2026-10-07T13:29:18Z https://api.figshare.com/v2/oai

oai:figshare.com:article/33972877 2026-09-23T11:52:46Z category_29200 item_type_3 month_year_09_2026

Java dataset and results peng (25096915) Software testing, verification and validation Unit test generation Agent skills This dataset contains the Java benchmark projects and experimental results used to evaluate a skill-based framework for LLM-assisted Java unit-test generation and iterative test repair.The benchmark collection comprises Maven-based open-source Java projects, including Apache Commons CLI, Commons CSV, Commons Lang, Google Gson, and additional benchmark projects. For each benchmark, the release includes project manifests and target-method metadata used to define the unit-test generation tasks.The experimental-results package contains one directory per experiment run. Each run includes: (1) summary-level effectiveness metrics; (2) per-class and per-target execution records; (3) baseline results without skill injection; (4) results obtained with skill selection and iterative skill evolution; (5) coverage and, where enabled, mutation-testing measurements; (6) execution status and logs; and (7) skill-evolution records. The reported metrics include compilation success, test pass rate, line coverage, branch coverage, mutation score, and token usage.The framework compares a no-skill baseline with a skill-enhanced setting. In the enhanced setting, the system selects test-generation skills based on observed failure signals and may evolve skills through iterative generation, execution, feedback, and repair cycles. The archive is intended to support reproduction, validation, and secondary analysis of experiments on skill-guided LLM-based Java unit-test generation. Generated build artifacts (for example, Maven target directories), credentials, and local configuration overrides are excluded. 2026-09-23T11:52:46Z Dataset Dataset 10.6084/m9.figshare.33972877.v1 https://figshare.com/articles/dataset/Java_dataset_and_results/33972877 CC BY 4.0