exploratory-data-analysis
Pass
Audited by Gen Agent Trust Hub on Jun 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill is designed to ingest and summarize content from untrusted external scientific data files (such as CSV, JSON, and sequence files), which are then presented to the agent in a markdown report.
- Ingestion points: File reading operations within
scripts/eda_analyzer.pyand suggested custom analysis workflows inSKILL.md. - Boundary markers: The generated reports do not use explicit security delimiters or "ignore instructions" warnings when incorporating data summaries.
- Capability inventory: The skill environment typically includes file system access and the ability to execute Python scripts.
- Sanitization: While the tool primarily extracts metadata and statistics, it reflects content from user-provided files that could contain malicious instructions.
- [COMMAND_EXECUTION]: The skill includes a dedicated Python script
scripts/eda_analyzer.pymeant to be executed within the agent's environment to automate the analysis process. - [EXTERNAL_DOWNLOADS]: The instructions and error handling in the skill recommend the installation of numerous third-party scientific libraries (e.g.,
biopython,rdkit,nd2reader,pandas) from standard package registries to support various file formats.
Audit Metadata