An Empirical Study of Code Clone Genealogies in Human–AI Collaborative Development

This repository contains the replication package for the paper “Code Clone Genealogy in Human–Agent Collaborative Software Projects”, which investigates how code clone genealogy evolves in projects where developers actively collaborate with coding agents powered by Large Language Models (LLMs).

Code clones are two or more identical or similar code snippets. Their presence can harm maintainability because changes often need to be applied consistently across multiple locations, increasing the risk of inconsistent modifications. To understand how clones evolve over time, we rely on code clone genealogy, which captures the evolutionary history of duplicated snippets by linking them across successive versions of a software system, forming clone lineages.

When comparing two consecutive versions within a lineage, we characterize how clones evolve through two complementary perspectives:

• Change Patterns: modifications to the code snippets inside an existing clone.

• Evolution Patterns: additions or removals of code snippets that compose a clone across versions.

We analyzed 350 code clone lineages across 6 human–agent collaborative projects. Humans introduced most clones (85.71%) compared to agents (14.29%). Although clone survival rates were similar (humans 80%, agents 76%), maintenance differs: humans predominantly maintain lineages created by agents, based on Change and Evolution Patterns. Overall, humans remain essential for the long-term evolution of agent-assisted code.

This replication package is intended to support reproducibility, enabling other researchers to re-run the analyses, validate the results, and extend the study to additional projects, datasets, or collaboration settings.