Agents are about to start dealing with other companies’ agents.
Not through a registry or a standard — the way the web already works: an organization publishes, at a /.well-known path, a plain description of what it can do and on what terms. Another company’s agent discovers it and acts. This page hands you the smallest concrete version of that shape — and lets you see what an agent sees on any domain.
Inspect a domain
Walk any domain the way an agent would, and see whether what it publishes fits the v0 shape.
Why an inspector?
Publishing is just files on your own domain — so how do you know an agent can actually read them? One missing field or wrong path and it silently breaks, and you only find out when someone’s agent fails. Paste your domain here and see exactly what an agent sees — the same way a feed validator or a robots.txt tester lets you check your work before shipping.
It stores nothing and lists no one. It is not a directory and agents never query it to find providers — that would be the central registry this whole idea avoids. Discovery stays decentralized; this is only a mirror for checking your own file against the v0 shape.
Describe, don’t register
Discovery stays decentralized: an agent reads robots.txt, follows the Agent: line to /.well-known/agent, then each capability. No central directory to join.
Meaning, not a global vocabulary
Capability ids are the organization’s own local names. A model matches a request to a capability by reading its description — so the semantics stay loose and only a thin envelope is pinned.
v0, on purpose
Two independent parties can’t interoperate over nothing, so there is a small shape. It is versioned and provisional — a published file carries a spec_version so it can evolve without breaking anyone.