2025/07/27 by Nazmul Hossain, Md Nazmul Hossain, Hossain, Md Nazmul +2 · 2 citations
Computer Science · #FOS: Computer and information sciences #Service-Oriented Architecture and Web Services #Software Engineering (cs.SE) #Software Engineering Research #Software System Performance and Reliability #cs.SE
paper · pdf · doi:10.48550/arxiv.2507.20402
openalex publication_date 2025/07/27 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28 · arxiv created 2026/08/03 · arxiv updated 2026/08/04
Continuous Integration (CI) configurations often need to be migrated between services (e.g., Travis CI to GitHub Actions) as projects evolve due to changes in service capabilities, usage limits, or service deprecation. Previous studies reported that migration across CI services is a recurring need in open-source development. However, manual migration is time-consuming and error-prone. The state-of-the-art approach, CIMig, addresses this challenge by analyzing past migration examples to create service-specific rules and produce equivalent configurations across CI services. However, its relatively low accuracy (0.49 Cosine Similarity for Travis->GHA) raises concerns about the feasibility of rule-based CI migration alone. Meanwhile, Large Language Models (LLMs) have demonstrated strong capabilities in code generation and transformation tasks, suggesting potential to improve the automation, usability, and generalizability of CI configuration migration. This paper presents an empirical study assessing whether CI migration can be improved using LLMs. We propose CIgrate, an LLM-based framework for automatically migrating CI configurations, and compare it with CIMig using (a) zero-shot/few-shot prompting and (b) fine-tuning on a dataset of established CI migrations. We also evaluate practical deployment by submitting pull requests to active open-source projects. Our results show that CIgrate substantially outperforms CIMig: the fine-tuned Gemma 3 12B achieves 0.90 Cosine Similarity and 0.74 CrystalBLEU for Travis CI (Travis)->GitHub Actions (GHA) migrations (+82.2% and +295.5% over CIMig), while producing syntactically valid, immediately parseable YAML in 100% of cases, compared to 5.6% for CIMig. Even zero-shot LLMs outperform the rule-based baseline, demonstrating that LLM-based approaches provide a more practical and accurate solution for CI configuration migration.