AI-Authored Commit History from Public GitHub Projects

AgentPack contains 1.3 million GitHub commits identified as co-authored by AI coding agents like Claude Code, OpenAI Codex, and Cursor Agent. The dataset was created by the nuprl research organization and captures commits from public projects between April and mid-August 2025. It provides a large-scale view of how AI agents contribute to real-world software development.

Use Cases

• Analyze commit metadata to identify patterns in AI agent usage across different projects and time periods.
• Study the co-authorship relationships between human developers and specific AI agents like Claude Code or Cursor Agent.
• Train classifiers to distinguish between AI-authored and human-authored commits based on features like commit messages and timestamps.
• Investigate the temporal evolution of AI agent contributions from April to August 2025.

Strengths

• 1.3 million commit records provides a substantial sample for analysis.
• Data is focused on a specific, recent time period from April to mid-August 2025.
• Explicitly identifies contributions from three major AI coding agents.

Limitations

• Specific column names and data schema are not publicly documented.
• Limited to public GitHub projects, excluding private repositories which may have different usage patterns.
• The 4.5-month collection window may not capture long-term seasonal or project lifecycle trends.

Provenance

Source

Public GitHub projects.

Collection Method

Commits were identified as co-authored by specified AI agents.

Time Range

April 2025 to mid-August 2025.

Freshness

Dataset was last updated on the platform in October 2025.