agent-consistency – a Python consistency layer for multi-agent workflows

I keep seeing AI agent workflows claim “task completed” even when the outcome was never actually verified.

I’m curious how people here deal with:

• stale state between agents
• incomplete handoffs
• outcome verification
• workflows that look clean in logs but are still wrong

Are you solving this with tests, traces, guardrails, custom validators, or something else?

I built a small MIT-licensed Python package around this problem called agent-consistency.
Repo/demo here: https://github.com/karimbaidar/agent-consistency-refund-demo

It’s not a framework replacement.
It’s a consistency layer meant to catch stale state, broken handoffs, and false success.

I’d love blunt feedback on where this approach breaks down.