Keep SemaPact governance independent from any specific AI model or agent framework

[ElliotSun]

User Story

As a platform owner,
I want SemaPact agent governance to remain model- and framework-independent,
so that governance remains stable while AI tooling evolves.

Business Value

Governance should not depend on:

• one LLM vendor
• one agent runtime
• one orchestration framework
• one prompt format

The stable product boundary is governed data/semantic state and machine-readable decisions.

Scope

Ensure agent-facing APIs operate on deterministic domain models.

LLMs may consume or propose information, but core governance must not require LLM inference.

Tasks

• Define model-independent agent contracts.
• Audit APIs for model-specific types.
• Keep GovernanceDecision deterministic.
• Keep policy evaluation non-LLM.
• Document optional AI enrichment boundary.
• Add architecture tests/docs.

Acceptance Criteria

• Core agent-safe APIs work without an LLM.
• GovernanceDecision does not depend on model output.
• Agent framework adapters remain replaceable.
• LLM-generated proposals enter normal governance workflow.
• No model can become the source of governance authority.

Dependencies

US-M7-005
US-M7-009

Out of Scope

• LLM hosting
• prompt orchestration
• model evaluation framework