Sign in to view source links and access this dataset
Description
SCR-Bench is a benchmark for evaluating security risks that emerge when individually benign skills are composed into agent workflows. The dataset, authored by kyle-X1e, was last updated on Hugging Face on June 11, 2026. It focuses on harmful outcomes arising from capability flow, trust transfer, or authorization confusion along composition paths.
Use Cases
Evaluating emergent security risks in AI agent workflows based on skill composition.
Testing for harmful outcomes from capability flow between individually safe skills.
Assessing risks from trust transfer across composed agent skills.
Identifying vulnerabilities from authorization confusion in multi-skill systems.
Strengths
Focuses on a specific, emerging risk area in AI safety: composition of benign skills.
Benchmark is designed to evaluate three concrete risk mechanisms: capability flow, trust transfer, and authorization confusion.
Limitations
Description metadata is limited; actual data quality requires manual inspection after download.
Column-level documentation is absent; field semantics must be inferred after download.
Row count and file formats are unknown, which may limit suitability assessment.
Provenance
Source
kyle-X1e on Hugging Face.
Freshness
Last updated 2026-06-11 18:59:43; freshness should be verified.
License is unknown; users should verify permissions before use.