data-analysis
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external files such as CSV, RDS, and DTA, creating a surface for indirect prompt injection. 1. Ingestion points: Data is loaded and summarized in the 'Pre-Flight Report' (SKILL.md) using head(), names(), and summary() functions. 2. Boundary markers: The skill uses a structured report format to separate analysis planning from data loading, but does not specify delimiters for data content within the report. 3. Capability inventory: The skill utilizes 'Write', 'Edit', and 'Bash' tools (SKILL.md) to generate and execute R scripts. 4. Sanitization: There is no explicit sanitization or escaping of data values before they are used to inform the agent's interpretation and code generation.
- [DYNAMIC_EXECUTION]: The skill dynamically generates R analysis scripts at runtime based on user-provided datasets and specific analysis goals, writing the resulting code to the local filesystem.
- [COMMAND_EXECUTION]: The skill uses the 'Bash' tool to execute generated R scripts and employs the 'Monitor' tool to track background processes and stream output.
Audit Metadata