evaluation-metrics
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides an evaluation framework that ingests external data (such as contexts, questions, and model outputs) and interpolates it into prompts for automated quality scoring.
- Ingestion points: Data is processed via the
eval_datastructure inSKILL.mdand theEvaluationSampleclass inscripts/llm_evaluator.py. - Boundary markers: The scoring prompts (e.g., in
RAGMetricsandHallucinationDetector) do not use specific delimiters or instructions to isolate potentially adversarial data from the evaluation instructions. - Capability inventory: The skill performs model generation and statistical calculations; it does not perform sensitive file access, credential harvesting, or unauthorized network operations.
- Sanitization: Input data is used as-is from datasets to ensure evaluation fidelity, which is standard practice for this task.
Audit Metadata