GH-AWTrace: A Large-Scale Dataset of GitHub Agentic Workflow Execution Logs
The dataset links 4.1 million workflow runs to execution logs that reveal failures and agent behavior beyond reported outcomes.
GH-AWTrace covers 25,014 verified GitHub Agentic Workflows across 5,051 repositories, with 562,877 nonempty execution log archives. Illustrative cases expose tool denials, provider failures, changes in approach and delegation among agents. The records support research into how reported run outcomes relate to visible task progress. Coverage varies, and GitHub’s log-retention limits mean the logs offer partial observations rather than complete agent trajectories.
HF Daily Papers' note
GH-AWTrace covers 25,014 verified GitHub Agentic Workflows across 5,051 repositories, with 562,877 nonempty execution log archives. Illustrative cases expose tool denials, provider failures, changes in approach and delegation among agents. The records support research into how reported run outcomes relate to visible task progress. Coverage varies, and GitHub’s log-retention limits mean the logs offer partial observations rather than complete agent trajectories.
HF Daily Papers' note
score 5