meta-forest-binary-plot
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill is centered around executing local R and Python scripts (forest_binary.R, forest_binary.py, and extract_criteria.py) via shell commands. The instructions provide the agent with the specific command patterns to use for these operations.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes external data (CSV files) and user-supplied strings which are used in subsequent commands and output generation. Additionally, it contains a metadata file (meta-forest-binary-plot_audit_result_v2.json) with self-referential audit claims that could be used to influence the agent's perception of the skill's safety.\n
- Ingestion points: User-provided CSV files and the outcome_name parameter.\n
- Boundary markers: There are no explicit instructions to the agent to treat CSV content as untrusted data or to avoid executing embedded instructions within that data.\n
- Capability inventory: Shell execution (python/R), local file read/write.\n
- Sanitization: The scripts use standard data science libraries for parsing CSV files, but there is no semantic sanitization of the natural language content within those files.\n- [EXTERNAL_DOWNLOADS]: The documentation suggests the use of the pip package manager to install well-known libraries (numpy, pandas, matplotlib) if they are missing. The R script also depends on several standard packages from CRAN (ggplot2, meta, metafor, gridExtra, grid, stringr).
Audit Metadata