MA-Mesh is a framework that builds a role‑based self‑healing layer on top of an existing collaboration pattern. It first derives resilience‑aware system specifications from task requirements and then synthesizes them into executable multi‑agent policy code. Two self‑healing mechanisms are provided: cross‑agent reflection enables each agent to evaluate not only its own outcomes but also peer performance, allowing runtime behavioral adjustments and collective policy refinement; role‑based failover dynamically reassigns roles when sub‑tasks fail or an agent underperforms, letting other agents substitute to maintain execution continuity. Experiments across software development, tool use, and scientific reasoning tasks show that MA‑Mesh markedly outperforms current systems in both task success and robustness to uncertainty.
Review