Turn your own merged PRs into executable agent evals: 79.6% verified task-construction success across five task families
arXiv 2607.28591·low signal
Change2Task converts repository changes — merged pull requests — into verified, executable coding-agent tasks with runnable environments, reporting 79.6% verified construction success across five task families. For a team with a long PR history, this is a route to a domain-specific eval set that reflects your codebase instead of SWE-bench's. Single-source, abstract-level read only.