# SENTINEL — Agent-Level Transaction Safety Oracle
Pre-execution transaction safety oracle for autonomous AI agents. Call it BEFORE signing any on-chain tx.

## Endpoint
POST /v1/guard
  Input:  {"chain": "base", "from": "0xAGENT", "tx": {"to": "0xCONTRACT", "data": "0x", "value": "0x0"}}
  Output: {"verdict": "SAFE|UNSAFE|UNKNOWN", "risks": [...], "reasons": [...], "signature": "base64-ed25519", "txDigest": "sha256:..."}

## Payment (x402 v1, USDC on Base)
  Price: $0.005 USDC per call.
  Flow: POST without payment -> receive 402 with 'accepts' (network=base, maxAmountRequired=5000, payTo=0xCf1d31020A7915421f6d66B9835Dcb6f422337E7).
  Then sign an EIP-3009 TransferWithAuthorization and retry with the X-PAYMENT header (use x402-fetch).
  SENTINEL verifies the signature locally (facilitator-free) and returns the signed verdict.

## Risk data (real, live)
  - Contract security: GoPlus token security API (honeypot, owner abuse, hidden owner, taxes, mintable, proxy).
  - Simulation: Alchemy eth_call from the agent's address (detects reverts).

## Discovery
  - Agent card: /.well-known/agent.json
  - Pricing: /pricing

## Siblings
  VeraData (verified LATAM data), Intelica (competitive intel), TrustBoost (PII sanitization).