Collaboration: Governance controls mapping for Agent Spec

[imran-siddique]

Hi @dhilloulinoracle

We consolidated our agent governance work into microsoft/agent-governance-toolkit and wanted to continue the conversation about governance controls in the Agent Spec context.

Background: Our toolkit implements runtime governance primitives - policy enforcement, capability sandboxing, audit trails, trust verification - covering 9/10 OWASP Agentic Top 10 risks. We've been thinking about how these map to specification-level controls.

Where we see potential alignment:

1. Capability declarations in agent specs - Our ring-based permission model (Ring 0-3) defines what an agent can access. This could inform a standard way to declare agent capabilities and restrictions in a spec.

2. Trust and identity requirements - Our inter-agent trust protocol uses DID-based identity. Agent Spec could define how agents declare trust requirements for collaboration.

3. Governance policy references - Specs could reference governance policies that agents must comply with, similar to how API specs reference auth schemes.

We noticed Agent Spec already considers controls at the component level - we'd love to understand your thinking on where runtime governance fits vs. specification-time declarations. A few questions:

1. Does Agent Spec have a concept of required capabilities or permission boundaries?
2. Is there a pattern for referencing external policy/compliance requirements?
3. Would governance metadata belong in the spec itself or in a companion artifact?

Happy to contribute a section on governance controls if that would be useful.

Repo: microsoft/agent-governance-toolkit

[dhilloulinoracle]

Hi @imran-siddique ,

Yes I think that could be quite interesting to have this kind of support in Agent Spec.

We will try to have a look in the following days (maybe week) and then maybe we can try to discuss here on what the support could look like?