scenario-model-comparison
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data including prompt text, brief definitions, and reference asset IDs, which are then processed by tools like
asset_analyzeandmodel_run. This design establishes a surface for indirect prompt injection where instructions embedded within the user-provided data could attempt to manipulate the agent's evaluation results or behavior. Evidence of ingestion occurs in the 'Normalization' and 'Judging' sections through interpolation of brief and asset content into tool calls without explicit sanitization or boundary markers mentioned. - [EXTERNAL_DOWNLOADS]: The skill documentation includes instructions for the user to install additional official skills from the
scenario-labs/skillsrepository using thenpxcommand. This is a legitimate mechanism for functional extension using resources provided by the skill's author.
Audit Metadata