BuildBench Benchmarks LLM Agents on Diverse Real-World OSS Compilation

BuildBench introduces a more realistic evaluation framework for LLM agents by targeting diverse open-source software projects with undocumented dependencies and missing instructions. The benchmark addresses the limitations of existing methods that rely on highly rated, easily configurable software subsets.