Executability of replication packages is a prerequisite for validating, reusing, and extending results in software engineering (SE) research. Despite substantial progress in artifact availability and badging, recent large-scale studies reveal that a majority of published replication packages remain partially executable or fail entirely, most often due to incomplete documentation, missing components, and unclear execution procedures. These issues impose a significant burden on reviewers and secondary researchers, undermining the reusability and long-term value of research artifacts. This paper introduces OpenART, a framework designed to systematically improve the completeness of SE replication packages to enhance executability. OpenART addresses a critical gap in current artifact practices by automatically deriving actionable artifact specifications directly from companion manuscripts. Leveraging both foundation and fine-tuned large language models, OpenART extracts explicit and implicit execution-relevant information from papers to generate structured artifact documentation, including required components, dependencies, and execution steps. From this documentation, OpenART further derives an expected artifact layout that enables (i) guided preparation of replication packages from scratch and (ii) automated identification of missing or inconsistent components in existing packages. We evaluate OpenART through a mixed-method user study with SE researchers across two realistic workflows: artifact creation and artifact completion. We expect that OpenART will substantially reduce preparation effort, improve perceived completeness, and enhance confidence in the executability of artifacts. OpenART will be the first framework to demonstrate the feasibility of manuscript-driven, automated artifact support and to offer a concrete step toward scalable, reviewer-friendly artifact evaluation in SE research.