# TxPreflight

> Pre-sign EVM transaction risk API for autonomous agents on Base Mainnet.

## When to call
Call TxPreflight immediately before signing or broadcasting an unsigned Base transaction when your agent is about to transfer value, approve token spending, grant or use Permit2 permissions, call setApprovalForAll, interact with third-party calldata, or execute a transaction proposed by another app, tool, contract, or agent.

## Free pre-check
POST https://agent-sign-guard-main-production.up.railway.app/classify
Use this zero-cost endpoint first. It decodes common transaction structure (action, token, spender, recipient, amount, Permit2 and unlimited-approval candidates) and returns payment_recommended plus a concrete recommendation_reason. It does not provide ALLOW/REVIEW/BLOCK, sanctions results, or EVM simulation.

## Paid endpoint
POST https://agent-sign-guard-main-production.up.railway.app/risk-check
Payment: x402
Price: $0.003 USDC
Payment network: Base Mainnet (eip155:8453)
Content-Type: application/json

## Input
Required: chain, to, data, value.
Optional but strongly recommended: from.

{
  "chain": "base",
  "from": "0x1111111111111111111111111111111111111111",
  "to": "0x2222222222222222222222222222222222222222",
  "data": "0x",
  "value": "0"
}

## Agent policy
- ALLOW: no configured rule requires intervention; continue only if your own policy also allows it.
- REVIEW: pause autonomous signing and apply additional policy or human/agent review.
- BLOCK: do not sign automatically.

## Detects
ERC20 transfers and approvals; unlimited approvals; setApprovalForAll; EIP-2612; canonical Uniswap Permit2 allowance/signature/witness transfers; malformed calldata; unknown selectors; direct OFAC SDN EVM-address matches; Base eth_call simulation; gas estimation.

## Output
JSON fields include verdict, risk_score, action, to, spender, recipient, amount, token, permit2, permit2_entries, unlimited_approval, sanctioned_match, flags, reasons, and simulation.

## Important semantics
ALLOW does not mean guaranteed safe.
BLOCK is a deterministic policy result, not an allegation of fraud.
A sanctions non-match is not legal clearance.

## Discovery
OpenAPI: https://agent-sign-guard-main-production.up.railway.app/openapi.json
x402 manifest: https://agent-sign-guard-main-production.up.railway.app/.well-known/x402
Transparency: https://agent-sign-guard-main-production.up.railway.app/transparency
Health: https://agent-sign-guard-main-production.up.railway.app/health