Open Agent Governance Specification (OAGS)

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

Overview

OAGS defines local-first governance primitives that work across any language or runtime:

• Deterministic Identity — content-addressable agent IDs derived from code + prompt
• Runtime Policy Enforcement — policies enforced during execution, not just declared
• Cryptographic Verification — all claims backed by Ed25519 digital signatures
• Mutual Authentication — agent-to-agent trust handshake protocol
• Signed Runtime Requests — optional session-backed signatures for protected HTTP endpoints
• Audit Trail — structured, tamper-evident audit logging

Specifications

• Start Here - Recommended reading order for first-time implementers.
• 5-Minute Quickstart - Fastest path from clean clone to first conformance run.
• OAGS v0.1 - The core specification: identity, policy, enforcement, audit, verification, trust headers, handshake, and A2A types (v0.1.1, Draft).
• sekuire.yml Schema - Canonical file format reference for sekuire.yml / sekuire.json. Defines the unified manifest + policy + SDK extension schema.
• Conformance Harness - Runner/adapters and profile execution contract.
• Conformance Matrix - SDK conformance status across reference implementations.
• Profile Schemas - JSON schemas for each OAGS profile (identity, policy, audit, trust headers, handshake, A2A).
• Test Vectors - Shared test vectors for cross-implementation conformance testing.

Website

The specification website is available at oags.sekuire.ai and lives in the website/ directory. It is a Next.js site that presents the spec content, conformance levels, implementation links, and community resources.

cd website
npm install
npm run dev

Repository Structure

├── specs/
│   ├── README.md                 # Start-here index
│   ├── quickstart-5-minutes.md   # First implementation path
│   ├── oags-v0.1.md              # Core specification
│   ├── sekuire-yml-schema.md     # File format reference
│   ├── conformance-harness.md    # Harness usage guide
│   ├── conformance-matrix.md     # SDK conformance status
│   ├── profiles/                 # JSON profile schemas
│   └── test-vectors/             # Shared conformance vectors
├── examples/
│   ├── customer-support/         # Policy with restrictions and custom rules
│   ├── research-assistant/       # Full-featured (manifest + policy + SDK)
│   ├── minimal-agent/            # Bare minimum Level 1 conformance
│   └── multi-provider/           # LLM provider configuration variants
├── website/                      # Next.js specification website
├── LICENSE                       # Apache 2.0
├── CHANGELOG.md                  # Spec/docs change history
├── CONTRIBUTING.md               # Contribution guidelines
└── CODE_OF_CONDUCT.md

Each example is a self-contained agent project with a sekuire.yml, system_prompt.md, and tools.json.

Examples

Example	Description
customer-support	Billing and ticket agent with network/filesystem restrictions, tool allowlists, rate limits, custom rules for refund approval, and compliance metadata on sensitive tools
research-assistant	Full-featured example showing all three zones: manifest with capabilities/discovery/deployment, policy, and SDK extensions (LLM, memory, logger)
minimal-agent	Bare minimum Level 1 conformance - just spec_version, project, identity, and one tool
multi-provider	Shows how the same agent can target different LLM providers by changing the SDK extension section

Contributing

We welcome contributions! Please read CONTRIBUTING.md before submitting.

For spec-related issues or pull requests, use the [OAGS-Spec] prefix in the title.

License

Licensed under the Apache License 2.0.

Copyright 2024-present Sekuire, Inc.