SkillsandAgents.co and Agentic Resource Discovery (ARD)

skillsandagents.co already had an Agent Mode: a clean Markdown directory of every published skill and agent, written for a crawler or agent. An agent could already see what the site has to offer to accomplish work. It just had to know to fetch that page. An emerging standard now makes it easier for search engines and AI tools to discover agents.

ARD (Agentic Resource Discovery) is a consortium specification for how agents discover resources across sites through a single common interface.

What ARD adds

ARD is an early open protocol for sites with agents to publish a discovery endpoint and a manifest at a well-known path. An AI client holds a list of trusted endpoints. When it needs a capability, it sends the endpoint a natural-language query and gets back ranked catalog entries: an identifier, a display name, a type, a URL to connect to, and a relevance score. Then the client checks trust on its own and connects over the resource's native protocol, whether MCP, an API, or a skill install. This makes ARD useful because it lets clients discover and rank resources without having to know where to look.

Whenever I publish a new agent, the build engine defines the agent and turns each one into a catalog entry: a URN identifier, a display name, a type, and a URL pointing to the actual skill file. The catalog gets served as a static manifest at a well-known path, and a pointer in robots.txt helps a crawler find it. That makes the site discoverable through ARD, not just readable through the page I already had.

Second, search. A small function handles the POST request to the /search endpoint. It takes a query, scores each catalog entry with plain keyword matching, applies filters, and returns ranked results. This gives the discovery endpoint a simple way to surface the most relevant entries.

The consortium behind it

ARD is relatively new. Google published the spec on June 17, 2026, at v0.9, with launch contributors that read like a roll call of the industry: Microsoft, Hugging Face, GitHub, Nvidia, Salesforce, ServiceNow, Snowflake, Databricks, Cisco, and GoDaddy. It builds on the AI Catalog data model from the Linux Foundation's AI Catalog Working Group. What they released is two primitives: a static ai-catalog.json manifest at a well-known path on your domain, and a registry API that crawls those manifests and returns ranked matches to natural-language queries. It's deliberately decentralized, and anyone can run a discovery service, and clients decide which ones to trust.

Adding the spec to skillsandagents.co furthers our goal of providing agents that help others save time on mundane tasks.

You can learn more about Agentic Resource Discovery in the official spec.