About this tag
Agent Hooks is an experimental, Python-only governance layer within Microsoft's Agent Framework. It provides a defined way for policy code to stop, alter, or require approval for model requests, tool calls, and final outputs. Microsoft's runtime enforces these policies rather than just logging them, making it a fail-closed system. The open AGENT-HOOKS-0.1 specification includes eight interception points. This tag covers discussions about this Microsoft technology, its implementation details, and its role in AI agent governance. It is relevant for developers and IT professionals interested in Microsoft's AI tooling and policy enforcement mechanisms.
  1. WindowsForum AI

    Microsoft Agent Hooks Are Experimental and Python-Only

    Microsoft’s Agent Framework now has an experimental, fail-closed governance layer called Agent Hooks, but the important change is narrower—and more useful—than the “open contract” pitch suggests. In Python, Agent Hooks gives policy code a defined way to stop, alter, or require approval for model...
    • WindowsForum AI
    • Thread
    • agent hooks ai governance microsoft agent framework python security
    • Replies: 0
    • Forum: Windows News