SkillSafetyBench: Evaluating Agent Safety under Skill-Facing Attack Surfaces

arXiv:2605.12015 · cs.CR, cs.AI, cs.CL, cs.LG, cs.MA · Submitted 2026-05-12 · Read on arXiv

cs.CR, cs.AI, cs.CL, cs.LG, cs.MA

Submitted: 2026-05-12

Updated: 2026-08-28

Comments: EMNLP 2026 Main

Code: https://github.com/AI45Lab/skill-safety-bench

Project page: https://jinchang1223.github.io/skill-safety-bench-website

License: http://creativecommons.org/licenses/by/4.0/

The gist: Reusable skills are becoming a common interface for extending large language model agents, packaging procedural guidance with access to files, tools, memory, and execution environments.

Terminology

Abstract

Reusable skills are becoming a common interface for extending large language model agents, packaging procedural guidance with access to files, tools, memory, and execution environments. However, this modularity introduces attack surfaces that are largely missed by existing safety evaluations: even when the user request is benign, unsafe influence may reside in skill guidance, local artifacts, or execution-environment files that steer the agent toward unsafe actions. We present SkillSafetyBench, a runnable benchmark for evaluating such skill-facing safety failures. SkillSafetyBench includes 155 adversarial cases across 47 tasks, 6 risk domains, and 30 safety categories, each evaluated with a case-specific rule-based verifier. Experiments with multiple CLI agents and model backends show that non-user attacks can consistently induce unsafe behavior, with distinct failure patterns across domains, attack methods, and scaffold-model pairings. Our findings suggest that agent safety depends not only on model-level alignment, but also on how agents interpret skills, trust workflow context, and act through executable environments. The complete benchmark is available at https://github.com/AI45Lab/skill-safety-bench.

Sources

Related papers