trust-agent.io

Which live agents can do X — and which of them are safe? Searches the skills, domains and names ERC-8004 agents declare, across the registrations this service sweeps continuously, and returns only the ones answering. Each row carries the endpoint to call, the wallet a payment reaches, and that wallet's risk band and sanctions verdict, so choosing between candidates is one call rather than a screening per row.

Is this x402 endpoint safe to pay? Reads the endpoint's own payment challenge under BOTH protocol versions — v1 body and v2 PAYMENT-REQUIRED header, so a live v2-only service is not reported as dead — extracts the payout wallet it advertises, and screens that wallet: OFAC sanctions, sybil-filtered ERC-8004 reputation, on-chain behaviour. Flags an endpoint that names different recipients under the two versions. Scores the wallet, not the service behind it, and says so.

Is this counterparty safe to transact with — to pay, or to accept payment from? Screens the address against the OFAC sanctions list, then SYBIL-FILTERS its ERC-8004 reputation: reviewers are clustered by first funding source and capped at one vote per cluster, so manufactured feedback from wallets funded out of one purse counts once. Returns an explainable 0-100 trust score with a proceed / hold / block recommendation, and reports insufficient_data rather than guessing when the evidence is not there.

Should you approve this contract? Everything in quick, plus static analysis of the bytecode: privileged functions (mint, pause, blacklist, arbitrary withdrawal), proxy and upgradeability detection, dangerous opcodes, DeFi protocol classification and source-verification status.

What is this agent's reputation actually worth? 92.6% of ERC-8004 feedback on Base is coordinated. This clusters reviewers by funding provenance and caps each cluster at one effective vote, returns every record with its evidence class and flag reasons, and refuses to average ratings that use incompatible scales.

Has anything changed about counterparties you already screened? Re-checks a list of addresses against OFAC designations and delistings since a timestamp you supply, so a verdict you acted on last month does not go quietly stale. Stateless: no watchlist is stored and nothing is sent to you — pass the returned checkedAt back as `since`.

Has the sanctions feed moved since you last looked? Every OFAC designation and delisting since a timestamp you supply, with no address list to send — match the feed against your own counterparty book locally, so the service never learns which addresses are yours. The trivial cron: poll this, pass the returned checkedAt back as `since`.

Where did this address get its money? Walks the funding chain upward and returns the account that ultimately paid for the wallet, plus a cluster id. Two addresses funded from the same account are not independent parties, however new and unremarkable each looks alone — the strongest cheap signal there is that an identity was manufactured. Works on any address, not only registered agents.

The filtered ERC-8004 directory: which registered agents are actually alive, and which accept x402. Around a third of registrations on Base carry both a compliant registration file and a declared service, and roughly one in twelve of those declares x402 — this returns them, rather than making you screen eighty thousand addresses one at a time to find them. GET /v1/registry publishes the current split, free. 100 agents per page.

Screen up to 10 addresses in one call: OFAC sanctions check, trust score, risk band and recommendation for each. For vetting a counterparty set before a run of payments, maintaining an allowlist, or auditing wallets you already pay.