seo-ecommerce (M18)

Conditional commerce module: it runs only when commerce signals are detected, then audits (a) product structured data for Search and (b) whether an AI shopping agent can actually read and act on the store. Reference: references/schema-tier1.md (Product + Offer, Review/AggregateRating rows). Platform-specific ids — Shopify's Catalog and UCP defaults among them — live in references/platforms/shopify.md and are emitted by the platform-conditional checks, not from here.

Inputs

Work from the PageSnapshot named in your dispatch envelope: read parsed from <run_dir>/pages/<slug>.json (jsonld[], metas[], text_sample) and vertical from the envelope — the commerce gate is decided there; <run_dir>/site/discovery.json carries the probes for /.well-known/ucp and /api/ucp/mcp (status, content-type, body); <run_dir>/profile.json names the platform. Grep pages/<slug>.html for verbatim evidence; site artifacts live in <run_dir>/site/{robots.json,sitemaps.json,discovery.json}. Deterministic findings already emitted by audit.mjs are listed in <run_dir>/findings.deterministic.json — do not re-emit those ids; add model-judged findings only. If invoked directly with a URL/path and no snapshot exists, first run node "${CLAUDE_PLUGIN_ROOT}/scripts/snapshot.mjs" <target> --out "${CLAUDE_PLUGIN_DATA}/runs" and use the printed snapshot path.

Audits

Working from the PageSnapshot (parsed_rendered when render.used is not none, else parsed):

1. Commerce gate: detect commerce signals (price/currency text, add-to-cart/buy controls, Product/Offer schema, og:type=product). If none, every M18 finding is not_applicable (severity 0) — do not force product schema onto non-commerce pages.
2. Product + Offer: Product has name, image; Offer has price, priceCurrency, availability (schema.org/InStock etc.). Per references/schema-tier1.md.
3. AggregateRating / Review: if ratings are visibly shown, expect AggregateRating (ratingValue, reviewCount/bestRating) and/or Review (author). Never mark up ratings that are not visible on the page.
4. Page → schema → feed consistency: schema price/priceCurrency/availability must match the visible on-page values and the merchant feed. Mismatches risk disapproval and lost rich results.
5. Faceted navigation: filter/sort URLs (?color=, ?sort=) should canonicalize to the clean product/category URL or be noindex, to avoid index bloat and duplicate-content dilution.
6. Variants / ProductGroup: multi-variant products should use ProductGroup with hasVariant and variesBy, not duplicate standalone Product blocks per SKU.

Agentic commerce readiness (ecommerce only)

Shopping agents do not read a product page the way a shopper does: they look for a machine-readable merchant profile, a catalog they are allowed to enumerate, and a feed whose fields parse. These checks run only when the ecommerce vertical is active, and they populate the conditional AI category for M18 (weight 6) alongside the Search category.

7. UCP merchant profile — /.well-known/ucp must be publicly reachable (200, no auth), valid JSON, with ucp.version as YYYY-MM-DD and services/capabilities as objects keyed by reverse-DNS names mapping to arrays of {version, spec, transport?, endpoint?, schema}. When the store is not an ecommerce vertical this file is M21's (seo-ai-discovery); on a store, M18 owns it.
8. Catalog eligibility — for each sampled product, check the properties a catalog needs to list it at all: a title, at least one image, a price greater than zero, a stable identifiable product URL, and a product that is not hidden/unavailable in the storefront. Report per-product results, not a single verdict.
9. Agentic-checkout feed (ACP) — when the user supplies a feed with --feed <path>, lint it: the required fields, a price written as an amount plus a currency code ("79.99 USD"), an availability value from the allowed enum, GTINs that pass the mod-10 check, absolute URLs, and no duplicate item_id. With no feed supplied the finding is needs_api — never a pass.
10. Feed ↔ page price — where a feed is available, compare its price and availability against the rendered page and the schema. Three sources, one truth.
11. Merchant Center / platform hints — a verified merchant feed is how Google receives authoritative product data; note whether one is plausibly present (structured data alone is not a feed) and whether the platform declares a catalog integration.

Fixes

• AUTO (fixable: auto): generate complete, valid Product/Offer/AggregateRating JSON-LD inferred from the DOM (name, image, price, currency, availability, visible rating) inside a single @graph with a stable @id. The block is a diff for fix.
• PROPOSED (fixable: proposed): canonical/noindex tags for faceted URLs, ProductGroup restructuring, and feed-field corrections — drafts requiring per-item accept.
• ADVISORY (fixable: advisory): price/availability mismatches between page, schema, and feed; UCP profile publication; catalog enrolment. The store backend, the feed, and the platform are the sources of truth, so the tool never writes these. Never invent prices, currencies, availability, ratings, GTINs, or endpoints — ask the user or leave a clearly-marked TODO placeholder.

Verification

• Offline: node "${CLAUDE_PLUGIN_ROOT}/scripts/validate-jsonld.mjs" --snapshot <pages/<slug>.json> (or --url <u>) (schema_validator) — JSON validity + required Product/Offer properties.
• Feed lint: node "${CLAUDE_PLUGIN_ROOT}/scripts/acp-feed-lint.mjs" --feed <path> (JSONL, CSV, or TSV) — required fields, price format, availability enum, GTIN mod-10, absolute URLs, duplicate item_id (schema_validator).
• UCP/catalog probes: node "${CLAUDE_PLUGIN_ROOT}/scripts/ai-discovery.mjs" --snapshot <pages/<slug>.json> re-reads site/discovery.json and validates the profile shape (schema_validator).
• Tier 1: Google Rich Results Test / schema.org validator (rich_results_api) for merchant-listing eligibility.
• When the required tier (feed, API, merchant account) is unavailable, status is needs_api, never a false pass.

Findings

Emit findings per schema/finding.schema.json. On a non-commerce page every id is not_applicable at severity 0. Severity policy: 5 is reserved for catastrophic, eligibility-killing facts at site/template scope; 4 major · 3 moderate · 2 minor · 1 cosmetic · 0 informational — and only an established severity-5 fail in an active category can cap a score (references/scoring-model.md).

Product data — axis both:

• M18.offer.missing_price — Product without Offer.price/priceCurrency (fail, severity 4, fixable: auto, confidence established).
• M18.offer.price_feed_mismatch — schema price differs from the on-page or feed price (warn, severity 3, fixable: advisory, confidence directional; needs_api when no feed is supplied).
• M18.facets.uncanonicalized — indexable faceted URLs with no canonical and no noindex (warn, severity 3, fixable: proposed, confidence directional).
• M18.variants.no_productgroup — per-SKU duplicate Product blocks instead of ProductGroup + hasVariant (warn, severity 3, fixable: proposed, confidence directional). Counts declarations only: a page that already models variants with hasVariant never fires it, however many variants it lists.
• Catalog eligibility judges the ProductGroup itself, and prices it from the variants' offers when the group carries none of its own — a group whose variants each carry a price is not "missing a price".

Agentic commerce readiness — ecommerce only:

• M18.agentic.ucp_profile_missing — no /.well-known/ucp on a store (warn, severity 3, axis ai, fixable: advisory, confidence directional).
• M18.agentic.ucp_profile_invalid — the profile exists but the version, services, or capabilities shape does not parse (warn, severity 3, axis ai, fixable: advisory, confidence directional).
• M18.agentic.ucp_not_public — the profile is behind auth, a redirect to login, or a non-JSON content type, so no agent can read it (fail, severity 3, axis ai, fixable: advisory, confidence directional).
• M18.agentic.catalog_eligibility — per-product catalog check: pass when title, image, price > 0, identifiable URL and visible storefront status all hold; warn when one is missing; fail when the product could not be listed at all (severity 3, axis both, fixable: proposed, confidence directional).
• M18.agentic.acp_feed_errors — feed lint failures, one row per error class with the offending line (fail, severity 3, axis both, fixable: proposed, confidence directional; needs_api whenever --feed was not supplied).
• M18.agentic.feed_page_price_mismatch — the feed price or availability disagrees with the rendered page (warn, severity 3, axis both, fixable: advisory, confidence directional).
• M18.agentic.merchant_center_hint — a verified merchant feed is the documented path for authoritative product data reaching Google's shopping and AI surfaces (pass, severity 1, axis both, fixable: advisory, confidence established; informational — we cannot see the merchant account).
• M18.agentic.shopify_catalog_declared — the platform declares a catalog integration in its UCP profile (pass, severity 1, axis ai, confidence directional — a platform default, not merchant merit).

Each finding: evidence.observed quotes the page, the profile, or the feed line verbatim (e.g. the rendered price text, the offending ?filter= URL, item_id and column); verification.reproduce is a runnable command; expected_impact is banded + confidence-tagged (no naked %).

Honesty

• AggregateRating/Review markup does not guarantee star rich results — Google shows them at its discretion and gates them by policy; never promise stars from markup alone.
• Faceted-URL hygiene reduces crawl and index waste but is not a direct ranking factor — directional, not a guaranteed gain.
• Agentic commerce is early and unevenly documented. UCP, catalog enrolment rules, and agentic checkout are vendor programmes that change fast, and no vendor publishes a "readiness score". Every M18.agentic.* finding is directional — the only established claim in the set is that a merchant feed is the documented channel for authoritative product data.
• On Shopify, /.well-known/ucp, agents.md, and the agentic discovery files ship by default. Their presence on a Shopify store is a platform default, not evidence the merchant did anything; their absence on a non-Shopify store is a gap, not a failure.
• A feed lint failure is a fact about the file, not about the store's ranking. Report the row, the field, and the rule; do not extrapolate a revenue or visibility number.
• Keep schema strictly consistent with the visible page and the feed. Fabricated or feed-mismatched values invite merchant disapproval, not a win — and never invent a GTIN.