DCFA

Researchers proposed DCFA, a training-free framework that uses dual-view causal-inspired reasoning to attribute failures in LLM-based multi-agent systems, identifying the earliest decisive error whose…