# Interacting with AgentFide (for agents)

AgentFide decides whether to trust a transacting agent. Call it before money moves.

## Quickest path

1. Discover this service via `/.well-known/agent-card.json` (AgentFide's signed A2A card)
   or the advertised MCP endpoint.
2. Call `POST /v1/assess` with the counterparty agent, the mandate you are acting under,
   and the action you intend to take.
3. Read the `decision`: `allow` (proceed), `review` (hold for a check), `deny` (do not proceed).
4. After the transaction, call `POST /v1/outcome` so the reputation graph learns from it.

## assess request (shape)

```json
{
  "agent": "https://counterparty.example/.well-known/agent-card.json",
  "acting_for": { "principal": "operator_123", "mandate": { "limit": 500, "currency": "USD", "expires_at": "2026-12-31T00:00:00Z" } },
  "action": { "type": "purchase", "amount": 42.0, "currency": "USD", "counterparty": "seller_789", "description": "API credits" },
  "context": { "marketplace": "vaultplane", "idempotency_key": "ord_abc" }
}
```

## What you get back

A `decision`, plus the four checks (`identity`, `authority`, `reputation`, `recourse`),
an `assessment_id`, and `expires_at`. Present the assessment to your counterparty as a
bona fide credential.

## Notes

- Reputation is public and portable. Transaction details are private by default.
- An agent can never buy a better score.
- Standards: A2A signed agent cards, Model Context Protocol, AP2 mandates.