completion-report.js and write both:

• execution/live-import-summary.json during/after live import for machine inspection
• execution/completion-report.json as the authoritative final outcome artifact

For each row, reconciliation is:

Any reconciliation mismatch, non-zero failed, non-zero deferred, or non-zero unexpectedSkipped count must appear in the report headline and in the deterministic human completion summary. Do not report a run as cleanly complete when source, import, and crosswalk counts do not reconcile.

Verifying Wix APIs

The primary control is the Target write contract above: call rp-target-wix‘s verified primitives when they exist. That adapter is where each stable shape is verified-once (by a real call) and where a Wix surface change is fixed in one place. When Wix has a native entity but rp-target-wix does not yet have a dedicated writer, codegen must generate a Wix REST call for that native entity via the adapter’s generic direct REST helper, log the missing writer, and call the RePlatform notification hook. Do not route to CMS merely because the writer is missing.

• If a Wix tool surface such as Wix MCP is available, prefer it to locate the endpoint and request/response shape. If the shape is common enough, add a dedicated primitive to resources/rp-target-wix/lib/wix-writers.js; otherwise generate a project-local native REST call through sendDirectRest and mark it UNVERIFIED.
• Confirm the shape with a real call, not a doc example — public examples have been wrong (e.g. lowercase Ricos plugin enums that 400). Cover the new primitive in tests/target-wix/contract-test.js so drift stays visible.
• Fallback when no Wix tool surface is available: rely on published Wix REST/SDK docs and conservative names. Mark the generated native call // UNVERIFIED: until a real call confirms it — never ship an unchecked Wix call to a user’s live site without surfacing it in the execution plan.

Runtime policy

Verify each Wix endpoint and field at codegen time. The // UNVERIFIED: marker is a fallback only for environments where no direct verification aid is available, not a way to ship unchecked calls that fail later on the user’s live site. Anything unverified must be surfaced explicitly in downstream artifacts before execution.

Missing writer policy

CMS fallback is for source concepts that do not have a suitable native Wix entity, or for native entities explicitly rejected because they cannot preserve fidelity or would cause unsafe side effects. CMS is not a fallback for a missing writer, and it is not a special default for coupons just because coupon scoping/restrictions need mapping.

When the mapping targets a native Wix entity and no dedicated rp-target-wix writer exists:

1. Generate project-local code that calls the native Wix REST endpoint through sendDirectRest.
2. Add a clear log line before the first use of that generated REST path.
3. Call notifyMissingWriter({ sourceEntity, wixEntity, method, path, reason }). The current implementation may be a no-op; the generated code must still call it.
4. Mark the path UNVERIFIED in execution/review/import-plan.md and the execution-plan report until a live/sandbox call promotes it.
5. Maintain the same idempotency rules as dedicated writers: crosswalk by source ID in local state and never dedupe by slug.

Codegen rules

• Keep reader and writer responsibilities separate.
• Keep setup, extraction, and import responsibilities separate.
• Do not generate a read-all-into-memory importer for the general case. The reader extracts to disk first; the importer consumes extracted files from disk.
• Make transforms deterministic and testable.
• Make generated write specs deterministic and declarative.
• Preserve source IDs for traceability, but do not assume native Wix target IDs can be preserved or client-assigned.
• Generate a stable selection layer before executing writes. Use the shared import-recovery.js runtime helpers where available, or vendor equivalent helpers into the project. Selection must:
  • apply --entity before entity write execution,
  • apply --source-type against deterministic source subtype fields such as WooCommerce type: "subscription",
  • exclude local-crosswalk hits in --missing-only mode and count them as alreadyPresent,
  • select latest failed attempt rows for --failed-only,
  • select latest deferred/needs-verification attempt rows for --deferred-only,
  • emit an execution selection summary before the first write.
• Native Wix creates must perform a local crosswalk lookup immediately before every create, even after pre-selection. If a crosswalk row is found, skip idempotently and count the record as already present instead of writing a duplicate.
• Every selective/resumed/partial run must append execution/recovery-log.json with a standard recovery entry containing: recovery id, timestamp, selection filters, reason, records selected, records attempted, imported, already present, failed, deferred, crosswalk changes, summary changes, operator-visible outcome, and links to detailed logs. Recovery artifacts are append-only; never overwrite or collapse prior recovery actions.
• Update execution/live-import-summary.json through a shared summary writer, not by hand-editing counters in recovery-specific code. The recovery-log entry must include the summary delta so final state can be reconstructed from standard execution artifacts.
• For public routed entities, consume mapping urlPolicy and initialize local URL preservation state before writes. Generate deterministic helpers, or vendor the shared url-preservation-state.js behavior, to:
  • write base-paths.json from entity-level route policies
  • derive sourceRelativeUrl from sourceUrlField or sourceBasePath + sourceSlugField
  • preserve the original sourceSlug before any normalization
  • record plannedTargetSlug before create/update
  • record actualTargetSlug and actualTargetRelativeUrl only after the Wix response or a safe target lookup proves them
  • append replacement rows to url-ledger.ndjson instead of editing rows in place
  • write redirects.ndjson only when both relative URLs are known and differ
  • write unresolved.ndjson with pending_target_route, target_url_missing, source_url_missing, or manual_review when a concrete preserved URL or redirect cannot be derived safely
  • never silently overwrite a source slug with a normalized slug
• URL ledger resume must replay the latest valid url-ledger.ndjson row by sourceStableKey + sourceRelativeUrl, derive expected rows from extracted source records and urlPolicy, merge target IDs from the local crosswalk, and query Wix only when the write spec declares the lookup safe.
• For native Wix entities, generate and use the shared local-state runtime (local-state.js) whenever the target API does not expose a client-controlled source-id field. Load state/crosswalk/crosswalk.ndjson at startup, rebuild missing or corrupt index caches from it, append a started row to state/attempts/write-attempts.ndjson before each native write, write the confirmed crosswalk row after success, then mark the attempt imported. Slug-based dedupe does NOT work — Wix rewrites slugs; never rely on it.
• Only fetch CMS ImportCrosswalk rows as a pre-execution seed for existing-site, legacy, recovery, or delta flows, and only when valid local crosswalk state does not already exist. Dedupe downloaded mirror rows by sourceStableKey and newest valid updatedAt. Never use CMS as the ordinary resume source after local state exists.
• Conditional-entity gating must match the verification artifact exactly. When an entity’s import is gated on a setup/setup-verification.json item (e.g. an unverified native path that setup must promote), the generated lookup must key on the same field the generated setup runner writes (requirement id vs step id — pick one and use it in both). A mismatched key silently reports “not verified” (fails safe into deferral, but would wrongly defer a passing entity).
• The optional CMS mirror collection name is ImportCrosswalk. If upstream artifacts still say MigrationRefs, normalize them to ImportCrosswalk in the generated code and note the normalization in execution/review/import-plan.md.
• Attach every related entity, don’t just create it. Creating a tag/category is not the same as linking it. Blog posts attach tags via tagIds (GUIDs) and categories via categoryIds on the draft-post create — collect the resolved ids and pass them, or the taxonomy exists on the site but postCount stays 0 (a builder bug we hit before).
• Taxonomy creates are not idempotent: treat 409 ALREADY_EXISTS as success AND resolve the existing entity’s id (e.g. listBlogTags) so it can still be attached — don’t drop it.
• Blog rich text is chunked for you. convertHtmlToRichContent transparently splits HTML over the 30k Ricos cap and merges the node arrays; pass full HTML, don’t pre-truncate or skip large posts. This is the blog path only — a blog post’s richContent really is a Ricos document. Stores products must NOT use it: their HTML goes in plainDescription and Wix converts it server-side (see the product-description rule above).
• Use the shared runtime for batching, retries, throttling, checkpoints, and audit/report emission rather than generating those mechanics ad hoc.
• Use the deterministic execution-state preparation contract before native writes. Do not let generated import code query CMS directly for resume; CMS mirror rows must be copied into local crosswalk state first.
• Do not hardcode secrets.

Headless storefront codegen rules

When the delivery mode includes a headless storefront (i.e. a wix-headless Astro project is generated alongside the import), apply these rules in addition to the general codegen rules above.

Cursor-paginated catalog queries (required). Generated shop and category pages must never call queryProducts().limit(N).find() once and stop. Wix Stores returns at most 100 items per page — a store with more than 100 products silently truncates the listing. Always generate a cursor loop in the shared catalog helper and call it from every page that lists products:

Apply the same pattern to category-filtered product queries (productsInCategory). This is a data-correctness requirement, not a UX preference.

Product count display (required). Every generated shop and category page must display the total product count (e.g. מציג X מוצרים for RTL Hebrew sites, or equivalent). The count is products.length after the cursor loop completes and must appear above the product grid.

client:load for primary CTA islands (required). Generated product detail pages must mount add-to-cart and cart-view React islands with client:load, not client:only. client:only omits the component from the server-rendered HTML entirely — it is invisible to crawlers and does not render until after full React hydration. client:load hydrates immediately on page load. client:only must never be the default for a primary call-to-action.

URL redirect file for WooCommerce sources (required). When the source platform is WooCommerce and the delivery includes a storefront, generate frontend/public/_redirects mapping old WordPress URL patterns to the new Astro routes. At minimum:

Also generate frontend/src/pages/product/[slug].astro (singular path) as a redirect shim to /products/[slug] (plural path). This prevents 404s on all inbound links from the old site. Derive the source URL pattern from the source profile captured during discovery; /product/ is the standard WooCommerce single-product path.

The _redirects file and redirect shim are required migration outputs, not optional polish.

When codegen must halt to LLM

Emit execution/llm-handoff.json with needsLlm: true when any of the following occur:

• a native Wix path exists but only an unverified write surface is available
• setup artifacts are insufficient to generate deterministic setup execution
• transform semantics remain unresolved
• a multi-pass relationship strategy cannot be derived deterministically
• the runtime contract required from rp-target-wix is missing for the mapped target

Output

Summarize which files were generated, which entities they cover, and any remaining implementation gaps.