SkillTrace: Multi-Trace Provenance Auditing for LLM-Agent Skill Reuse

The paper, posted on arXiv (2608.05204v1) in the cs.AI category, argues that existing clone detection methods are insufficient for LLM-agent skills because these skills mix multiple modalities—metadata, natural language instructions, code, tools, references, and operational workflows. SkillTrace proposes a multi-trace provenance auditing approach that examines distributed evidence across authored components, unlike single-modality or whole-package similarity detectors. This addresses the practical challenge of monitoring how skills are reused and propagated once they become marketplace artifacts in the rapidly expanding LLM-agent ecosystem. The work implies a shift toward provenance-aware governance for agent skill sharing, potentially impacting future marketplace design and compliance mechanisms.