An Agent Lands on Your Homepage — What Can It Actually Read?

When an AI agent visits a vendor's website, it swiftly scans for specific data formats rather than lengthy human-readable content. The vendor agentbadge.xyz has recently improved its readiness by generating a set of machine-readable manifests, all with a 200 OK status, that are free to access and serve without authentication. In just 200 milliseconds, the agent can retrieve eleven of these manifests, providing crucial information about the site's identity, capabilities, and payment methods.

These manifests include /.well-known/agent-card.json, containing the agent's name, provider, supported interfaces, and skills; /.well-known/api-catalog RFC 9727 linkset, which lists every API entry; /.well-known/erc8004-agent.json, an on-chain identity registration for the agent; /.well-known/mcp/server-card.json, detailing MCP capabilities and tools; /.well-known/agent-evaluation.json, a verification ladder of claims and references; /.well-known/owner-questions.json, answering questions about the site's operators and jurisdiction; /.well-known/did.json, a document linking the site to its DID; /.well-known/did-configuration.json, signed domain linkage using VC-JWT; /.well-known/jwks.json, a real Ed25519 key for secure communication; /.well-known/security.txt, an RFC 9116 document containing security information; and /.well-known/security.txt, an llms.txt file with machine-readable entry points for LLMs.

The manifests are generated from live sources, such as openapi.ts, route config, blog-data.ts, and SKU catalog, ensuring they are always up-to-date and accurate. CI drift-check ensures that any manual edits to the manifests will fail the build, preventing drift from the source code. This automated approach guarantees that the vendor's website always presents accurate and up-to-date information to AI agents, enabling them to quickly assess the vendor's identity, capabilities, and payment methods.