bio-spatial-transcriptomics-image-analysis
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS] (SAFE): The skill documentation suggests installing standard, reputable scientific Python packages including
squidpy,scanpy,scikit-image, andcellpose. These are trusted libraries in the research community. - [DATA_EXFILTRATION] (SAFE): No network operations (e.g.,
requests,curl,socket) or attempts to access sensitive system files (e.g.,.ssh,.aws) were detected. The skill interacts exclusively with local biological data files. - [COMMAND_EXECUTION] (SAFE): Analysis of the Python code and scripts confirms there is no use of
os.system,subprocess.run,eval, orexecto run arbitrary shell commands or code. - [PROMPT_INJECTION] (SAFE): The instructions and examples do not contain any patterns typical of prompt injection, such as instruction overrides, safety filter bypasses, or requests to reveal system prompts.
- [INDIRECT_PROMPT_INJECTION] (LOW): While the skill ingests external data files (H5AD and TIF), the capabilities of the skill are limited to image processing and numerical analysis, providing no direct mechanism for embedded data instructions to escalate privileges or exfiltrate information.
- Ingestion points:
sc.read_h5adinexamples/extract_features.pyandsq.im.ImageContainerinSKILL.md. - Boundary markers: Absent (common for binary data formats).
- Capability inventory: File reading/writing and standard matrix/image operations; no network or shell access.
- Sanitization: Relies on the underlying scientific libraries for data parsing.
Audit Metadata