xgboost-analysis
Pass
Audited by Gen Agent Trust Hub on Jul 6, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill uses robust parameter validation and file existence checks in
scripts/utils.Rto ensure it only operates on valid, user-provided data files.\n- [EXTERNAL_DOWNLOADS]: The skill facilitates the installation of necessary R libraries from the official R Project repository (cloud.r-project.org), which is a trusted and well-known source for software distribution.\n- [SAFE]: Data processing steps, such as one-hot encoding and the automatic exclusion of identifier columns, follow standard statistical methodology and do not involve unauthorized data access or network transmission.\n- [PROMPT_INJECTION]: The skill ingests untrusted data from user-provided CSV or TXT files, which constitutes an indirect prompt injection surface (Category 8). While feature names or values could theoretically contain malicious instructions intended to influence an agent reading the output, the risk is inherent to data analysis tools and is mitigated by the skill's technical constraints.\n - Ingestion points: Data is loaded via the
read_datafunction inscripts/utils.R.\n - Boundary markers: No specific delimiters or warnings to ignore embedded text are present in the instructions.\n
- Capability inventory: The skill performs model training, writes tabular data, and generates plot images.\n
- Sanitization: Feature names are sanitized internally using R's
make.namesfunction.
Audit Metadata