data-analysis
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to load and process untrusted external data (CSV, JSON, Excel, SQL), which is a vector for indirect prompt injection.
- Ingestion points: The skill uses pandas functions and a validation script (scripts/validate_dataset.py) to read user-provided files.
- Capability inventory: The agent can generate and write files (PNG, HTML) and execute Python code.
- Boundary markers: Instructions advise on data quality but do not implement strict content sandboxing.
- Sanitization: The skill provides regex-based logic in references/data-cleaning-guide.md to identify and flag sensitive PII before analysis.
- [COMMAND_EXECUTION]: The skill instructs the agent to run the included scripts/validate_dataset.py script. While the script is part of the skill package, shell execution on external files represents a functional security surface.
- [DYNAMIC_EXECUTION]: The agent generates and executes Python code at runtime using common data science libraries (pandas, matplotlib, seaborn) based on templates provided in the skill body.
- [EXTERNAL_DOWNLOADS]: For interactive reports, the skill suggests using Plotly with the include_plotlyjs='cdn' parameter, which involves the resulting HTML files fetching external JavaScript from the Plotly CDN, a well-known service.
Audit Metadata