Skip to main content
Search

Zongrui Yang

1 related articles

Subscribe to updates
  1. arXiv

    MCRI turns agent-skill evaluation into a pre-execution four-dimensional score, lifting top-1 skill selection by 17.7 to 22.8 percentile points across three benchmarks

    The authors propose the four-dimensional MCRI Framework grounded in information gain and behavioral constraint and operationalize it as MCRI-Eval, a large language model-based evaluation method, validated on 63,812 public skills from the OpenClaw skill Hub and 58,275 skill-conditioned model executions across BigCodeBench, BFCL-Fundamental, and Mind2Web: MCRI-Eval scores are positively associated with community popularity signals, achieve the highest downstream ranking agreement among the evaluated methods, and improve top-1 skill selection over each benchmark's strongest baseline by 17.7, 22.8, and 19.6 percentile points, indicating it can serve as a pre-execution signal for prioritizing promising skills.

You’re all caught up