muse-sdk-agentic-commerce

muse-commerce-audit opens your shop in real headless Chromium (Playwright) and checks whether a standards-following AI shopping agent can discover, understand and act on the page, using only public web standards. It can also ask a real Muse Code agent to browse and shop the page itself through the official @muse-code/sdk (developer preview), as a second opinion from a frontier agent.

Honest caveat, up front: Meta has not published the rules its consumer Muse shopping agent applies, so nothing here reproduces its internal logic. Two things this measures instead: (1) machine-readability against public standards (robots.txt RFC 9309, schema.org, the ARIA accessibility tree), which is what any well-behaved shopping agent builds on; (2) what Muse Code, a real frontier agent driven through the public SDK, can actually do with the page. Treat the Muse verdict as a second opinion, not as the shopping product's verdict.

No Universal Commerce Protocol endpoints are needed or checked.

---

What it checks

Check	What it looks at
bot-wall, http-status	403/429/CAPTCHA/Cloudflare/DataDome when loaded by headless Chromium
robots-*	Whether robots.txt blocks Meta's documented crawlers: meta-externalagent, meta-externalfetcher (error), facebookexternalhit, FacebookBot (warning). The shopping agent's own browsing UA is unpublished, so this is a proxy.
product-schema, offer-*, product-*	Product/Offer JSON-LD: name, price, currency, availability, image, sku/gtin
price-mismatch	JSON-LD price is visible in the rendered text
no-js, no-js-text	Product data present in the server HTML without running JavaScript
purchase-control, unnamed-control	An "Add to cart / Buy" control findable by ARIA role + accessible name
open-graph	Open Graph metadata