For agent builders

Where Human Seal sits in an agent harness

A harness is everything around the model: what it may see, what it may touch, what it remembers, what counts as proof, and when it must stop. Human Seal covers two of those jobs, and only the part of them that has to be believed by someone outside your system.

The jobs, and which one is ours

Most write-ups name about seven jobs for a harness: define the task, choose the context, give the model tools, keep state, check results, authorize actions, and keep a trace. You build the first five yourself, and we do not try to do them for you.

Authorize. Before an agent does something that matters, a named person approves that exact action. Human Seal records the approval and binds it to the action by hash, so the approval cannot be moved onto a different action.
Trace. After the action, there is a signed record of who approved what, when, and how long they looked. That record is a credential you can hand to someone.

The gap this fills

Authorization and traces normally live inside the builder's own system. They are true for you, and nobody else can check them without trusting you or asking your server. A credential signed with a published key can be checked by anyone with one small file and no call to us. You can sign with your own key, and the format is open.

You do not need to replace what you already log. Keep it. Add the receipt for the moments where someone outside your project has reason to ask whether a person approved.

See one, then check it

The specimen page holds a real approval with its credential, the published issuer key and the exact action. Your browser checks it and nothing is sent anywhere. The outside verification page has the standalone verifier and what is not built yet.

This page describes where the product sits. It is not an endorsement from anyone, and FLOCORE both issued the specimen and wrote the standard, so read the limits before you rely on it.