exploratory-data-analysis
Pass
Audited by Gen Agent Trust Hub on Mar 3, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions direct the agent to run a local Python script (
scripts/eda_analyzer.py) to automate the analysis of user-provided scientific files.- [EXTERNAL_DOWNLOADS]: The skill documentation and script reference numerous third-party scientific Python libraries, such asbiopython,pandas,rdkit, andscipy. These are well-established, standard packages within the scientific community.- [PROMPT_INJECTION]: The skill processes untrusted external data from scientific files, creating a surface for indirect prompt injection where malicious instructions embedded in file headers or data cells could influence the agent's behavior during report generation. - Ingestion points: User-provided scientific files (e.g., FASTA, CSV, HDF5) processed by the skill scripts.
- Boundary markers: The script uses JSON serialization and markdown code blocks for some output, but data content is largely interpolated into the final reports without rigorous isolation.
- Capability inventory: The agent has the ability to read local files and execute Python scripts on the host system.
- Sanitization: There is no evidence of explicit content sanitization or filtering to prevent the execution of instructions hidden within the data files.- [REMOTE_CODE_EXECUTION]: The reference files (e.g.,
references/chemistry_molecular_formats.md) suggest using the Pythonpicklemodule to read.pklfiles. Deserializing data from untrusted sources usingpickleis a known security vulnerability that can lead to arbitrary code execution, although the documentation does include a brief note regarding security validation.
Audit Metadata