Adding runtime guardrails to AutoGen workflows

[ishita-0301]

I've been experimenting with AutoGen for long running multi agent tasks, and one pattern I've noticed is that failures often happen during execution rather than planning.

For example:

• An agent gets stuck calling the same tool repeatedly
• It keeps retrying a failing action without making progress
• A bad intermediate state cascades through the rest of the workflow

To explore this, I started using FailproofAI, an open source runtime layer that focuses on detecting issues like loops and other execution failures so agents can recover instead of burning time and tokens.

Repository: https://github.com/FailproofAI/failproofai

I think this complements frameworks like AutoGen nicely. AutoGen excels at orchestration, while a runtime safety layer can help keep long autonomous workflows on track. Curious if others here are using similar guardrails or have built their own recovery mechanisms.