nature-figure
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests external user data (CSV/TSV) and private template collections to generate and execute plotting scripts, which introduces a risk of processing embedded instructions.
- Ingestion points: User data files and private template directories are accessed as described in SKILL.md and references/r-template-index.md.
- Boundary markers: There are no explicit instructions for using delimiters or ignoring embedded instructions within the processed data.
- Capability inventory: The skill is capable of file system modification, shell command execution (find, rg), and full Python/R script execution.
- Sanitization: No explicit sanitization or filtering of external content is mandated before interpolation into code.
- [COMMAND_EXECUTION]: In references/r-template-index.md, the skill recommends using shell commands like find and rg to explore and search local directories for template adaptation. This grants the agent the ability to use command-line tools for file system inspection.
Audit Metadata