Replies: 2 comments
|
Correction/qualification: The workflow bridge described below is a reasonable workaround, but I could not verify that it is the only possible solution for the exact Agent Framework 1.3 package set in the question. My original categorical statement about transparent nested approval propagation was therefore too strong. For current Agent Framework packages, the documented multi-agent approval path is a workflow orchestration: keep the approval-required tool on the participating agent, observe the workflow's That does not by itself establish the behavior of a plain agent-as-tool composition in version 1.3. If you must stay on 1.3, the next useful evidence is a minimal reproduction plus the exact resolved If direct nested propagation is unavailable in that package set, an explicit workflow bridge—or moving the approval-required tools to the top-level agent—is a sound fallback. It should be presented as a workaround pending version-specific confirmation, not as a proven universal requirement. Current workflow HITL documentation: |
Uh oh!
There was an error while loading. Please reload this page.
Before the newest release of Agent I was able to have a triage agent that had sub agents that had MCP tools that required approval and I handled all of this by using AsDeclarationOnly for the sub agents and then manually calling the agent if needed and rebuilding the message stack to prevent 400 bad requests going to OpenAI.
With the newest version I cannot seem to get that to work any more.
Has anyone been able to have an agent use another agent as a tool and then have tools in the sub agent generate approvals that flow back up?
I am using .NET (currently Agent 1.3)
All reactions