SKILLLITE: Evidence-Guided Malicious Skill Auditing with Compact LLMs
SKILLLITE audits third-party Agent Skills for hidden malicious behavior using compact local LLMs.
SKILLLITE is an evidence-guided framework for auditing third-party Agent Skills with compact, locally deployable LLMs. It extracts security-relevant behaviors, infers intended functionality, and then asks a small model to judge maliciousness in that context. Experiments show better detection than representative auditing baselines across compact backbones, including behaviorally confirmed in-the-wild malicious Skills, with low inference latency.
- Third-party Agent Skills create a supply-chain risk for LLM agents.
- Compact models miss implicit malice hidden in complex Skill packages.
- SKILLLITE scores skills from extracted behaviors and inferred functionality.
- Detection improves across compact backbones and confirmed in-the-wild malicious Skills.
From the original article
1 min read at arxiv.orgAs LLM-based agents perform increasingly complex tasks, Agent Skills have emerged as a flexible mechanism for extending their capabilities. An Agent Skill packages task-specific instructions with executable components and auxiliary resources to provide specialized functionalities. However, the growing adoption of third-party Skills introduces a new supply-chain attack surface.
By Haoran Ou · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.