v0.1.1 Draft

Open Agent Governance Specification

An open specification for identity, policy, enforcement, audit, and verification of autonomous AI agents.

Core Primitives

Five pillars of agent governance

OAGS defines five composable primitives that work together to make AI agent behavior auditable, predictable, and verifiable.

Deterministic Identity

Content-addressable agent IDs via BLAKE3 hash of model, prompt, and tools. Same configuration always produces the same identity.

Policy Declaration

Declarative constraints on network access, filesystem, tools, models, APIs, and rate limits defined in sekuire.yml.

Runtime Enforcement

Policies enforced at execution time with allow, deny, and warn decisions. Not just documentation - actual runtime checks.

Audit Trail

Structured, tamper-evident event logging with optional Ed25519 signing. Every policy decision is recorded.

Cryptographic Verification

Multi-stage security scoring with compliance framework support. Verify agent behavior against declared policy.

How It Works

Three steps to governed agents

01

Define

sekuire.yml

Declare your agent’s identity, permitted tools, network access, and runtime constraints in a single configuration file.

02

Enforce

Runtime Checks

The SDK enforces declared policies at execution time. Every tool call, network request, and API access is checked against policy.

03

Verify

Audit + Signatures

Generate tamper-evident audit trails with Ed25519 signatures. Third parties can independently verify agent compliance.

Conformance

Three levels of compliance

Implementations can conform at different levels, allowing incremental adoption.

L1

Identity + Policy

  • Agent computes SekuireID from BLAKE3 hash
  • Declares policy in sekuire.yml
  • Policy schema validates correctly
L2

Runtime Enforcement

  • Everything in Level 1
  • Runtime enforcement of declared policies
  • PolicyDecision events for every action
  • Deny/warn on policy violations
L3

Verifiable Audit Trail

  • Everything in Level 2
  • Ed25519 signed audit events
  • Third-party verification support
  • Compliance framework scoring
Example

sekuire.yml

A single file defines your agent's identity, policies, and audit configuration.

sekuire.yml
agent:
  name: "my-agent"
  version: "1.0.0"
  model: "claude-sonnet-4-20250514"

identity:
  signing_key_path: "./keys/agent.ed25519"

policy:
  network:
    allowed_domains:
      - "api.openai.com"
      - "api.anthropic.com"
    deny_all_other: true

  tools:
    allowed:
      - "web_search"
      - "file_read"
    require_confirmation:
      - "file_write"
      - "shell_execute"

  rate_limits:
    max_requests_per_minute: 60
    max_tokens_per_hour: 100000

  audit:
    level: "full"
    sign_events: true
Implementations

Get started with any language

Official SDKs and tools that implement the OAGS specification. TypeScript and CLI are available now. Python and Rust are coming soon.

TypeScript SDK

Available

Multi-provider LLM support, local policy enforcement, structured audit logging.

npm install @sekuire/sdk

Python SDK

Coming Soon

Reference Python implementation is in progress and will be released soon.

Installation command will be published at launch.

Rust SDK

Coming Soon

Reference Rust implementation is in progress and will be released soon.

Installation command will be published at launch.

CLI

Available

Init projects, generate Ed25519 keys, compute identity hashes, publish and verify agents.

cargo install sekuire-agent-cli