tooluniverse-image-analysis

Pass

Audited by Gen Agent Trust Hub on Mar 9, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The test script test_image_analysis.py downloads dataset archives from huggingface.co. Hugging Face is an established and trusted repository for scientific datasets and models, and these downloads are used exclusively for verification and testing purposes.
  • [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface due to its core functionality of processing external data files.
  • Ingestion points: Data enters the system context from untrusted external CSV, TSV, and image files via pandas.read_csv, tifffile.imread, and PIL.Image.open as seen in SKILL.md, scripts/batch_process.py, and various reference guides.
  • Boundary markers: There are no explicit delimiters or instructions to the agent to disregard potentially malicious commands embedded within the data tables or image metadata.
  • Capability inventory: The skill includes capabilities for file system write operations (results.to_csv in scripts/batch_process.py), network communication (requests.get in test_image_analysis.py), and dynamic evaluation of statistical formulas (statsmodels.ols in references/statistical_analysis.md).
  • Sanitization: The skill lacks validation or sanitization of ingested content before it is passed to analytical functions or used in formula construction.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 9, 2026, 12:28 PM