ajnas-skills-registry

Signed enterprise skill manifest registry with policy and provenance hooks for Ajnas runtimes.

ajnas-skills-registry is the package that lets an Ajnas deployment decide which skills can be installed, reviewed, trusted, and replayed. It is designed for enterprise agent infrastructure where skills need explicit capability metadata, permissive license evidence, deterministic digests, signature verification, policy gates, and audit receipts.

The package is original Ajnas implementation work. It was shaped by prior public OSS research, but it does not vendor or copy third-party source code, examples, docs, or branding.

Install

npm install ajnas-skills-registry

ajnas-runtime is an optional peer dependency. The registry policy helper is structurally compatible with the runtime PolicyEngine contract and can be used without importing runtime types.

createSkillInstallPolicy() is deliberately scoped to the configured installation tool and denies every other tool. Compose unrelated tool rules in a separate explicit policy; never use an allow-on-miss policy as a runtime-wide fallback.

Quick Start

import { generateKeyPairSync } from "node:crypto";
import {
  SkillAuditLedger,
  SkillRegistry,
  createSkillInstallPolicy,
  signSkillManifest
} from "ajnas-skills-registry";

const manifest = {
  schemaVersion: "ajnas.skill.v1",
  id: "com.ajnas.research-brief",
  name: "Research Brief",
  version: "1.0.0",
  description: "Turns approved research notes into a concise internal brief.",
  license: "MIT",
  entrypoint: { type: "module", value: "./dist/research-brief.js" },
  compatibility: { ajnasRuntime: "^0.2.2" },
  capabilities: [
    {
      name: "research.summarize",
      description: "Summarize reviewed research notes.",
      risk: "medium"
    }
  ],
  permissions: {
    network: "none",
    filesystem: "read",
    externalPublish: false,
    humanApproval: "on_high_risk"
  },
  trust: {
    source: "first-party",
    publisher: "ajnas",
    repository: "https://github.com/ajnas/examples",
    licenseEvidence: {
      kind: "repository",
      checkedAt: "2026-07-05T02:00:00.000Z"
    }
  }
};

const { privateKey, publicKey } = generateKeyPairSync("ed25519");
const signed = signSkillManifest(manifest, {
  algorithm: "ed25519",
  keyId: "ajnas-release-key",
  privateKey
});

const audit = new SkillAuditLedger();
const registry = new SkillRegistry({
  audit,
  keyResolver: (keyId) => (keyId === "ajnas-release-key" ? publicKey : null)
});

registry.register(signed, {
  actor: "release-automation",
  source: "internal-review"
});

const installPolicy = createSkillInstallPolicy({
  trustedPublishers: ["ajnas"]
});

CLI

ajnas-skills validate fixtures/research-brief.skill.json
ajnas-skills digest fixtures/research-brief.skill.json

validate prints validation status, issue codes, and the deterministic digest for valid manifests. digest prints only the canonical sha256:<hex> digest and exits non-zero if the manifest fails validation.

Core API

• validateSkillManifest(manifest): validates schema version, package identity, permissive license, runtime compatibility, capability risk metadata, permissions, and trust evidence.
• computeSkillDigest(manifest): produces a deterministic sha256:<hex> digest from canonical JSON.
• signSkillManifest(manifest, options): signs the digest envelope with Ed25519 key material.
• verifySkillSignature(envelope, publicKey): verifies digest integrity and Ed25519 signature material.
• SkillRegistry: registers signed or unsigned manifests, rejects conflicting duplicates, and records audit events.
• SkillAuditLedger: emits replayable hash-chained audit receipts for registry activity.
• createSkillInstallPolicy(options): returns a runtime-compatible policy engine for skills.install tool calls.
• SkillApprovalQueue: serializes review requests and approval resolutions for resumable human review workflows.

Safety Model

The registry treats skill installation as a governed supply-chain action. A skill must declare what it can do, what it needs, who published it, which permissive license evidence was checked, and whether human approval is required for high-risk behavior.

Validation rejects non-permissive license identifiers and high-risk manifests that disable approval. Runtime policy can deny invalid manifests, require approval for untrusted publishers, and require approval for high-risk permissions.

Local Development

npm install
npm test
npm run typecheck
npm run build
npm run example:basic
npm run pack:dry

Release Status

Version 0.2.2 is selected for the maintained Node 22/24/26 metadata and runtime peer-floor patch. It is not a public release until the trusted workflow and registry verification succeed.