meta-baujat-plot
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data via CSV files for statistical analysis. 1. Ingestion points:
SKILL.mdworkflow,scripts/baujat_plot.R, andscripts/baujat_plot_fallback.py. 2. Boundary markers: No explicit prompt delimiters for untrusted data are provided. 3. Capability inventory: The agent executesRscriptandpythonand the scripts write to the local file system. 4. Sanitization: Scripts perform numeric type casting which partially mitigates injection risks. - [INDIRECT_PROMPT_INJECTION]: The package includes
meta-baujat-plot_audit_result_v2.json, which contains a "SAFE" verdict and other safety assertions. This is a self-referential pattern targeting security analysis. - [EXTERNAL_DOWNLOADS]: The skill references standard R packages from official registries for heterogeneity analysis, which is categorized as safe.
Audit Metadata