REPRO-Bench: Can Agentic AI Systems Assess the Reproducibility of Social Science Research?