Skip to content

BLOG / llms-txt-vs-agents-md-vs-sku-md.mdx

llms.txt vs agents.md vs SKU.md: choose the right responsibility

A practical matrix for Sitemap, Product JSON-LD, llms.txt, agents.md, SKU.md, live catalogs, MCP, UCP, and Checkout.

Published

By SKU.md Editorial Team

Original specification version: v0.10

Do not choose one machine-readable file to do every job. Use each surface for the responsibility its publisher and runtime can actually support, then connect them with explicit links.

SKU.md means merchant-hosted product discovery and stable product knowledge: agents can find products, understand sourced facts, and know where to recheck live commerce. It is not a new checkout protocol.

Responsibility matrix

Surface Discovery Summary or instructions Product semantics Stable knowledge Live state Transaction authority
Sitemap URL inventory No No No No No
Product JSON-LD Page entity Product fields Product/Offer vocabulary Limited to page data Observed page values No
llms.txt Curated readable links Site overview Not commerce-specific Unstructured No No
agents.md Repository or agent entry Operating instructions Not commerce-specific Instructions, not product evidence No No
SKU.md Product document graph Sourced product context ProductGroup, Variant, SKU, GTIN Yes Points out where to recheck No
Catalog / MCP / UCP Capability discovery Structured responses Current Product and Variant Only as the live system provides Yes Action-dependent
Checkout / order system Purchase entry Binding disclosures Selected line items No Revalidated Yes

Google’s Product structured-data guide explains how product page markup can qualify for richer search appearances. That is a different contract from a SKU.md knowledge document and still does not guarantee display.

Discovery does not establish authority

A Sitemap can reveal a URL without describing its meaning. llms.txt can curate documentation without proving the linked claims. agents.md can tell a coding agent how to test a repository without becoming a catalog. Product JSON-LD can express an Offer found on the page, but the checkout system can still return a different contextual total.

SKU.md connects stable identity and reviewed knowledge to these surfaces. It should never convert a discovery hint into permission, freshness, or transaction authority.

A selection workflow

  1. List the question: URL discovery, instructions, product identity, evidence, live conditions, or purchase.
  2. Choose the narrowest source that owns that question.
  3. Link stable knowledge to current Catalog or storefront verification.
  4. Fail closed if Product, Variant, SKU, or GTIN cannot be matched.
  5. Recheck conditions and obtain user confirmation before Checkout or an order action.

This workflow permits several files to coexist. Duplication is acceptable only when ownership stays explicit and values are checked for consistency.

Common combinations

Publisher need Minimal combination
Documentation discovery Sitemap + llms.txt + Markdown pages
Coding-agent instructions Repository README + agents.md
Product page semantics Visible product page + Product JSON-LD
Sourced durable product context SKU.md + source URLs + live verification link
Agent-driven purchase Stable knowledge + negotiated Catalog/MCP/UCP + Checkout

Shopify’s own 2026 resources illustrate the need for precise attribution: its /agents.md, /llms.txt, and /llms-full.txt behavior is documented for Shopify stores. That evidence must not be generalized into Shopify support for SKU.md.

Further reading

Read What is SKU.md? for the category definition and AI commerce authority handoffs for runtime ownership.

Next step

Review the project scope