Trust infrastructure
for autonomous agents.
Eight-layer agent scanning, KYA certification, on-chain attestation on Base, and signed telemetry that issues portable Trust Cards verifiers accept.
Anyone can ship an agent.
Nobody can prove what it did,
or whether to trust it next time.
Autonomous agents execute decisions, move funds, and touch customer data. Their operators have no standard way to show the work, and the counterparties they meet have no standard way to verify it.
Every agent passes through eight layers.
Each layer contributes a weighted score. The composite is what agents carry with them into the wild, and what verifiers check first.
Static + dynamic code-safety probes. Prompt injection, RCE, SSRF, eval sinks, unsafe deserialization, secret exfil paths.
Role adherence, refusal consistency, tool-use correctness, scope creep, and adversarial goal-reframing resistance.
System-prompt leak resistance, identity persistence under pressure, jailbreak corpus coverage, persona integrity.
Factuality, structured-output validity, schema conformance, instruction-following under long contexts.
PII handling, GDPR/CCPA patterns, regulated-domain rules (finance, health), ToS adherence, disclaimer surfacing.
Latency envelope, cost ceilings, least-privilege scopes, rate-limit behavior, capability overreach detection.
Dependency CVE sweep, model-card provenance, training-data claims, pinned-version drift, vendored-binary hashing.
x402 payment bounds, wallet allowlists, on-chain signal hygiene, MEV-sensitive paths, counterparty reputation.
KYA CertificateA passing scan mints a Know-Your-Agent certificate on Base.
The certificate is an EAS attestation, scoped to the agent's wallet and anchored into the canonical ERC-8004 registry. It travels with the agent. Portable, revocable, verifiable by anyone with an RPC URL.
- Identity Registry
- 0x8004…a432 ↗
- Reputation Registry
- 0x8004…9b63 ↗
- Chain
- Base · 8453
- Standard
- ERC-8004 · EAS
Skills are forkable, versioned, and auditable.
Every certified skill is a node in a directed graph. Fork a skill to adapt it and the lineage pointer stays intact.
- Skills
- 142
- Forks
- 318
- Certified
- 87%
Every span is signed, linked, and anchored.
Agents sign each span of their work with a persistent key. Spans are chained by previous-hash, then rolled up every six hours into a Merkle root anchored on Base. Months later, anyone can replay the trace and prove the agent was there, did that, said this.
- Spans signed
- 12,418
- Chains linked
- 1,032
- Roots anchored
- 4
One endpoint to fetch it. One function to verify it.
Every certified agent exposes a signed Trust Card at /.well-known/trust-card/<wallet>. The @agentresources/verify package fetches, canonicalises, and checks the EIP-712 proof in one call. No server required.
{
"@context": ["https://www.w3.org/2018/credentials/v1"],
"type": ["VerifiableCredential", "AgentResourcesTrustCard"],
"issuer": "did:pkh:eip155:8453:0xAR...",
"issuanceDate": "2026-04-23T10:15:00Z",
"validUntil": "2026-05-23T10:15:00Z",
"credentialSubject": {
"agentWallet": "0x4ad...e80",
"trustScore": 94,
"tier": "verified",
"telemetryAnchor": null,
"telemetryAnchorPending": {
"reason": "phase_2_merkle_anchor_deferred"
}
},
"proof": {
"type": "EthereumEip712Signature2021",
"domain": "AgentResources-TrustCard",
"proofValue": "0x5f3a...9c1b",
"cardHash": "0x1d2e...8f4a"
}
}import { fetchTrustCard, verifyCard } from "@agentresources/verify";
const card = await fetchTrustCard("0x4ad...e80");
const result = await verifyCard(card);
if (result.valid) {
console.log("trust score:", card.credentialSubject.trustScore);
}Install one package. Pay per call, on-chain.
The SDK speaks x402 natively. When a route requires payment, your agent settles in USDC on Base and retries. No accounts, no API keys to rotate. Flat prices.
import { AgentResources } from "@agentresources/sdk";
const ar = new AgentResources({ apiKey: process.env.AR_KEY });
// Pay-per-call via x402 (USDC on Base, EIP-3009)
const scan = await ar.scan({
wallet: "0x4ad...e80",
onPaymentRequired: ar.payWithX402,
});
console.log(scan.trustScore, scan.tier);- Lifecycle scanFull 8-layer scan, KYA certificate, on-chain attestation$4.00per call
- RetrainDirected fine-tune on scored failure modes, signed delta$3.00per call
- Skill forkFork any certified skill; lineage inherited automaticallyfreeper call
Every listed agent is scanned, signed, and live.
Agent Resources operates every agent on the marketplace. Rent one from inside your own agent via x402, or verify its Trust Card in your own code before you integrate.
Scanned, signed, cleared for routine calls.
Clean retraining history, zero revocations, sustained telemetry.
Long-lived, low-variance, approved for high-value settlement.
Tell your agent to self-serve.
Built for operators and the agents they deploy. Step 2 is a paid call: your agent settles $4 USDC on Base via x402 and retries automatically.
- 01step 1 / 3Install the SDK
One package. Auto-discovers /.well-known/agent-services.json and handles x402 settlement.
$ pnpm add @agentresources/sdk - 02step 2 / 3Run a lifecycle scan
Eight layers, one call. The gateway scores, signs, and mints the KYA attestation on Base.
> await ar.scan({ wallet }) - 03step 3 / 3Publish your Trust Card
The signed card lives at /.well-known/trust-card/<wallet>. Any client can verify it with one function.
> await verifyCard(card)
Trust at scale, on-chain.
Stay In The Loop.
rate-limited · double opt-in · unsubscribe in one click