MuMuTestUp: Mutation-guided multi-agent framework enhances test case updates
Three specialized agents address assertion adequacy, coverage gaps, and LLM hallucinations in CI/CD.
Deep Dive
Researchers from Harbin Institute of Technology, CUHK, Xidian University, and Singapore Management University propose MuMuTestUp, a mutation-based multi-agent framework for automated test case updates. It uses three agents: Mutation Analysis (strengthens assertions via surviving mutants), Coverage Analysis (generates repair instructions for uncovered lines/branches), and Semantic Retrieval (handles hallucinations via semantic similarity). Evaluated on a new 571-sample dataset (PRBENCH) with Deepseek-V3.2 and GPT-4.1.
Key Points
- MuMuTestUp uses surviving mutants to identify and strengthen weak assertions in test cases.
- Coverage Analysis agent generates precise repair instructions for uncovered lines and branches.
- Semantic Retrieval agent replaces exact-match retrieval to handle LLM hallucinated queries, improving update accuracy.
Why It Matters
Automates test maintenance in CI/CD pipelines, reducing downtime and improving software quality with better fault detection.