data-analysis
Pass
Audited by Gen Agent Trust Hub on Aug 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill processes untrusted external data (PDF, DOCX, CSV, MD, images) and uses it to generate and execute code (Python, SQL, JS). This creates an attack surface where malicious content within processed files could influence agent behavior.
- Ingestion points: SKILL.md specifies reading data via
\docx,\pdf, andpandas.read_csvfrom user-supplied files. - Boundary markers: Instructions focus on 'safe probing' for resource management but do not explicitly detail sanitization for embedded instructions within the data content.
- Capability inventory: Capability to execute Python (Pandas/Matplotlib), SQL, R, and JavaScript (Node.js) based on analysis findings.
- Sanitization: Mentions data quality checks (nulls, types) but lacks specific filtering or delimiting for adversarial text patterns embedded in input data.
- [COMMAND_EXECUTION]: The skill is designed to generate and execute code in multiple environments (Python, SQL, Node.js). While central to its purpose, this provides a broad execution capability that could be misused if the generated code is influenced by malicious input data.
Audit Metadata