Commit authorship disambiguation for Claude Code and OpenAI Codex contributions

Determine which commits in the AIDev dataset were actually authored by the AI coding agents Claude Code and OpenAI Codex, given that these agents attribute their contributions to human developers and thereby obscure commit-level authorship attribution.

Background

The study analyzes AI-generated Python code from the AIDev dataset at the commit level. For some agents (Copilot, Cursor, Devin), commits can be reliably attributed to the agent identity, enabling code proficiency analysis of added code.

However, Claude Code and OpenAI Codex attribute their contributions to human developers, which prevents the researchers from definitively identifying which commits were actually produced by these AI agents. As a result, the study excludes these agents from the analysis, leaving an unresolved authorship attribution challenge that affects the completeness of comparative evaluations across AI agents.

References

Nonetheless, Claude Code and OpenAI Codex attribute their contributions to human developers. Thus, the authors of the commits using them were varied, and we could not precisely decide which commits were actually made by agents.

— When is Generated Code Difficult to Comprehend? Assessing AI Agent Python Code Proficiency in the Wild  (2604.00299 - Temkulkiat et al., 31 Mar 2026) in Section 3.1 Data Preparation (Study Design)