# Goal

## Primary objective

Produce a comprehensive, source-grounded report on the governance-toolkit landscape and its implications for HPE Cray Agentic Framework (CAF) v2.

## Scope

- Research existing AI-agent governance toolkits and partial building blocks.
- Research generic governance toolkits applicable with or without AI agents.
- Analyze Microsoft `agent-governance-toolkit` and locate it within the industry landscape.
- Evaluate whether CAF v2 should reuse, integrate, or build governance capabilities.
- Deliver detailed research, comparison tables, and strategic recommendations.

## Constraints

- Project files and report are in English.
- Separate verified facts, interpretation, and recommendations.
- Ground current market claims in cited sources.
- Treat CAF as thin, replaceable, and developer-first; host services own hard enforcement and system telemetry.
- CAF covers the managed path and must not overclaim visibility into downstream host effects.

## Success criteria

- Broad market map covering complete products, partial tools, standards, and enabling infrastructure.
- Detailed Microsoft toolkit fit and maturity assessment.
- CAF build-versus-adopt recommendation by capability layer.
- Complete report with comparison tables, research detail, citations, and strategic recommendations.

## Current result

The research and report are complete. The principal recommendation is to build CAF's semantic contract, runtime adapter SDK, admission/conformance suite, HPC facades, and evidence schema; reuse mature host mechanisms; and evaluate Microsoft AGT/ACS only as an optional provider behind CAF-owned interfaces.

## Open questions

- Whether to proceed with the proposed bounded AGT/ACS versus minimal OPA/Cedar technical spike.