agentic-commerce-kit

A Claude Code skill that audits an ecommerce store for agentic commerce readiness — whether ChatGPT, Claude, Perplexity, and Gemini's commerce agents can find, parse, and recommend its products.

/audit-agentic-commerce https://your-store.com

Output: score / 9, per-criterion findings, ranked fix list.

---

Why

Agentic commerce already routes billions of shopping queries. The agents that do the routing (ChatGPT Shopping, Perplexity Buy with Pro, Google AI Mode) use a different stack than classic SEO:

• They identify themselves via specific user agents (GPTBot, PerplexityBot, ClaudeBot, etc.) that a merchant's robots.txt might silently block
• They rely heavily on structured data — Product, AggregateRating, FAQPage, Organization JSON-LD
• They cannot execute arbitrary JavaScript on a merchant's PDPs
• They favor pages where title, description, and schema together answer "does this product match what the shopper is asking for?"

The rubric this kit uses was developed for the Nohi Agentic Commerce Readiness Index Edition 1 (50 Premium DTC Beauty brands, April 2026). Median score: 3/9. Zero brands scored 7+. This tool lets you audit yours.

Install

# Clone and place the skill in your project:
git clone https://github.com/zhao-hanbo/agentic-commerce-kit.git
cp agentic-commerce-kit/skills/audit-agentic-commerce/command.md \
   ./.claude/commands/audit-agentic-commerce.md

# Or install globally (user scope):
mkdir -p ~/.claude/commands/
cp agentic-commerce-kit/skills/audit-agentic-commerce/command.md \
   ~/.claude/commands/audit-agentic-commerce.md

Requires Claude Code. No other dependencies.

Use

/audit-agentic-commerce https://www.summerfridays.com

Takes ~60-90 seconds.

The rubric

Nine criteria, each binary (0 or 1). Maximum 9.

#	Criterion	Area
C1	AI-crawler access in robots.txt	Crawlability
C2	PDP server-side renders (no JS required)	Crawlability
C3	Sitemap exists, fresh (<90d), lists products	Crawlability
C4	Product JSON-LD on PDPs (name, image, brand, offers, identifier)	Structured data
C5	AggregateRating or Review JSON-LD with real counts	Structured data
C6	FAQPage or HowTo JSON-LD on at least one page	Structured data
C7	Organization JSON-LD on homepage (with sameAs socials)	Structured data
C8	PDP titles include use case or differentiator	Content quality
C9	PDP descriptions answer ≥3 of: skin type, comparisons, ingredient rationale, sizing, usage, counter-indications	Content quality

Tier labels:

• 8-9 → Agent-Ready
• 6-7 → Partially Ready
• 3-5 → Early Stage
• 0-2 → Not Ready

Full methodology: Nohi Readiness Index — Edition 1 Methodology.

Sample audit: Summer Fridays

Run on https://www.summerfridays.com, April 2026. Default 3-PDP sample. Schema checks via source-level curl fetch of raw HTML.

Score: 5/9 — Strong Foundations, Variable Schema Coverage

Area	Score	Criteria passed
1. Crawlability	3/3	C1, C2, C3
2. Structured data	1/4	C7
3. Content quality	1/2	C8

What works (source-level verified):

• ✅ C1 Crawler access — robots.txt allows all 8 tested AI bots (GPTBot, PerplexityBot, ClaudeBot, anthropic-ai, Google-Extended, Amazonbot, CCBot, FacebookBot). No matching Disallow rules.
• ✅ C2 PDP server-side render — across the 3 sampled PDPs, raw HTML contains product title, price, and description without requiring JS.
• ✅ C3 Sitemap freshness — valid sitemap index with <lastmod> within the past 90 days.
• ✅ C7 Organization schema — homepage ships a complete Organization JSON-LD block: name="Summer Fridays", url, and sameAs array with 5 social-profile URLs.
• ✅ C8 PDP titles — across the 5 sampled titles, 4+ include specific use cases or differentiators ("Jet Lag Mask", "Cloud Dew Gel Cream Moisturizer Deluxe Sample").

What's missing (source-level verified):

• ❌ C4 Product schema — across the 3 sampled PDPs, 1 of 3 (/products/cloud-dew) has no Product JSON-LD at all — only Organization schema ships on that page. The other PDPs sampled (e.g. /products/jet-lag-mask) ship a valid Product block with name, image, brand, offers.price, offers.availability, sku, and gtin. The finding is variance between templates, not absence everywhere. Fix: audit the PDP Liquid templates to identify which variant ships without Product schema and restore the schema snippet. Verify across a broader sample with --confidence=high.
• ❌ C5 AggregateRating schema — across the 3 sampled PDPs, aggregateRating presence varies. On /products/jet-lag-mask it appears in a separate Product block (ratingValue 4.5, reviewCount 2771) rather than inside the main Product block — machine parsers may or may not associate them. On /products/cloud-dew no aggregateRating is emitted. Fix: consolidate the split Product blocks so ratings live on the same Product as the product fields, and ensure aggregateRating emits on every PDP where the review count is non-zero.
• ❌ C6 FAQPage / HowTo schema — /pages/faq was fetched and contains only Organization schema, no FAQPage. Homepage and other auxiliary pages also lack FAQ/HowTo markup. Fix: Summer Fridays publishes extensive "How to use" content per product — wrap this content in HowTo JSON-LD on PDPs and FAQPage JSON-LD on /pages/faq. See schema.org/HowTo.
• ❌ C9 PDP descriptions (evaluated on visible-text via WebFetch) — across the 5 sampled PDPs, descriptions are largely marketing-led with limited buyer-context cues (comparison to adjacent SKUs, skin- type rationale, counter-indications). This finding is based on rendered-response text; the descriptions may include these cues below the fold or in expandable sections that weren't fetched. Fix: expand PDP descriptions with explicit buyer-context blocks (who it's for / how it compares / when to use / what it isn't).

Schema hygiene flags (present but malformed / worth fixing):

• Split Product blocks on /products/jet-lag-mask — one block holds product fields (name, sku, gtin), a separate block holds aggregateRating. Google's Rich Results parser may not associate the rating with the product. Recommended: consolidate into a single block.

Priority top 3 (biggest leverage first):

1. Restore Product + AggregateRating schema on every PDP. Consolidating the split blocks and adding Product schema to the PDPs currently missing it would move C4 and C5 toward PASS across the sample. Lightweight if the Liquid template already has the fields; heavier if the missing-schema PDPs use a different template variant.
2. Add FAQPage / HowTo schema (C6). Summer Fridays has strong how-to content on PDPs; the markup layer is the gap. Effort depends on whether how-to content is structured in a reusable page section.
3. Expand PDP descriptions with buyer-context cues (C9). The highest-leverage change because it affects every product, but also the only one that requires editorial rather than template work.

Implementing 1-3 would move this brand materially upward on the framework. Final score depends on implementation quality and broader content improvements.

Disclosure: this audit scored 3 PDPs (via source-level curl fetches on 2026-04-19), 2 auxiliary pages (homepage, /pages/faq), plus robots.txt and sitemap.xml. Findings are sample-based. It does not audit uniqueness (C10, deferred) or GEO (geographic/local retrieval). Schema variance in sampled PDPs does not guarantee absence sitewide — rerun with --confidence=high or --full for broader coverage.

Related work

• Nohi Readiness Index — Edition 1 — full 50-brand scoring + notable findings.
• Finland Agent Test — the qualitative counterpart: what ChatGPT actually returns when you ask it to shop in an underserved market.
• Nohi blog — ongoing research on agentic commerce infrastructure.

Further reading on the agentic commerce stack

The protocols and surfaces this audit targets:

• OpenAI Agentic Commerce Protocol (ACP)
• Google Agent Payments Protocol (AP2)
• A2A — Agent-to-Agent protocol
• Stripe Agent Toolkit
• Shopify Roast — Shopify's agent workflow framework
• MCP specification

Contributing

This is v1: one skill, one flagship use case. Planned additions (depending on community feedback):

• /fix-product-schema — walks the user through applying C4 fixes in their Shopify theme
• /generate-faq-jsonld — takes a store's existing FAQ page and produces FAQPage JSON-LD ready to paste
• Cursor and Cline rule variants of the readiness audit

File an issue with what you'd find useful. No CONTRIBUTING.md yet — reach out directly.

License

MIT. See LICENSE.

Author

Built by Zhao Hanbo while building Nohi — the marketplace connecting DTC merchants to agentic commerce surfaces.