SkillScreen: High-Recall Screening of Malicious Agent Skills with Decision Models

Summary

SkillScreen is a compact decision model that scores whether an agent skill is malicious directly from its SKILL.md file in a single forward pass. On MaliciousSkillBench it reaches 99% recall on unseen attack families while flagging 20-25% of benign skills, fewer than the word-level classifier and prompted LLM judges tested. It also generalizes better than a fully fine-tuned encoder to unseen sources, though unfamiliar benign sources sharply raise false positives.

Related items

• HighCVE-2026-86063: Claude Code is an agentic coding tool. Prior to 2.1.179, the Claude Code and Claude Desktop plugin installation flow…Similar attack · NVD/CVE Database
• MediumOver 1,500 npm packages exposed by Trusted Publishing validation flawSimilar attack · CSO Online
• MediumCVE-2026-108958: TensorLayer HDF5 model loading deserialization via model_config argumentSimilar attack · NVD/CVE Database
• InfoHow to Secure AI Models in Production Environments | Endor LabsSimilar attack · Endor Labs Blog
• InfoIntroducing Security for AI Coding Agents and Workstations | Endor LabsSimilar attack · Endor Labs Blog