Skills governance

The dedicated skills-policy packages live under openagent-sdk/crates/openagent-skills-policy. They parse policy/skills manifests and evaluate invocation context. The unified SDK's small SkillsPolicy interface is a separate adapter boundary; it does not automatically enable every rule in the dedicated engine.

Policy dimensions

TypeScript engine

The source exports SkillsPolicy, InvocationContext, Did, SkillsManifest, typed errors, a rate limiter and audit helpers. For example, this creates an explicit policy and records an invocation under it:

import { Did, InvocationContext, SkillsPolicy } from '@openagentid/skills-policy';

const policy = SkillsPolicy.fromYaml(`
version: 1
agent: did:oas:example:agent:docs-bot
skills:
  read-documentation:
    allow: true
    audit_level: hash
default:
  allow: false
`);
const context = new InvocationContext({
  agentDid: Did.parse('did:oas:example:agent:docs-bot'),
  sessionId: 'local-docs-example',
  arguments: { path: '/essentials/quickstart' },
});
const receipt = policy.recordInvocation('read-documentation', context);

Install this package from the source path in the catalog. The example configures policy only; it does not authenticate the DID or invoke a remote tool. The application must ensure the policy decision precedes the actual action and that context comes from authenticated state.

Audit and enforcement

A hash chain can verify linkage relative to retained history. It does not make in-memory receipts durable or prove every invocation was recorded. Configure persistence and redaction. Full argument capture may contain sensitive data; choose it deliberately.

See exact TypeScript policy exports, Rust policy modules, and framework hooks.