Skip to content

Python: Hosted Agent Framework: human-in-the-loop tool approval not working #5654

@cs609

Description

@cs609

Summary

We are testing the hosted Agent Framework sample here:

https://github.com/microsoft-foundry/foundry-samples/blob/main/samples/python/hosted-agents/agent-framework/responses/02-tools/main.py

When we change the tool decorator from:

@tool(approval_mode="never_require")

to

@tool(approval_mode="always_require")

we see the following warning in the logs:
WARNING agent_framework_foundry_hosting._responses: Content type 'function_approval_request' is not supported yet. This is usually safe to ignore.

We are using:

  • agent-framework-foundry-hosting 1.0.0a260429
  • agent-framework = 1.2.2

At the same time, we do not get a usable response in the playground, so the human-in-the-loop approval flow does not appear to work in the hosted scenario.

What we understand

From the warning and the playground behavior, our current understanding is that the hosting adapter layer may not yet support the function_approval_request content type required for human-in-the-loop tool approval.

What we are trying to achieve

We want to support human-in-the-loop approval for tool execution in our Agent Framework orchestrations.

For orchestration scenarios involving plan review and tool approval, it appears these human-in-the-loop interactions are not currently supported in the hosted flow. If correct, this means hosted orchestration cannot yet support these approval steps end-to-end.

Question

Is there a planned timeline or roadmap for support of human in the loop in hosted adapter Agent Framework scenarios?

Metadata

Metadata

Labels

Type

No type

Projects

Status

No status

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions