SkillTrustBench: Benchmark for AI Agent Skill Security Analysis

SkillTrustBench is a benchmark dataset for evaluating the security analysis of AI agent skills. It was created by author cuhk-zhuque and last updated on June 15, 2026. Each case follows an agent-skill-style layout with a SKILL.md entrypoint defining usage.

Use Cases

• Benchmarking security analysis tools based on the structured skill definitions.
• Evaluating AI agent vulnerability detection based on natural-language skill instructions.
• Testing skill trustworthiness assessment based on provided tool-use guidance and reference assets.

Strengths

• Dataset is specifically designed as a benchmark for a defined task: evaluating security analysis of agent skills.
• Each case follows a structured layout with a defined entrypoint (SKILL.md), suggesting consistent formatting.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

cuhk-zhuque on Hugging Face

Freshness

Last updated 2026-06-15 06:06:16; freshness should be verified.