Ready for Agentic Commerce: SKINBOT Publishes Its Agent Card

Commerce is gaining a new participant. Not a new channel and not a new format, but a new buyer-side actor: the AI agent that acts on a person's behalf. It holds the conversation, gathers requirements, calls services, compares options and brings back a justified result. The person still decides. The agent does the work.

For every product on the seller side, this raises one question: when an agent comes instead of a customer, what exactly will it find?

Today SKINBOT gives its answer. We have published our Agent Card: a machine-readable description of SKINBOT as a callable sub-agent, at the address where agents are taught to look.

/.well-known/agent-card.json

What the card says

SKINBOT is callable by AI agents as a skin analysis and selection sub-agent: an A2A skill with a verifiable verdict.

Translated from protocol language: when a universal agent, a marketplace assistant or a brand's own AI needs a skincare selection for a user, it does not need to become a skincare expert. It delegates. The agent keeps the conversation; SKINBOT runs the skin analysis inside the session, selects from the partner's live catalog, explains every product and returns a structured verdict.

Why the verdict is the point

An agent that relays a recommendation inherits responsibility for it. This is the part of agentic commerce that gets less attention than the protocols, and it is the part SKINBOT was built around.

The basis of every selection is declared in the response itself:

"ranking_basis":  "session skin profile + live catalog match"
"sponsored":      false
"brand_weighting": "none"

A calling agent can verify what the selection was built on instead of trusting a description of it. SKINBOT's revenue is not coupled to any brand's outcome, so the layer is indifferent to which product wins. For the partner and for the calling agent this reads as selection accuracy: the verdict is driven by the session's skin profile and the live assortment, nothing else.

And because SKINBOT is stateless, the delegation carries no data debt: the skin analysis exists inside the session and disappears with it. No profiles, nothing stored, aligned by construction with GDPR, EU AI Act, 152-FZ and UAE PDPL expectations.

Why this required no rebuild

The honest part of this announcement is how little we had to change. SKINBOT was designed as a callable layer from day one: one verdict engine behind four doors, API, QR, iframe and widget. In live retail deployments the engine already answers machine callers; a shelf QR code and an API call reach the same logic.

The Agent Card does not add a capability. It makes an existing one discoverable. The architecture was designed for agent-to-agent delegation; the card is the address plate on a door that was already there. The next integration surface is a JSON-RPC endpoint for direct agent calls, wrapping the same engine.

What to do with this

If you run a marketplace, a retailer or a brand and you are building your own AI agent: your agent does not need to learn skincare. It can call a neutral sub-agent and verify the verdict. The card is live, the engine is in production, and the details live on one page: skinbot.beauty/en/for-agents.

Commerce spent two years asking whether agents will buy. The more practical question is what they will find when they arrive. As of today, when a skincare selection is needed, they find a published card, a callable layer and a verdict they can check.

FAQ

What is agentic commerce? A mode of commerce where an AI agent acts on a person's behalf: it holds the conversation, gathers requirements, calls specialized services and assembles the result. The person still decides; the agent does the work.

What did SKINBOT publish? An Agent Card at /.well-known/agent-card.json: a machine-readable description of SKINBOT as a skin analysis and selection sub-agent, following the A2A discovery convention.

Is SKINBOT already reachable over the A2A protocol? The Agent Card is published and the architecture is designed for agent-to-agent delegation. The verdict engine already runs behind API, QR and iframe in live retail deployments; a JSON-RPC endpoint for direct agent calls is the next integration surface.

Why would an agent choose a neutral sub-agent? Because it inherits responsibility for the recommendation it relays. SKINBOT declares the basis of every selection in machine-verifiable fields, so the agent can check the basis instead of trusting a claim.