03-ref-hallucination-arena
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill requires installing external Python packages
py-openjudgeandmatplotlibviapip. It also references an official dataset hosted on HuggingFace. - [COMMAND_EXECUTION]: Provides instructions to execute evaluation logic using the
python -msyntax and specific CLI flags like--configand--save. - [INDIRECT_PROMPT_INJECTION]:
- Ingestion points: Reads queries and constraints from JSON/JSONL files specified in the
dataset.pathconfiguration. - Boundary markers: The provided system prompt templates interpolate user-supplied data without explicit boundary markers or instructions to ignore embedded directives within the query data.
- Capability inventory: The skill performs extensive network operations (calls to various LLM APIs and academic databases) and writes results to the local file system (output directory).
- Sanitization: There is no mention of sanitization or validation of the input queries before they are processed by the agent and forwarded to target LLM endpoints.
Audit Metadata