@agenticprimitives/identity-directory-adapters

The binding layer that keeps the directory honest. identity-directory can promise "indexer proposes, on-chain confirms" only because its core never touches a real data source — it declares ports, and this package implements them. That separation is enforced, not aspirational: this is the one package allowed to import agent-naming (spec 100 §4 / ADR-0015), so every naming read in the identity stack flows through one auditable seam. Adapters translate sources into ports and add no policy of their own.

Part of agenticprimitives — the trust substrate for the agent economy: one canonical Smart Agent identity with custody, delegation, naming, credentials, and audit evidence designed as one system.

What's here

makeNamingPort({ client, chainId })   wrap agent-naming + eip155 Address↔CanonicalAgentId lift
makeOnChainReadPort(readers)          assemble OnChainReadPort from { exists, confirmsCredential }
viemExists(client)                    a contract-agnostic exists (bytecode-at-address)
createInMemoryIndexer(entries)        an in-memory IndexerPort (demo + tests)
toCanonicalAgentId / addressOf        eip155 CAIP-10 glue

Usage

import { createDirectory } from '@agenticprimitives/identity-directory';
import {
  makeNamingPort, makeOnChainReadPort, viemExists, createInMemoryIndexer,
} from '@agenticprimitives/identity-directory-adapters';

const dir = createDirectory({
  naming: makeNamingPort({ client: namingClient, chainId: 8453 }),
  onChain: makeOnChainReadPort({
    exists: viemExists(publicClient),
    // app-wired to the real membership getter (isCustodian/isTrustee):
    confirmsCredential: async (id, p) => publicClient.readContract({ /* … */ }),
  }),
  indexer: createInMemoryIndexer(seedEntries),
});

How it's different

The competing pattern is the monolithic indexer/resolver SDK — one client that talks to every source and bakes resolution policy into the data access. The cost shows up later: you cannot swap a source, cannot audit which mechanism answered a query, and the index silently becomes authoritative. The ports-and-adapters split here means:

• Sources are swappable. Naming, on-chain reads, and the indexer are independent factories; a new source for an existing port is one new factory, not a rewrite.
• Authority placement is visible in the types. confirmsCredential is a live on-chain membership check, the indexer is a proposer — the interfaces make it impossible to confuse the two.
• No adapter ever escalates. Each port is one mechanism; there is no try naming, catch → on-chain fallback anywhere (ADR-0013), and reads are readContract/getCode only — never eth_getLogs (ADR-0012).

Boundaries

• confirmsCredential is app-wired — the on-chain membership getter is contract- and credential-kind-specific (an EOA address vs a passkey credentialIdDigest), so this package does not guess it. It MUST reflect the current set so a revoked credential returns false (the directory drops it). Never wire it to a cached or event-derived list that could revive a revoked credential.
• The in-memory indexer is non-authoritative — it proposes candidates; the directory confirms them on-chain. The production indexer is a SPARQL/GraphDB adapter (spec 225 §7), landing later.
• CAIP-10 here is eip155-only glue; the canonical multi-namespace builder is @agenticprimitives/agent-profile (/caip10), deliberately kept out of this adapter's dependency set.

Validation

pnpm --filter @agenticprimitives/identity-directory build   # core dist (vitest resolution)
pnpm --filter @agenticprimitives/identity-directory-adapters typecheck
pnpm --filter @agenticprimitives/identity-directory-adapters test
pnpm check:forbidden-terms

Status

Testnet/pilot-ready. Production launch is gated on the public checklist in the root README.md — including third-party contract audit and governance key rotation. Track every security finding live in docs/audits/findings.yaml. The shipped indexer is the in-memory reference (demo + tests); the production SPARQL/GraphDB adapter is roadmap.

License

UNLICENSED (internal monorepo, not published).