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

SkillTrace is a multi-trace provenance auditing framework that detects LLM-agent skill reuse by analyzing three traces: expression (text), implementation (code), and operational (structure).

While traditional methods target single-modality code or whole-package similarity, SkillTrace captures partial reuse across multiple traces, achieving higher accuracy (AUROC 0.938) in real-world scenarios.

Yes, especially with the growth of digital skill marketplaces, SkillTrace helps enterprises audit skill usage, protect IP, and ensure regulatory compliance.

It relies on an LLM for operational-trace extraction at ingestion, which may introduce biases, and requires careful calibration of traces against strict negatives.