SafeGate v6.5 / Agent-Ready Payment Verification Direction

Make receipt verification safe for AI agents.

v6.5 defines how an AI agent should request payment verification without receiving secrets, overbroad data, or unsafe unlock authority.

Agent-ready does not mean agent-controlled money movement. It means bounded, auditable, fail-secure receipt verification that an AI agent can safely consume.

30-Second v6.5 Summary

SafeGate agent-ready verification means: agent asks → SafeGate verifies → bounded response returns → audit event is recorded → unsafe states fail secure.

1. Why Agent-Ready Comes After Audit Trail

An AI agent should not be trusted just because it asks for access. SafeGate must first have audit trail, edge-case behavior, safe error handling, rate-limit direction, and privacy boundaries.

v6.0 to v6.4 created that foundation. v6.5 defines how agents can use receipt verification without bypassing the trust layer.

2. Agent Verification Flow

Step 1

Agent Requests Verification

The agent sends a receipt ID and limited context. It does not send or receive backend secrets.

Step 2

SafeGate Validates Input

Receipt format, known record, merchant boundary, and request scope are checked.

Step 3

SafeGate Verifies State

Payment state, receipt proof, access state, and merchant evidence are checked server-side.

Step 4

Bounded Response Returns

The agent receives only what it needs: result, code, state, boundary, and next allowed action.

Step 5

Audit Event Is Recorded

Every agent verification request becomes an audit-relevant event.

Step 6

Unsafe State Fails Secure

Invalid, ambiguous, replay-like, unauthorized, or rate-limited requests do not unlock access.

3. Proposed Agent Request Envelope

{
  "agentId": "agent_demo_001",
  "agentType": "merchant_assistant",
  "intent": "verify_receipt",
  "receiptId": "SG-RCPT-1781465776152",
  "requestedAction": "confirm_access_state",
  "merchantContext": "demo_merchant",
  "requestMode": "agent_ready_direction",
  "mustNotReceive": [
    "backend_secret",
    "service_role_key",
    "wallet_passphrase",
    "raw_provider_payload",
    "private_user_data"
  ]
}

4. Proposed Agent-Safe Success Response

{
  "ok": true,
  "statusCode": 200,
  "code": "AGENT_RECEIPT_VERIFIED",
  "receiptId": "SG-RCPT-1781465776152",
  "orderId": "TP-DEMO-273551",
  "network": "Pi Testnet",
  "verification": {
    "paymentState": "PAYMENT_FINALIZED",
    "receiptProof": "CREATED",
    "accessState": "UNLOCKED",
    "merchantRecord": "FOUND"
  },
  "allowedAgentAction": "CONFIRM_ACCESS_DELIVERED",
  "blockedAgentActions": [
    "CREATE_PAYMENT_CLAIM",
    "BYPASS_ACCESS_CONTROL",
    "REQUEST_SECRET",
    "OVERRIDE_MERCHANT_RECORD"
  ],
  "auditEvent": {
    "eventType": "AGENT_RECEIPT_VERIFIED",
    "actorType": "ai_agent",
    "result": "OK",
    "safeBehavior": "CONTINUE",
    "createdAt": "mock_timestamp"
  },
  "boundaries": {
    "productionPublicApiClaim": false,
    "productionApiKeyIssued": false,
    "backendSecretExposed": false,
    "piMainnetSettlementClaim": false
  }
}

5. Bad Agent / Unsafe Request Handling

Agent Behavior	Response Code	Safe Behavior	Audit Event
Agent asks for backend secret	403 AGENT_ACTION_FORBIDDEN	Deny secret access	AGENT_SECRET_REQUEST_BLOCKED
Agent probes unknown receipt IDs	404 RECEIPT_NOT_FOUND / 429 RATE_LIMITED	No enumeration hints; throttle if repeated	AGENT_RECEIPT_PROBING_DETECTED
Agent retries same receipt too aggressively	409 DUPLICATE_OR_REPLAY_ATTEMPT	Preserve state; no new unlock	AGENT_REPLAY_ATTEMPT
Agent asks to override access control	403 AGENT_ACTION_FORBIDDEN	Block override	AGENT_ACCESS_OVERRIDE_BLOCKED
Payment state is ambiguous	503 VERIFICATION_AMBIGUOUS	Fail secure; no unlock	AGENT_VERIFY_AMBIGUOUS_FAIL_SECURE
Internal error occurs	500 INTERNAL_ERROR_SAFE	No stack trace, no secret, no unlock	AGENT_VERIFY_INTERNAL_ERROR_SAFE

6. Agent Policy Rules

Least Data

Agent receives only the minimum fields required for verification outcome.

No Secret Access

Agent never receives service role keys, API keys, tokens, wallet passphrases, raw provider payloads, or internal logs.

No Autonomous Override

Agent cannot override access state, merchant record, receipt proof, or payment status.

Fail Secure

Ambiguous, delayed, replay-like, unauthorized, or suspicious requests do not unlock access.

Rate-Limited

Agent requests should be scoped by agent ID, merchant, API key, receipt ID, and behavior pattern.

Audit-Logged

Every agent verification request and blocked action becomes a minimal audit event.

7. Agent-Ready Does Not Mean Production API

This page defines agent-ready direction. It does not claim a production public API or unrestricted AI access.

Agent-ready means SafeGate can define how AI agents should consume verification safely within bounded contracts, rate limits, audit trails, and privacy rules.

8. V6 Progression

v6.5 Agent-Ready Direction

Current stage: AI agent receipt verification boundaries.

9. Public Boundary

SafeGate v6.5 is a public direction page. It does not claim a production public API, does not issue real API keys, does not create production merchant accounts, does not expose backend secrets, and does not claim Pi Mainnet settlement proof.

Current evidence remains Pi Testnet controlled post-payment trust flow.