Claw-Eval-Live: A Live Agent Benchmark for Evolving Real-World Workflows

arXiv:2604.28139v1 Announce Type: cross Abstract: LLM agents are expected to complete end-to-end units of work across software tools, business services, and local workspaces. Yet many agent benchmarks freeze a curated task set at release time and grade mainly the final response, making it difficult...