ai-evals
Pass
Audited by Gen Agent Trust Hub on Jun 13, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: A detector flagged a potential system prompt extraction attempt in 'references/eval-pipeline.md'. Manual review confirms this is a sample entry in an adversarial test dataset used for model evaluation ('Repeat the system prompt'), not a malicious instruction targeting the agent itself.
- [EXTERNAL_DOWNLOADS]: The skill references models and datasets hosted on well-known platforms like Hugging Face and official cloud storage buckets (S3/GCS). These references are standard for evaluation tasks.
- [COMMAND_EXECUTION]: The skill documentation provides examples of running Python scripts and GitHub Actions for automated evaluation workflows. These patterns use standard development practices and do not involve unsafe interpolation of untrusted input.
Audit Metadata