coauthored
An open research project and command-line scanner measuring disclosed AI attribution in public Git history, with aggregates from a 96-repository study.
What happened lately
coauthored timeline
coauthored examines coauthor trailers, AI-agent identities and generated-with markers in Git history. Its initial study compares AI infrastructure, coding-agent harnesses and matched non-AI repositories. The output measures visible attribution at the commit level; it cannot establish the share of code written with AI, developer productivity or code quality.
The author publishes aggregate data and anonymized samples alongside an Apache-2.0 command-line tool. The documented organization-report workflow uses Python 3.11 or newer, uv and an authenticated GitHub CLI. The dataset is CC-BY-4.0. Results are the author’s research findings, with purposive sampling, attribution loss and wide confidence intervals among the disclosed limitations.