← all datasets

Skill-Inject

Emerging
2papers using it
2026first seen

The 'Skill-Inject' dataset/benchmark contains a collection of agent skills used to evaluate the effectiveness of skill-poisoning attacks on large language model agents, specifically measuring the Attack Success Rate of these injections while maintaining the appearance of legitimate task execution.

Papers using Skill-Inject (2)

Skill-Inject β€” datasets β€” cybersecurity