Identify bug-introducing commits from bug oracles rather than fix commits

Determine whether bug-introducing commits can be identified from the outputs of bug oracles—such as fuzzer crash reports, failing unit tests, or monitoring alerts—instead of requiring fix commits, enabling agents to supply bug-introducing changes as context for automated program repair.

Background

The methods introduced in the paper take a fix commit as input and trace back to the bug-introducing commit. For broader utility, especially in automated program repair, it would be valuable to locate bug-introducing commits directly from evidence produced by bug oracles, prior to any fix being written.

If feasible, this capability could allow agents to both localize and repair bugs by providing the bug-introducing changes as context to patch generation systems.