Proposal: Governance Policy Schema for Agent Spec

Problem

Oracle's Agent Spec provides a powerful declarative format for defining agents, flows, and their compositions. However, the current specification doesn't include a standard way to express governance policies - the safety constraints, resource limits, and trust requirements that agents must operate within.

As Agent Spec aims to be the universal agent configuration format, governance should be a first-class schema component.

Proposed addition

A governance section in the Agent Spec schema:

yaml kind: agent apiVersion: v1 metadata: name: researcher spec: model: gpt-4o governance: policy: name: production-safety version: "1.0" max_tokens_per_request: 4096 max_tool_calls_per_request: 10 blocked_patterns: - pattern: "rm -rf" type: substring - pattern: ".*password.*=.*" type: regex trust: min_delegation_score: 0.7 require_identity_verification: true audit: level: full merkle_chain: true

Why this matters

1. Portable governance - Policies travel with the agent spec, not buried in framework-specific config
2. Cross-framework enforcement - Any runtime that reads Agent Spec can enforce the same policies
3. Schema-validated - Governance constraints are validated at spec parse time, not runtime
4. Composable - Flow-level policies can tighten (never loosen) agent-level policies

Reference implementation

We've built a working governance engine in Agent-OS:

• GovernancePolicy with YAML import/export
• PatternType enum (substring, regex, glob)
• Policy diff/comparison (is_stricter_than)
• 700+ tests

Happy to contribute a governance schema section and reference validator.