@absolutejs/agent-reputation

Portable, evidence-based reputation for AI agents without a gameable universal trust score.

Reputation is always assessed for an explicit scope. Evidence has a stable subject and issuer, bounded value and confidence, issuance and expiration, optional receipt digest, verification status, and revocation state. Assessment uses a neutral Bayesian prior, time decay, configurable issuer trust, and a cap per issuer so one party cannot inflate a score by repeating claims.

import {
  createAgentReputation,
  createPostgresAgentReputationStore,
} from "@absolutejs/agent-reputation";

const reputation = createAgentReputation({
  store: createPostgresAgentReputationStore({ client: postgres }),
  verifyEvidence: (evidence) => verifyIssuerReceipt(evidence),
  issuerWeights: { "did:web:certifier.example": 1 },
});

await reputation.record(signedConformanceEvidence);
const assessment = await reputation.assess(
  "https://agents.example/calendar",
  "calendar.scheduling",
);

Unsigned or unverifiable evidence is denied by default. allowUnverified is an explicit local-development escape hatch. Revoked and expired evidence is never scored, evidence from each issuer is capped, old evidence decays, and evidence from one scope is never silently reused in another.

createAgentReputationCredential() maps an assessment to the W3C Verifiable Credentials Data Model 2.0. Signing and verification remain adapters so callers can use the W3C JOSE/COSE Recommendation, Data Integrity proofs, SD-JWT, a DID stack, or an enterprise trust service without provider lock-in. The package publishes a stable remote JSON-LD context at https://absolutejs.github.io/agents/contexts/reputation/v1.json.

Standards:

• https://www.w3.org/TR/vc-data-model-2.0/
• https://www.w3.org/TR/vc-jose-cose/

License

MIT