Show and Tell: LoopGrid — Signed decision and action evidence for Microsoft Agent Framework

[Cybertech (cybertechsoft)]

LoopGrid — Signed decision and action evidence for Microsoft Agent Framework

We've released LoopGrid Microsoft Agent v0.1.0, an open-source Python integration that connects Microsoft Agent Framework's native middleware to independently verifiable agent decision evidence.

It uses native chat and function middleware while preserving the existing Microsoft Agent Framework execution lifecycle.

The integration supports:

• Agent decision and model lifecycle evidence
• Application policy and authorization evidence
• Human approval and rejection evidence
• Tool execution and observed outcome records
• Ed25519/SHA-256 tamper-evident verification through LoopGrid Core
• Fail-closed checks for unauthorized tool execution

Validation: 28 automated tests passed. We also tested real Microsoft Agent Framework execution against LoopGrid Core, including blocked policies, missing or rejected approvals, evidence-write failures, and native always_require approval/resumption.

The successful sandbox workflow achieved 100% applicable evidence coverage and valid cryptographic verification. The approval examples use synthetic host reviewer information and do not establish authenticated production identity.

The integration is maintained independently and requires no changes to Microsoft Agent Framework or LoopGrid Core.

Resources:

• GitHub: https://github.com/loopgridio/loopgrid-microsoft-agent
• PyPI: https://pypi.org/project/loopgrid-microsoft-agent/
• Documentation: https://loopgrid.io/integrations/microsoft-agent/

We'd welcome feedback from the Agent Framework community, particularly on the native middleware integration and whether a focused evidence-verification example would be useful in the official samples repository.