Abstract
Multi-agent AI systems are increasingly utilized to automate software engineering tasks such as requirements analysis, architecture design, test generation, and traceability linking. When these agents function in a sequential pipeline over shared software artifacts, errors and low-confidence decisions from upstream agents propagate downstream, resulting in orphaned requirements, contradictory links, and compliance gaps that pose significant risks in safety-critical domains.
We propose a trust-aware coordination framework where a shared knowledge graph acts as both a centralized semantic memory and a coordination surface, allowing agents to assess and build upon each other's contributions using calibrated confidence scores. Our approach introduces a two-stage traceability link prediction pipeline that combines embedding-based retrieval with LLM-based multi-criteria analysis, a traceability seeding mechanism for comparing derivation-time and validation-time confidence, and a consistency protocol governing pipeline interactions through confidence threshold gating, confidence divergence detection, and conflict resolution.
We evaluate our framework on an automotive software engineering case study measuring link prediction calibration, protocol effectiveness, threshold sensitivity, and the impact of traceability seeding. Ablation studies confirm that confidence calibration is essential for effective pipeline coordination.
Blogger's Review: The proposed trust-aware framework significantly enhances traceability and consistency in multi-agent systems for software engineering through knowledge graphs and confidence calibration mechanisms, showcasing notable potential in safety-critical applications. Future research could explore implementation and optimization in more complex scenarios.